Nghiên cứu giải pháp tối ưu hoá quá trình điều khiển và vận hành trạm sạc tích hợp điện mặt trời tại Việt Nam

Nghiên cứu giải pháp tối ưu hoá quá trình điều khiển và vận hành trạm sạc tích hợp điện mặt trời tại việt nam Nghiên cứu giải pháp tối ưu hoá quá trình điều khiển và vận hành trạm sạc tích hợp điện mặt trời tại việt nam Nghiên cứu giải pháp tối ưu hoá quá trình điều khiển và vận hành trạm sạc tích hợp điện mặt trời tại việt nam Nghiên cứu giải pháp tối ưu hoá quá trình điều khiển và vận hành trạm sạc tích hợp điện mặt trời tại việt nam Nghiên cứu giải pháp tối ưu hoá quá trình điều khiển và vận hành trạm sạc tích hợp điện mặt trời tại việt nam Nghiên cứu giải pháp tối ưu hoá quá trình điều khiển và vận hành trạm sạc tích hợp điện mặt trời tại việt nam Nghiên cứu giải pháp tối ưu hoá quá trình điều khiển và vận hành trạm sạc tích hợp điện mặt trời tại việt nam

Motivation for research

COP26 and PDP VIII – the commitments of Vietnam to sustainable

At the COP26 conference, Vietnam made strong commitments and responsible contributions to tackle global climate change Accordingly, Vietnam has committed to bring net emissions to zero by the middle of the century and joined the Global Coal to Clean Power Transition Statement In line with these commitments, the government has outlined a comprehensive roadmap with eight key tasks aiming at achieving sustainable and low-emission economic development [143]

These tasks involve promoting the transition from fossil fuel to green/clean renewable energy sources (RESs), reducing greenhouse gas (GHG) emissions in energy, transportation, and other sectors Notably, the reduction in the use of fossil fuel vehicles, the encouragement of electric vehicle (EV) research, EV development and adoption are also promoted The Ministry of Transport needs to study the feasibility of phasing out fossil fuel vehicles by 2040 and develop a roadmap for the transition to clean energy transportation

Worth mentioning, on May 15 th , 2023, the Vietnamese government adopted the Power Development Plan VIII (PDP VIII), showing a strong commitment towards decarbonization

The PDP VIII sets a new RES development direction by increasing the amount of renewable power generation capacity (i.e., up to 48 percent of the total capacity by

2030, and 65.8-71 percent by 2050) while significantly reducing coal power share in the electricity distribution plan (i.e., from 20 percent of the total capacity to 0 percent by 2050) The PDP VIII no longer prioritizes grid-connected solar power projects It strongly promotes the development of solar energy for self-consumption (i.e., solar power on rooftops of residential houses and buildings for on-site consumption, without injection into the electricity grid) Specifically, it sets a target of 50 percent of office buildings and residential houses using rooftop solar power for self- consumption by 2030

The commitments at the COP26 and the PDP VIII demonstrate the Vietnamese government's determination toward sustainable development across various sectors, especially energy and transportation Follow the orientation, this research is proposed to study solutions to promote the development of both RESs and clean transportation in the context of Vietnam.

The transition to electric two-wheeler mobility in Vietnam’s urbans

Currently, in Hanoi, Ho Chi Minh City (HCMC), public transport has not been able to satisfy the travel demand, constituting about 15 % and 9 % of travel need in Hanoi and HCMC, respectively Urban traffic heavily depends on private vehicles, in which gasoline-powered motorcycles take a dominant role of about 80 % [25] The overwhelming majority of private fossil fuel vehicles (Figure 1) leads to traffic congestion (Figure 2), increased GHG emissions and air pollution [134]

Figure 1 Private vehicle ownership in Vietnam and other countries

Figure 2 Traffic congestion in Hanoi Research shown that in developing countries as in Vietnam, poor traffic infrastructure [133], [146], inadequate public transport [43], [133], [151], low- and middle-income level and weather condition are factors which cause most people to choose private motorcycles as the preferred means of transportation in urban traffic

The high number of private vehicles puts a burden on urban traffic, causing traffic congestion, noise, and air pollution (Figure 3) To address these issues, solutions had been proposed such as applying stricter exhaust emission standards, limiting private vehicles, developing public transport, and encouraging fewer polluting vehicles Among the solutions, transport electrification can contribute to both air pollution reduction and energy diversification Benefits include zero tail-pipe emissions, higher efficiency than vehicles using internal combustion engines (ICE), high potential for

GHG emissions reduction when coupled with a low- carbon electricity sector, reducing fossil fuels dependence, lesser noise and being able to provide ancillary services to the energy system [99]

Regarding electric mobility transition in Vietnam, in the first periods, with low purchase price (about 400

USD [99]), low speed, no driver's license and vehicle registration requirements, electric bicycles had been widely used by students and elders However, these vehicles were not attractive because of their low quality, and they couldn’t perform as well as gasoline- powered counterparts [134]

Figure 3 The fifteen most polluted cities in Southeast

Electric two-wheelers (E2Ws) gradually attracted the public’s attention when major manufacturers such as Honda, Yamaha, Piaggio, Vinfast etc engaged in the market Modern designs, good quality, diverse features (e-Sim, anti-theft, cruise control, operation history record, waterproof) and reasonable purchase price were factors that had propelled E2Ws to the forefront of the public interest

In 2019, many high-quality E2W products were debuted in Vietnam A remarkable growth from 0.9 million E2Ws by 2017 to five million units by 2019 was recorded by the International Association of Public Transport (UITP) By 2019, there were eleven enterprises producing E2Ws with the output volume reaching 52,938 units

(Registry Department) Annually, the growth rate of the E2W market is up to 30 - 40

The transition from traditional motorcycles to E2Ws might be the result of current high rate of motorbike adoption, socio-economic conditions, and limited transport infrastructure [13], [49] E2Ws still retain remarkable features of motorcycles in urban traffic, while providing benefits of electric mobility and having potential of developing intelligent transportation systems, effectively improving connection to the public transport services at transit hubs

However, the continuing growth of these emerging vehicles has been projected to result in an accelerated burden on the distribution grid which is designed to meet the expected growth rate of traditional load demand Thus, in order to accommodate the transition, research on solutions addressing this issue should be considered.

Rooftop solar power development in Vietnam and its impacts

With a high average annual total radiation, Vietnam is considered a place with special potential for developing solar power (Figure 4)

Figure 4 Map of average daily global horizontal irradiance (GHI) in Vietnam

Figure 5 Top 10 countries by PV installed capacity in 2020

China United States of America

Japan Germany India Italy Australia Vietnam Republic of Korea United Kingdom

In 2015, only 4 MWp of installed solar capacity for power generation was available, of which about 900 kWp was connected to the grid [2] In 2018, the total solar power capacity was 106 MWp However, by 2019, the capacity increased dramatically to around 5 GWp, including nearly 0.4 GWp from rooftop photovoltaic (PV) systems By the end of 2020, the country's total PV capacity reached approximately 16,500 MW [14], placing it among the top 10 nations globally (Figure

5) With regards to rooftop solar, there were 105,212 systems installed throughout the country by December 2020, with a total capacity of 9,730.87 MWp [142]

However, it has been generally believed that once PV penetration exceeds a certain limit, problems and challenges could arise, affecting the operation and security of the grid High PV penetration level affects the grid stability, triggering frequency and voltage anomalies, overloading the existing infrastructure and mismatching demand and supply [6], [74], [85], [86], [98], [100] According to [14], about 365 GWh of solar output was curtailed in 2020 as a result of balancing supply and demand

In order to increase the hosting capacity of the grid, the utility could introduce mitigation techniques or adopt grid optimization solutions to coordinate PV operation with the rest of the grid Besides, the power grid needs to be upgraded and flexibly managed to better accommodate the RESs

Another method to promote PV usage without affecting the distribution grid operation is to encourage self-consumption, which was clearly mentioned in the PDP VIII With the target of 50 percent of buildings and residential houses utilizing rooftop solar power for self-consumption by 2030, it would definitely be a challenge that needs comprehensive solutions to reach.

PV-integrated charging stations – A solution for both E2W and rooftop solar

Generally, for the urban distribution grid in Vietnam, there are two factors that might have significant impacts in the immediate future Firstly, it is the remarkable emergence of charging loads which are deferrable and not planned for the current grid infrastructure Secondly, it is the continuing popularity of distributed RESs which possess stochastic and intermittent nature and are encouraged to self-consume The emergence of these two factors needs solutions to deal with and also leads to opportunities as well as challenges for the energy system

Studies show that any form of EV such as

HEV (Hybrid EV), PHEV (Plug-in hybrid EV),

PEV (Plug-in EV) has lower emissions than ICE counterparts Noteworthily, the amount of emissions depends on the proportion of clean energy supplied to the vehicle [42], [45], [50]

Figure 6 A PV-integrated E2W charging station

If EVs are charged from the grid and if the grid electricity is primarily generated by fossil fuels such as coal or natural gas, then the emissions are significant Thus, the adoption of EV only contributes significantly to the reduction of GHG emissions if EVs are charged from RESs or from the grid with high share of renewable electricity

Wind power, solar power, hydropower, biogas, or tidal energy can be considered as sustainable energy sources to power EVs Regarding urban areas in Vietnam, rooftop PV could be an attractive option because of the following factors:

1) Vietnam has a large potential of PV power with solar radiation reaching up to 5kWh/m 2 and annual sun hour being from 1600-2600 hours [55] Rooftop PV is encouraged as it does not consume many land resources [119], [126]

2) The costs of PV module, hardware and inverter are continuously decreasing

By Q1/2019, the cost of PV module was less than $0.3/Wp [89]

3) Since PV modules can be installed on the roofs close to the charging stations or placed on/used as the parking lot’s roof, PV power can be accessed easily

4) The combination of EV charging, and solar power can reduce the amount of

PV power injected into the distribution grid and, at the same time, satisfy charging needs Thus, it can help mitigate unwanted impacts of charging load and high penetration level of RESs on the grid

5) EV batteries can be used as energy storage devices, which could contribute to optimal exploitation of solar energy [9], [19]

6) Charging cost from PV power is cheaper than from the grid Electricity is locally generated and consumed, which can be seen as a solution for the decrease of feed-in tariffs (FIT) [55]

7) PV systems have low operation and maintenance costs

8) PV-integrated charging stations can help reduce GHG emissions and contribute to sustainable development

Figure 7 Charging station block diagram

In Figure 7, the block diagram of a PV-integrated charging station is illustrated Electricity, which is from solar power or from the grid, is supplied to the vehicles’ batteries If the amount of PV power is greater than the charging demand, the surplus electricity will be injected into the grid In case of the charging demand being greater than the PV output, the grid would compensate for the insufficient power by providing additional supply If AC/DC converters are bidirectional, EV batteries can be used as energy storage devices which can provide ancillary services.

Research goals, scope, and research questions

This work conducts research on optimal charging solutions for PV-integrated E2W charging stations The objectives, scope, and questions of the research are described belows

1) This work aims at coordinating hundreds of E2Ws charging considering both V2G and non-V2G

2) The proposed solutions in this work aim at load leveling, valley filling and peak shaving Thus, it can support the grid and can help reduce the impacts of E2W charging on the distribution grid and on other loads

3) Uncertainties such as charging behavior should be considered

4) The proposed solutions in this work should consider the high number of vehicles in the E2W charging stations

1) Since E2Ws tend to adopt Li-ion batteries, this work focuses on charging/discharging Li-ion batteries Battery capacity and charging power should be consistent with popular E2Ws in Vietnam Futhermore, batteries are considered ideal energy storage systems Thus, the transient performance of batteries during charging/discharging are not considered

2) The desired SOC and expected departure time of vehicles are assumed to be available to the scheduler This information can be inputted through a user interface which is out of scope of this thesis

3) Because E2Ws are typically equipped with portable single-phase chargers which appear as residential single-phase powering and unidirectional type, charging at the residential voltage level of 220 V should be considered

4) Although the battery capacity and charging power of E2Ws are small (capacity: 0.2–5 kWh; power: 0.25-4 kW [53]) compared to electric cars (average capacity: 50 kWh [18]; power: 7.2-19.2 kW for AC level 2 charging and 50-400 kW for DC charging [51]), hundreds of E2Ws charging can result in a significant aggregated power which impacts on other loads, on the distribution grid, and on the system efficiency Thus, charging in such situations should be considered

5) The charging station is powered by a rooftop PV system and the grid However, this study does not address issues such as island mode operation or optimizing power dispatch Optimal PV sizing is also out of scope of this study

6) In this research, it is assumed that forecasted data of PV generation and conventional load are available and sufficiently accurate To be explicit, in this study, the PV power data is retrieved through the simulation of fixed solar panel systems with locations and weather conditions in Vietnam Furthermore, because the simulation does not consider the shading effect or the uncertainty of weather conditions, the solar irradiation on different PV panels is assumed to be uniform Conventional load data is retrieved from the dataset of Mendeley Data [75] The uncertainties in PV generation and load data forecast are not considered in this work

With the mentioned objectives and scope, the following questions may need further clarification:

1) The integration of charging loads, RESs, conventional loads, and the grid introduces microgrids Thus, it is necessary to develop smart and flexible solutions for these systems

2) The charging algorithm needs to meet the charging requirements while reducing impacts on the distribution grid and conventional loads

3) The algorithm should also be tailored to facilitate ancillary services or grid support

4) The proposed solutions should take into account that the number of E2Ws in the charging station is often greater than the number of EVs in a typical electric car charging station

5) Uncertainties such as charging behaviors should be taken into account Furthermore, the proposed algorithms should be verified through diverse case studies and empirical test.

Research methodology

The dissertation is carried out based on the following research methods:

1) Synthesis Research: This method involves in gathering and synthesizing information in literature, and implementing data collection related to technical specifications

2) Modeling and Simulation: Mathematical models and simulations are leveraged to investigate various operation scenarios of PV-integrated E2W charging stations Simulation can provide insights in system operation and the performance of scheduling algorithms

3) Empirical Testing: This work plans to perform practical tests and gather empirical measurement This helps in research validation practically

4) Expert Consultation: Consultation from experts in the field can provide valuable insights for the research findings.

Research contributions and outline of the thesis

In this work, a research on the scheduling algorithms for E2W charging stations is conducted The main contributions of this work can be expressed in the following points:

1) This work provides solutions to meet new load demand and encourages the development of rooftop PV power in urban areas while mitigating the adverse impacts of EVs and solar power on the distribution grid Thus, reducing the need to upgrade/reinforce the current grid

2) In this work, two scheduling algorithms are proposed The effectiveness of the algorithms in terms of improving load profile, filling the valleys, and shedding peak loads are verified

3) An empirical test bench has been successfully set up for evaluating the real- time responses of E2W charging station following long-term charging plan from the scheduler

The dissertation consists of five chapters The introduction section mentions the motivation, objectives, scope and contributions of the research Chapter 1 reviews the architectures and control algorithms for EV charging stations In Chapter 2, a mathematical model and a simulation model for the station are developed Chapter 3 proposes a power allocation algorithm aiming at load leveling An optimal algorithm based on receding horizon framework is proposed and verified in Chapter 4 Finally, a test bench is established in Chapter 5 to verify real-time responses of E2W charging to scheduled commands

This section is mainly based on:

[1] Huu D.N and Ngoc V.N (2021) Analysis Study of Current Transportation Status in Vietnam’s Urban Traffic and the Transition to Electric Two-Wheelers Mobility Sustainability, 13(10), 5577

[2] Ngoc Van Nguyen and Huu Duc Nguyen (2022) PV-Integrated Electric Two- wheeler Charging Stations: A Solution towards Green Cities TNU Journal of

OVERVIEW OF ELECTRIC VEHICLE CHARGING

Charging station architectures

Generally, charging a group of EVs can be implemented by centralized, decentralized, or hierarchical architectures (Figure 1.1)

In the charging station, an EV aggregator is the entity responsible for controlling the charging/discharging process directly or indirectly [98] A direct aggregator will decide the charging strategy for each EV while an indirect aggregator will broadcast information-carrying signals to the EVs to coordinate charging

Indirect Aggregator Indirect control signal

Local controller a) Centralized b) Decentralized – Type 1 c) Decentralized – Type 2

Indirect control signal Local controller d) Hierarchical e) Hierarchical

In centralized architecture, each EV charging schedule is determined by a direct aggregator who collects charging requirements and specifications of EVs, solves an optimization problem (OP) to determine the charging profile then communicates to the vehicle owners

This architecture requires vehicle owners to give up some autonomy over vehicle charging When comprehensive system information is available, it often produces optimal solutions It allows for easy consideration of constraints on the EVs, the grid, and others However, drawbacks include concerns about privacy (e.g., charging starting/ending time, charging behavior, travel distance, etc.) Additionally, system may collapse if a failure occurs at the aggregator

In a centralized architecture, scalability is a significant challenge As the number of EVs increases or the scheduling time is divided into smaller timeslots, the OP complexity rises This leads to high demand of computational capability

In decentralized architecture, EVs determine their charging schedule using local controllers While decentralized solutions may not always produce optimal strategies, this approach offers scalability and practical implementation benefits By contrast to the centralized one, the decentralized model is more resilient against system collapse when network failures, especially if the controllers are programmed to operate during such case Based on communication network structure, the decentralized architecture can be categorized into two types

- Decentralized architecture - Type 1 (Figure 1.1b): EVs calculate their own charging schedule and communicate with other EVs until a global equilibrium is reached This approach requires EVs to continuously exchange their charging schedule with other vehicles, resulting in a substantial communication burden in scenarios with a high number of EVs

- Decentralized architecture - Type 2 (Figure 1.1c): An indirect aggregator is introduced to collect specific information and broadcast it to the EVs This approach reduces the communication burden and, consequently, significantly diminishes the communication infrastructure needs compared to the Type 1 architecture

Generally, in a decentralized architecture, the local controllers are responsible for scheduling Charging profiles are determined based on information received from the indirect aggregator or from other EVs If the local controllers receive the same information and pursue a common objective, they tend to output a common charging schedule to the EVs For instance, if the controllers optimize charging cost based on time-of-use (TOU) electricity prices, all EVs might charge during low-price periods and stop charging during high-tariff times Thus, it could affect the global optimization goal

The hierarchical approach combines the benefits of both centralized and decentralized architectures Using a tree-like structure, it assigns control and computation tasks to direct/indirect aggregators Each aggregator manages a group of EVs and impacts the decision of other aggregators These groups could be EVs in the same area, like a parking lot, an apartment complex, or a transit hub (Figure 1.1d-g)

A direct central aggregator is responsible for calculating the charging plan for all sub-aggregators to achieve the overall optimal goal Based on the determined charging plan, the sub-aggregators will calculate the charging profile for each EV it manages On the other hand, an indirect central aggregator broadcasts information to sub-aggregators who then determine the charging schedule for each EV (Figure 1.1d)

In Figure 1.1e, a central aggregator calculates the charging plan for all sub- aggregators who broadcast information, transfer calculation task to the local controllers at EVs

In Figure 1.1f-g, the central aggregator is eliminated, and a communication network links sub-aggregators If there's a disconnection between two aggregators, another connection will assure system resilience However, if an aggregator fails, the EVs it manages will lose control

1.1.4 Proposal of E2W charging station architecture

In the decentralized architecture, the charging/discharging coordination of EVs is often achieved through signal broadcasting The vehicles then actively plan their charging To be specific, an EV aggregator can offer and broadcast a price signal to the vehicles which indirectly change their charging behavior The change in charging behavior of each vehicle can help the aggregator achieve system-level objectives This architecture needs to continuously update and broadcast new signals because EVs can choose to charge during low-price periods simultaneously, causing new peak loads and affecting the aggregator's objectives

The decentralized architecture for electric car charging stations (high power consumption) may make sense because prices would have a significant impact on charging behavior However, this approach may not be effective for E2W charging as the price signal may not be attractive enough to affect the charging behavior since the energy and power consumption of E2Ws are usually much lower than electric cars Therefore, when considering hundreds of E2W charging in a parking lot, it is necessary to have a centralized controller to manage E2W charging The vehicle owner should only be responsible to a certain extent, such as providing charging requirements, departure time, etc., to the controller or choosing available options recommended by the aggregator

For centralized architecture, studies show that scalability is a major drawback As the number of vehicles increases, the demand for computing capability increases and solving optimization problems becomes very complex and time-consuming [69],

[117] Because of this, the centralized approach may be impractical when implementing a large-scale and real-time application [4] Therefore, algorithms based on centralized architecture should pay attention to computational complexity and computational time

EV charging station control algorithms

Currently, research on algorithms for E2W charging stations is limited Researchers mainly focus on algorithms for electric car charging However, reviewing the control algorithms for electric car charging stations can be referred to and applied to E2W charging stations to some extent

Most studies consider EV charging problem as a constrained OP, with charging rates and durations being seen as decision variables The OP includes various constraints imposed by the grid operator, aggregator, vehicle, and vehicle owner

Realizing an optimal algorithm is challenging, especially for non-convex OPs However, certain OPs were solved using diverse techniques as in [122], [128], [136],

Charging station control problems can be classified in terms of technical and economic aspects (Table 1.1) Each aspect can be considered from the point of view of different entities such as: the grid operator; EV aggregator; vehicle owner However, the objectives achieved from a technical standpoint can also help improve the economic indicators to some extent and vice versa

Table 1.1 Classification of charging station problems

- Maximizing convenience for vehicle owners

1 Minimizing cost of electricity generation

3 Maximizing revenue of EV aggregator

1.2.1 Algorithms focus on technical aspects

Control algorithms focusing on technical goals have been widely discussed in research The technical objectives are follows:

In load regulation algorithms, numerous studies focused on flattening the aggregated load profile (including both conventional load and charging load) By reducing peak load, the overloading problem of transformers, transmission lines, and other electrical infrastructure can be reduced On the other hand, flattening the load profile helps to reduce the need for sudden ramp up and down generators, maintaining generators to operate at their highest efficiency and stability

The control of EV charging to fill low load periods during nighttime was widely addressed in literature Although the impact of a single EV charging/discharging on the distribution grid is relatively minor, the aggregated impact of a high number of EVs can be significant

When considering load regulation, the constraints include energy requirement, charging/discharging power, overloading constraints, and voltage constraints at the point of common coupling (PCC)

The most intuitive approach for load leveling (or valley filling) is to minimize the total load variance with the objective function as in (1.1) [108] min

Where: 𝐷 𝑡 𝑆 is non-EV load

𝑃 𝑡 𝑖 denotes the charging power of 𝐸𝑉 𝑖 at timeslot 𝑡

𝑀 is the total number of timeslots of Δ 𝑇 during the scheduling time Subject to, ∑ 𝑀 𝑡=1 𝜂 𝑖 𝑃 𝑡 𝑖 Δ 𝑇 = (𝐹𝑆𝑂𝐶 𝐸𝑉 𝑖 − 𝐼𝑆𝑂𝐶 𝐸𝑉 𝑖 )𝐴 𝑖 (1.2)

𝑃 𝑚𝑖𝑛 𝑖 ≤ 𝑃 𝑡 𝑖 ≤ 𝑃 𝑚𝑎𝑥 𝑖 (1.3) Where 𝐼𝑆𝑂𝐶 𝐸𝑉 𝑖 , 𝐹𝑆𝑂𝐶 𝐸𝑉 𝑖 , 𝐴 𝑖 , 𝑃 𝑚𝑖𝑛 𝑖 , 𝑃 𝑚𝑎𝑥 𝑖 , 𝜂 𝑖 are the State of Charge (SOC) at arrival time, the SOC at departure time, battery capacity, minimum charging power, maximum charging power, and charging efficiency of 𝐸𝑉 𝑖 , respectively

The load regulation approaches are diverse Study [108] managed to minimize the variation of the total load to flatten the load curve and fill the valleys, while [93] affected the charging behavior of vehicle owners through electricity pricing Accordingly, the aggregator could broadcast control signals, such as electricity price signals that varied based on total load demand Each vehicle would then try to minimize charging costs by scheduling to charge at low-price electricity times, thereby filling low-load periods At each iteration, each EV would update its charging profile Similarly, the study [111] focused on EV charging at discrete charging rates Game theory is also a promising tool used to coordinate charging of EVs by optimizing the charging behavior of individual vehicles In [117], a non-coordinative game was established to coordinate a high number of EVs that were weakly linked through a common electricity price Notably, the low-load valley was filled at the

Nash equilibrium (NE), where no EV benefited if it unilaterally deviated from the selected charging strategy [93]

Based on the decentralized control architecture, [108] presented an online algorithm to regulate EV load by broadcasting a reference signal based on total real- time load After receiving the signal, each EV decided whether to charge or not by comparing its SOC with the reference This solution was online, did not rely on forecasting, and as a result, it was not affected by forecasting errors

Using dynamic programming and game theory, the valley filling and peak shaving algorithm was developed in [123] The charging schedule of each EV was found using a forward induction dynamic programming algorithm

In [81], a non-iterative approach was proposed for sequential scheduling of each

EV at a time The algorithm aimed to minimize both the variance and the peak of total load This algorithm only determined the charging profile of an EV once it was connected However, the drawback was the waiting time in case multiple EVs were connected simultaneously

In [104], the authors used a sequential scheduling approach to design a decentralized scheme that aimed to minimize the mean squared error between real- time aggregated load and a reference point estimated offline based on data from both non-EV and EV loads

Several load regulation algorithms managed to consider dual objectives by integrating constraints such as transformer overload constraints [95]; constraints on the power supply capability of the grid [37], [111]; constraints on voltage at PCC

[108], [109] or both overload and voltage constraints [131] Study [102] integrated the transformer load level into the price signal Another method was to use an ant- based swarm algorithm to provide signals to EVs whenever the total load exceeded the transformer's capacity [66] Using the bisection method, study [120] determined the load level that met the energy demand of EVs, in which transformer overload constraints were handled by reducing the energy demand of EVs by a certain proportion

Another important aspect is to improve operational efficiency The objective function could be to maximize the efficiency of using RESs at the charging station, reduce electricity consumption from the grid, or coordinate charging/discharging to follow the day-ahead energy plan [139]

Research [131] presented a decentralized algorithm and proposed a token-based IT infrastructure This infrastructure provided energy services through generation and consumption tokens to maximize the average utilization of generation while ensuring that the actual power consumption of EVs was less than the total power allocated for charging

Leveraging game theory, real-time electricity generation plan balance was investigated in [139] In the first stage, EV owners engaged in a non-cooperative game to determine their day-ahead anticipated electricity demand which minimized their cost Based on this, the aggregator decided the power generation or electricity purchasing plan for the next day In the second stage, the vehicle owners participated in a real-time game to adjust their charging patterns to align with the previously anticipated demand

Summary

Chapter I conducts research on the architectures and control algorithms for EV charging stations, then proposes appropriate architecture for E2W charging stations in the context of Vietnam

It can be seen that each architecture has its own advantages and drawbacks While a centralized station can easily achieve global optimization, it has drawbacks in terms of scalability, vulnerability to system failure, and concerns of privacy By contrast, a decentralized architecture has the advantage of being scalable, reducing the risk of losing control when system failure occurs However, it is difficult to achieve global optimization, and signal broadcasting also requires constant updating until a global equilibrium is reached

Regarding E2W charging, due to the small charging power and modest energy requirements of E2Ws compared to electric cars, a decentralized architecture based on price signal broadcasting may not be effective in changing the charging behavior Therefore, it is much more practical if E2W charging stations use a central controller to coordinate vehicle charging However, to reduce the computational burden in the case of hundreds/thousands of vehicles charging, the charging station should be divided into multiple sub-stations managed by sub-controllers

Regarding algorithms for E2W charging stations, current research is still quite limited However, algorithms that have been developed for electric car charging stations can also serve as a reference for E2W charging stations to some extent Besides, algorithms for E2W charging stations should also consider characteristics of E2W charging such as high number of vehicles, modest battery capacity and small charging power compared to electric cars

This chapter is mainly based on:

[1] Ngoc Van Nguyen, Khac Nhan Dam, and Huu Duc Nguyen (2022) Research on the architectures and control algorithms for electric vehicle charging stations and electric two-wheeler charging stations in the context of Vietnam Journal of Science and Technology - HaUI, 58(4), 55–64

CHAPTER II: MODELING OF PV-INTEGRATED ELECTRIC-

Chapter objectives

In this chapter, a PV-integrated E2W charging station is modeled to simulate and assess the operation of the stations as well as components such as E2W batteries, PV system, power converters The models can also be utilized to deploy control algorithms Thus, this chapter focuses on:

- Modeling power electronic converters such as grid-tie solar inverter, boost/buck converters, battery chargers/dischargers

The following features should be included:

- Maximum Power Point Tracking (MPPT)

Both real-time and long-term models are considered in this chapter, allowing to investigate real-time responses and integrate long-term scheduling algorithms.

Charging station block diagram

Figure 2.1 depicts an EV charging station connected to the grid, a PV system, and conventional loads

DC AC SOLAR INVERTER WITH MPPT

SOC/SOH ev(N) SOC/SOH ev01

Figure 2.1 Charging station block diagram

The station is powered by a PV system and the distribution grid Electricity which is generated from PV panels is converted to AC power through a solar inverter then feeds to the AC bus The bus links other components such as conventional loads and E2Ws It should be noted that, if V2G feature is permitted, E2Ws can absorb energy and can also discharge energy to the AC bus Thus, the chargers should be bi- directional type.

Realtime model

This model is suitable for investigating the charging/discharging response of E2Ws in real-time The model consists of components as follows:

2.3.1 PV module and PV array

To model PV modules, either a one- diode model or a two-diode model can be exploited [72] Figure 2.2 illustrates the equivalent circuit diagram for the one-diode model

Figure 2.2 The one-diode model and

Thevenin equivalent circuit One-diode model is established based on the following equations [72]:

𝑅 𝑠ℎ (2.4) Output current of the module:

𝑘 𝑖 represents short-circuit current of the cell at 25°C and 1000 W/m 2

𝑇 𝑛 is the nominal temperature which is 298 (K)

𝑞 is the charge of one electron which is approximately 1.6×10 -19 coulombs (C)

𝑛 is the ideality factor of the diode

𝐾 is the Boltzmann constant which is 1.38×10 -23 (J/K)

𝐸 𝑔0 is the bandgap energy of the semiconductor material (eV)

𝑁 𝑠 is the number of cells connected in series

𝑁 𝑃 is the number of PV modules connected in parallel

𝑉 𝑡 is the diode thermal voltage (V)

For example, a PV system can be modeled based on PV panel specifications as in Table 2.1

Voltage at maximum power point 𝑉 𝑚𝑝 (V) 35.6

Current at maximum power point 𝐼 𝑚𝑝 (A) 4.65

Based on the one-diode model, PV module can be built on MATLAB/Simulink as shown in Figure 2.3

For example, in the case of a 150 kWp PV system, it is possible to use 900 panels connected in strings and arrays The panels can be arranged in parallel with 36 strings, with each string consisting of 25 solar panels connected in series

To maximize the potential of solar energy, Maximum Power Point (MPP) tracking algorithms are adopted By continuously tracking the MPP, solar panels will operate at their optimal power output regardless of variations in irradiance and temperature

Battery modeling plays a crucial role in EV studies They are used by EV designers to optimize the storage system and predict the battery system behavior Battery models serve in the Battery Management System (BMS) to estimate real-time SOC and battery performance As most EVs utilize high-energy-density Li-ion batteries, accurate models of this battery type need to be investigated

Battery modeling can be classified into electrochemical models, artificial neural network models, and equivalent circuit models [50] Electrochemical models describe chemical reactions within batteries, employing multiple parameters to simulate the polarization However, these models can be difficult to establish due to the impact of environmental conditions on electrochemical processes Furthermore, its application in real working conditions is limited [50]

Artificial neural network models utilize the non-linear and self-learning properties of neural networks combined with experimental data to establish relationships between different parameters of the battery The drawback is that neural networks require a large amount of experimental data to predict battery operations

On the other hand, equivalent circuit models use circuit elements such as resistors, capacitors, voltage sources, to simulate the dynamics of batteries Commonly used equivalent circuit models include Rint, RC, PNGV (Partnership for a New Generation of Vehicles), and Thevenin models The RC model describes battery polarization using capacitance without reflecting resistance The PNGV model aims at simulating complex relationships within the battery during charge/discharge, but its complexity makes simulation challenging The Thevenin model reflects both capacitance and resistance of the battery It is relatively simple and easy to implement [50]

The battery models mentioned in [84] described various mathematical representations of batteries Equivalent circuit models contain relatively few parameters and are easy to obtain state-space equations [65] Thus, they are widely used in simulation and real-time control systems Numerous experiments have shown that the equivalent circuit model is suitable for LiFePo and LiMnCo batteries [26] Figure 2.4 shows the equivalent circuit model of a battery

Figure 2.4 The equivalent circuit model of a battery

The open-circuit voltage VOCV and internal resistance of the battery 𝑅 𝑏𝑎𝑡 𝑖𝑛𝑡 vary with SOC and are obtained from the manufacturer's specifications [56] The output voltage of the battery is determined by [104], [113]:

𝑉 𝑂𝐶𝑉 , 𝑑𝑖𝑠𝑐ℎ𝑎𝑟𝑔𝑖𝑛𝑔 (2.10) Based on equations 2.6-2.10, the open-circuit voltage and internal resistance of the battery (consisting of multiple cells) can be determined

In Figure 2.5, the open-circuit voltage, discharge resistance

(Rdis) and charge resistance (Rchg) are obtained from the manufacturer's data for the Saft

Lithium Ion Battery 7 Ah, based on the working temperature and

SOC The specific data for the battery can be referred to as in

2.3.3 DC-DC boost converter and maximum power point tracking (MPPT) algorithm

The boost converter is a power electronic converter used to step up the voltage from a lower level to a higher level It consists of an inductor, a switch, a diode, and a capacitor The model of boost converter is shown in Figure 2.6

Figure 2.6 DC-DC boost converter block

The RefGen function is responsible for tracking the maximum power point based on the voltage and current values from PV array

2.3.3.2 Maximum power point tracking algorithms

Maximum Power Point Tracking (MPPT) is a control algorithm that utilizes the variations in voltage and current of PV array to search for the maximum power point of the PV system This section investigates two common MPPT algorithms: the P&O and Incremental Conductance (INC)

▪ The Perturb and Observe (P&O) algorithm:

The basic principle of this algorithm is to perturb the operating point of PV panels and observe the change in power

Based on this observation, the algorithm adjusts the operating point to increase the power output

The P&O algorithm is simple and widely used, although it may have some drawbacks, such as oscillations around the MPP and slower tracking speed in rapid environmental condition change Figure 2.7 Flowchart of P&O algorithm

The flowchart of the P&O algorithm is shown in Figure 2.7 [132]

The Incremental Conductance (INC) algorithm operates based on the following principles [59] (Figure 2.8)

- The slope of the power curve is zero at the

- The slope is positive on the left side of the

- The slope is negative on the right side of the

, at the MPP , at the left of MPP , at the right of MPP

Figure 2.9 Flowchart of INC algorithm

, at the MPP , at the left of MPP , at the right of MPP

Figure 2.9 illustrates the flowchart of the INC algorithm The MPP can be tracked by comparing the instantaneous conductance (𝐼/𝑉) to the incremental conductance

The inverter control blocks are established in Simulink as follows [83]:

Voltage transformation and current transformation blocks to alpha-beta coordinate system and d-q coordinate system are shown in Figure 2.10

Figure 2.10 Transformation to the 𝑑𝑞 coordinate system

Position of the rotating frame (𝑤𝑡) coordinate system (Figure 2.11)

Control signal block of the PWM generation block (Figure 2.12)

Figure 2.12 Control signal block for generating PWM signals

The PWM signal generation block and power circuit of the inverter are as in Figure 2.13

Figure 2.13 PWM signal generation and the inverter power circuit

To implement battery charging/discharging, a buck/boost converter can be used The PWM control signal for the buck/boost circuit is generated from a charger controller as in Figure 2.14

Long-term model

This model is suitable for scheduling the charging of EVs over multiple periods Accordingly, a working day or a shift (also known as scheduling horizon) can be discretized into multiple time intervals The model acquires necessary information and carries out scheduling to achieve certain optimal objectives over the entire scheduling horizon

Assuming that the charging station serves 𝑁 electric bicycles/electric motorcycles, where 𝐸𝑉 𝑖 represents the 𝑖 𝑡ℎ EV at the charging station Let 𝐶 𝑡 𝑆 be the aggregated energy of batteries at time 𝑡, 𝐷 𝑡 𝑆 be the total power of non-EV loads subtracting the power generated by the PV system at time 𝑡 (also known as netload), and 𝐸 𝑡 𝑆 be the total load demand (including charging load and netload) Assuming that a working day is divided into timeslots of Δ 𝑇 𝑃 𝑡 𝑆 represents the total charging power of all EVs at time 𝑡 It is also assumed that all chargers have Vehicle-to-Grid (V2G) capability Thus, the charging station can be described by the following equations [66], [75], [119]:

As chargers have V2G capability, energy exchange between the grid and the EVs can perform in both directions The charging behavior of vehicles can be determined by the sign of 𝑃 𝑡 𝑆 , where 𝑃 𝑡 𝑆 > 0 if the batteries receive energy, and 𝑃 𝑡 𝑆 < 0 if the batteries discharge energy

The power constraints for charging/discharging can be described as follows:

𝐶ℎ𝑎𝑟𝑔𝑖𝑛𝑔: 𝑃 𝑡 𝑆 ≤ 𝑚𝑖𝑛{∑ 𝑁 𝑖=1 𝑃 𝑚𝑎𝑥 𝐶 𝑖 , ∑ 𝑁 𝑖=1 𝑃 𝑚𝑎𝑥 𝐵 𝑖 } (2.16) 𝐷𝑖𝑠𝑐ℎ𝑎𝑟𝑔𝑖𝑛𝑔: 𝑃 𝑡 𝑆 ≥ 𝑚𝑎𝑥{∑ 𝑁 𝑖=1 𝑃 𝑚𝑖𝑛 𝐶 𝑖 , ∑ 𝑁 𝑖=1 𝑃 𝑚𝑖𝑛 𝐵 𝑖 } (2.17) Where, 𝑃 𝑚𝑎𝑥 𝐶 𝑖 , 𝑃 𝑚𝑖𝑛 𝐶 𝑖 are the maximum allowable charging/discharging power for the 𝑖 𝑡ℎ charger, respectively

𝑃 𝑚𝑎𝑥 𝐵 𝑖 , 𝑃 𝑚𝑖𝑛 𝐵 𝑖 are the maximum allowable charging/discharging power for the battery of the 𝑖 𝑡ℎ EV

The total charging/discharging power of all EVs in the station must not exceed the supply capacity of the grid, which is expressed by the constraints:

𝐶ℎ𝑎𝑟𝑔𝑖𝑛𝑔: 𝑃 𝑡 𝑆 ≤ 𝑃 𝑚𝑎𝑥 𝑓𝑒𝑒𝑑𝑒𝑟 − 𝐷 𝑡 𝑆 = 𝑃 𝑚𝑎𝑥 𝑒𝑥 (2.18) 𝐷𝑖𝑠𝑐ℎ𝑎𝑟𝑔𝑖𝑛𝑔: 𝑃 𝑡 𝑆 ≥ −𝑃 𝑚𝑎𝑥 𝑓𝑒𝑒𝑑𝑒𝑟 + 𝐷 𝑡 𝑆 = −𝑃 𝑚𝑎𝑥 𝑒𝑥 (2.19) Where 𝑃 𝑚𝑎𝑥 𝑒𝑥 is the maximum power that the microgrid can exchange with the EVs and is equal to the difference between the maximum allowable exchange power between the microgrid and the distribution grid (𝑃 𝑚𝑎𝑥 𝑓𝑒𝑒𝑑𝑒𝑟 ) and the netload

It is also assumed that the non-EV loads are not dispatchable In this case, the charging station needs to adjust the total charging load demand of the EVs to ensure that the constraints (2.18) and (2.19) are not violated

Constraints on the energy requirements of the EVs:

Where, 𝑅 𝑆 denotes total required energy of E2Ws in the station

𝐹𝑆𝑂𝐶 𝐸𝑉 𝑖 is the SOC of 𝐸𝑉 𝑖 at the departure time

𝐼𝑆𝑂𝐶 𝐸𝑉 𝑖 is the initial SOC of 𝐸𝑉 𝑖

𝑡 0 is the starting time of the working day

𝑡 𝑚 is the end time of the working day

𝑃 𝑡 𝑖 is charging/discharging power of 𝐸𝑉 𝑖 at time 𝑡.

Summary

In this chapter, a PV integrated E2W charging station in real-time and long-term operational mode is modeled

The real-time model can be used to study the real-time operation of the charging station while the long-term model can be utilized as the framework for integrating optimal charging scheduling algorithms in the next chapters.

CHARGING POWER ALLOCATION ALGORITHM FOR

Chapter objectives

The objective of this chapter is to propose and develop a smart charging strategy for E2W charging stations comprising of an PV system, the distribution grid, and conventional loads The proposed algorithm aims to improve the aggregated load profile, contribute to grid support, and mitigate the drawbacks of the high penetration levels of rooftop PV and EV loads into the distribution grid The algorithm outputs charging profiles during the scheduling horizon Through diverse case studies and scenarios, the proposed algorithm demonstrates its effectiveness both in the improvement of load profile and the reduction of computational requirement

Since E2W charging stations typically serve a high number of EVs, each with small battery capacity and charging power, a grouping approach is proposed in which E2Ws are grouped based on their energy demand during the scheduling horizon The charging scheduling is carried out through three stages:

- Stage 1: Find a common profile for the entire charging station This profile is determined based on the objective of improving the total load curve

- Stage 2: From the profile of the station, allocate the power at each timeslot to the groups to determine the group charging profiles

- Stage 3: Based on the group profiles, determine individual charging profile considering relevant constraints

With the grouping approach and three stages of power allocation, solving the constrained OP multiple times to find each E2W charging profile is eliminated, thereby reducing computational demand when scheduling hundreds of E2Ws [49].

Input data requirements

In order to verify the performance of a charging scheduling algorithm, certain input data needs to be provided The data includes vehicle technical specifications, charging behaviors, forecasted load and PV power output

3.2.1 Electric bike and electric motorcycle specifications

According to the European Commission (EC), 2-wheel, 3-wheel vehicles and quadricycles are classified into L-category (Table 3.1)

It could be realized that an e-bike can travel up to 25 km/h using a motor of up to

250 W (up to 1000 W for high-speed e-bikes) and uses a 12–48 V battery with an energy capacity of 0.2–1 kWh Electric mopeds, on the other hand, can go up to 45 km/h using a motor of 1–4 kW and typically uses a 48 V battery of 1–5 kWh

Furthermore, e-bikes have an extremely low energy consumption in the range of 5–

15 Wh/km depending on the drivetrain efficiency, riding behavior, tire characteristics, and the combined weight of the bike and rider This is much lower than the 150–200 Wh/km energy consumption of an electric car [12]

Table 3.1 The family of L-category vehicles [3]

Being consistent with the EC, e-bikes in Vietnam often adopt 48 V-12 Ah batteries with motor power of below 500 W and maximum travel distance of 50-60 km Electric mopeds use larger batteries (60 V-20 Ah), higher motor power (800-1200 W) and have 70-80 km travel range (Table 3.2)

Maximum travel distance (km) e-bikes

Regarding charging scheduling problem, there are two major challenges: the uncertainty of EVs’ charging behaviors and the overlong solving time for the optimal solutions in the scheduling problem of large-scale EVs [69]

There are numerous studies, both theoretically and statistically, on charging behavior In general, charging behavior is characterized by the arrival time (connection time), departure time (disconnection time), charging duration, average distance traveled since the last charge, initial SOC as well as final SOC, etc

In [23], the charging behavior was determined by charging starting time and charging duration In this context, the daily private vehicle driving ending time was fitted to a normal distribution (also known as a Gaussian distribution), and the daily driving distance was fitted to a log-normal distribution This is also consistent with studies such as [11], [139], [148] According to [144], the probability density function for the first drive time and last return time follows Gaussian distribution with (μ; σ) being (7.5; 3.24) and (17.5; 3.41), respectively The daily driving mileage distribution conforms to the lognormal distribution pattern with (μ; σ) taking (3.37; 0.5)

A study based on driving pattern data of a large number of customers from the National Household Travel Survey (NHTS) [130] shown that the probability distribution of arrival time to the office or home and departure time from the office or home could be represented by a normal distribution [24] Other works such as [15],

[26], [41], [102], [147] also stated that home arrival/departure times and the daily trip distances turned out to be quite similar to a normal distribution

At home and workplace, authors in [16] claimed that on a larger scale, for a certain area with a high number of EVs, the arrival and departure time behavior could be statistically modeled with proper probability density functions (PDFs) like normal distribution function The normal distributions parameters could be extracted from the available historical data of the study region

In some other works, charging behavior is proposed to follow different distribution forms such as the gamma function, Birnbaum–Saunders distribution [9], extreme- value distribution [100] etc

In general, most studies on charging behavior consider the arrival and departure times of EVs to follow a probability distribution in the form of a normal distribution or a log-normal distribution Therefore, in this study, the E2W arrival and departure time distributions are assumed to be the normal distribution which is expressed as:

(3.1) Where 𝑓 is the probability of arrival/departure time, and à and σ are mean value and standard deviation, respectively

Regarding SOC behavior, the authors in [100] concluded that the majority of charging behaviors involved fully recharging EVs (over 90 %) regardless of the initial SOC value EV owners usually required their batteries to be fully charged as soon as possible to increase the vehicle’s autonomy [145] Specifically, the final SOC analysis revealed that a significant portion of the sampled EVs (87.5 %) were fully recharged, while less than 2.4 % of all charges were terminated prematurely before reaching a final SOC of less than 50 % Approximately 10.1 % of all charges resulted in a final SOC ranging from 50 to 90 % This charging termination occurred when the EV was required to depart even though the battery was not yet fully recharged [21]

Study [69] revealed that 75 % of the EV owners prefer home-charging and disconnect their vehicles for driving to work in the next day morning Home-charging was often implemented overnight Moreover, most EV owners tended to charge their vehicles as soon as they arrived home [144], which often coincided with the peak load period in the evening [41] Furthermore, EVs typically remained idle for a longer duration at night than the necessary charging time If charging started early, it would end early as well This resulted in limiting the potential for "valley filling" during the late-night and early-morning periods [144]

Regarding home-charging behaviors, in [69], it was observed that EV owners arrived home at around 18:00 and left for work at around 7:30, which was in accordance with the trip characteristics of people on weekdays (Figure 3.1) It was also observed that, probability densities of home arrival and departure times were quite different between weekdays and weekends [27] Sharp peaks were observed on weekdays, while more flattened curves were observed on holidays (Figure 3.2)

Figure 3.1 Arrival/departure distribution function - trip home [63]

Figure 3.2 Probability densities of home arrival and departure times [27]

According to [144], the probability density function for the first drive time and last return time followed normal distribution with (μ; σ) being (7.5; 3.24) and (17.5; 3.41) respectively In [107], (μ; σ) took (18; 5) as a result of the observation that most EV users started charging after returning home from work at 18:00 and the starting time for charging of more than 90 % of EV users was between 13:00 and 23:00 In [17], the means and standard deviations were (19.55; 1.67), (7.47; 0.383) and (39,5; 15,8) for home arrival time, departure time and daily trip distance These parameters in [69] were slightly different, being (19.89; 1.92), (7.56; 2.33) for the connection time (arrival) and disconnection time (departure), respectively

In general, home-charging behavior follows normal distribution function and EVs can be assumed to stay parked at home till the next morning departure time and occasional evening trips are ignored Nevertheless, this assumption would not change the performance of the scheduling solution if the actual mobilities of EVs are updated in real time

Charging power allocation algorithm for E2Ws

Since the parking duration is typically longer than the required charging time, charging loads such as EVs can be considered as deferrable loads which have high flexibility Thus, if properly dispatched, charging load can contribute to increasing the economic and technical benefits for charging station owners and distribution operators

With the aim of improving total load profile, this study proposes a charging power allocation algorithm that satisfies the energy needs of the E2Ws, charging/discharging constraints, and at the same time shapes the total load profile to match a predefined one

This approach is inspired by the concept of "water filling," where the allocatable power is determined for charging at each time-step This allocatable power is calculated based on the predetermined energy requirements over the scheduling horizon, assuming that accurate and forecasted data of conventional load, renewable power generation, and data of the vehicles are available

The "water filling" approach for power allocation to electric cars had been discussed in [56], [105], [149] This algorithm injected the power (charging demand) to the valley just liked pouring water into it, and the main objective was to minimize the load variance considering both conventional load and charging demand [56],

[149] However, unlike those studies, this research performs power allocation through three stages to determine the charging profile of individual E2W in the charging station

The proposed algorithm consists of three stages, which are described as follows:

▪ Stage 1: This stage aims at determining the total charging profile of the station The profile is determined by subtracting the desired total load profile (including netload and charging load) from the netload

▪ Stage 2: Based on the total charging profile of the station, stage 2 allocates that power to E2W groups Grouping criterion is based on vehicle’s energy requirements Thus, vehicles in the same group will have similar charging power profiles

▪ Stage 3: After obtaining group profiles, stage 3 allocates the group charging power at each timeslot to the individual E2W

All three stages must satisfy constraints relating to the energy requirements, charging/discharging power limits, depth of discharge (DOD) etc

If E2W groups are classified and the energy requirement difference between vehicles is negligible, the power allocation might stop at stage 2 As a result, the group profile can be applied to the vehicles within the group without the need of the third- stage algorithm However, applying the group profile to the vehicles in the group may result in small errors in the charging requirement of E2Ws, which might deteriorate the vehicle owners’ satisfaction

3.3.1 Mathematical formulation of the algorithm

Inherited from the long-term model in Chapter 2, in this section, a charging power allocation algorithm is developed with the aim of improving the total load profile Assuming that the charging station is connected to the grid, to the conventional loads of a building and to a rooftop PV system, and all chargers have V2G feature The average power of total load during a working day can be estimated by the following equation:

The variance of the total load (including netload and charging loads) can be calculated by:

𝑀∑ 𝑀 𝑡=1 (𝐷 𝑡 𝑆 + ∑ 𝑁 𝑖=1 𝑃 𝑡 𝑖 − 𝜇 𝑎𝑣𝑒 𝑆 ) 2 ≥ 0 (3.3) Thus, for the purpose of load leveling, an ideal charging reference vector or preliminary total charging power at timeslots could be defined as:

In order to refine the preliminary total charging power at each timeslot The refinement process must consider the constraints for the aggregated charging load as in equations (2.16) - (2.19)

If the total charging power of E2Ws in the station follows the ideal charging reference vector, then (2.20) would be satisfied Therefore, the refinement process is essentially the process of reallocating the preliminary total charging power at timeslots such that the constraints (2.16) - (2.19) are satisfied

After the reallocation, the final charging reference vector is obtained as in (3.5):

In the next stage, the charging reference vector is allocated to groups of E2Ws Assuming that 𝑁 vehicles in the charging station are divided into 𝐾 groups The vector of EV group number is:

𝒏 = [𝑁 1 𝑁 2 ⋯ 𝑁 𝐾 ] 𝑇 , ∑ 𝐾 𝑗=1 𝑁 𝑗 = 𝑁 (3.6) The charging power matrix of the groups which is called charging reference matrix is defined as:

The allocation of the charging reference vector to E2W groups should consider constraint on the equality between aggregated group charging power and total charging power of the station as in (3.8)

Each column vector in the charging reference matrix corresponds to the charging profile of one group For the 𝑗 𝑡ℎ group, the constraint on the amount of required energy is:

Constraint on group charging/discharging power:

𝑃 𝑡 𝐺 𝑗 ≥ 𝑚𝑎𝑥 {∑ 𝑁 𝑖=1 𝑗 𝑃 𝑚𝑖𝑛 𝐶 𝑖,𝑗 , ∑ 𝑁 𝑖=1 𝑗 𝑃 𝑚𝑖𝑛 𝐵 𝑖,𝑗 }, 𝑗 = 1,2, … , 𝐾; 𝑡 = 1,2, … , 𝑀 (3.10) Maximum group battery capacity constraint:

Deriving from the charging reference vector, stage 2 algorithm determines the charging reference matrix considering the constraints (3.8) - (3.11) In the next stage, the charging reference matrix is then exploited to determine individual charging pattern

It is also noted that the selection of vehicles in a group is based on the required charging energy Assuming that all E2Ws are fully charged at the end of the scheduling horizon Thus, the vehicles in a group would roughly have the same initial SOC For simplicity, the algorithm in stage 3 estimates the charging power for each vehicle of the group based on the group charging pattern as in (3.12)

𝑁 𝑗[𝑃 1 𝐺 𝑗 𝑃 2 𝐺 𝑗 … 𝑃 𝑀 𝐺 𝑗 ] 𝑇 , 𝑗 = 1,2, … , 𝐾; 𝑖 = 1,2, … , 𝑁 𝑗 (3.12) After preliminarily assigning as in (3.12), a correction process should be performed to meet the following constraints:

At all time-steps, the aggregated charging power of vehicles in the group is approximately equal to the total charging power of the group:

Constraint on the amount of required energy for each vehicle:

= ∑ 𝑀 𝑡=1 𝑃 𝑖,𝑗 𝑡 ∆ 𝑡 , 𝑗 = 1,2, … , 𝐾; 𝑖 = 1,2, … , 𝑁 𝑗 (3.14) Individual charging and discharging rate constraint:

Permitted depth of discharge constraint:

Based on the mathematical formulation in Section 3.3.1, the stage 1 algorithm consists of two functions (Figure 3.5) The purpose of function 1 is to find the charging reference vector based on the information of vehicles at the station, forecasted conventional load data, and data of forecasted PV power output Function 2 is a constraint checking function These functions are described as follows:

Stage 1 – Total charging pattern finding

Function 1 - Charging reference vector finding This function includes the following steps:

Acquisition of E2W parameters, non-EV load profile, PV power output It is assumed that the forecast is accurate enough

Calculation of netload (as in 2.15), total charging energy demand (2.20), average total load demand (3.2) and ideal charging reference vector (3.4) or preliminary total charging power at each time-step

Check the power constraints (2.16) (2.17), (2.18) (2.19) at each timeslot If there are any violations, implement corrections then allocate charging power at the unsatisfactory period to the others The allocation should not make the charging power at any periods violate the power constraints After this step, a more improved total charging power pattern, which is then assigned to the charging reference vector, is obtained

Function 2 - Total battery capacity checking This function includes the following steps:

The first step in this function is to calculate total energy stored in batteries then check the maximum battery capacity violation at each timeslot If violation occurs at any periods, reallocate charging energy at the unsatisfactory period to the others The reallocation process should consider charging power limits

After the correction and reallocation, recalculate the total charging power pattern which is utilized as charging reference vector for stage 2 algorithm

Stage 2 – Group charging pattern finding

Based on the charging reference vector obtained after stage 1, stage 2 manages to allocate the total charging power at each timeslot to EV groups (Figure 3.6) The stage

2 algorithm consists of the following functions:

Function 3 - Preliminary group charging pattern finding

Summary

In Chapter 3, a novel power allocation algorithm consisting of three stages is proposed and simulated Compared with other methods based on the centralized approach [27], [29], [60], [64], [65], [82], the proposed solution does not derive from solving multivariable optimization problems but from the desired total load profile then implement power allocation The process includes three stages in which preliminary charging profiles are found then constraint checking, and reallocation is implemented until no violation occurs This approach helps to reduce computational complexity as well as computational time in solving optimization problems, especially when the number of E2Ws in a charging station is often much higher than those in electric car charging stations, leading to higher number of optimal variables

The total load profile is designed aiming at load leveling, peak clipping and valley filling However, because the total load profile can be pre-computed for various objectives, the proposed idea can completely adapt for many different optimization purposes

The simulation results in different scenarios show that, by applying the proposed algorithm, the total load profile is significantly improved, dramatically narrowing the load fluctuation, filling the valley load as well as reducing peak load

This chapter is mainly based on:

[1] Huu D.N and Ngoc V.N (2021) A Two-Level Desired Load Profile Tracking Algorithm for Electric Two-Wheeler Charging Stations Eng Technol Appl Sci

[2] Huu D.N and Ngoc V.N (2022) A Three-Stage of Charging Power Allocation for Electric Two-Wheeler Charging Stations IEEE Access, 10, 61080–61093.

OPTIMAL CHARGING ALGORITHM BASED ON

Chapter objectives

It can be observed that although the 2 or 3-stage power allocation algorithms can address the charging scheduling problem for a high number of EVs (specifically for E2W charging stations), the algorithms still have some limitations

Firstly, the charging scheduling problem is assumed to follow the average total load profile over the scheduling horizon Mathematically, this solution cannot be claimed as optimal

Secondly, the algorithm is verified based on deterministic data of vehicles in the charging station Parameters such as arrival time and departure time at the charging station are assumed to be fixed However, since the charging behavior is not a deterministic data, the dynamic impacts of the behavior should be further addressed Especially when new vehicles connect to the charging station or when vehicles depart, the charging schedule needs to be adjusted adaptively

However, the prominent advantages of the algorithm are the grouping and power allocation approaches, which help release computational demand compared to solving an optimization problem for each vehicle Based on this approach, this section develops a real-time scheduling algorithm in which the vehicles are dynamically grouped at the current timeslot E2W data is updated whenever any change in vehicle existence occurs Moreover, scheduling is performed whenever the current timeslot changes

With the grouping approach, computational complexity, a crucial aspect in coordinated charging of large-scale EV populations can be minimized [30], [40] A low computational complexity of decision-making processes is necessary to achieve the real-time performance of charging coordination for practical applications The scheduling framework also should be capable of handling realistic dynamics and uncertainties [68]

Regarding desired SOC and drivers’ expected departure time, it would require a user interface (UI) to collect necessary information Such information should be set by the user every time a connection between his vehicle and the station is formulated However, the UI design is out of the scope of this research Thus, the desired SOC and expected departure time of vehicles are assumed to be available to the scheduler

To summarize, the proposed algorithm in this Chapter has the following characteristics:

1) Grouping approach: This approach allows group level optimization, reducing computational complexity Noteworthily, grouping is implemented whenever the current timeslot changes Thus, it is dynamic

2) Reduction of computational complexity: By optimizing at group level, the algorithm contributes to releasing computational complexity, making it suitable for practical applications

3) Real-time performance: The algorithm considers dynamic changes such as the vehicle arrival/departure time It updates the scheduling data and re-schedules

4) Handling dynamics and uncertainties: The framework is capable of handling realistic dynamics and uncertainties associated with charging coordination Adaptive charging plans are implemented when new vehicles connect/leave.

Mathematical formulation, control framework and algorithm flowchart

In most studies on EV scheduling, a scheduling day is typically discretized into 96 timeslots equally and the length of the timeslot (∆ 𝑇 ) is 15 minutes A 15-min resolution was one of the most common temporal resolutions for demand side management (DSM) schemes [16], [31], [39], [59], [60], [65], [147] It is understandable that if a vehicle arrives at time 𝜏 𝑎 and departs at time 𝜏 𝑑 , it would be assigned to arrival timeslot 𝑡 𝑎 and departure timeslot 𝑡 𝑑 as follows:

To be explicit, for a particular EV, the charging process must be scheduled from the timeslot next to the arrival time 𝜏 𝑎 to the timeslot right before the departure time

As far as the accuracy of the charging scheduling is concerned, the timeslot should be as short as possible [58] While, if the time slot is too short, it will increase both the complexity of the problem and the re-scheduling frequency [57]

Figure 4.1 Scheduling and implementing timeline

For the purpose of load leveling, the optimization objective aims to minimize load fluctuation This can be interpreted as total load variance minimization [16], [28],

[33], [34], [35], [41], [56], [66], [75] Thus, the objective function would be as follows: min 𝐹 𝑜𝑏𝑗 = 1

𝑀∑ 𝑀 𝑡=1 (𝐷 𝑡 𝑆 + ∑ 𝑁 𝑖=1 𝑃 𝑡 𝑖 − 𝜇 𝑎𝑣𝑒 𝑆 ) 2 (4.3) The scheme aiming at the objective function (4.3) will attempt to make the total load at each timeslot be as close as possible to the mean load during the parking period, resulting a maximally flat load profile [20]

The problem of load variance minimization was proved to be a convex optimization problem [135] Thus, it possesses two prominent advantages The first advantage is that the problem can be solved quickly and efficiently using commercial solvers, which is essential for real-time dispatch The second is that these objectives can be easily integrated as constraints in other EV charging objective functions such as minimizing system operating costs or maximizing aggregator profits [135] It is also well known that the global optimal solution of quadratic programming (QP) problem can be efficiently solved in polynomial time [136] Thus, the resulting OP is polynomial time-space complexity

Solving the QP problem (4.3) gives the optimal charging profile for EVs during the scheduling horizon, usually one day, for the given netload forecast

In this study, the optimization problems were solved using MATLAB Quadratic Programming (quadprog) with the interior-point-convex algorithm

The standard form of a QP is: min𝑥

Note that 𝑥 𝑇 denotes the transpose of 𝑥, and 𝐴𝑥 ≤ 𝑏 means that the inequality is taken elementwise over the vectors 𝐴𝑥 and 𝑏

The quadprog function expects a problem of the above form, defined by the parameters {𝐻, 𝑓, 𝐴, 𝑏, 𝐴 𝑒𝑞 , 𝑏 𝑒𝑞 , 𝑙, 𝑢}; 𝐻 and 𝑓 are required, the others are optional Alternate QP formulations must be manipulated to conform to the above form; for example, if the inequality constraint was expressed as 𝐴𝑥 ≥ 𝑏, then it can be rewritten

Considering 𝐸𝑉 𝑖 , if 𝑥 𝑡 is the charging power of 𝐸𝑉 𝑖 at timeslot 𝑡, then the objective function aiming at load variance minimization over the scheduling horizon can be expressed as: min ∑ (x 𝑡 − (𝜇⏟ 𝑎𝑣𝑒 𝑆 + 𝑃 𝑡 𝑃𝑉 − 𝑃 𝑡 𝑛𝑜𝑛𝐸𝑉 )

𝑡 𝑎 is the arrival timeslot of 𝐸𝑉 𝑖

𝑡 𝑑 is planned departure timeslot of 𝐸𝑉 𝑖 , and

𝑎 + 𝐴 𝑖 (𝐹𝑆𝑂𝐶 𝐸𝑉𝑖 −𝐼𝑆𝑂𝐶 𝐸𝑉𝑖 ) Δ 𝑇 ) (4.11) With the assumption that the forecast data for conventional load and solar power is accurate enough, 𝜇 𝑎𝑣𝑒 takes on a definite value at each timeslot

After finding {𝐻, 𝑓, 𝐴, 𝑏, 𝐴 𝑒𝑞 , 𝑏 𝑒𝑞 , 𝑙, 𝑢} from above equations, it is also needed to set what MATLAB solver to use A general choice is to use the interior point method which is also the selected method in this study

The complexity of the optimization problem depends on the number of timeslots and the number of E2Ws in the charging station The proposed algorithm adopts a group-based approach, finding an optimal charging profile for the representative E2W of each group and correcting it to fit with the remaining E2Ws Consequently, the scale of the optimization problem is influenced by the number of groups and timeslots

To address the issue of real-time charging scheduling to accommodate the behavior uncertainty of E2Ws, the proposed solution is to use a receding horizon framework Following this framework, the charging station updates the status of the current E2Ws in the station, including existing E2Ws and any new arrival or departure at the current timeslot

Research [94] shown that receding horizon control (RHC) provided a method to extend the static optimization approaches to real-time scheduling so that the dynamic changes of the system could be taken into account Generally, RHC works by solving optimization over the next 𝑇 time-steps, implementing the decision for the first time- step, and then re-solving the optimization problem for the subsequent 𝑇 time-steps by incorporating new measurement data available at that moment (Figure 4.2) In short, RHC scheme repeats the process of optimization, execution, and adaptation

Figure 4.2 Illustration of receding horizon time window [64]

At each iteration, the optimization problem will be updated by considering the changes of loads For example, the entire time horizon is assumed to be [1, 24] for 24- hour period The first iteration solves the optimization in time horizon [1, 24], the second iteration considers optimization horizon [2, 24], and so on

Two schemes can be considered: the complete RHC algorithm and the partial RHC algorithm The former scheme implements scheduling for all E2Ws in the station at each iteration The latter scheme keeps the original schedule decision for the existing loads, and only schedules newly arrived charging requests It could be seen that the second scheme is suboptimal compared to the first one with the merit of light computational burden

One major benefit of the receding horizon approach is the ability to incorporate updated information In E2W scheduling, updates may include newly arrived charging requests as well as unexpected charging termination For example, a customer may plug his EV to the station at any time in the scheduling horizon On the other hand, it is possible that an existing process is unexpectedly terminated, for instance, in the middle of charging for an unplanned trip

It is worth noting that the chargers’ efficiency varies with both the operating power and the working voltage However, it has been demonstrated that they don’t change fiercely in magnitude [121], thus herein they are considered being constant for simplicity Another simplification involved in the scheduling is that the onboard batteries are deemed as ideal energy storage systems Thus, their transient performances during charging/discharging are not considered

In order to extend the lifetime of batteries, upper limit and lower limit for the SOC value should be set to avoid over charging and deep discharging, which both can harm the physical constitution of batteries Most studies agree that the minimum SOC value should be 20 %, while the maximum SOC value ranges from 80 - 100 % [14], [25],

[32], [36], [41], [48], [105] In this study, it is assumed that E2Ws do not discharge if their SOC value is below 20 % and E2Ws can be charged up to a maximum of 100 %

It is also assumed that the smart charging station is able to recognize the nominal capacity of the onboard battery and the connection SOC, permitted charging/discharging power via a vehicle information system (VIS) which is responsible for providing E2W information to the station controller [67], [72] After communicating with the VIS and E2W owner, an E2W can be characterized by a vector of parameters as in (4.12)

Case study and simulation results

In this section, several case studies are proposed and analyzed to assess the algorithm’s effectiveness These case studies examine charging stations at a university with two non-continuous studying periods, an office charging station operating from 8:00 to 17:00, an apartment charging station where EVs are typically charged overnight, and a charging station for a factory operating in three consecutive shifts In comparison to the case study in Chapter 3, an additional scenario, where the E2Ws do not allow discharging, is also included This is because battery wear and tear cost may be a crucial factor preventing V2G acceptance of EV owners Hence, each case study includes the following

Figure 4.3 Flowchart of the algorithm scenarios:

▪ Scenario 1 - Charging loads do not participate in the microgrid This scenario is analyzed to evaluate netload variance in case of RESs participation

▪ Scenario 2 - Max speed charging (or uncontrolled charging): E2Ws are charged at the maximum possible power as soon as they are connected and stop charging when reaching their required SOC

▪ Scenario 3 - Average charging: E2Ws are charged at a constant rate At the departure time, E2Ws reach their required SOC

▪ Scenario 4.1 - Smart charging: E2Ws follow the proposed algorithm However, only G2V is permitted

▪ Scenario 4.2 – Smart charging: E2Ws follow the proposed algorithm and allow bi-directional power exchange (both charging and discharging are permitted) Regarding input data for the case studies, it is assumed that the day-ahead forecast of the non-EV load and PV power output are available and sufficiently accurate To be explicit, load profile data is retrieved from the dataset of Mendeley Data for industrial, commercial, and residential end-use sectors [75] This data can be referred to Appendix 3 Solar power output data can also be forecasted using various methods However, this is also not the focus of this study Thus, in this study, the PV power data is retrieved through the simulation of fixed solar panel systems with locations and weather conditions in Vietnam Furthermore, because the simulation does not consider the shading effect or the uncertainty of weather conditions, the solar irradiation on different PV panels is assumed to be uniform Appendix 4 reveals the PV power profiles (15-minute resolution) used for case studies

Regarding SOC distribution, the initial SOC of E2Ws was created based on research data about charging behavior as in [21] For each case study, a random generator is adopted to generate several sets of E2W populations as in Appendix 2

In Chapter 3, It is elucidated that the E2W arrival and departure time distributions can be assumed to be the normal distribution The distributions parameters could be extracted from the available historical data of the study region Thus, in each case study, patterns for the arrival time and departure time are generated as in Appendix 1

With the purpose of verifying the effectiveness of the real-time RHC based algorithm and comparing with the power allocation algorithm (which is studied in Chapter 3), two case studies have been developed: a) Charging station for university staff (full day working): The station can accommodate 150-170 vehicles b) Charging station for students (two studying periods in a working day): The station can accommodate up to 250 vehicles

The case study a) is investigated with the same assumptions as in Chapter 3 However, case study b) is supposed to service two studying periods: the morning period from 7:00 to 11:30 and the afternoon period from 12:30 to 17:00

Other input data such as PV power, conventional load and E2W specifications are similar to the case study in Chapter 3

4.3.1.1 Charging station for university staffs

This case study is conducted to compare with the previously proposed power allocation algorithm However, smart charging scenario includes both V2G and non- V2G instances

The simulation results are as follows:

Figure 4.4 Average charging pattern Figure 4.5 Max rate charging pattern a) Neload profile b) Total load profile - max rate charging c) Total load profile - average rate charging

Figure 4.6 Total load profile in scenarios 1, 2, 3

Figure 4.7 Charging profile – RHC based algorithm (scenario 4.1)

Figure 4.8 Total load profile – RHC based algorithm (scenario 4.1)

Figure 4.9 Charging profile – RHC based algorithm (scenario 4.2)

Figure 4.10 Total load profile – RHC based algorithm (scenario 4.2)

In max speed charging scheme (scenario 2), it can be noticed that because E2Ws start charging at a high power immediately at the connection time (Figure 4.5), a peak load is created at the beginning of the working hours (near the peak periods of netload), resulting in a significant difference between peak and off-peak loads (Figure 4.6b) In January, the peak of 89.1 kW (around 7:00) is 13.33 times higher than the valley load of 6.69 kW at timeslot 21 (12:00)

Besides, since E2Ws complete charging early in the morning, the utilization of PV power for charging is less efficient In months of high solar radiation (e.g., May, June, July), total load at periods of high PV power output might be negative, which implies a PV power injection into the distribution grid

In scenario 3, by evenly distributing the charging power from the arrival time to the departure time, the total load shape remains unchanged compared to the netload (Figure 4.6c) However, the overall load increases as it includes both the conventional load and the charging load

The total load profile only exhibits a significant improvement in the smart charging scheme In scenario 4.1, due to the E2Ws not allowing discharging, this scheme does not contribute to reducing the peak load (Figure 4.8) During timeslots when the netload reaches its peak (around timeslot 5 in the morning and timeslot 37 in the afternoon), it can be observed that the E2Ws do not charge to avoid increasing the peak load (Figure 4.7) The E2Ws perform charging at a high charging rate during the off-peak hours when coincide with high solar power generation hours (Figure 4.7) The effect of valley filling is clearly demonstrated

Compared to scenario 4.1, in scenario 4.2, peak shaving effect occurs as the E2Ws discharge energy during peak periods (Figure 4.9), thereby contributing to reducing the overall peak load Scenario 4.2 demonstrates the most improvement in the total load profile compared to other scenarios However, during the final timeslots of the working day, the E2Ws do not discharge to assure the required SOC level by the departure time As a result, higher power demand can be observed in these last timeslots than in the earlier ones Peak shaving and valley filling are clearly observed in Figure 4.10

The simulation results also indicate that compared to the three-stage power allocation algorithm, scenario 4.2 produces a total load profile that closely resembles

Table 4.1 Load variance in different scenarios

Scenario 1 Scenario 2 Scenario 3 Scenario 4.1 Scenario 4.2

Jan 111,071,069 496,987,981 111,071,069 13,190,716 651,266 Feb 148,246,204 580,699,035 148,246,204 23,622,939 1,458,466 Mar 145,483,222 556,286,750 145,483,222 21,170,792 4,380,515 Apr 161,290,670 576,055,083 161,290,670 26,484,600 8,871,876 May 182,823,353 620,486,235 182,823,353 36,571,416 16,478,368 Jun 176,123,091 606,921,755 176,123,091 34,269,423 17,557,115 Jul 139,741,318 554,705,620 139,741,318 22,239,232 10,175,335 Aug 149,118,945 550,593,284 149,118,945 25,958,136 15,714,134 Sep 177,426,398 568,487,987 177,426,398 39,495,142 29,439,929 Oct 206,897,743 568,674,450 206,897,743 59,060,602 50,952,454 Nov 193,474,507 571,720,137 193,474,507 47,711,259 36,736,745 Dec 156,611,432 555,227,993 156,611,432 27,477,741 13,319,280

Figure 4.11 Load variance in different scenarios

Considering the working hour period, Table 4.1 and Figure 4.11 depict the load variance in different scenarios Among the scenarios, the max speed charging scheme triggers the worst load fluctuation, while the average charging scheme does not

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec L oad v ar ian ce ( W 2 )

Load variance in different scenarios

Scenario 1 Scenario 2 Scenario 3 Scenario 4.1 Scenario 4.2 improve the overall load profile The load profile shows the most notable improvement when implementing the RHC based algorithm Among scenarios 4.1 and 4.2, the load variance in scenario 4.2 is lower, resulting in a better overall load profile This is because in scenario 4.2, E2Ws permit discharging to reduce the peak load during high- demand hours

Figure 4.12 Load variance in the two proposed algorithms

Figure 4.12 illustrates the load variance in scenario 4.2 compared to the load variance using the three-stage power allocation algorithm It can be observed that there is no significant difference between the two algorithms in terms of load variance Besides, the RC-based algorithm has the advantage of dynamic and real- time scheduling As a result, it can effectively handle dynamics and uncertainties, making it more practical for scheduling large-scale E2W population

Appendix 6.1 depicts the charging profile and battery capacity of several E2Ws in scenarios 4.1 and 4.2 The results show that E2Ws charge at a high power during timeslots with high solar power generation The significant difference can be observed in the charging power profiles between scenario 4.1 and scenario 4.2, where E2Ws in scenario 4.2 can discharge energy to support the micro grid

This case study is proposed to investigate the performance of the RHC based algorithm considering the dynamic factors of arrival/departure times and studying periods within the workday

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec L oad v ar ian ce ( W 2 )

Three-stage power allocation RC based algorithm (Scenario 4.2)

Summary

In this chapter, an optimal charging algorithm for PV-integrated E2W charging stations is proposed and validated The performance and effectiveness of the algorithm have been analyzed through various case studies

Compared to the work [10], which also adopted receding horizon framework to minimize the overall daily peak power, this work schedules E2W charging iteratively with updated E2W data then performs load leveling and outputs individual charging profiles while the work [10] only focused on minimizing the overall daily peak power

To be specific, receding horizon policy in [10] tried to coordinate the vehicle charging to maintain the power consumption equal to the current or future peak power Besides, individual charging profiles were not clearly demonstrated in [10]

The proposed RHC-based algorithm inherits the advantages of the group-based approach and, at the same time, leverages the RHC framework and mathematical optimization tools to solve the real-time optimal charging problem RHC framework adoption enables the algorithm to handle uncertainties in charging behavior Scheduling is performed at the current timeslot As the current timeslot changes, the charging station updates the EV table, implements grouping dynamically, solves the

OP, assigns group profiles to the individuals then carries out correction This significantly releases the computational burden and makes it practical to schedule a high number of E2Ws, which is an important characteristic of the E2W charging stations

Research results under various case studies have revealed interesting conclusions

To be specific, a significant improvement in the aggregated load profile can be observed under smart charging schemes Compared between V2G and non-V2G smart charging scenarios, non-V2G scenario does not contribute to reducing the peak load while V2G scenario can help peak shaving because it enables E2Ws to discharge during high demand periods Both V2G and non-V2G schemes can fill the valley and perform load leveling effectively compared to other scenarios such as uncontrolled charging and average charging scheme

In the case of non-continuous working shifts, such as charging stations for students, because there is low charging load between two shifts, a dramatical load fluctuation can occur at that time This effect is more serious if the time between shifts coincides with the time of high solar output In the case of charging stations for shift-working factories, since the shifts are consecutive, attention should be paid during the shift transition because of the sudden high number of vehicles at that time Besides, because load characteristic of shift-working factory is a relatively flat, average charging might be a practical option since it is simple and doesn’t affect the total load profile dramatically In the case of charging stations for apartment buildings, since E2Ws are parked overnight and leave in the next morning, the load profile under smart charging schemes only improves during nighttime when the vehicles are available Such schemes neither affect the total load profile during daytime nor effectively utilize solar energy for charging purposes These issues would raise the need for other solutions (such as BESS) to address

Generally, through various case studies such as university charging with two non- continuous studying periods, office charging, overnight charging at an apartment building, and three-working shift factory charging, the effectiveness of the algorithm in load leveling, valley filling, and peak shaving has successfully validated

This chapter is mainly based on:

[1] Huu D.N and Ngoc V.N (2022) A Three-Stage of Charging Power Allocation for Electric Two-Wheeler Charging Stations IEEE Access, 10, 61080–61093

[2] Huu D.N and Ngoc V.N An Optimal charging algorithm based on receding horizon framework for electric two-wheeler charging stations (ongoing submission)

[3] Ngoc V.N and Huu D.N (2023) Optimal valley filling algorithm for Electric Two-wheeler Charging Stations Eng Technol Appl Sci Res (Accepted).

REALTIME RESPONSES OF E2W CHARGING AND

Chapter objectives

This chapter aims at establishing a testing workbench to implement observation, measurement, and validation of the charging/discharging responses of E2Ws in the charging station

As mentioned in the previous chapters, the coordination of charging for E2Ws involves optimizing the charging/discharging schedules with the output being the charging/discharging power at timeslots over the entire scheduling horizon However, to realize those long-term schedules, studies on the real-time responses following commands derived from long-term operation should be considered To investigate this response, in this chapter, an E2W charging station is simulated in real-time mode, then a test bench establishment is carried out A control program to charge/discharge E2Ws at predetermined powers is also developed

The contributions of this chapter are as follows:

1) Real-time charging/discharging response simulation

2) Developing an empirical test bench for an E2W charging station

3) Conducting experimental tests to assess the feasibility of long-term scheduling algorithm

4) Measuring and evaluating the real-time charging/discharging.

Real-time charging/discharging simulation

Most of the results obtained from long-term charging schedule are charging profiles of individual vehicle at timeslots Using the real-time model introduced in Chapter 2, in this section, real-time charging/discharging responses of E2Ws are investigated Charging/discharging commands can be considered as predetermined power values at timeslots If the power value is positive, the E2W battery will be charged By contrast, the battery will discharge

Figure 5.1 illustrates the total charging power profile of the charging station on a typical day

This profile shows the charging power at each time-step during the scheduling horizon To observe the charging responses, three charging power values of 5.6 kW, -1.95 kW, and -6.56 kW at three timeslots are considered (positive values when the station consumes energy and negative values when the station discharges energy)

Figure 5.1 Typical total charging profile

Figure 5.2 Total charging current at 5.6 kW charging power command

Figure 5.3 Charging response when power command changes from 5.6 kW (charging) to -1.95 kW (discharging) Figure 5.2 and Figure 5.3 illustrate the changes in the current supplied to the charging station when charging command abruptly changes from 5.6 kW (charging) to -1.95 kW (discharging) It is observed that the current response follows the reference current

Figure 5.4 Total charging power at charging commands of 5.6 kW, -1.95 kW, -6.56 kW

Figure 5.5 PV power, grid power, conventional load and charging load

In Figure 5.4, the charging response to the charging commands of 5.6 kW, -1.95 kW, and -6.56 kW is shown Figure 5.5 illustrates the real-time characteristics of the solar power (P_PV), conventional load (P_Load), power exchange with the distribution grid (P_Grid), and power consumption of the charging station (P_Batt)

It can be realized that, at any time, a power balance is maintained between the solar power, conventional load, charging load, and power consumed from the grid Within the observation window, the PV power is capable of supplying power to both the conventional load and the charging station

Figure 5.6 SOC and battery voltage of a typical E2W

Figure 5.6 shows the variations in the SOC and battery voltage of a typical battery in the charging station Depending on the high or low charging/discharging power, it determines the charging rate or steep slope of the SOC characteristic over time During the charging/discharging process, the battery voltage and current are monitored to assure they do not exceed the maximum allowable voltage

The simulation results demonstrate that with the charging/discharging power commands obtained from the long-term charging schedule, the charging station is fully capable of meeting the plan This affirms the feasibility of the long-term charging scheduling algorithm when considering real-time responses.

Testing workbench set up

5.3.1 The technical scope of the test bench

The first step in the test bench development is to determine the scale and technical scope of the prototype This aims to define the experimental scale, charging/ discharging power, and necessary features to ensure the proper functionality, reliability, and safety of the test bench This step is also crucial in pointing out technical specifications, measurement parameters, and testing procedures

The main components of the experimental charging station are depicted in Figure 5.7 These components include DC-DC converters, inverters, batteries, and solar panels Besides, measuring equipment are also needed to acquire measurement data such as SOC, charging/discharging power, battery voltage, and current

Figure 5.7 Test bench block diagram

In the test bench, PV panels are fixed on the roof of a building in Hanoi (location: 21.01, 105.8) Futhermore, the shading effect and the uncertainty of weather conditions are not investigated in the test Thus, the solar irradiation on PV panels is considered to be uniform

The investigated specifications in Chapter 3 indicate that the majority of electric bicycles in Vietnam use batteries with a voltage of 36-48 V and a capacity of 12 Ah E-bikes adopt motor power below 250 W and have a design speed not exceeding 25 km/h (according to QCVN 75:2019/BGTVT, QCVN 68:2013/BGTVT) E-bikes can travel a maximum distance of approximately 50-60 km On the other hand, electric motorcycles use larger batteries (48-60 V; 20 Ah) with higher motor power (800-

1200 W), enabling them to travel a maximum distance of around 70-80 km

Generally, most chargers for E2Ws work with single-phase AC power from the residential grid The charging power of these chargers is typically around 400 W, and the charging time ranges from 3 to 5 hours On the other hand, the discharge power can be up to 1200 W

Since the test bench is designed to investigate the real-time charging/discharging responses, in the experimental model, the range of charging/discharging power is selected from 0 W to 400 W, and the batteries are chosen having specifications of 12.8 V, 30 Ah The battery type is LiFePO4, allowing continuous discharge current up to 30 A (maximum continuous discharging power of 380 W), instantaneous discharge current up to 100 A, and maximum charging current of 10 A It's also worth noting that to optimize the lifetime of cells, the battery pack is integrated with an active voltage balancing circuit

For Lithium battery charging, the most popular method is the Constant Current- Constant Voltage (CCCV) because of its simplicity and easy implementation [7],

[110] It should be noted that more than 80 % of the battery capacity is filled during

CC stage [47] However, about 50 % of the total charging time is taken during CV stage [62], [47] To address the long charging time of CCCV method, solutions (such as multistage constant current (MCC) method; pulse charging; boost charging; variable current profiles) manage to regulate the charging current in the CC phase Higher current levels are usually chosen for the earlier CC stages

It is obvious that the charging rate increases with charging power The control of charging rate or charging power is primarily performed during the CC stage Therefore, in the test bench, tests are conducted to observe different charging responses within the CC stage

The converters are selected so that the rated power, voltage range, and current ratings are compatible with the batteries The charging and discharging power can be changed by regulating the charging and discharging current

5.3.2 Test bench design and operation

Figure 5.8 illustrates the design of the test bench The charging station employs separate buck/boost converters and utilizes two inverters: a grid-tie solar inverter, and a single-phase grid-tie inverter converting DC link voltage into AC voltage

In case of battery discharging, because the input voltage of the inverter must be greater than a minimum value specified by the inverter’s specifications, the test bench must equip with proper buck/boost converters to convert the battery voltage to the suitable input voltage level of the inverter In case of battery charging, these converters are adopted to convert the DC voltage (from the DC link or AC/DC converter) to the appropriate level for the battery

For a small-scale E2W charging station, because of modest permitted charging power and battery capacity, a single-phase grid-tie inverter is a suitable option However, in the case of larger scale, it is possible to add multiple sub-stations being connected to different phases By this way, the charging station can be easily expanded in terms of power and scale, allowing it to serve higher number of vehicles However, the addition of new sub-stations to the three-phase grid should consider phase load balancing and cooperation between sub-controllers which are out of scope of this dissertation

In the experimental setup, because of the small-scale pilot prototype and available equipment limitations, the single-phase grid-tie inverter is unidirectional Thus, an AC/DC converter is adopted to convert AC electricity from the grid for DC charging

Besides, an additional path of charging can be realized from the DC link as in Figure 5.8 If charging is executed from the DC link, the total charging power must be lower than the total discharging power of the remaining batteries Worth mentioning, in the case of using bidirectional grid-tie inverters, the employment of AD/DC converter is not necessary

To charge Battery 1, the grid voltage is converted into DC voltage through the AC/DC converter It then goes through R12 and a programmable buck converter to regulate the voltage and current before passing through R14 to Battery 1

Summary

In this chapter, real-time charging/discharging response research is conducted to test the operation of the charging station with charging/discharging commands obtained from the long-term scheduling algorithm Both simulation results from the Simulink model and experimental results retrieved from the test bench demonstrate that the optimized charging algorithms for E2Ws over a scheduling horizon are feasible under real conditions The response time for charging/discharging commands is negligible compared to the duration of one timeslot

The charging/discharging voltage and current setpoints are represented in the form of data tables in the controller’s memory and sent to the programmable power converters via Modbus RTU communication This solution leverages the advantages of Modbus communication and the reliability of industrial control systems while still providing scalability The setup and operation of the test bench have demonstrated the feasibility of the proposed solution in real-world conditions

However, the test bench still has some limitations that need to be improved, such as:

1) In the test bench, due to limitations in the power capacity and available commercial equipment, the grid-connected converter is unidirectional However, in case of using a bidirectional grid-connected converter, which allows power exchange in both directions (from DC link to the grid and vice verse), the AC/DC converters can be eliminated Additionally, if the bidirectional converter supports control and data acquisition through industrial communication standards, the operation of the charging station may become more flexible

2) At present, the test bench does not utilize monitoring devices for charging voltage, SOC, and State of Health (SOH), which would enable recording and observation of data over a long period The SOC and battery voltage are only measured in real-time

3) The test bench primarily focuses on the constant current (CC) charging stage The constant voltage (CV) stage has not been evaluated Further research and evaluation of the CV charging stage could provide valuable insights into the overall charging performance

4) The coordination of battery power control, PV power, and grid power has not been investigated in the test bench This issue is important as it determines how these different power sources interact and complement each other during charging and discharging Further research in this concern would be beneficial to optimize the overall performance and efficiency of the charging station, taking advantage of the combined power sources effectively

However, because the test bench is set up as a pilot prototype to verify the feasibility of optimal charging algorithms in practical conditions, it meets the goal of the research Other limitations mentioned above will be further investigated in the future test bench development

This chapter is mainly based on:

[1] Ngoc Van Nguyen and Huu Duc Nguyen (2022) A validation of real-time responses following to long-term charging schedule of PV-integrated electric two-wheeler charging stations Journal of Science and Technology - HaUI, 58(5),

[2] Ngoc Van Nguyen and Huu Duc Nguyen (2023) Empirical verification of real- time charging responses following long-term scheduling for electric two-wheeler charging stations TNU Journal of Science and Technology (Accepted)

The dissertation implements research on optimal solutions for photovoltaic integrated charging stations in the context of Vietnam The main research outcomes of the thesis can briefly be summarized as follows:

1) Centralized architecture is suiable for E2W charging stations In the case of larger scale, a hierarchical architecture which coordinates mutilple centralized charging stations may be adopted

2) Real-time and long-term models are considered in the study with the former aiming at investigating real-time charging responses while the latter aiming at integrating optimal charging algorithms

3) In the dissertation, two scheduling algorithms aiming at improving total load profile are proposed and verified The algorithms can accommodate high numbers of E2Ws and handling dynamics and uncertainties

4) Simulation results reveal that attention should be paid to charging stations depending on specific locations

5) In the study, a testing workbench has been set up for empirical verification Quantitatively, this work has addressed the following issues: 1) it has clarified the need for PV-integrated E2W charging stations as well as operational solutions of charging stations to promote sustainable development in the transportation and energy sectors in Vietnam, 2) proposed scheduling algorithms with objective of load leveling, valley filling, and peak shaving In the thesis, two scheduling algorithms have been proposed and verified: the charging power allocation algorithm and the optimal charging algorithm based on the receding horizon framework The power allocation algorithm can schedule a large number of E2Ws without solving constrained OPs for each E2W The optimal charging algorithm based on the receding horizon framework employs a group-based approach, optimization tools and the receding horizon framework to generate optimal charging plans The framework dynamically considers arrival/departure behaviors as well as energy level variations of E2Ws at timeslots The effectiveness of the algorithms is compared with average charging scheme, uncontrolled charging scheme, and quantitatively assessed through load variance 3) The thesis has empirically assessed the real-time responses of E2W charging following long-term charging schedules The experimental measurements have demonstrated that long-term charging plans are feasible in practical implementation

As an extension of this thesis, several tasks for future work are possible and listed below:

1) Exploring additional uncertainties that could impact on the algorithms These uncertainties could be technical specifications variations of E2Ws and the accuracy of forecasted PV power generation and non-EV loads Further addressing the uncertainty in charging requirements of vehicle users

2) Addressing other issues such as: island mode operation, optimal power dispatch of power sources in the station; optimal PV sizing

3) Further quantitatively evaluating the computational time of the proposed algorithms with a certain harware infrastructure

4) Further investigating detailed design of converters as well as sizing components quantitatively MPPT algorithms for small-scale rooftop PV systems should also be in detail

5) Further developing the test bench and conducting experiments under various scenarios

[1] Ngoc Van Nguyen and Huu Duc Nguyen (2022) A validation of real-time responses following to long-term charging schedule of PV-integrated electric two-wheeler charging stations Journal of Science and Technology - HaUI,

[2] Ngoc Van Nguyen, Khac Nhan Dam, and Huu Duc Nguyen (2022) Research on the architectures and control algorithms for electric vehicle charging stations and electric two-wheeler charging stations in the context of Vietnam Journal of Science and Technology - HaUI, 58(4), 55–64

[3] Huu D.N and Ngoc V.N (2022) A Three-Stage of Charging Power Allocation for Electric Two-Wheeler Charging Stations IEEE Access, 10, 61080–61093

[4] Ngoc Van Nguyen and Huu Duc Nguyen (2022) PV-Integrated Electric Two- wheeler Charging Stations: A Solution towards Green Cities TNU Journal of

[5] Heckmann W., Duc N.H., Granford Ruiz D et al (2022) Smart Energy Buildings: PV Integration and Grid Sensitivity for the Case of Vietnam

Proceedings of the 11th International Conference on Smart Cities and Green ICT Systems, SCITEPRESS - Science and Technology Publications, 117–124

[6] Huu D.N and Ngoc V.N (2021) A Two-Level Desired Load Profile Tracking Algorithm for Electric Two-Wheeler Charging Stations Eng Technol Appl Sci

[7] Huu D.N and Ngoc V.N (2021) Analysis Study of Current Transportation Status in Vietnam’s Urban Traffic and the Transition to Electric Two-Wheelers Mobility Sustainability, 13(10), 5577