The power split of different types of energy storage systems within an EV can be concisely described as follows. Considering the block diagram of Figure 2.2, the contribution of power to meet a particular load requirement is split between two energy storage types. W1 and W2 represent the weighing factors corresponding to the proportion of energy extracted from the two storage units. Due to the difference in Power to Energy ratios of Type 1 and Type 2 systems, a strategy to coordinate power flow by dynamically varying the weighting factors is required. For successful operation of the vehicle, the power availability must at least meet the power requirement. This has to be done with further consideration to the system constraints, for example the depletion level of the energy storage units. Figure 2.3 illustrates a typical power split of Type 1 and Type 2 Energy Storage Systems (ESS) to fulfil the load demands.
Energy Storage System. Type 1
Energy Storage System. Type 2
W1
W2
Σ LOAD
Figure 2.2 Power Split between two energy sources
State of ChargeState of Charge t
t
Load Demand
Power from Type 1 ESS
Power from Type 2 ESS
Energy level of Type 1 ESS
Energy level of Type 2 ESS Power & Energy Management
Time
Time
Time
Power PowerPower
Figure 2.3 Power split and energy expenditure between two energy sources
This vexing issue of controlling the power flow of two or more sources has been addressed through various approaches. Jalil, Kher and Salman [19] suggested a rule-based framework for power split between a battery pack and an internal combustion engine. The proposed strategy ensured that both power sources operate at maximum efficiency whenever possible.
The concept demonstrated an increase in efficiency in terms of fuel economy. Recognising that the battery energy expenditure as well as the system power split requires a controlled intervention, Caratozzolo, Sera and Riera [20] also suggested an energy management strategy derived from a heuristically composed rule-base. Due to the highly non-linear nature of EV and HEV drivelines, the authors suggested a rule-base approach to provide an employable scheme for arbitration of power flow under various operating modes of the vehicle.
Steinmauer and Del Rel [21] stated that techniques that use a fixed controller structure and then searches for optimal parameters to minimise a cost function yields only a solution that is a consequence of the selected structure. They proposed to tackle the dual source power split problem in terms of optimal control using statistical data of vehicle power demands for known drive cycles. Their procedure addressed the problem by deriving optimal solutions for a fixed set point, which was then extrapolated to various power demand profiles. The authors demonstrated optimal power split between a battery and generator. The analysis
showed that the battery State of Charge (SoC) at the beginning of the drive cycle equalled the SoC at the end of the cycle. However, the negative effects of rapid deep charge and discharge cycles imposed on the batteries were not considered.
According to Langari and Won [22], optimal control methods, due to its dependency on the drive cycles used to generate the control actions may not yield optimal power split for misclassified or arbitrary drive cycles. As an alternative, they proposed a concept of a fuzzy logic (FL) based energy management to capture driving situational awareness. Details of their study can be found in Won’s Ph.D dissertation [23]. Similarly, Hellgren and Jonasson [24] conducted a comparison of a fuzzy logic approach and an analytical formula for a hybrid powertrain. Their findings showed that the FL method proved more flexible but required three times as many design variables.
The DC-Link voltage control method suggested by Lohner and Evers [25] uses a voltage reference as the power management control parameter. Given that multiple power delivery systems share a common DC-link in the vehicle power system architecture, the principle behind this method is to regulate the DC link voltage within a tolerance band around a set point reference voltage. Using band pass filters and proportional-integral (PI-type) loops to control the current drawn and delivered to several energy storage systems, the authors showed that the DC-link voltage control method limits the DC-link voltage dips that occur during vehicle acceleration and the voltage rises that occur during decelerations. In effect, the technique indirectly arbitrates the power sharing of several electrical power delivery systems.
West, Bingham and Schofield [26] introduced a Model Predictive Control (MPC) method to coordinate the power flow from two sources in a pure electric vehicle. Employing a constrained MPC with zone control, they demonstrated that the net energy expenditure of a battery bank in a battery-ultracapacitor system was significantly less compared to a DC-Link voltage control method. Also along the lines of predictive control, but for a HEV application, Salman, Chang and Chen [27] proposed a theoretical framework for a predictive energy management strategy. Although termed as ‘predictive’, the strategy still depends on
previewed information about the mission profile. However, the leaning energy management strategy [28], also by Chen and Salman, lends itself more of an implementable method.
Leading more towards the practical implementation of power split strategies, which require instantaneous management of power flow, Paganelli et al. [29] introduced a general supervisory control policy. Although the formulation of the policy was intended for charge- sustaining HEVs, the proposed power split algorithm is generic and may be adapted to pure electric vehicles with more than one energy storage type.
Moreno et al.[30] reported valuable experimental results for a test vehicle that incorporated optimal control methods with an artificial neutral network (ANN). The ANN was trained offline for a set of driving cycles followed by a series of field-testing. Compared to a fixed strategy to regulate the ultracapacitor SoC, the ANN strategy was reported to yield a 4.9%
theoretical improvement in efficiency (km/kWh) when simulated and a 3.3% improvement during field-testing.
Also using ANNs, Papadimitropoulos et al. [31] evaluated their energy management concept on a test vehicle developed at the University of Patras. Their test vehicle, (the E-240) followed an energy management strategy to trace a maximum motor efficiency map regardless of the arbitrary driving patterns. The authors used a trained ANN to predict the battery state of charge and the motor temperature, which was then computed for maximum efficiency determination. In conclusion of their work, the authors commented that although energy economy of electric vehicles can be achieved by using more efficient energy storage systems, an energy management system could provide significant efficiency gains instead.
A strategy that uses knowledge of subsystem efficiency maps and then computes a reference power split following a minimisation function was proposed by Pisu and Rizzoni [32]. Based on a concept of Equivalent Consumption Minimization Strategy (ECMS), this generic strategy addresses the energy optimisation problem of multiple energy sources by replacing the global criteria of energy expenditure with a local criterion. The authors also drew attention to the fact that energy optimisation strategies that require priory information about
by Guzzella and Sciarretta [18] for sub-optimal but implementable techniques due to the causal control nature of the method. In addition, the authors of [18] demonstrated that non- causal methods that strongly depend upon the precision of future power profile can lead to an energy management strategy that causes excessive deviation to energy storage system target state of charge.
Exploring several energy management strategies, Koot et at. [33] demonstrated that the general concept of energy management is warranted since even the most basic of strategies yields a reduction in net energy usage. For a fixed vehicle drive profile and subsystem architecture, the authors of [33] evaluated five energy management strategies. Since the outcome of their work also concurred that implementable strategies do not have the drive profile horizon as priory knowledge, they suggested a dynamic programming approach that uses a short horizon length rather than the complete driving cycle. Although dissimilar in implementation method, the strategy bares fundamental similarities to the ECMS proposed by Pisu and Rizzoni [32], which replaces a global criteria of energy expenditure with a local criterion.
Recognising the stochastic nature of the energy management problem, Lin, Peng and Grizzle [34] proposed a strategy using stochastic dynamic programming (SDP). Representing the vehicle power demand as transition probabilities over an unknown mission profile, the authors formulated the power split decision rules as a time-invariant infinite horizon SDP problem. Although the method was intended for a HEV application, the technique is transferable to EVs. The SDP technique was also examined by Min et.al [35]. Modelling the vehicle driver power demand as a Markov chain, the authors of [35] developed a strategy to split power delivery between a fuel cell and battery system. By constructing a transition probability function based on several driving scenarios, the SDP method was used to map the observed states to the control of power split decisions.
In a recent publication, Cacciatori et al.[36] provided a basic classification of energy management strategies. The authors categorised energy management strategies into two groups. Strategies that require a priori knowledge about the mission profile and those that have no or limited knowledge in that regard. For the first group, three approaches are
generally used. They are, optimal control theory, dynamic programming in which the control problem is recast into a multi criterion decision process and solved using Bellman’s principal of optimality and a third approach that uses an optimal design technique. The second group of energy management strategies can be employed for real-time control. The general approaches used in this group are; heuristic rule-base control, fuzzy logic inference engines and cost-based suboptimal control. Similar to the ECMS for charge sustaining systems put forward by Pisu and Rizzoni [32] and also demonstrated Guzzella and Sciarretta [18], cost- based suboptimal techniques are based on representing the energy consumption as a cost function which is minimised in a very short period and can be implemented in real time controllers.
Miller et al.[37] suggested a method to determine power split ratios between batteries, ultracapacitors and ICE by means of power spectral decomposition and frequency banding.
Using discrete wavelet transforms (DWT), the power splits are discerned simultaneously in time and frequency by utilising the DWT adaptive windowing characteristic. Decomposing the power spectrum into designated low, mid and high frequency bands correspondingly determine the power splits between the ICE, battery pack and ultracapacitors. A similar wavelet-based load sharing algorithm was later adopted by Uzunoglu and Alam [38] to determine the power split between fuel cells and ultracapacitors for a HEV application.
Gielniak and Shen [39] provided a very different perspective to the vehicular power management problem in suggesting a power split strategy based on game theory. Classifying the power sources and the load demands as game ‘players’, the authors explored the possibilities of adopting game theory to achieve high efficiency and performance payoffs, where the payoffs are represented by utility functions of each of the power sources. The general concept behind this approach is to assume that the energy systems are one set of players in a game and alters its strategy in order to place itself in a state that yields a high utility. The load demands, with its non-stationary fluctuating power demand and transients can be seen as the opposing player or the adversary in a two-player game.
Evidently there are various methods and approaches to manage multiple energy storage units in a vehicular propulsion system. Variations in approaches and methods provide interesting
insights to the problem description. As discussed, the problem is sometimes addressed solely as a ‘power’ management issue and sometimes as a topic of ‘energy’ management. Both problem descriptors are valid since energy is simply the time integral of power. However, when multiple energy storage systems that have very different specific power (kW/kg) to specific energy (kWh/kg) ratios and also different peak power handling capabilities are combined, the problem is best addressed jointly as power and energy management issue. The problem of designing a complete power and energy management system could be stated as a problem encompassing energy resource planning, power delivery and an effective architecture design for a real-time system
On a rather theoretical level, several researchers have developed energy management and power management techniques that apply priory information regarding the vehicle propulsion power demands. These methods do provide a means to identify the maximum obtainable improvements in terms of energy efficiency and performance benefits. The findings also clearly support the grounds for further research in this area. However, in spite of significant contributions, there have not been many attempts to address the complete implementation process of a working system.