Ứng dụng trí tuệ nhân tạo trong việc dự đoán lũ lụt ( English version) The artificial neural network (ANN) has been applied in many hydrological models in recent years and pays attention thanks to the performance of the model. This report focuses on using the application of the ANN based on artificial intelligence, to estimate floods in Australia. This report also presents the principle of the operation of the artificial neural network model as well as its prominent features. Comparing the performance of the artificial neural network (ANN) model with a traditional model indicates that using the ANN model in flood estimation results in a better performance.
Report title The application of artificial intelligence in estimating floods in Australia Author full name: Thi Hanh Vu Student ID number: 1730464 Date of submission: 22/6/2017 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] TABLE OF CONTENTS List of Illustrations………………………………………………………………… ii Glossary…………………………………………………………………………… iii Abstract……………………………………………………………………………….v Introduction………………………………………………………………………1 The Artificial Neural Network and its application…………………………….3 2.1 Overview of the artificial neural network…………………………………… 2.2 Model of an network………………………………………….3 2.3 artificial neural Data selection………………………………………………………………….5 2.4 Research process……………………………………………………………… Evaluation of the model’s performance……………………………………… Conclusion……………………………………………………………………… Reference List……………………………………………………………………….10 Bibliography……………………………………………………………………… 11 2 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] List of Illustrations List of figures Figure 1: Multi-layer perception………………………………………………………4 List of tables Table 1: Description of regions……………………………………………………….6 Table 2: Average error values (%) for ANN based models and QRT model…………7 Table 3: Coefficient of efficiency values for ANN based models and QRT model… 3 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] Glossary Algorithm Artificial intelligence Artificial neural network Catchment area Database Error back-propagation Flood Flood quantile Hydrology Irrigation Layer 4 n A process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer (Oxford Advanced Learner’s Dictionary 2015) The theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages (Oxford Advanced Learner’s Dictionary 2015) A computing system that is designed to simulate the way the human brain analyzes and process information (Kantardzic 2011) n The area from which rainfall flows into a river, lake, or reservoir (Oxford Advanced Learner’s Dictionary 2015) n A structured set of data held in a computer, especially one that is accessible in various ways (Oxford Advanced Learner’s Dictionary 2015) A common method of training a neural net in which the initial system output is compared to the desired output, and the system is adjusted until the difference between the two is minimized (Kantardzic 2011) n An overflowing of a large amount of water beyond its normal confines computer (Oxford Advanced Learner’s Dictionary 2015) The flood peak discharge magnitude corresponding to a specified exceedance probability The symbol used in this report is Q n The branch of science concerned with the properties of the earth's water, and especially its movement in relation to land computer (Oxford Advanced Learner’s Dictionary 2015) n The supply of water to land or crops to help growth, typically by means of channels (Oxford Advanced Learner’s Dictionary 2015) n The organisation of programming into separate functional components that interact in some sequential and hierarchical way, with each layer usually having an interface only to the layer above it and the layer [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 Multi-player perception (MLP) Neuron Non-linear Parameter Quantile regression technique (QRT) Signal Variable 5 [Type text] below it (Kantardzic 2011) A multilayer perception is a feedforward artificial neural network that generates a set of outputs from a set of inputs (Kantardzic 2011) n A nerve cell that carries information between the brain and other parts of the body(Oxford Advanced Learner’s Dictionary 2015) Used to describe a process, series of events, in which one thing does not clearly and directly follow from another (Kantardzic 2011) n A numerical or other measurable factor forming one of a set that defines a system or sets the conditions of its operation (Oxford Advanced Learner’s Dictionary 2015) QRT is a standard linear regression technique that summarises the average relationship between a set of regressors and the outcome variable based on the conditional mean function (Kantardzic 2011) n A gesture, action, or sound that is used to convey information or instructions, typically by prearrangement between the parties concerned (Oxford Advanced Learner’s Dictionary 2015) n A variable is a value that can change, depending on conditions or on information passed to the program computer (Oxford Advanced Learner’s Dictionary 2015) [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] Abstract The artificial neural network (ANN) has been applied in many hydrological models in recent years and pays attention thanks to the performance of the model This report focuses on using the application of the ANN based on artificial intelligence, to estimate floods in Australia This report also presents the principle of the operation of the artificial neural network model as well as its prominent features Comparing the performance of the artificial neural network (ANN) model with a traditional model indicates that using the ANN model in flood estimation results in a better performance 6 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 7 [Type text] [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] Introduction A flood is usually a natural disaster that results from a number of deferentially interacting factors Specifically, floods are caused not only by heavy rainfall, but also can result from a cyclone, tsunami, extremely high tide or climate change (Sharifi 2012, p.534) Floods cause a range of damage to crops and properties, sometimes threatening human lives (Aziz Rai & Rahman 2015, p.805; Dawson 2006, p.392) The diversity of terrain in Australia as well as climate change makes the evolution of floods become more complex and unpredictable Therefore, flood estimation is necessary to minimise flood damage to infrastructure and human beings as well as to offer the optimal design for drainage infrastructure, flood risk management and irrigation systems in future (Middelmann-Fernandes 2010, p.89) Various predictive models have long been used in Australia to estimate flood levels, so as to help mitigate against flooding However, with diversified hydrology, as well as the changes between catchments in different areas in Australia, traditional models such as non-linear models are not longer effective Therefore, a non-linear model, specifically the artificial neural network (ANN) model, based on the artificial intelligence theory, has recently been applied as an alternative method of estimating floods The ANN model has been used successfully in predicting many hydrological factors such as extreme rainfall, streamflow forecasting, rainfall forecasting, and water quality (Aziz, Rai & Rahman 2015, p.807; Campolo, Soldati & Andreussi 2003, p 381) 8 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] The artificial neural network (ANN), as one model of artificial intelligence theory, is a useful tool to simulate complex models in a diverse range of fields including engineering, economics and medicine However, in this report, ANN is used to focus on exploring the relationship between a range of catchment descriptors to produce the flood index (predicted estimated level of flooding) The advantage of using an artificial intelligence model is that it presents flexible model structures to the data (Aziz et al 2016, p.2) In addition, it can easily account for non-linearites between model input and output, and their complex interactions in regional flood modeling The aim of this report is to outline the potential for more widespread application of artificial intelligence to the problem of flood estimation in Australia This report shows that using the artificial neural network model (ANN), based on artificial intelligence theory in flood estimation, is more effective than traditional models The results of the experiments of ANN models in Australia have been taken from research over the past five years The Artificial Neural Network and its application 2.1 Overview of the artificial neural network based on artificial intelligence theory 9 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] The artificial neural network (ANN) based on artificial intelligence theory is an information-processing model that is based on simulated activities of the human brain An ANN includes an enormous amount of neurons that are connected with each other to process information Similar to prominent features of artificial intelligence such as learning and problem solving, the ANN can learns experiences through training, and has the ability to store learning experiences and, using this knowledge, to predict unseen data (Kantardzic 2011, p.200) Therefore, the artificial neural network is a form of artificial intelligence that can be applied in many fields It has been used in electronics, medicine as well as the military to solve problems that are complicated and require high accuracy such as automatic control, data mining or identification Another outstanding feature of ANN is that it can present the flexible model structure and be able to easily calculate non-linear models between input models and output models with flexible interactions (such as various parameters and a big database) In this study, ANN has been trained to illustrate the relationship between inputs (basin descriptions) and outputs (flood estimation index) 2.2 Model of an artificial neural network In this study, the ANN model is based on the structure of multi-layer perceptions to build an application to forecast flood levels The ANN consists of three layers of neurons: an input layer, a hidden layer and an output layer More specifically, an input layer is a set of connecting links from different inputs Each input or neuron refers to one attribute of a data pattern A hidden layer summarises or receives the input signals from the previous layer, then transmits these input signals to the next processing layer One or more hidden layers can exist in an artificial neural model An output layer 10 10 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] produces one solution for one problem The processing of the ANN is that, initially, data goes directly into the first layer, and then is transmitted through the hidden layer to pass to the output layer According to the principles of the ANN operation, the artificial neural network is trained by adjusting its connections using a technique called error back-propagation (Error back-propagation is a method to train a neural network in which the system output is adjusted to fit the desired output) In flood estimation, predicted outputs are compared with observed data and are evaluated through the standard of error If the parameters of the model are not satisfied, the external weights are adjusted The processing must be repeated many times until the evaluated criteria meet an acceptable standard to produce the relationship model Figure 1: Multi-layer perception model (Willey, 2011) 2.3 Data selection The database is definitely the most important factor in all models to produce accurate results In this report, the ANN model in flood estimation also requires two main types 11 11 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] of data: streamflow data, and climatic and catchment characteristics data However, not all databases are enough high quality to use for the data input of the model Therefore, before running the artificial neural network model, all input data must be standardised to a pre-determined level (Bowden, Maier, & Dandy 2005, p 97; Maier & Dandy 2000, p 103) Poor data quality will be excluded from a set of predictor variables of the model because it can have a negative influence on calculations to produce accurate results of the model, as well affect the performance of the model 2.4 Research process The operation of the model is carried out using the following steps The first step is finding the set of predictor variables and then, to evaluate the quality of data, choosing the best set of data to forecast a flood model Next, the artificial neural network model and related algorithms are used to train the model to make output predictions During the period of training and processing for the model, the variables are adjusted This process is repeated until the evaluation criteria are met to give a final forecast result known as the flood estimation index After calculating the model’s result, the last step is to evaluate the model’s performance Evaluation of the model’s performance To evaluate the performance of the ANN model in flood estimation, the different results, using the divergent techniques of the two models (the ANN model and 12 12 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] Quantile Regression Technique, which is a traditional model to forecast floods) were used in forecasting floods in the research by Aziz et al (2014) (see Table 1) They used an extensive Australian database, which included information concerning 542 catchments in four states: New South Wales, Victoria, Queensland and Tasmania Table 1: Description of regions (From Aziz et al 2014, p 548) Aziz et al (2014) built and developed an ANN model to estimate floods from sets of databases, being Flood Quantiles 2, 5, 10, 20, 25, 50, and 100 years (Q 1, Q2, Q5, Q10, Q20, Q25, Q50, Q100) Average Recurrence Intervals (ARI) that shows the possibility of flood events (see Tables and 3) A Flood Quantile (Q) is the discharge of the flood peak corresponding probability exceeds the specified level Table 2: Average error values (%) for ANN-based models and QRT models 13 13 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] (From Aziz et al 2014, p 551) In terms of average error values (Table 2), the results show that when using ANN models, average error values are always lower than when using a QRT model For Flood Quantile 2, (Q2), the average result of error values in regions using ANN models was 37.56 %, while when using QRT the error values was 65.38 % This illustrates how much more efficient the ANN model can be Table 3: Coefficient of efficiency values for ANN based models and QRT model (From Aziz et al 2014, p 551) 14 14 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] Regarding coefficient of efficiency values (Table 3), the results when using the ANN model are almost twice as effective as the QRT model For Q2, the efficiency value for the ANN model was 0.73, while 0.35 was the obtained result when using the QRT model The worst result was found in case Q50 using the QRT model with the coefficient of -8.42 compared with 0.68 of the same coefficient when using the ANN model From the results of average error values and the coefficient of efficiency values (Tables and 3), it can be seen that using the ANN model is more effective and accurate than using the QRT model in flood estimation 15 15 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] Conclusion This report demonstrates the application of the Artificial Neural Network model based on artificial intelligence theory to estimate flood levels in Australia Applying the ANN model to estimate floods is more effective than the traditional model (QRT), because of the strong, more accurate features of the ANN model Moreover, the ANN model has the capability of forecasting floods with relative accuracy just relying on hydrological data without needing details of the geological terrain Finally, changing climate conditions can influence variables in flood estimation, so the Artificial Neural Network will be the most accurate model for the future 16 16 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] Reference list Aziz, K, Haque, MM, Rahman, A, Shamseldin, AY, and Shoaib, M 2016, ‘Flood estimation in ungauged catchments: application of artificial intelligence based methods for Eastern Australia’, Stochastic Environmental Research and Risk Assessment, vol 1, no.1, pp 1-16 Aziz, K, Rahman, A, Fang, G, Shrestha, S 2014, ‘Application of artificial neural networks in regional flood frequency analysis: a case study for Australia’, Stochastic Environmental Research and Risk Assessment, vol 28, no 3, pp 541-554 Aziz, K, Rai, S, Rahman, A 2015, ‘Design flood estimation in ungauged catchments using genetic algorithm-based artificial neural network (GAANN) technique for Australia’, Natural Hazards, vol 77, no 2, pp 805–821 Bowden, GJ, Maier, HR, Dandy, GC (2005), ‘Input determination for neural network models in water resources applications, Parts Case study: forecasting salinity in a river’, Journal of Hydrological, vol 301, pp 93-107 Campolo, M, Soldati, A, Andreussi, P 2003, ‘Artificial neural network approach to flood forecasting in the River Arno’, Hydrological Sciences Journal, vol 48, no 3, pp 381-398 Dawson, CW, Abrahart, RJ, Shamseldin, AY, Wilby, RL (2006) ‘Flood estimation at ungauged sites using artificial neural networks’ Journal of Hydrology, vol 319, no 1, pp 391 – 409 Kantardzic, M (2011) Data Mining Concepts, Models, Methods, and Algorithms Hoboken, Wiley Maier, HR, Dandy, GC 2000, ‘Neural networks for the prediction and forecasting of waster resources variables: a review of modeling issues and applications’, Environmental Modeling and Software, vol 15, no.1, pp 101-124 Middelmann-Fernandes, MH 2010, ‘Flood damage estimation beyond stage-damage functions: an Australian example’, Journal of Flood Risk Management, vol 3, no 1, pp 88-96 Oxford Advanced Learner’s Dictionary, 9th edn, 2015, Oxford University Press UK, Hornby, UK Sharifi, F, Samadi, SZ, Wilson, CAME 2012, ‘Causes and consequences of recent floods in the Golestan catchments and Caspian Sea regions of Iran’, Natural Hazards, vol 61, no 2, pp 533-550 17 17 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] Bibliography Abbot & Marohasy 2012, ‘Application of artificial neural networks to rainfall forecasting in Queensland, Australia’ Advances in Atmospheric Sciences, vol 29, no.4, pp 717-730 Ahmad, Kamruzzaman, and Habibi, 2012, ‘Application of artificial intelligence to improve the quality of service in computer networks’, Neural Computing and Applications, vol 21, no 1, pp 81-90 Aziz, K, Haque, MM, Rahman, A, Shamseldin, AY, & Shoaib, M 2016, ‘Flood estimation in ungauged catchments: application of artificial intelligence based methods for Eastern Australia’, Stochastic Environmental Research and Risk Assessment, vol 1, no.1, pp 1-16 Aziz, K, Rai, S, Rahman, A 2015, ‘Design flood estimation in ungauged catchments using genetic algorithm-based artificial neural network (GAANN) technique for Australia’, Natural Hazards, vol 77, no 2, pp 805-821 Aziz, K, Rahman, A, Fang, G, Shrestha, S 2014, ‘Application of artificial neural networks in regional flood frequency analysis: a case study for Australia’, Stochastic Environmental Research and Risk Assessment, vol 28, no 3, pp 541-554 Babovic, V, Keijzer, M 2000, ‘Genetic programming as a model induction engine’, Journal of Hydroinformatics, vol 2, no 1, pp 35 -60 Bowden, GJ, Maier, HR, Dandy, GC (2005), ‘Input determination for neural network models in water resources applications, Parts Case study: forecasting salinity in a river’, Journal of Hydrological, vol 301, pp 93-107 Campolo, M, Soldati, A, Andreussi, P 2003, ‘Artificial neural network approach to flood forecasting in the River Arno’, Hydrological Sciences Journal, vol 48, no 3, pp 381-398 Dawson, CW, Abrahart, RJ, Shamseldin, AY, Wilby, RL 2006 ‘Flood estimation at ungauged sites using artificial neural networks’ Journal of Hydrology, vol 319, no 1, pp 391 – 409 Flavell, D 2012, ‘Design flood estimation in Western Australia’, Australian Journal of Water Resources, vol 16, no 1, pp 1-20 Fleming, Sean W, Bourdin, Dominique R, Campbell, Dave, Stull, Roland B, Gardner, Tobi 2015 ‘Development and Operational Testing of a Super‐Ensemble Artificial 18 18 [Type text] [Type text] Student’s name: Thi Hanh VU (Hana) ID: 1730464 PEP 17 [Type text] Intelligence Flood‐Forecast Model for a Pacific Northwest River’, Journal of the American Water Resources Association, vol 51, no 2, pp 502 – 512 Kantardzic, M (2011) Data Mining Concepts, Models, Methods, and Algorithms Hoboken, Wiley Maier, HR, Dandy, GC 2000, ‘Neural networks for the prediction and forecasting of waster resources variables: a review of modeling issues and applications’, Environmental Modeling and Software, vol 15, no 1, pp 101-124 Middelmann-Fernandes, MH 2010, ‘Flood damage estimation beyond stage-damage functions: an Australian example’, Journal of Flood Risk Management, vol 3, no 1, pp 88-96 Oxford Advanced Learner’s Dictionary, 9th edn, 2015, Oxford University Press UK, Hornby, UK Sayers, W, Savic, D, Kapelan, Z, Kellagher, R 2014, ‘Artificial Intelligence Techniques for Flood Risk Management in Urban Environments’, Procedia Engineering, vol 70, no 1, pp 1505 – 1512 19 19