Research On Power Battery State Of Charge Estimation Algorithms

Recent oil and energy crisis has led the world to think for an alternative source of energy that is neither harmful to the environment nor effects the ozone layer.This need for alternative energy has driven the individuals from academia and industry,government institutes,automotive industries and researchers from around the world to invest and develop vehicles with efficient energy systems.Hybrid electric vehicles(HEVs)and battery powered Electric Vehicles(EVs)are developed as the most promising solution for an alternative source of energy.One of the major challenges faced by current EV industry is the overall driving range that is much lesser compared to the internal combustion engine vehicles.Adding to the problem of overall driving range is a lack of battery management system that can estimate and predict the actual remaining power of a battery,i.e.,to predict the residual driving range.Therefore preventing EVs to run out of charge or leaving the passengers stranded is the main concern.In order to predict the residual range of the electric vehicle one of the parameters directly involved in its calculation is state of charge(SOC).State of charge estimation is important as it plays a critical role in the operation of an electric vehicle power battery.This thesis focuses on filtering techniques and data driven based machine learning algorithms for state of charge estimation of a lithium ion battery for an electric vehicle.As a prerequisite step for filtering techniques,battery model has a particular importance to guide the state estimation algorithms.Electric equivalent circuit model was established and Thevenin battery model parameters were identified using the least square method.?3470160‘LiFePO4 battery cells in combination with Advanced Vehicle Simulator(ADVISOR)were used to collect and validate the models and SOC estimation strategies.Three different estimation algorithms,i.e.,Extended Kalman filter(EKF),Sigma Point Unscented Kalman filter(SPUKF)and the application of Sequential Monte Carlo,i.e.Particle filter(SMOPF)is proposed.The first part of the thesis considers comparison studies based on the least square parameter identification method for state of charge estimation of a LiFePO4 battery pack using three model-based algorithms for electric vehicles.The main outcome of these studies is comparing the performance of different model based algorithms while using different open circuit voltage-state of charge curves in the presence of Gaussian noise.Later,a novel SOC estimation method based on Sequential Monte Carlo based particle filter is introduced to deal with different types of noises and numerical complexity.The experimental result shows that the presented SPUKF and SMOPF developed in this research are more robust,efficient and works better when an initialization error is introduced.However,to overcome the deficiencies in the filtering based state of charge algorithms,a novel framework for data driven SOC estimation using subtractive clustering based neuro-fuzzy method(SC-ANFIS)under diversified driving cycles for battery state estimation is proposed.The proposed framework contains data processing,SC-ANFIS training and testing step in complying with the estimation accuracy and performance of SC-ANFIS based model while applying it on ten different unseen drive cycles proving the effectiveness of the proposed algorithm.In comparison with other advanced algorithms,including back propagation neural network and Elman neural network,our proposed algorithm yielded better SOC estimation results and predicted the SOC accurately with maximum absolute estimation error lower than 0.1% for all ten cycles.In addition,results of sensitivity analysis indicated that battery module temperature and heat removed from the battery are the most important parameters in modeling the SOC estimation by SC-ANFIS model.The rest of the thesis focuses on promoting the practical usage of state of charge estimation models by applying the proposed models to the data set acquired during the data acquisition process to include the effects of noise,different combinations of open circuit voltage and state of charge curves,state estimation algorithms behavior under varying conditions using diversified drive cycles.