Research On SOC Estimation Of Power Battery Based On Improved Neural Network

SOC(State of Charge)is a core parameter in the BMS(Battery Management System),the accuracy of its estimation will directly affect the performance of the BMS,affecting the driver's grasp of the battery status and driving experience.However,the current algorithm for SOC is yet not mature,and there are still have improvements in practicability and applicability.In order to solve the problem of SOC estimation accuracy,this paper based on neural network to estimate SOC,and according to the characteristics of the battery under static and dynamic conditions,the neural network has been optimized to improve the accuracy of SOC estimation.The main work is as follows:Firstly,this paper briefly introduces the current development status of electric vehicles,then expounds the research status of BMS system and SOC at home and abroad,and then analyzes the common estimation methods of SOC and compares its advantages and disadvantages.Relevant battery experiments were designed.The temperature characteristics,cycle characteristics and multiplier characteristics of the battery were deeply studied by combining the experiments.The effects of battery multiplier,temperature and open-circuit voltage on SOC of the battery were analyzed.Secondly,this paper based on the characteristics of SOC,designed an estimation model use BP neural network.After comprehensively analyzing the influence of each variable on SOC,three variables of voltage,current and temperature are selected as the network input,and a three-layer network structure model is designed.The training samples were normalized and randomly cross-ranked,and the network was trained and tested by gradient descent learning method.Then,the 1C discharge data at 25 °C was used to verify the experiment under static offline environment.Then,aiming at the problem that BP neural network is not accurate in estimating SOC,a method of using immune genetic algorithm to optimize BP neural network is proposed.The method optimizes the weight and threshold of BP neural network by using the fast convergence and global optimal characteristics of immune genetic algorithm.Through the memory unit in immune genetic algorithm,it can form the dynamic feedback effect during network training,accelerate network training and improve network learning efficiency.Finally,CFA-WNN estimation algorithm based on chaotic firefly optimization wavelet neural network is proposed for the dynamic change of battery parameters under real dynamic conditions of automobiles.This method searches for the optimal solution by simulating the foraging behavior of fireflies,and further optimizes the optimization process by mapping chaotic sequences.In addition,it is also cited in the wavelet neural network.The momentum term gradient descent method is introduced to improve the learning efficiency of the network.Then,a SOC network estimation model is designed for actual operating conditions,and the effect is verified on the experimental platform by using the battery operating conditions data.