Research On The Prediction Of Lithium Battery SOC Based On Improved Artificial Neural Network Model

China has recently become the fifth country in the world to announce the phase-out of conventional vehicles for fossil fuels.China has vigorously promoted the adoption of electric vehicles and plans to use electric vehicles as a truly sustainable mode of transportation.The battery is one of the key technologies for electric vehicles,and the prediction of the state-of-charge(SOC)is very important for the battery.First of all,this paper confirms the research content of this paper through the research and analysis of lithium battery SOC prediction.Then it outlines the working principle of lithium-ion batteries and the factors that affect their SOC.This paper proposes two prediction models based on artificial neural network: one is the introduction of the traditional artificial life model of cellular automata in artificial neural networks,and a combination of the two methods to construct a cellular neural network model.One model is to combine the NEAT algorithm with BP neural network,and then introduce the Quantum-behaved Particle Swarm Optimization(QPSO)algorithm in the artificial life algorithm to optimize the NEAT-BP model,and then obtain an improved artificial neural network model of QPSOBP-NEAT.In order to verify whether the model presented in this paper is effective,the simulation model data experiment and the actual discharge experiment were carried out respectively to test the validity and reliability of the model.In the simulation model experiment,three different working conditions of UDDS,10-15 and FTP75 were set using the ADVSOR and lithium-ion battery models to obtain the voltage,current,and SOC values under the corresponding operating conditions.Finally,five different types were obtained.The data in the case of rate discharge is normalized to the sample data and then the data is used to train the proposed model.The predicted results under various conditions indicate that the improved artificial neural network model has low learning efficiency because it has no learning algorithm,while the QPSOBP-NEAT model has higher efficiency and the highest average accuracy is as high as 98.7%.It is predicted that in a single cell measurement data experiment,the discharge voltage,discharge current,and SOC of a certain type of lithium-ion battery at different magnifications are collected and then normalized,and then the model is imported for corresponding training.The highest average prediction accuracy is obtained.Up to 98%,the results show that the method in the practical application of the feasibility and accuracy.