SOC Estimation For Lithium Ion Batteries Based On Recurrent Neural Network

With the rapid development of new energy vehicles around the world,battery-related technologies have also become the focus of research.The State of Charge(SOC)is one of the most important parameters of the battery and represents the remaining battery capacity.The accurate estimation of the battery SOC can avoid the excessive charging and discharging of the battery and ensure the healthy use of the battery,which is of great significance.This paper takes power battery as the object,and deeply studies the SOC estimation of lithium battery under different temperatures and working conditions.The specific work is as follows:The definition and research status of SOC are introduced.The structure and principle of lithium battery are analyzed.The charge and discharge characteristics of lithium battery and several common factors affecting SOC are explored.The temperature,discharge rate and internal resistance of battery and SOC are studied.This paper finally takes the discharge temperature of the battery and the discharge rate of the battery as factors affecting the SOC of the lithium battery.The NARX neural network is used as the model of SOC estimation,and the NARX network structure is designed.The NARX network is built for different temperatures.In the selection of data sets,taking into account the impact of battery discharge rate on SOC,the data of driving cycles is used as the training set and test set of the network,including DST,FUDS and US06,which makes the data closer to the actual use of the battery.The SOC estimation of the lithium battery is carried out by the cyclic neural network GRU,and the battery discharge voltage,current and temperature are used as the input data of the network.Compared with the NARX network,the model only needs to train the model once to estimate SOC on the driving cycles data at different temperatures.SOC estimation can be performed on the battery charging process at the same time.The results prove that the GRU is effective in estimating the SOC of the power battery,but at the same time,the prediction of the network also has the problem of insufficient stability.Finally,The problems existing in the GRU network were improved.The UKF was used to improve the GRU network,and the GRU-UKF model was proposed.The improved model is used to estimate the SOC of the test set data.The results show that the performance of GRU-UKF is significantly improved compared with the GRU network,and the stability and accuracy of the network are greatly improved.