Research On Wide Temperature Soc Estimation For Lithium-ion Batteries Based On Gated Recurrent Unit Neural Network

In recent years,the electric vehicle industry has developed vigorously.The battery energy management system(BMS),the core function of electric vehicles,is particularly important for the development of electric vehicles,and the battery state of charge(SOC)estimation function is even more important.With the rapid development of deep learning related technologies,a large amount of data generated during batteries use can be combined with it for SOC estimation.At present,the SOC estimation methods used in the industry are complex,and the estimation accuracy is uneven and the applicability is poor.At the same time,the power batteries used in electric vehicles have a large temperature-affected characteristic change,which will also affect the SOC estimation accuracy.Therefore,there is an urgent need for a high-precision SOC estimation algorithm that can be used in a wide temperature range in engineering.In view of the above situation,this paper designs a pure data-driven deep learning algorithm—Recurrent Neural Network with Gated Recurrent Unit(GRU-RNN)to estimate the SOC of lithium-ion batteries at a wide temperature.Because the deep learning network requires a lot of data,this topic uses the battery electrochemical model to obtain battery characteristic data.The specific work is as follows:Firstly,perform temperature correction of the electrochemical model of the lithium-ion batteries.The improved single-particle model(SP+ model)of the existing lithium-ion batteries is inaccurate at high and low temperatures,and therefore needs to be corrected at different temperatures.Relevant parameters are corrected for the solid-liquid phase diffusion and concentration polarization process of the battery.After the correction,a temperature-corrected improved single-particle model(TC-SP+ model)is obtained.The model will have high accuracy and can simulate battery characteristics to produce battery data that are basically close to the true value at different temperatures.Secondly,design a wide-temperature SOC estimation method for lithium-ion batteries based on GRU-RNN.Set the battery characteristic variables(temperature,voltage,current)as network input and SOC as network output,and establish a basic algorithm framework.The large amount of data produced by the above-mentioned TC-SP+ model is used for algorithm training and testing,and the network structure is gradually optimized as the battery working conditions become more complicated.Finally,verify the effectiveness of the SOC estimation method proposed in this topic.Use measured battery data at different temperatures for network validity verification to verify the performance of the method in a wide temperature range;perform model-based Adaptive Extended Kalman Filter(AEKF)SOC estimation,and compare with GRU-RNN-based SOC estimation method.Comprehensively measure the performance of the SOC estimation method designed in this subject.