Joint Estimation Of SOC And SOP For Lithium Battery Based On Long And Short Term Memory Networks

With people’s increasing consideration to the ecological environment,the development speed of the new energy industry is also gradually increasing,and the use of power lithiumion battery is growing rapidly and is diffusely used in miscellaneous fields.Its precise battery modeling and state prediction can undertake the responsible start-up and steady work of the BMS.SOC and SOP are two important parameters of lithium-ion batteries,and precise estimation of State is the crux precondition for competent control of lithium-ion batteries.It is hard to accurately estimate the state of charge and peak power of power batteries under different conditions,the following studies are carried out.(1)Based on the inner reaction theory of lithium batteries,the working characteristics of power lithium-ion batteries are analyzed through relevant experiments.The working characteristics of power batteries are tested at different temperatures and times of cyclic discharge,and the vital characteristics of battery parameters under different test conditions are summarized.According to the test data,the vital characteristics of the battery are obtained by analyzing and summarizing,and the discrete equation model is used.Different parameter identification methods are compared.Finally,the FFRLS algorithm is used to come true the parameter identification.The maximum voltage error of the discrete equation model is 17.39 m V in the experiment under complex conditions.(2)Aiming at the conundrum of poor stability of neural network prediction,long and short term memory network improved by statistical regression model is introduced to improve the prediction accuracy of lithium-ion battery state of charge.By comparing different intelligent algorithms,the long and short term memory network which is more suitable for SOC prediction is selected,based on which statistical regression model is used to improve it.Moreover,combined with multi-scale conditions,it can maintain high accuracy under different temperatures and aging conditions,which improves the robustness of the system.(3)Aiming at the conundrum of lithium-ion battery state cooperative estimation,a joint estimation system of lithium-ion battery SOC and SOP based on LSTM and discrete equation model is structured.Because of the completion of SOC prediction for lithium battery,a peak power estimation model given the effect of SOC is constructed by considering the battery characteristics and terminal voltage,and the predicted value of SOP at the current moment is taken as the input of SOC prediction at the next moment.The collaborative estimation of SOC state and peak power based on LSTM network is realized.(4)On behalf of check on the capability of the improved algorithm under different environments,the predicted SOC and SOP are verified through dynamic stress testing conditions of Beijing buses under different temperatures and aging conditions.The maximum root-mean-square error of LSTM-JSR algorithm is 0.77% for estimation of SOC and 13.22 W for estimation of peak power under different aging and temperature conditions.Experimental results show that the algorithm can achieve accurate estimation of SOC and SOP of lithium-ion batteries under different temperature and number of discharge cycles.Aiming at the problem that the LSTM network is prone to deviation in the long term prediction of lithium battery,this thesis constructs the LSTM network optimizes based on statistical regression model by obtaining the working characteristics of the power lithium-ion battery,and uses the LSTM-SR algorithm to achieve accurate estimation of the state of charge.Furthermore,the discrete equation model and the state of charge are combined to predict the peak power of the power lithium-ion battery,and a joint estimation system is built to achieve the collaborative estimation of the state of charge and the peak power.The experimental verification results of different aging and different temperatures show that compared with other algorithms,the joint estimation system of state of charge and peak power can realize the estimation of state of charge and peak power of power lithium-ion batteries more effectively and has strong universality.However,the prediction of lithium-ion battery pack is still lacking,and it may be necessary to find more suitable characteristics to improve the accuracy of the model prediction.