Prediction Of Lithium Battery Capacity Decay Based On GA-BPNN-LSTM Fusion Model

Lithium ion batteries are widely used in various fields of life due to their non pollution,stable performance,long service life,and high energy density.However,lithium batteries will undergo natural aging during use,and their internal discharge capacity will continue to decrease.When the discharge capacity of the battery reaches the critical life span,the probability of accidents will increase,affecting the normal operation of the equipment.Therefore,accurately predicting the residual discharge capacity and remaining service life of lithium batteries has high practical value.This work applies partial data from the CS2 battery pack in the CALCE team to propose a fusion model to predict the residual discharge capacity of lithium batteries and further obtain the remaining service life of the batteries.The specific research content is as follows:(1)Perform data conversion,missing values,abnormal values,and standardization processing on the original data,and then use four methods for comprehensive feature selection: variance screening,mutual information,correlation coefficient,and XGboost.The features selected simultaneously by these four methods are used as the final feature set.(2)In the selection of battery capacity prediction models,traditional prediction models such as Differential Integrated Moving Average Autoregression(ARIMA),Multiple Linear Regression,Ridge Regression,Support Vector Regression(SVR),and Adaboost Regression are first used for modeling.Then conduct model evaluation based on the regression model evaluation indicators MAE,MSE,and MAPE.Finally,through the prediction effects of BP neural network and LSTM on experimental data sets,the performance differences between traditional prediction models and neural network models are compared.(3)Because the parameters of BP neural network may not reach the global best in the training process,this paper tries to use genetic algorithm to optimize and make experimental comparison.(4)Due to the lack of consideration for data timing in BP neural networks,the features used by LSTM neural networks are prone to lack physical significance.Therefore,this work fuses BP and LSTM neural networks to obtain BPNN-LSTM and GA-BPNN-LSTM fusion models.The fusion strategy is to weight the prediction results of BP and LSTM neural networks based on the generalized error(MAE)of the two models.Subsequently,the predicted value of the remaining service life of the battery can be obtained based on the prediction results of the optimal effect model.The empirical results show that the GABPNN-LSTM prediction model can accurately predict the remaining capacity and remaining service life of lithium batteries,with a relative prediction accuracy of 98.38%for the remaining capacity and 97.63% for the remaining service life.Based on the data and attempted methods used in this work,the optimal model for predicting the remaining discharge capacity and remaining service life of batteries is the GA-BPNN-LSTM fusion model,which can provide a reference for the maintenance and replacement of lithium batteries and ensure the operational safety of equipment.