Remaining Useful Life Prediction Method Of Lithium-ion Batteries Based On KELM

As the most important energy supply link in the new energy industry,lithium-ion batteries(LIBs)have been widely used in various fields.However,as an energy storage device,LIBs have complex electrochemical reactions inside themselves.As the batteries continue to charge and discharge,the capacity of the batteries will decrease due to the weakening of the electrochemical reactions inside the batteries.There are a large number of LIBs in the electric vehicle market,and most of them will decommission and recycle.Predicting the life of the LIBs in time is conducive to the maintenance,recycling and replacement of the battery,and ensures the safe and reliable operation of electric vehicles.The main contents of this article for predicting the remaining cycle life(RUL)of LIBs include the following:(1)The performance degradation mechanisms and degradation factors that affect battery life are analyzed.LIBs degradation data and charge-discharge process data is obtained from the charge-discharge cycle life experiment of LIBs,which is used to support subsequent life prediction work.(2)Indirect health factors that can characterize battery life degradation are extracted from the charge-discharge process.The health factors are selected mainly from the charging process.Three indirect health factors are extracted in the constant voltage stage,constant current stage and static stage.The Pearson and Spearman correlation analysis methods are used to analyze the relationship between health factors and capacity to verify the feasibility of indirect health factors.(3)The Kernel Extreme Learning Machine(KELM)algorithm is built to predict the RUL of LIBs.The Whale Optimization Algorithm(WOA)is used to optimize the parameters of the KELM algorithm for improving the accuracy of the RUL prediction.The WOA-RBF-KELM algorithm with the RBF kernel function as the kernel function of the KELM algorithm and the WOA-poly-KELM algorithm with the poly kernel function as the kernel function of the KELM algorithm are built respectively,and the selected health factors are used to verify the results.The established WOA-KELM algorithm can predict the RUL of LIBs accurately.(4)The role of the kernel function is very important for the KELM algorithm.In order to make the KELM algorithm have strong learning ability and generalization ability at the same time,the kernel function has been improved.The RBF kernel function with strong local performance and the poly kernel function with strong global performance are functionally fused,and the WOA-MKELM algorithm is built.Verification by selected health factors and compared with the ELM algorithm and the WOA-KELM algorithm shows that the WOA-MKELM algorithm improves the prediction accuracy,and can predict the RUL of LIBs more accurately.The WOA-MKELM algorithm built in this article uses indirect health factors instead of capacity or other direct health factors to predict the RUL,which is useful to the application and development of battery management systems.