A multi-feature fusion model based on differential thermal capacity for prediction of the health status of lithium-ion batteries

Abstract

Accurate prediction of the SOH (state of health) of lithium-ion batteries is still a key problem in the safe application of lithium-ion batteries. A new multi-feature fusion SOH prediction method, SRLF-CHI-AdaPSOELM, is proposed based on differential thermal capacity (DTC). DTC is a data representation that integrates the capacity, surface temperature, and voltage information that directly affect the health of the Li-ion battery. But the DTC has a complex trajectory, strong nonlinearity, and large noise. Therefore, a new function model (S-RLF) is established, which is expressed as a sigmoid model with an exponential and a Lorentz-rational function model to represent DTC in different voltage segments. As verified by public datasets, the proposed S-RLF can well express the key features of DTC, such as peak value and area. Then two new lithium-ion battery health indicators (HI) are extracted from the RLF parameters, further a fusion health indicator (CHI) is established by canonical correlation analysis. CHI eliminates the redundancy of the RLF model parameters and can be better used to accurately predict SOH. Then a new SOH prediction model (Ada-PSOELM) based on an extreme learning machine (ELM) is established. The input parameters of the ELM are optimized by the particle swarm optimization (PSO) algorithm, and an AdaBoost algorithm is introduced to integrate multiple PSO-ELM weak predictors to enhance the generalizability of SOH prediction. Finally, the prediction results are compared with other models on different lithium-ion battery datasets. The results show that the MAE of the SRLF-CHI-AdaPSOELM model is below 0.5%, which verifies the high accuracy and robustness of the model on both large-cycle and small-cycle lithium-ion battery datasets.