Lithium-Ion Battery RUL Prediction Using CNN-LSTM Optimized by Improved Artificial Lemming Algorithm

Accurate prediction of remaining useful life is paramount for effective lithium battery health management. This paper introduces a comprehensive prediction model that integrates an improved artificial lemming algorithm (ALA), a vonvolutional neural network (CNN), and a long short-term memory (LSTM) neural network. First, four sets of indirect health features were extracted from lithium battery charge-discharge cycle experiments to serve as predictive characteristics: constant current charge time, constant voltage rise time, and time integral of the constant voltage charge current curve during the charge phase, as well as constant voltage discharge time during the discharging phase. To mitigate the ALA’s inherent susceptibility to local optima, an improved ALA algorithm was designed. This was achieved by incorporating the Tent chaotic map to enhance population diversity and integrating the mutation and crossover operations from the differential evolution algorithm, thereby improving both local search accuracy and convergence speed. Finally, a hybrid CNN-LSTM neural network model is proposed, capitalizing on the complementary strengths of both architectures. The model’s predictive capability is further enhanced by employing optimization algorithms to address the sensitivity issues related to neural network hyperparameters.