Optimization Design of Lithium-Ion Battery Remaining Useful Life Prediction Based on Particle Swarm Optimization

Abstract

Accurate remaining useful life (RUL) prediction is the core support for battery health management in new energy vehicles and energy storage systems. Aiming at the accuracy bottleneck caused by empirical parameter setting in neural network (NN) prediction models, this paper constructs a coupled RUL prediction model optimized by particle swarm optimization (PSO). Based on the NASA 18650-type lithium-ion battery degradation dataset(B0005-B0018), three key parameters including hidden layer nodes, learning rate and regularization coefficient are selected as design variables. A weighted root mean square error (WRMSE) is proposed as the objective function to strengthen prediction accuracy at the end-of-life stage. An improved PSO algorithm with dynamic inertia weight and Gaussian perturbation is developed to avoid local optimum. Experimental results show that the optimized model reduces test set RMSE from 3.49 cycles to 2.35 cycles, achieving a 32.7% error reduction, and the prediction deviation near the 70% capacity attenuation threshold is reduced by 41.2%. This study provides a complete technical path of “data-driven-intelligent optimization-model verification” for engineering battery RUL prediction.