Hybrid Particle Swarm and Gravitational Search Optimization for Intelligent Battery Health Estimation

Abstract

This paper presents a novel hybrid optimization framework for data-driven estimation of the state of health (SoH) of lithium-ion batteries (LIBs). Existing data-driven SoH estimation methods struggle to select suitable model hyperparameters that consider capacity regeneration phenomena and to identify meaningful data samples from LIB parameters. The specific contribution to this research lies in the integrated optimization strategy, which bridges the limitations mentioned. The dataset has 31 features, comprising MIT-Stanford lithium-ion battery profiles. It employs an intelligent hybrid approach that combines the gravitational search algorithm (GSA) with particle swarm optimization (PSO) to fine-tune recurrent neural network (RNN) parameters, such as the hidden layer neurons and learning rate. The combined GSA and PSO algorithms integrated with RNN improve the SoH estimation accuracy through better search efficiency and faster convergence. The 31 LIB parameter samples are closely linked to capacity degradation and are well-suited to form the data framework. The proposed model demonstrates high accuracy in SoH estimation, particularly when applied to cell c33 from MIT-Stanford lithium-ion battery profiles. Results show that the RNN optimized with the GSA-PSO algorithm for the c33 dataset achieved a mean squared error (MSE) of 1.30×10−8, a root mean square error (RMSE) of approximately 0.0142, and a mean absolute percentage error (MAPE) of 0.0096.