Parameter identification of lithium-ion electrochemical model using fully connected neural network and adaptive switching swarm intelligence algorithm

Accurate sensing of the internal state of lithium-ion batteries and the identification of key electrochemical parameters are essential for reliable evaluations of the battery health. However, parameter identification using swarm intelligence algorithms (SIAs) alone is time-consuming, and purely neural network-based methods often require extensive experimental data. To overcome these challenges, this paper proposes a parameter identification approach, referred to as fully connected neural network and adaptive switching swarm intelligence algorithm (FCASIA), for an electrochemical model of lithium-ion batteries. The proposed approach integrates deep learning with dual neural networks: a convolutional neural network (CNN) adaptively selects the suitable SIA, and a fully connected neural network (FCNN) refines the parameter estimation. This dual-network framework substantially improves both accuracy and speed, thereby addressing the limitations of single solution. In addition, an adaptive online training mechanism coupled with the FCNN's ability to generate and inject high-potential new candidates, enable a high-precision parameter identification without relying on large volume of experimental data. By constructing an efficient parameter identification framework for pseudo two-dimensional (P2D) model, the proposed method further accelerates convergence and enhances accuracy. Comparative experiments indicate that FCASIA outperforms direct SIA-based methods, achieving an average parameter error of only a 5 % deviation from ground truth and demonstrating its potential for accurately and efficiently handling complex battery models.