Journal Article

Deep Learning Approach for Equivalent Circuit Model Parameter Identification of Lithium-Ion Batteries

Liu, Yi-Hua, Khanh, Dat Nguyen, Ho, Kun-Che, Wang, Shun-Chung, Hsueh, Yu-Fang

2025

Overview

This study proposes a deep learning (DL)-based method for identifying the parameters of equivalent circuit models (ECMs) for lithium-ion batteries using time-series voltage response data from current pulse charge–discharge experiments. The application of DL techniques to this task is presented for the first time. The best-performing baseline model among the recurrent neural network, long short-term memory, and gated recurrent unit achieved a mean absolute percentage error (MAPE) of 0.52073 across the five parameters. Furthermore, more advanced models, including a one-dimensional convolutional neural network (1DCNN) and temporal convolutional networks, were developed using full factorial design (FFD), resulting in substantial MAPE improvements of 37.8% and 30.4%, respectively. The effectiveness of Latin hypercube sampling (LHS) for training data generation was also investigated, showing that it achieved comparable or better performance than FFD with only two-thirds of the training samples. Specifically, the 1DCNN model with LHS sampling achieved the best overall performance, with an average MAPE of 0.237409. These results highlight the potential of DL models combined with efficient sampling strategies.

• Publisher: MDPI AG
• Subject: Accuracy / Analysis / Artificial neural networks / Batteries / Computational linguistics / Datasets / Deep learning / Efficiency / Electric vehicles / Equivalent circuits / Factorial design / Heuristic / Hypercubes / Kalman filters / Language processing / Latin hypercube sampling / Lithium / Lithium-ion batteries / Machine learning / Natural language interfaces / Neural networks / Optimization algorithms / Parameter estimation / Parameter identification / Recurrent neural networks