A Hybrid LSTM-Transformer Approach for State of Health and Charge Prediction in Industrial IoT-Based Battery Management Systems

Abstract

In this paper, we propose a hybrid model combining Long Short-Term Memory (LSTM) and Transformer networks for predicting the state of charge (SOC) and state of health (SOH) of batteries within Industrial Internet of Things (IIoT) based Battery Management Systems (BMS). Our approach leverages the temporal modeling capabilities of LSTM and the self-attention mechanism of Transformers. Using the NASA battery dataset, we demonstrate that our hybrid model significantly outperforms conventional methods such as SVM and Kalman filtering. Specifically, the MSE for SOC prediction is reduced from 0.0271 to 0.0107 (a 59.8% reduction), and the MAE for SOH prediction is decreased from 0.161 to 0.08 (a 50.3% reduction). These improvements are achieved through a more sophisticated handling of temporal dependencies and nonlinear relationships in the battery data.