Capturing Temporal and Spatial Dynamics in Battery SOH Prediction with a T-GNN Model

Abstract

Lithium-ion (Li-ion) batteries are widely used in industrial products. Predicting the state-of-health (SOH) of Li-ion batteries is a critical component of Prognostic and Health Management (PHM) systems. SOH prediction has been extensively researched, and many methods have been proposed for this task, including gated recurrent units, Bayesian statistics, and Long Short-Term Memory (LSTM) networks. However, there has been limited exploration of Graph Neural Networks (GNNs) for this purpose. To address this gap in the literature, this study introduces a Temporal GNN (T-GNN) model that integrates LSTM with GNN to combine the strengths of both models for accurate SOH prediction of Li-ion batteries. The proposed model focuses on capturing the degradation patterns in battery cells over time. T-GNNs are specifically designed to process both temporal and spatial dynamics, capabilities that most employed models often do not fully leverage. This model shows promise in detecting evolving patterns of usage and degradation in battery cells, which is crucial for precise SOH predictions. The approach emphasizes the dynamic nature of T-GNNs, allowing continuous adaptation to changes in battery cell conditions. The performance of the T-GNN model is evaluated on two publicly available datasets and compared to three benchmark models. It achieves high predictive performance, as demonstrated by low Root-Mean-Squared Error (RMSE) and Mean-Absolute Error (MAE) values.