Improving early warning of lithium-ion battery thermal risk using multi-scale residual attention and IoT integration

In recent years, early identification of abnormal thermal behaviour in lithium-ion batteries is essential for improving electric-vehicle safety and reliability. This study proposes a novel Internet of Things-enabled Multi-Scale Residual Attention-enhanced Convolutional Dilated Network for continuous battery temperature prediction and residual-based thermal-risk assessment. The framework utilised temperature, voltage, current, and state-of-charge measurements arranged into 600-sample temporal sequences. A one-dimensional temporal convolutional network is designed to extract informative temporal patterns through temporal attention. The Pearson Correlation-Coefficient-based feature selection is applied to retain relevant learned features while reducing weakly informative and redundant representations. The selected features are processed by a Residual Self-Attention-based Multiscale Dilated Network, where the Multi-Scale Temporal Network employs a multi-scale dilated convolution network to represent temporal behaviour at different receptive-field ranges, followed by Residual Self-Attention for temporal dependency refinement. The sliding-window residual analysis is applied to evaluate prediction deviations using residual mean and standard deviation to identify normal, warning, and critical thermal-risk states. The model achieved an average mean squared error of 0.0195 and a coefficient of determination of 0.990. The evaluated low-temperature charging cases provide reported warning lead times of 14.18–27.45 s, demonstrating the potential of the framework for continuous battery thermal-condition monitoring and early risk indication.