State of Charge Estimation for Electric Vehicle Batteries Using Temporal Convolutional Networks

Accurate state of charge (SoC) estimation of batteries is a critical requirement for energy management systems in electric vehicles operating under real-world driving conditions. Recent deep learning approaches have shown promising performance; however, most studies rely on recurrent architectures with limited comparative evaluations. In this paper, we present a deep learning-based framework for SoC estimation based on Temporal Convolutional Networks (TCN) under identical experimental conditions using the IEEE Battery and Heating Data in Real Driving Cycles dataset, featuring real-world EV driving data across summer and winter scenarios. The proposed TCN model leverages dilated temporal convolutions to capture long-term dependencies in battery dynamics without relying on recurrent structures. Performance is evaluated using standard metrics, including RMSE, MAE, MSE and the coefficient of determination ($\mathbf{R}^{2}$). Comparative results demonstrate that the TCN-based approach in some cases, improved SoC estimation accuracy compared to the baseline deep learning model, particularly under challenging seasonal operating conditions. These findings highlight the potential of TCN architectures as an efficient and robust alternative for SoC estimation in real-world electric vehicle applications.