Battery State of Charge Estimation with Separable Temporal Convolutional Networks

Accurate State of Charge (SOC) estimation is paramount for the efficient and safe management of battery systems. Traditional SOC estimation techniques often face challenges due to the nonlinear and dynamic behavior of batteries and errors in the assumptions about initial battery states. Machine learning, on the other hand, offers an opportunity to build black box SOC estimation models that do not require exact knowledge about the behavior of the battery and initial states. This paper proposes a SOC estimation approach utilizing Separable Temporal Convolutional Networks (STCNs). We evaluate our model using the Panasonic 18650PF Li-ion Battery dataset. STCNs leverage the causal and temporal nature of battery data while maintaining computational efficiency. Our model estimates SOC based on a segment of consecutive time steps, providing an approach independent of a battery's starting point.