PINN with Memory: A Novel Methodology for State of Charge Estimation of Lithium-ion Batteries Under Dynamic Load Profile

Batteries are the enabling force behind electric vehicles (EVs), which require energy storage for operation and mobility. The battery packs are complex systems, and their operation must be regulated within the specified limits to ensure safety and reliable operation. The battery management system (BMS) of EVs continuously estimates the state of charge (SOC) and the state of health (SOH) of the cells, enabling available energy, power, and range calculations. The SOC depends on various factors, but one critical factor that influences the current SOC is the load influenced by the driving conditions. SOC estimation techniques can be broadly categorized as model-based or data-driven, with each approach offering distinct advantages and disadvantages. While data-driven techniques can yield accurate estimations for the data they are trained on, they often encounter challenges such as overfitting and limited generalizability due to their detachment from the physical model. This paper developed a hybrid data-driven and model-based technique to estimate the SOC of a cell. The proposed model uses the equivalent circuit model with memory in a physics-informed neural network to accurately predict the SOC of a cell when exposed to a Worldwide Harmonized Light Vehicles Test Procedure (WLTP) class 3 drive profile. The proposed model achieves a mean absolute error and the mean squared error of 0.7240% and 0.0106%, respectively, outperforming the base LSTM models.