State-of-Charge Estimation Algorithm for Li-ion Batteries using Long Short-Term Memory Network with Bayesian Optimization

As Electric Vehicles begin to dominate the new era of the automotive industry, it has become imperative to ensure that their Li-ion batteries can be run under dynamic load conditions while ensuring a long life. Among the most critical metrics that ensure long battery life and protection from catastrophic failure of the battery pack is the Battery State-of-Charge (SOC) estimated by the Battery Management System (BMS). Therefore this metric must be estimated accurately in real-time without delays. This paper proposes a method to achieve the same by utilising a Long Short-Term Memory (LSTM) network (A type of gated-RNN architecture), coupled with the Bayesian Optimization (BO) algorithm to tune the network hyperparameters. First, the dataset is pre-processed and compiled into training, testing, and validation subsets. Next, key hyperparameters are recognized and optimized using the BO algorithm with the objective to minimize the Root Mean Squared Error (RMSE) of the validation data. Finally, the trained BO-LSTM network is evaluated using 4 test datasets with various ambient temperatures. It achieves an RMSE of 0.872% and a Mean Absolute Error (MAE) of 0.645% when tested at the ambient temperature of 40°C.