SoC Estimation Model for Regenerative Braking System in Three-Wheeled Electric Vehicles Using LSTM and BPNN-Based Coulomb Counting

The advancement of electric vehicle technology calls for accurate estimation of battery state of charge (SoC), particularly in systems equipped with regenerative braking. This paper presents a comparative modeling study of state of charge estimation for a three-wheeled electric vehicle equipped with regenerative braking, using two different approaches: a data-driven long short-term memory model (LSTM) and a hybrid backpropagation neural network-based Coulomb counting (BPNN-CC) method. Experimental data were collected from a prototype three-wheeled electric vehicle with a Votol EM-150 controller. The dataset includes real-world operating parameters such as voltage, current, rotational speed, and measured state of charge from the battery management system (BMS). The LSTM model employs a two-stage architecture to predict regenerative current and SoC sequentially, while the BPNN-CC model integrates physical Coulomb counting with neural correction. Results show that the LSTM model achieves lower root mean square error (RMSE) and mean absolute error (MAE) on test data, with coefficient of determination values above 0.90. The study highlights the effectiveness of deep learning for capturing nonlinear energy recovery behavior in electric vehicle systems.