Machine Learning-driven Battery SOC Estimation for Electric Vehicle Application

The batteries are the main energy source in electric vehicles (EVs). Estimating the State of Charge (SOC) batteries is crucial for maximizing their performance and optimizing their range. This study proposes a unique approach using an Artificial Neural Network (ANN) to estimate the SOC of batteries. This ANN model incorporates multiple hidden layers to capture the complex and nonlinear relationship between the terminal voltage and the SOC. The training is performed using the Levenberg Marquardt algorithm, known for its effectiveness in nonlinear regression and curve-fitting problems. The simulation is performed in MATLAB for a 12.8V, 7.5Ah battery, which considers variable current, temperature, and voltage as the input factor and provides SOC for different conditions. It is found that the ninth hidden layer in ANN achieves a minimal Root Mean Square Error (RMSE) of 0.0027, which provides the optimal layer for training. Two charging methods are considered to check the effectiveness of the ANN model. The result shows predicted SOCs for Multi-step Constant Current and Constant Current Constant Voltage charging methods in this paper. Other parameters of batteries such as SOH, aging, safety, and protection can be predicted using ANN for EV applications.