Artificial Neural Network-based State of Charge (SOC) Estimation of a Lithium-Ion Battery under Different Temperatures Conditions

Lithium-Ion batteries are rechargeable batteries used to store energy. State of Charge estimation is necessary to increase the operational viability and reliability of the battery to reduce cost and for proper maintenance. Using a feed-forward backpropagation neural network (FFBPNN) technique combined with the Levenberg Marquardt Training (LM) algorithm, an artificial neural network (ANN) model is proposed in this research that can estimate the state of charge of a lithium-ion battery with greater precision. In the TANSIG transfer function's first layer, the quantity of hidden neurons is compared. The ANN model has been trained and achieved MSE is 0.00039609, RMSE is 0.019902, and R2 of 0.99815 for training, validation, and testing. Using MATLAB software, the effectiveness of the proposed ANN Model scheme has been evaluated for some statistical tools that are signal-based condition indicators based on time domain variables, such as Root Mean Square Error (RMSE), Gradient, Mu, Variance, error histogram, and Regression plot.