State of Health Prediction of Lithium-Ion Batteries Under Varying Operating Temperatures Using Deep Neural Networks

Estimation of State of Health (SoH) of lithium-ion batteries plays a very crucial role in the efficient running of the battery and performance assessment of the battery. This paper proposes a Deep Neural Network (DNN) model for estimation of the SoH of batteries using an experimental data set containing 1945 data points with variation in battery type, temperature, load current, minimum voltage (Vmin), voltage variance (VVar), and cycles. The input features are normalized through Min-Max scaling, and the DNN structure is tuned through the activation functions, number of hidden neurons, number of hidden layers, batch sizes, optimizers, drop out rates, and kernel L2 regularization parameters. In terms of generalization capabilities, LOTO and LOBO tests are conducted, from all the configurations, 43 °C temperature the activation function of sigmoid attains the testing MAE of 2.16 with only one hidden layer having 12 neurons, drop out rate of 0.2, a kernel L2 of 0.000001, a batch size of 1, and Adam optimizer. Alongside comparisons of results from the other machine learning algorithms, such as Linear Regression, Decision Tree, Random Forest, and XGBoost. The DNN structure obtains the smallest Kullback-Leibler divergence from all the battery types and temperatures tested, implying high similarity in the distributions of SoH from the predictions and reference data. In addition, actual versus predictions charts of 43 °C indicate that the DNN is able to capture the battery SoH trend.