Machine Learning Application in Energy Storage System’s State Estimation: State of Health (SOH)

Abstract

There exists an increasing demand for the modern prognostics and health management system for the Li-ion batteries under real-world operation, specifically for the electric vehicle (EV) applications. Since the estimation of battery state of health (SOH) is critical for the safety and the decision making such as warranty analysis, the battery SOH should be estimated accurately. In this work, we have employed measurable data such as current, voltage, and temperature towards developing different deep learning (DL) models to estimate the cell’s SOH cycled under a variety of extreme fast charging protocols. The results obtained from the different DL models have been compared with those obtained from the conventional feed forward neural networks (FFNNs). The accuracy of all the developed DL models with long short-term memory (LSTM), convolutional LSTM (ConvLSTM), and deep convolutional neural network (DCNN) architecture are acceptable by industry standards, with mean absolute percentage error (MAPE) less than 3%. The promising results obtained in this study indicate that the presented DL models in this work can be implemented in future battery management systems (BMSs).