Optimizing State-of-Health Estimation for Electric Vehicles using Neural Networks and Transfer Learning Techniques

Accurate State of Health estimation is one of the most important tasks of a BMS, especially in the EV and Renewable energy storage system applications, where battery reliability, safety, and lifetime are highly valued. Thus, accurate estimation of SOH helps in the proper utilization and efficient use of battery resources and ultimately the energy system’s efficiency. This paper investigates three deep learning methods for SOH estimation: Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN) with Long Short Term Memory (LSTM), and Combined CNN, RNN, LSTM(CCRL) model. By comparing the accuracy of the forecasts and the time required for calculation, this paper proves that using the combined model allows for capturing intricate patterns in the data. The dataset employed in this work involves features that may include voltage, current, and temperature, corresponding to the target labels known as the SOH. The data is pre-processed and the features are normalized with the help of Min-Max Scaler to ensure that all the features are falling within the same range required for deep learning models. The normalized data is then split into train and test sets, a common technique observed by most researchers to check the robustness of the models.