State of health estimation method based on real data of electric vehicles using federated learning

Accurately estimating the state of health (SOH) of lithium-ion batteries (LIBs) is of paramount importance for the development and operation of electric vehicles, as it directly impacts driving performance. Currently, there are numerous methods for estimating SOH, but only some are suitable for real-world data, most existing methods for estimating SOH use data of lithium-ion battery cells from laboratory operations to train and test. Firstly, obtaining data usage of battery from the real world takes work. Secondly, the model for estimating SOH must be representative and generalized because they are trained on separate data for each electric vehicle. Lastly, collecting data for centralized training may lead to huge data uploads This study goes beyond that by using battery data from real-world vehicle operations. This study is based on 10 in-service electric trucks over 3 years, adopting methods such as utilizing data acquisition technology from battery management systems (BMS) and other controllers, as well as vehicle networking data reporting technology, to propose a SOH estimation method utilizing a new machine learning approach based on artificial neural network (ANN) and federated learning (FL) to predict the SOH of lithium batteries in real-world scenarios. This study developed an ANN local model and compared it with the popular long short-term memory (LSTM) model, finding that the ANN model is relatively more suitable as a local model for electric vehicles. In addition, we apply the FL framework to train the model to reduce the amount of uploaded electric vehicles' battery data. We verified the results of the model on experimental data. Meanwhile, we analyzed the model on actual data by comparing its mean absolute and relative errors. The results show that federated training method achieves better generalization in predicting SOH compared to centralized training method and another traditional machine learning models. This study provides an effective solution of electric vehicle batteries SOH prediction and offers new ideas and methods for the generalization of traditional machine learning models.