Chinese and American researchers collaborate to predict battery parameters, which is expected to improve the safety and efficiency of electric vehicles

According to foreign media reports, the batteries that power electric vehicles have several key characterization parameters, including voltage, temperature, and state of change (SOC). Since battery failure is related to abnormal fluctuations in such parameters, being able to effectively predict such parameters is crucial to ensuring the safe and reliable operation of electric vehicles in the long term.

(Image source: techxplore.com)

Researchers from Beijing Institute of Technology, Beijing Co-Innovation Center for Electric Vehicles and Wayne State University in the U.S. recently developed a new deep learning-based method that can simultaneously predict multiple parameters of electric vehicle battery systems. The new method is based on a long short-term memory (LSTM) recurrent neural network, a deep learning architecture that can process both single data points (such as images) and entire data sequences (such as voice recordings or video clips).

The researchers trained and evaluated the LSTM model on a dataset collected by the Beijing Electric Vehicle Service and Management Center (SMC-EV), which includes battery-related data stored in an electric taxi over a one-year period. The model takes into account the three main characterization parameters of electric vehicle batteries, namely voltage, temperature, and SOC, and has a unique structure and design, including hyperparameters that are pre-optimized and can also be trained offline.

In addition, the researchers developed a method to perform weather-vehicle-driver analysis. This method takes into account the impact of weather and driver behavior on battery system performance, which can ultimately improve the prediction accuracy of the model. In addition, the researchers also adopted an early exit method to prevent the LSTM model from overfitting by confirming the most appropriate parameters before training.

After evaluating and testing the LSTM model, very good results were obtained. The new method does not require additional time to process data and performs better than other battery parameter prediction strategies. The results collected by the researchers show that the model can be used to identify various battery failures and notify drivers and passengers in time to avoid fatal accidents.

The researchers found that after completing offline training, the LSTM model could quickly and accurately complete online predictions. In other words, offline training did not reduce the speed and accuracy of the model's predictions.

In the future, the battery parameter prediction model developed by the researchers will help improve the safety and efficiency of electric vehicles. At the same time, the researchers plan to train the LSTM network on more data sets to further improve its performance and versatility.