A Method of Developing Quantile Convolutional Neural Networks for Electric Vehicle Battery Temperature Prediction Trained on Cross-Domain Data

Abstract

The energy consumption caused by battery thermal management of electric vehicles can be reduced using predictive control. A predictive controller needs a prediction model of the battery temperature, for example for different battery cooling and heating thresholds. In the proposed method, cross-domain data from simulation, vehicle fleet and weather stations were analyzed and processed as training data for a Convolutional Neural Network (CNN). The CNN took data from previous road segments and predictions for following road segments as input and predicted the change in battery temperature as quantile sequences over a prediction horizon. Properties of the collected cross-domain data sets were analyzed and considered during preprocessing, before 150 models were trained, of which the best performing model was further analyzed. Point-forecast metrics and quantile-related metrics were used for model comparison and evaluation. For example, the median prediction achieved a mean absolute error (MAE) of 0.27 °C and the true values were below the median prediction in 47% of the test data. Possible improvements of the method such as increasing data size, using more complex architectures as well as optimizing the horizon sizes were discussed. In conclusion, the method was able to well predict battery temperatures for different battery cooling thresholds.

Keywords

• Battery temperature
• convolutional neural network
• cross-domain data
• machine learning
• quantile forecasting