Predictive battery thermal management using quantile convolutional neural networks

Abstract

An improvement in energy efficiency of Battery Thermal Management Systems (BTMS) can increase range and reduce well-to-wheel emissions of Battery Electric Vehicles (BEV). In this work, the potential of a predictive BTMS using Quantile Convolutional Neural Networks (QCNN) was examined. The QCNN provided quantile predictions of battery temperature based on input data from both previous and following drive segments. The predictive control was designed to choose battery cooling thresholds based on a weighted sum of battery cooling, ageing and derating costs derived by the quantile predictions. The predictive BTMS was analyzed concerning its adaptability to different routes ahead, tunability of cost weights as well as robustness to uncertainty of inputs. A setup with unchanged ageing costs reduced average cooling costs by 9% compared to a fixed threshold strategy in a set of 18 scenarios. Simplifications and limitations were discussed to provide a base for further improvements, for example concerning the limited freedom of cooling threshold choice. In conclusion, the developed framework was able to use QCNN predictions to increase the BTMS energy efficiency while taking ageing and derating effects into account.

Keywords

• Battery thermal management
• Machine learning
• Predictive control
• Quantile convolutional neural networks