Driving power prediction of heavy commercial vehicles based on multi-task learning

Abstract

With the continuous development of vehicle intelligence technology, developing predictive control strategy has become a research hotspot, and the prediction of vehicle driving power is crucial for developing such strategies. In the field of short-term power prediction, most approaches indirectly predict vehicle driving power by predicting speed, road grade, and other information based on deep learning single-task methods. To reduce the calculation scale of the prediction model and solve the multivariable coupling prediction problem required for power prediction, this paper proposes a three-parameter prediction model for road grade, speed, and acceleration based on multi-task learning (MTL) network. The development and validation of the prediction model were based on actual vehicle data. Compared with traditional single-task prediction methods, the proposed model improves the accuracy of power predictions by more than 10%.