Intelligent Control Strategy for Automotive Regenerative Braking Based on Deep Learning
TL;DR
A two-layer hierarchical control setup based on deep reinforcement learning and its main elements is looked at, and an organized forecast for the future of the field is offered, with special attention to four possible developments: multi-objective coordinated optimization, transfer learning, distributed vehicle-road cooperation, and multi-mode hybrid control.
Abstract
The global new energy vehicle industry is developing rapidly, but range anxiety remains a key bottleneck hindering its widespread adoption. Among all methods of improving energy efficiency, the control of regenerative braking is by far the one with the lowest cost—hence its central importance for vehicle performance. Control strategies used in the past do not adjust well to actual working conditions and have difficulty in treating complex nonlinearities or multiple coupled constraints, which rules out the possibility of a fully coordinated multi-objective solution. The present work gives a structured account of basic principles-both of automotive braking energy recovery and of deep learning algorithms. It looks at a two-layer hierarchical control setup based on deep reinforcement learning and its main elements, and considers separately the performance of various approaches under uniform criteria, as well as the engineering obstacles now limiting such systems. Moreover, it offers an organized forecast for the future of the field, with special attention to four possible developments: multi-objective coordinated optimization, transfer learning, distributed vehicle-road cooperation, and multi-mode hybrid control.