Eco-Driving Strategy for Series Hybrid Electric Vehicle Based on Multi-Objective Deep Reinforcement Learning

One of the research hotspots in eco-driving is the collaborative optimization of vehicle energy management and adaptive cruise control (ACC). This article proposes a novel machine learning-based eco-driving strategy for a series hybrid electric vehicle (SHEV), with an emphasized consciousness of multi-objective weight adaptive adjustment. First, the conditioned network (CN) algorithm, which incorporates reward weights into neural network parameters, for the first time, is used to optimize the power allocation of SHEV under a car-following scenario. Second, in the framework of the proposed algorithm, the cost of battery degradation is embedded to improve the management quality. Third, the agent implements the longitudinal control of the vehicle by reorganizing two completely different theoretical car-following models to avoid excessive attention diversion beyond energy management. The proposed strategy is tested under different driving cycles to validate its superiority over state-of-the-art techniques in terms of training efficiency and optimization performance.