Energy Management Strategy for Plug-in Hybrid Electric Vehicles Based on Twin Delayed Deep Deterministic Policy Gradient Algorithm and Condition Recognition

This article explores the impact of driving condition information on the energy efficiency of deep reinforcement learning (DRL)-based energy management strategies (EMS) for plug-in hybrid electric vehicles (PHEVs). Specifically, a simulation model for the PHEV with a P1 + P3 configuration is developed based on experimental test data. Furthermore, a driving condition recognition strategy is proposed, which combines principal component analysis (PCA) for feature dimensionality reduction with k-means clustering and a fully connected neural network to achieve real-time classification. To address the high-dimensional continuous control requirement of the system's two-dimensional action space, the twin delayed deep deterministic policy gradient (TD3) algorithm is employed. By systematically incorporating the recognized driving conditions as environmental state parameters, an EMS integrating TD3 and condition recognition (TD3 + CR) is ultimately established. Simulation results demonstrate that considering driving conditions significantly enhances the energy efficiency of the EMS based on DRL, achieving a 2.72% reduction in total fuel consumption versus to the TD3-based EMS without driving condition consideration. The proposed methods show promise in improving fuel economy and maintaining battery state of charge more effectively.