Research on energy management optimization of hybrid electric vehicles based on improved curriculum learning

With the rapid development of big data and artificial intelligence technologies, reinforcement learning has emerged as a viable methodology for energy management optimization of hybrid electric vehicles. This study explores the technical feasibility of using curriculum learning technology for global optimization of energy management strategies. An improved curriculum learning method is constructed, extending the application scenarios of curriculum learning methods to time series datasets. When integrated with random action injection, the proposed method achieves a 72.7 % reduction in reinforcement learning training duration while enhancing computational efficiency. Utilizing the plant model developed previously, the enhanced curriculum learning technique is employed to simulate and optimize vehicle energy consumption. The optimization results demonstrate that, without increasing electricity consumption, the engine's operational time could be reduced by 5.92 %, and achieved 42.8 % reduction in specific fuel consumption, improved curriculum learning method could also effectively saving the training convergence duration for reinforcement learning by roughly 72.7 %.