Time Series Prediction Integrated RNN-DRL for Energy-Emission Management in HEVs

Learning-based strategies have gained widespread adoption in the field of energy management strategy (EMS) for hybrid electric vehicles (HEVs). However, existing algorithms often struggle to maintain the effectiveness in different driving scenarios. Additionally, achieving a fully observable Markov decision process in real-world situations is impractical due to the limitation in sensor sensitivity and noise. Therefore, it is crucial to design a stable EMS for HEVs that can adapt to various driving scenarios. This paper proposes a novel approach called the twin-delayed deep deterministic policy gradient with a long-short-term combined algorithm. It aims to enhance the generalization performance and EMS capability of HEVs’ learning-based EMS by integrating temporal predicted information using a recurrent neural network. Comparing this algorithm with other established deep reinforcement learning algorithms, the results demonstrate significant improvements in EMS performance for HEVs, including state of charge maintenance, energy conservation, and emission reduction. Notably, the algorithm achieves these improvements while ensuring robust generalization across diverse driving cycles and scenarios.