Distributed Reinforcement Learning Framework for Collaborative Energy Management in Connected Hybrid Electric Vehicle Ecosystems

The technical parameters of HEV energy management systems increase in interaction with connected and dynamic environments, therefore demanding superior control algorithms. This work proposes an asynchronous RL paradigm for energy management in driving HEV ecosystems. The framework uses distributed RL agents with each agent responsible for local policy optimization for energy control of individual electric vehicle and with sharing information through V2V and V2I mechanisms. A general optimization layer superposes local data to coordinate personal interests with global objectives, for example, the general level of emissions and traffic density. The proposed method is capable of effectively mitigating the periodic decision making issue through using high accuracy RL algorithms for instance the MAPPO alongside low latency communication. OpenPowertrainLib and SUMO are utilized for traffic modelling integrating stochastic traffic characteristics and renewable energy inputs as well as complicated vehicle dynamics. Research findings establish the efficiency in terms of energy savings and emissions compared to single optimized energy management solutions as well as improved traffic flow. The research puts emphasis on distributed RL frameworks in developing intelligent transportation systems that positively impact scalability and sustain ability of connected HEV networks. This study sets the ground for potential future developments in co-energy management and its incorporation into smart cities.