Shuo Zhang, Haiping Liang, Xiaoqing Guo
Photovoltaic Systems, Energy Storage, Electric Vehicle Charg-ing, Federated Learning, Deep Reinforcement Learning, Coupled Transportation-Energy Systems
The rapid expansion of integrated photovoltaic (PV) generation, bat- tery energy storage systems (BESS), and electric vehicle (EV) charg- ing stations within coupled transportation–energy networks presents unprecedented challenges in achieving energy efficiency, cost mini- mization, and operational scalability. Traditional centralized opti- mization methods struggle to accommodate the distributed, privacy- sensitive, and dynamic nature of such multi-node infrastructures. This paper proposes a novel hybrid optimization framework com- bining Federated Learning (FL) with Transformer-based Deep Re- inforcement Learning (T-DRL) to address these challenges. The FL module enables decentralized collaborative training across dis- tributed PV–Storage–Charging (PSC) nodes without sharing sensi- tive local data, preserving privacy and ensuring scalability. Concur- rently, the Transformer-DRL agent models complex temporal depen- dencies in solar irradiance, EV arrival patterns, and grid demand to dynamically optimize charging schedules and storage dispatch. The proposed framework operates within a co-simulation environment in- tegrating MATLAB/Simulink for energy system modeling, SUMO for transportation network simulation, and Python PyTorch for learn- ing algorithm implementation. Experimental results across multiple urban scenarios demonstrate a 36.5% reduction in peak grid load, a 31.2% improvement in PV utilization rate, and a 27.8% decrease in overall operational costs. Furthermore, the hybrid FL–T-DRL frame- work achieves a 12.4% faster convergence rate and 9.7% higher adapt- ability under variable traffic and weather conditions compared to baseline centralized DRL or conventional rule-based methods. These results confirm the framework’s efficacy for scalable, privacy-aware, and intelligent optimization of distributed PV–Storage–EV charging stations in modern smart cities. North China Electric Power University, China; e-mail: Henan Normal University, China; e-mail: Corresponding author: Zien Li Recommended by: Zhenling Liu
Important Links:
Go Back
