Cooperative Pricing for Vehicle-Traffic-Power Nexus to Support City Decarbonization: A Federated Deep Reinforcement Learning Framework
Yi Wang, Qinglai Guo, Yang Yu
Abstract
The traffic network (TN) and the power distribution network (PDN) are major contributors to urban carbon emissions. The prevalence of electric vehicles interconnects the two networks, which leads to the emergence of a vehicle-traffic-power nexus. This physical interdependency compromises the traditional separative scheduling method, where price signals are separately designed for decarbonization. Further, data barriers between the PDN and TN block the operators from collaborating, enlarging the challenges. Thus, we propose the Fed-DDQN, a privacy-preserving, dynamic, and cooperative pricing method for city decarbonization, which innovatively integrates vertical federated learning (VFL) with deep reinforcement learning (DRL). DRL approaches are adopted to handle time-varying traffic demand and electric load, whereas the novel VFL framework that introduces differential privacy is utilized to preserve sensitive information. Results demonstrate that compared with a separative pricing scheme, our proposed method reduces carbon emissions by 16% and improves the operational conditions of both networks.