Two-Timescale Voltage Regulation for Coupled Distribution and Flash-Charging-Enabled Public Transit Systems Using Deep Reinforcement Learning

As one of the most promising charging solutions for electric buses (EBs), the pantograph charger features fully automated high-power charging, allowing it to cater to EB flash charging at bus stops. By “charging less but more often,” the technical viability and economic competitiveness of EBs can be greatly enhanced. However, such a charging arrangement inevitably threatens the secure operation of distribution networks (DNs), which is already complicated by the growing renewables. Hence, this article proposes a data-driven two-timescale voltage regulation method to tackle the operation challenge in coupled distribution and flash-charging-enabled public transit systems (CDFPTSs). Concretely, remotely controlled switches (RCSs) and soft open points (SOPs), both functioning such as “valves” to optimize power flow distribution, are coordinated by multiple intelligent agents to essentially enhance voltage security heavily impacted by abrupt and excess EB charging demand. Due to the different operation timescales of RCSs and SOPs, a two-timescale Markov game is dedicatedly formulated to enable a model-free and decentralized control of agents for accelerating decision-making and reducing communication reliance. An action-persistence multiagent soft actor-critic (AP-MASAC) algorithm is proposed to effectively handle the hybrid action space of RCSs and SOPs, ensure their operational constraints, and, more importantly, mitigate nonstationary issues appearing in two-timescale learning to further boost regulation performance. Numerical results reveal that AP-MASAC can outperform various benchmarks in voltage regulation tasks for the CDFPTS, especially in relieving voltage violations in a data-driven manner.