Deep reinforcement learning based voltage regulation in edge computing paradigm for PV rich distribution networks

As the penetration of renewable energy in distribution networks continues to rise, the reverse power flow caused by peak outputs from photovoltaic (PV) generation is increasingly leading to voltage limit violations. These issues require mitigation through effective voltage regulation strategies. Previous studies have typically modeled the coordinated operation of PV inverters as optimization problems or neural network computation problems, but they often overlooked the consideration of the overall computing system's performance. In this study, we propose a low-power edge computing voltage regulation framework by integrating a gated control policy network, value evaluation function, and a gated Markov decision process. The update gate and reset gate are used to control the hidden state of information flow in the distribution network, ensuring that it is updated and passed onto the next time step at each time interval. The update gate and reset gate are used to control the hidden state of information flow in the distribution network, ensuring that it is updated and passed on to the next time step at each interval. Compared to traditional DDPG, this architecture demonstrates superior performance in both training convergence and voltage regulation. This edge computing voltage regulation framework, when tested on an IEEE distribution feeder with real historical data, has been shown to use less than 20% of the total resources of the Transformer Terminal Unit (TTU), making it feasible to deploy on edge devices with limited computational capacity.