Two-Stage Multi-Mode Voltage Control for Distribution Networks: A Deep Reinforcement Learning Approach Based on Multiple Intelligences

Photovoltaic output and load uncertainty are prone to cause voltage fluctuations in a distribution network, which can affect the power quality. With a particular focus on the distribution network voltage overrun caused by this problem, we propose a two-stage multi-mode voltage control strategy for distribution networks based on deep reinforcement learning to mitigate voltage fluctuations while realizing coordinated operation among different devices. First, the on-load regulator transformer and capacitor bank are effectively controlled using a centralized control strategy in the day-ahead phase to obtain the optimal tidal current for the distribution network, in order to achieve optimal operation of the long time-scale equipment of the system. Then, a distributed control strategy based on deep reinforcement learning with multiple intelligences is applied to regulate the reactive power of the photovoltaic inverters, based on local observations in the intraday phase, to provide fast, flexible, and reactive power support for the distribution network. Finally, a multi-mode data-driven conversion strategy based on deep reinforcement learning is used to coordinate and convert between different control modes, to enable the system to effectively reduce voltage offsets and reactive power losses under varying operational demand. The proposed method is experimentally validated on an improved IEEE 33-node distribution system, and the results show that it can effectively solve the voltage overrun problem and reduce the reactive power loss in the distribution system.