Abstract:After a large number of distributed photovoltaics are connected to the distribution network, the system voltage exceeds the limit due to the peak of photovoltaic output in the noon hours, which seriously threatens the safe and stable operation of the distribution network. The centralized control mode faces problems such as large calculation and multiple control equipment. Therefore, an active distribution network cluster voltage regulation strategy based on multi-agent deep reinforcement learning is proposed. Firstly, reactive power balance index, net load balance index and modularity index are divided into distribution network clusters. Then, based on the historical operation data of the distribution network, deep reinforcement learning algorithm is used to train the agent, and the trained agent is deployed to the cluster as the manager, and the real-time observation information of the cluster is used to generate online adjustment device action instructions, so as to achieve the goal of minimum internal loss, minimum node voltage deviation and minimum adjustment times of reactive power compensation equipment in the scheduling cycle. The improved IEEE33 distribution system is taken as an example to verify the validity of the model.