With the increasing integration of distributed photovoltaic systems into the distribution network

issues such as voltage limit violation and network losses have become more prominent. However

traditional voltage control methods fail to address the voltage fluctuations caused by rapid changes in the output of new energy sources

making it challenging to ensure the safe and stable operation of distribution networks. To address this

this paper proposes a zonal voltage control strategy for the distribution network based on graph multi-agent deep reinforcement learning with a spatial-temporal pseudo-twin network. First

the reactive power regulation range of photovoltaic inverters is defined under dual constraints. Next

the zonal voltage control problem is formulated as a distributed partially observable Markov decision process (POMDP). The algorithm incorporates a spatial-temporal pseudo-twin network that consists of a dynamic graph attention network and a long short-term memory (LSTM) network

to generate a spatial-temporal fused feature vector. Finally

the proposed method is validated through an example by using a modified IEEE 141-node distribution network. The results demonstrate that

compared with traditional voltage control methods

the proposed algorithm effectively reduces voltage deviations and power losses

and also presents strong generalization ability and real-time performance. This approach provides a flexible and efficient solution for achieving zonal voltage control in distribution networks.