Attention-enabled reinforcement learning for inverter-based voltage regulation in distribution network

The increasing penetration of distributed generations (DGs) makes the distribution network (DN) voltage control strategy different from the traditional DN. To address the lack of coordination between photovoltaic (PV) in local voltage control, this paper proposes an attention-enabled multi-agent deep reinforcement learning (MADRL) method for inverter-based voltage regulation in DN. Firstly, the Markov decision process (MDP) is designed according to the voltage regulation model. And then, the model is solved by the multi-agent actor-attention-critic (MAAC) algorithm, which can help each agent selects important information by the attention mechanism. The method can fully exploit the reactive power regulation potential of each inverter, giving the voltage control strategy in real time and dealing with violent voltage fluctuations. The test results on IEEE 33-bus system prove the superior performance of the proposed method in voltage regulation.