基于混合强化学习的主动配电网故障恢复方法

Fault Recovery Method for Active Distribution Network Based on Hybrid Reinforcement Learning

针对高比例新能源接入的配电网故障恢复问题,提出一种基于混合强化学习的主动配电网故障恢复方法。首先,以故障损失最小为恢复目标、配电网安全运行要求为约束条件,构造主动配电网的故障恢复模型;其次,建立用于故障恢复的强化学习环境,根据状态空间和动作空间特点,提出一种混合强化学习方法,该方法使用竞争架构双深度Q网络算法处理离散动作空间,进行开关动作;然后,使用深度确定性策略梯度算法处理连续动作空间,调节电源出力;最后,通过IEEE33节点系统仿真实验验证所提方法的可行性和优越性。

In response to the fault recovery problem of distribution network with high-proportion new energy connection,a fault recovery method for active distribution network based on hybrid reinforcement learning is proposed.First,a fault recovery model of active distribution network is constructed with minimizing the fault losses as its recovery objective and the safe operation requirements of distribution network as constraints.Second,a reinforcement learning environment for fault recovery is established,and a hybrid reinforcement learning method is put forward based on the characteristics of state and action spaces.This method uses a dueling double deep Q network(D3QN)algorithm to handle the discrete action space and perform switch actions.In addition,it uses a deep deterministic policy gradient(DDPG)algorithm to handle the continuous action space and adjust the power output.Finally,a simulation experiment is carried out on an IEEE33-node system,and the results verify the feasibility and superiority of the proposed method.