随着可再生能源在有源配电网中的渗透比例逐年增加，其带来的随机性、间歇性对已有调度策略产生了重大挑战。文章提出了一种基于多智能体深度强化学习的有源配电网经济调度策略,构建多区域能源自治框架，每个新能源自治区域对应一个智能体，应用多智能体深度强化学习（multi-agent deep reinforcement learning,MADRL）算法解决各区域的协同经济调度问题，并对包含风机、储能设备的有源配电网进行区域建模，设定经济优化目标及运行约束条件，在多智能体深度确定性策略梯度（multi-agent deep deterministic policygradient,MADDPG）算法基础上，采用BiGRU（bidirectional gated recurrent unit）代替全连接层，进行新能源的出力预测，有效降低新能源波动性带来的影响，以改进的IEEE33测试系统进行算例分析，验证了所提策略的有效性和对比同类算法的优越性。
英文摘要:
With the increasing penetration of renewable energy sources in the active power distribution network, their inherent randomness and intermittency pose significant challenges to existing scheduling strategies. This paper proposes an economic dispatching strategy for active distribution network based on multi-agent deep reinforcement learning. Firstly, it constructs a multi-area framework for renewable energy autonomy, with each autonomous region corresponding to an intelligent agent. The multi-agent deep reinforcement learning (MADRL) algorithm is applied to address the collaborative economic dispatching issues across regions. Secondly, it models the active distribution network containing wind turbines and energy storage devices at the regional level, with economic optimization objectives and operational constraints set. Finally, building upon the multi-agent deep deterministic policy gradient (MADDPG) algorithm, the bidirectional gated recurrent unit (BiGRU) is employed instead of fully connected layers for renewable energy output prediction, effectively mitigating the impact of renewable energy fluctuations. The effectiveness of the proposed strategy and the superiority of similar algorithms are validated through using an improved IEEE 33-node test system.