基于深度强化学习架构的多能互补微网日前经济调度研究

Day‑ahead economic dispatching of multi‑energy complementary micro‑grid based on deep reinforcement learning architecture

可再生能源出力和负荷的不确定性给多能互补微网的优化调度带来挑战,传统方法如随机优化和模型预测控制等需要精确的模型参数。针对含源荷不确定性的多能互补微网日前优化调度问题,提出了基于柔性行动器-评判器框架的深度强化学习方法,实现了自适应源荷不确定特征的微网经济调度。首先,考虑设备非线性效率,建立了多能互补微网的优化调度数学模型;其次,基于柔性行动器-评判器构建了智能体和环境交互的深度强化学习框架,并设计了状态空间、动作空间、奖励函数和神经网络结构;最后,通过算例仿真验证了算法的有效性。

The uncertainty of renewable energy output and load brings challenges to the optimal scheduling of multi-energy complementary micro-grids.Traditional methods such as stochastic optimization and model predictive control require accurate model parameters.For the day-ahead optimal scheduling problem of multi-energy complementary micro-grid with source-load uncertainty,a deep reinforcement learning method based on soft actor-critic(SAC)framework is proposed,and the economic scheduling of micro-grid with adaptive source-load uncertainty characteristics is realized.Firstly,an optimal scheduling mathematical model for the multi-energy micro-grid is established considering the nonlinear efficiency of the equipment.Secondly,a deep reinforcement learning framework for interaction between agent and environment is constructed based on soft actor-critic,and the state space,action space,reward function and neural network structure are designed.Finally,the effectiveness of the algorithm is demonstrated through numerical simulation.