基于双层强化学习的热电联供型微网群经济调度

Economic Dispatch of CHP Multi-Microgrid Based on Double-Layer Reinforcement Learning

摘要: 热电联供型微网群系统对于推动新型电力系统的构建和低碳化具有重要意义。本文提出了一种面向热电联供型微网群系统的双层强化学习调度优化方法，首先设计了一种双层强化学习优化框架，对优化任务进行了分解，上层由智能体求解微网间交互功率策略和各微网的储能充放电策略，下层各微网基于内部信息采用求解器对微网内设备出力自治优化，通过上下层协同完成热电联供微网群系统的全局优化，各微网之间无需信息交互，有效保护了各微网内部的数据隐私。最后通过算例分析以及与单层强化学习方法和传统集中式优化对比分析验证了本文方法的有效性和优越性。

Abstract: The CHP multi-microgrid system is of great significance to promote the construction and low carbon of the emerging power system. In this paper, a two-layer reinforcement learning scheduling optimization method for CHP multi-microgrid is proposed. Firstly, a two-layer reinforcement learning optimization framework is designed, and the optimization task is decomposed. In the upper layer, the interaction power strategy between microgrids and the energy storage strategy of each microgrid is solved by agents, and the lower microgrids use solvers to autonomously optimize the processing of devices in the microgrid based on internal information. Through the double- layer cooperation to complete the global optimization of the multi-microgrid system, there is no need for information exchange between the microgrids, and the data privacy within each microgrid is effectively protected. Finally, the effectiveness and superiority of this method are verified by example analysis and comparative analysis with single-layer reinforcement learning method and traditional centralized optimization method.