反映实时供需互动的Stackelberg博弈模型及其强化学习求解

A Stackelberg Game Model of Real-time Supply-demand Interaction and the Solving Method via Reinforcement Learning

为充分挖掘供给侧发电机和需求侧柔性负荷的联合优化调度空间，实现分布式自律计算与集中协调的互动框架，满足供需互动快速决策的需求，最大化系统的整体效益，搭建了基于stackelberg博弈的电力系统实时供需互动模型，并提出了一种全新的深度迁移强化学习（deep transfer reinforcement leaming，DTRL）算法。该算法通过对历史优化任务的有效信息进行知识存储，利用深度学习实现高精度的非线性迁移学习，并借助分布式计算优势，可快速获得高质量的最优解。算例仿真表明：DTRL在保证最优解质量的同时，其求解速度可达其他6种对比算法的419倍以上，适合求解大规模电力系统的供需互动快速决策问题。

A Stackelberg game model of real-time supply-demand interaction was built to fully excavate the joint optimized dispatch space for the generator on the supply side and the flexible load on the demand side. Under the framework of distributed self-discipline calculation and centralized coordination, the model can maximize the overall efficiency of the system by making quick decisions. To effectively solve this problem, a deep transfer reinforcement learning （DTRL） algorithm was proposed. The algorithm can store the effective information of historical tasks, so that the deep learning can be used to carry out high-precision nonlinear transfer learning. In addition, it can take advantage of distributed computing to obtain a high-quality optimal solution quickly. The simulation results show that the speed of DTRL can reach more than 419 times of the other six comparative algorithms while ensuring that the quality of the optimal solution is good. Therefore, the algorithm is suitable for making quick decisions of supply-demand interaction in a large-scale power system.