基于分布式多步回溯Q（λ）学习的复杂电网最优潮流算法

摘要

As for the problem that usual optimal power flow algorithm can not meet the timely demand of the complex power grid.,this paper presents a novel distributed Q（λ） learning algorithm based on complex districted power grid,which deals no auxiliary process with the optimal power flow（OPF） mathematical model and whose internal agent independently undertakes each district＇s learning duty with the standard multi-step Q（λ） learning algorithm,and then coordinately cooperate to reach the optimization of the whole system.The result of the application in IEEE118 bus bar demonstrates that the distributed Q（λ） learning algorithm provides a new feasible and effective method to the complex grid OPF problem.%针对传统最优潮流算法对复杂多目标函数的不适应性以及常规算法难以满足大规模电网计算实时性的要求,本文中提出一种新颖的基于复杂电网分区的最优潮流分布式Q（λ）学习算法,该算法无须对最优潮流数学模型进行辅助处理,不依赖于对象模型,其内部各Agent使用标准的多步Q（λ）算法独立承担各分区子系统的学习任务,通过统一协作从而形成整体意义上的最优,并在IEEE 118节点等标准算例中进行了验证,取得了良好的效果,为解决复杂电网多目标最优潮流问题提供了一种新的可行、有效的方法。