A MAS learning framework for power distribution system restoration

In this paper a multi agent system (MAS) framework with learning capability for power distribution system restoration is introduced. The power restoration takes place after the fault location and isolation in power distribution system to restore as much as loads possible. In this framework although agents have the capability of obtaining the optimal reconfiguration using the restoration algorithm they use Q-learning algorithm in conjunction with restoration algorithm to take the advantage of restoration experiences and making more accurate decisions. Using this framework agents only solve the restoration optimization problem when they don´t have enough knowledge about a special scenario. It means agents can do the restoration process in less time while it´s more accurate. Simulations are used to initialize the Q-Learning primary knowledge about the power distribution system. Q-Matrixes are developed in this work to keep track of previous restoration scenarios performance and are updated as the system is running. Proposed framework is applied to West Virginia Super Circuit and the results demonstrate how the learning algorithm can improve the performance of MAS for power restoration.