Reinforcement learning in power system scheduling and control: A unified perspective

Reinforcement Learning (RL) has been applied to various scheduling and control problems in power systems in the last decade. However, the area is still in its infancy. In this paper, we present various research works in this area in a unified perspective. In most of the applications, power system problems — control of FACTS devices, reactive power control, Automatic Generation Control, Economic Dispatch, etc — are modeled as a Multistage Decision making Problem and RL is used to solve the MDP. One important point about RL is, it takes considerable amount of time to learn a control strategy. However, RL can learn off line using a simulation model. Once the control strategy is learned decision making can be done almost instantaneously. A major drawback of RL is most of the application does not scale up and much work need to be done. We hope this paper will generate more interest in the area and RL will be utilized to its full potential.