Quantity or Quality? Data Enabled Online Energy Dispatch

The increasing penetration of renewable energy in the power system calls for the control paradigm shift from preventive control to online control. To better facilitate the online system control, system level prediction, as well as the necessary data, is crucial to the control performance. In this paper, we study the online economic dispatch problem for microgrids. Specifically, we cast this problem into the smoothed online convex optimization framework, which enables us to examine how data quantity and data quality affect the online dispatch efficiency. In this paper, we refer to data quantity as the value of training data for prediction and data quality as the window size in the online dispatch problem. We identify the empirical power law relationship between data quantity and forecast error, which is the key to understand the role of data in online energy dispatch. Such theoretical understanding is further justified by numerical studies.