Close-loop Predict-and-Optimize Method for Power System Operation: Concepts, Rationality, Applications, and Prospects

The closed-loop predict-and-optimize (CPO) method, a machine-learning-based forecasting paradigm which aims at less decision error instead of less prediction error, has drawn an increasingly attention in recent years. However, the problem, why a less accurate prediction can lead to a better decision so that using CPO is rational, is still obscure and lacks a theoretical explanation, which restricts the wide application of CPO. In this paper, we systematically introduce the concepts of CPO. Then, using simplex method, multi-parametric linear programming, and multi-parametric mix-integer linear programming, we prove the rationality of CPO while offering a set of analytical applicable conditions for CPO. Moreover, a theoretical framework for CPO’s rationality is established. Finally, the applications and prospects of CPO in power system operation have been reviewed. We believe as the technology matures, the CPO can play a significant role in the transition towards smarter and greener energy infrastructures.