Effective End-to-End Learning Framework for Economic Dispatch

Abstract

Conventional wisdom to improve economic dispatch effectiveness is to design the load forecasting method as accurately as possible. However, this approach can be problematic due to the temporal and spatial correlations between system cost and load prediction errors. This observation motivates us to jointly treat the two forms of correlations by adopting the notion of end-to-end machine learning. Thus, we first propose a task-specific learning criterion to conduct economic dispatch for maximal economic benefits. To reduce the task-specific approach's computational burden and over-fitting issues, we design an efficient optimization kernel to speed up the learning process. Additionally, we propose a more practical and robust model-free end-to-end learning framework and offer theoretical analysis and empirical insights to highlight the effectiveness and efficiency of our three proposed learning frameworks. Our numerical study highlights that the model-free framework decreases the additional cost from an inaccurate prediction by 5.49% in the IEEE 39-bus system compared with the conventional approach.