Improving economic values of day‐ahead load forecasts to real‐time power system operations

Abstract

Unit commitment (UC) is an important decision-making tool in power system operations. As a critical input to UC, inaccurate day-ahead load forecasts could result in UC solutions of high operation cost, or even trigger involuntary load shedding and jeopardise system operational security in real time. Existing researches usually explore advanced forecasting techniques for reducing statistical load forecast errors and providing better inputs to UC, while load forecasting and UC are exclusively considered as two sequential tasks. This study intends to close the loop between the two tasks by effectively incorporating certain UC information back to load forecasting models, so that day-ahead load forecasts can be improved for deriving better UC solutions in terms of a more cost-effective real-time operation with respect to actual loads. Two approaches are investigated: (i) design asymmetric error penalty functions for individual forecasting models, based on asymmetric economic impacts of under- and over-predicted load forecasts on UC solutions; and (ii) build a combining forecast framework in which weights of individual independent forecasting algorithms are determined by their economic impacts on UC solutions. Numerical cases are analysed to illustrate the effect of the proposed load forecasting models.