Generation uncertainties, especially during the unit startup and shutdown (SU/SD) processes, pose uncertainties for the real-time market clearing process, and they are often underestimated. This paper proposes two approaches to predict generator SU/SD trajectories in the real-time operations of independent system operators or regional transmission organizations (ISO/RTOs). We first collect and pre-process raw market data from state estimation. Then we investigate two approaches to account for the uncertainty in MW of generation SU/SD in the real-time market clearing. The first is an offline approach that leverages a machine learning technique, gradient boosting tree, to effectively capture the nonlinear relationship between the SU/SD curves and selected feature maps. The offline approach works for predicting generator trajectories in the real-time Look Ahead Commitment (LAC) process, based on historical data. We also investigate an online approach using a long-short-term memory network that can learn from the last-interval error information and enhance the current prediction, potentially applicable for the real-time economic dispatch process. We validate the benefit of the proposed approach with a full-day rolling LAC framework on MISO-size test cases. The result shows that using the predicted curves could help system operators achieve better results in real-time commitment and dispatch processes.
Citation
Working paper 09/01/2021
Article
View Capturing Unit Startup and Shutdown Uncertainties in the Real-time Commitment Process