Selecting and Evaluating Representative Days for Generation Expansion Planning

This work applies feature engineering and machine learning to design a rigorous algorithm to select representative daily profiles of net load for Generation Expansion Planning (GEP). This algorithm incorporates Principal Component. Analysis and clustering techniques to produce multiple sets of representative days. Guidelines to determine the proper number of representative days and the number of times the clustering process should be repeated are also proposed. These sets are then assessed using a rolling horizon unit commitment (RHUC) on a test system. Using RHUC as a metric to select the best set of representative days ensures that the selection is based on a criterion that closely approximates the operation of a power system with a large penetration of renewable generation.