Load Sampling for SCUC Based on Principal Component Analysis and Kernel Density Estimation

This paper studies how to sample load more realistically and efficiently for security constraint unit commitment (SCUC) problems in order to achieve a high degree of robustness of the unit commitment (UC) solution. For example, given the UC solution, 95% of load profiles can be supplied. Principal component analysis (PCA) is introduced to find a clear feature of the historical load in two-dimensional space rather than the original high-dimensional load space, whose feature is hard to capture. Kernel density estimation (KDE) with Gaussian kernel is applied to form the probability density function (PDF), from which hourly load profiles are sampled. The load profiles based on sparse sampling are used to find the UC solution with high robustness while the load profiles based on dense sampling are used to verify the robustness of the obtained UC solution.