Analysis of Large Scale Power Systems via LASSO Learning Algorithms

摘要

In this paper, a class of modern machine learning methods is utilized for estimating the transient stability boundary characterizing the large-scale power system grids. The boundary characteristic is viewed as a highly multidimensional response of power system variables. The proposed estimation methods based on various forms of the LASSO algorithm lead to simultaneous variable selection and function recovery yielding models of the reduced complexity. The obtained models have a clear interpretation and exhibit a smaller prediction error compared with known machine learning techniques used in the existing literature on modelling of large-scale power engineering systems. The performance of our method is assessed based on the real data generated from the 470-bus power system.