Chapter 5 - Moving Toward Agile Machine Learning for Data Analytics in Power Systems

Abstract

This chapter starts with a brief discussion on classical supervised and unsupervised learning paradigms. The focus is not to give an extensive review of the field, which is impossible due to its many ramifications, but rather to equip the readers with fundamental ideas and popular approaches for regression, classification, dimension reduction, etc. The chapter moves on with a discussion on the important issue of model selection (hyperparameter tuning), which is pivotal to the performance of those off-the-shelf machine learning (ML) tools. Because all ML tools are “garbage in garbage out,” the next section of this chapter is devoted to the problem of feature selection (FS), in which existing FS methods and their recent variants are presented in some depth. The rest of this chapter is devoted to the introduction of recent schemes of ML that seem promising for power system data analysis applications. The topics include but are not limited to semisupervised learning, multitask learning, transfer learning, multiview learning, etc. Overall this chapter aims at providing power system practitioners with basic knowledge of ML tools and their proper usage, as well as motivating researchers to develop new models and methods by combining their expertise from both ML and power system fields.