Unsupervised Learning for Operational State Identification

Abstract

This chapter describes some representative unsupervised machine learning approaches for process operational state identification. Whether a machine learning approach is regarded as supervised or unsupervised depends on the way it makes use of prior knowledge of the data. Data encountered can be broadly divided into the following four categories: (1) Part of the database is known, i.e., the number and descriptions of classes as well as the assignments of individual data patterns are known. The task is to assign unknown data patterns to the established classes. (2) Both the number and descriptions of classes are known, but the assignment of individual data patterns is not known. The task is then to assign all data patterns to the known classes. (3) The number of classes are known but the descriptions and the assignment of individual data patterns are not known. The problem is to develop a description for each class and assign all data patterns to them. (4) Both the number and descriptions of classes are not known and it is necessary to determine the number and descriptions of classes as well as the assignments of the data patterns.