A new PCA similarity factor based on elliptical principal angle

摘要

Measuring the similarity between two different PCA models is a key step in pattern matching, monitoring of processes with multiple operating conditions and other applications. This paper first explains the limitations of several existing methods for measuring the similarity of PCA models through a comprehensive comparison based on simulation, then defines the elliptical principal angle, based on which, a new PCA similarity factor capable of reflecting both the gradual rotation and the gradual eigenvalue change of a PCA model sensitively and smoothly is proposed.