Wavelet-based confirmatory factor analysis for monitoring of system factors: estimating goodness-of-fit measures with the aid of self-organizing feature maps

摘要

Under consideration is further development of the wavelet-based confirmatory factor analysis intended for monitoring of factors responsible for evolution of technical and other systems. According to the proposed approach, the samples of coefficients resulted from discrete wavelet transform of initial parameter time series under study and responsible for different observation periods are considered as values of observed variables in the subsequent confirmatory factor analysis to reveal time history of factor influences and estimate factor interaction. Identification of free factor model parameters (usually factor variances and covariances) is carried out by a new direct (noniterative) procedure, which is an alternative to traditional local iterative optimization procedures based on the maximum likelihood criteria. The main issue under consideration is a new approach to the goodness-of-fit factor model analysis that is based on the capabilities of self-organizing feature maps and makes it possible to avoid tight restrictions on observation data inherent in the traditional model identification procedure. A technique for estimating significance of factor model components is discussed. Applications to the analysis of aircraft damage accumulation and psychological investigations are given.