Dynamic Laplacian eigenmaps for process monitoring

Abstract

This paper proposes a process monitoring approach for dynamic systems based on Lapalacian eigenmaps. Aimed at the “out of sample” issue, we adopt radial basis function neural network instead of the traditional linear transformation, which is able to discover the accurate nonlinear functional relationship between the raw data and the low-dimensional data. Besides, in order to utilize temporal information, time-lagged embedding is employed to extract more meaningful information and dynamic characteristics. Thus, the proposed approach can be actually applied to nonlinear dynamic systems. Eventually, a numerical case demonstrates the effectiveness of the proposed approach.