Enhanced process monitoring via time-space coordinated-locality preserving projection

Abstract

In this paper, a novel dimensionality reduction method, time–space coordinated-locality preserving projections (TSC-LPP) is proposed based on locality preserving projection (LPP). In practical process, except the data correlation in spatial scale, there exists data correlation in time scale as well for the short sampling interval. To considering the correlation of sampling points in time and spatial scale simultaneously, TSC-LPP constructs the adjacency graph by selecting adjacent points in time sequence and Euclidean distance, respectively. Furthermore, the importance of the time-sequential neighbours is measured by the computed weight based on time distance. A dual objective function with a weight index coordinating the relationship between time and space is constructed to compute the transformation matrix. Hotelling's T2 and squared prediction error (SPE) are established for process monitoring. A numerical case and the Tennessee-Eastman process (TEP) are employed for the experimental verification.