Research of an improved PCA method for abnormality diagnosis in synchronous multi-dimensional data stream

摘要

An improved PCA method is presented which combined the PCA technology with data mining technology. In this method, the problem of the original data stream variation tendency is mapped to the eigenvector space, and the steady-state eigenvector is solved, then the abnormal changes can be diagnosed by analyzing the relationship between the instantaneous eigenvector and the steady-state eigenvector. The method is applied to analyze the synchronous multi-dimensional data stream for tunnel strain monitoring. Result shows this method can reflect the changes of the aperiodic variables timely and realize the anomaly monitoring for multi-dimensional data stream effectively.