Abstract:
Aiming at the problem of performance degradation after the upgrade of telecommunication systems, a cross correlation function-partial least squares (CCF-PLS) method is proposed to monitor the system performance and find the variables that results in the performance degradation. For the telecommunication systems with large sampling interval and strong linearity, the PLS model can well establish the regression model between key performance indicator (KPI) and other observation variables. Although the observation variables have many uncorrelated components with KPI, CCF can select those variables that are strongly related with KPI and can thus constitute a pre-processing procedure for PLS. Furthermore, by combining the contribution plot method, the CCF-PLS method can give the root cause of system abnormality. Finally, it is shown via simulation results that the method proposed in this paper can attain good practical effect.