Generation Method of Dynamic Alarm Baseline for Cloud Server Based on XGBoost and Tolerability

In order to solve the problem that massive server alarm management of cloud data centers in electric power enterprises cannot adapt to the personalized operation of cloud business and lean equipment management, we propose a dynamic baseline generation method for cloud server alarms based on XGBoost and alarm tolerability. Firstly, based on the historical data of each server operation indicators, we apply XGBoost algorithm to predict the operation status value of a performance indicator in a certain period range in the future. Then, by comprehensively considering multi-dimensional factors such as the importance of operation time interval, the levels of business systems, the number of historical alarms, and the number of users as constraint parameters, we quantitatively calculate different alarm tolerability ranges and generate the initial curve of alarm baseline. Finally, we use the Savitzky-Golay filter method to smooth the threshold of initial alarm baseline curve and dynamically generate post-processed alarm baselines for different servers. Through case analysis of the cloud servers operation data, this method can effectively learn the historical operation data of different servers and obtain the alarm threshold under their tolerability constraints, dynamically adapt to generate hierarchical alarm baseline of massive servers, and improve the efficiency of large-scale cloud server monitoring alarms.