System Health Assessment Model Based on Anomaly Detection

Abstract

Accurately assessing system health and taking timely interventions play a crucial role in promoting intelligent system operation. Traditional system health assessment techniques depend on human-set thresholds, which are only appropriate for tracking a few indicators and do not accurately represent the system’s health overall. Therefore, keeping track of a large amount of data and a diverse variety of indicators will be extremely challenging. To address this issue, we develop the AD_Select anomaly detection algorithm and suggest a system health assessment model based on indicator anomaly detection. By fusing anomaly detection with system health assessment, this algorithm increases the scientific rigor and precision of health assessment. The experiments of AD_Select anomaly detection algorithm and healthiness assessment model are conducted on the 2018AIOps dataset and industrial dataset respectively, and the results show that both the AD_Select algorithm and the healthiness assessment model proposed in this paper achieve good performance.