Erma Algorithm for Time Series Anomaly Detection and the Exploration of its Application in Financial Information System

The IT operations in financial industry has characteristics of wide variety of KPIs (Key Performance Indicators), large data volume, and high demand for business continuity and stability, bringing great challenges to business monitoring for finance enterprises. Conventional monitoring methods based on fixed thresholds are difficult to adapt to frequently changed business developments that abnormal false report and under-reports frequently occur, which leads to poor monitoring performance and huge manpower consumption. In view of the characteristics and current situation of the anomaly detection in above financial scene, this paper proposes a solution to apply the energy statistical algorithm (ERMA) to the financial scene for the first time, constructs a time series data set for anomaly detection with both the experience of business experts and algorithm experts are incorporated while bringing up the associated anomaly labeling criterion considering the difficulties of labeling in AIOps, and finally draw comparison between different intelligent algorithms of their performances of application in the financial industry. The results prove generally superior performance, high efficacy, and universality of Erma, guaranteeing the operational stability of IT system in financial industry, and exhibit capability of Erma for supporting real-time anomaly detection on data at million scale, which provides the possibility of developing a generic solution in the field.