Time sequence reusable anomaly detection method and system for service data

Abstract

The invention provides a method and a system for detecting a reusable anomaly of a time sequence of business data, and the method comprises the steps: 1, obtaining business historical data, and detecting the business data according to the stability, periodicity, correlation and time granularity of the data; 2, matching a proper time sequence algorithm for the service data according to the detection result to form a mapping relation; 3, storing the mapping relation between the time sequence and the algorithm and parameters in a sample library; 4, acquiring real-time data of the same service, predicting the service data according to the mapping algorithm and the optimal parameter in the sample library in the step 3, and determining a prediction interval according to a prediction result; and step 5, comparing the relationship between the actual value and the prediction interval of the same time period, and giving an alarm when the prediction range is continuously exceeded for more than 10 times. The dynamic threshold value can be monitored in real time, excessive human participation is not needed, false alarm and missing alarm of alarm are reduced, and the labor cost is effectively saved.

Description

Technical field [0001] The present invention relates to the field of big data anomaly detection methods, specifically, to a time series of business data may be reused anomaly detection methods and systems. Background [0002] In traditional monitoring operations, static thresholds are usually set according to expert experience for monitoring and alarming. The threshold is too low, the alarm is too frequent, and the operation and maintenance personnel need to spend a lot of time to determine whether the system is really a problem, and over time, the "wolf is coming" phenomenon is highlighted, and the significance of the alarm is lost. The threshold is too high, the quality hazard is difficult to find, resulting in the omission of many performance failures in the early stage, and when the problem is serious, perhaps the user has perceived the system abnormality, but the operation and maintenance engineers do not know anything. [0003] In anomaly detection, time series analysis is ...

AI Technical Summary

Problems solved by technology

[0005] The patent document "A Time Series Prediction Method Based on Quantitative Screening Time Series Prediction Model" with the publication number CN104866930A solves the problem of low quantitative screening accuracy of existing time series forecasting models

Images