Statistical Model Generation for Time Series Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems management technologies face challenges in efficiently detecting and diagnosing problems in complex data processing systems, particularly due to manual encoding, limited adaptability to unknown anomalies, and reliance on manual analysis of time series data.
Innovation Solution
A method that involves capturing snapshots of time series data during events, generating statistical models based on feature vectors, and comparing these models to identify similar events in the future, thereby automating the detection of recurring issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual encoding and analysis of time series data are used for problem detection, then systems management can identify known issues, but the process requires significant manual effort and cannot adapt to unknown anomalies
Solution Approach 1:
The system automatically generates statistical models from time series data snapshots without requiring manual encoding or analysis. The performance monitoring subsystem autonomously extracts feature vectors, generates statistical models representing events, and compares these models to detect similar events, eliminating the need for manual intervention while maintaining reliable problem detection
Solution Approach 2:
The patent replaces manual mechanical analysis processes with automated statistical modeling and pattern recognition systems. Instead of manual encoding of problem patterns, the system uses statistical models generated from data to automatically detect and diagnose issues, substituting human analysis with computational methods
2Adaptability or versatility
If manual analysis methods are used to detect problems in complex systems, then existing issues can be identified, but the system cannot adapt to unknown anomalies or recurring issues
Solution Approach 1:
The system performs preliminary actions by capturing snapshots of time series data during events and generating statistical models in advance. These pre-generated models serve as reference patterns that enable the system to adaptively recognize and detect similar unknown anomalies when they occur, without requiring real-time manual analysis
Solution Approach 2:
The patent transforms time series data into statistical models by changing the parameter representation from raw data points to extracted feature vectors and statistical characteristics. This parameter transformation enables the system to adapt to various types of anomalies by comparing fundamental statistical patterns rather than requiring explicit knowledge of each specific anomaly type
3Measurement precision
If comprehensive time series data analysis is performed to detect problems, then accurate diagnosis can be achieved, but the process is time-consuming and reduces productivity
Solution Approach 1:
The system extracts essential feature vectors from comprehensive time series data snapshots, isolating the most relevant characteristics that define events. This extraction process maintains diagnostic accuracy by capturing key patterns while reducing the data volume that requires analysis, thereby improving detection efficiency
Solution Approach 2:
The patent creates statistical model copies representing events from time series data. These model copies serve as efficient representations that can be rapidly compared against new data to detect similar events, maintaining measurement precision while significantly reducing the time required for analysis compared to examining raw comprehensive data
Data Source
AI summary
Software that generates statistical models of events impacting computer systems and uses those models to detect similar events in the future. The software performs the following operations: (i) receiving a snapshot of a first event impacting a computer system, where the snapshot includes a first set of values for a plurality of metrics occurring over a first time period corresponding to the first event; (ii) extracting a first set of feature vectors from the first set of values; (iii) generating a first statistical model representing the first event based, at least in part, on the extracted first set of feature vectors; and (iv) determining that a second event is similar to the first event by comparing the first statistical model to a second set of values for the plurality of metrics occurring over a second time period corresponding to the second event.