Global Change Point Detection in Multi-Time Series Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques are inadequate for detecting overall changes in large-scale systems represented by multiple time series, as they can only detect change points for individual time series, not simultaneously changing time points across multiple series.
Innovation Solution
An information processing system that learns time series models for each time series, divided into segments at change point candidates, and detects global change points by comparing parameters between segments, identifying simultaneous changes across multiple time series.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical methods based on regression analysis are used to approximate observed time series to detect change points, then measurement precision of individual change points is improved, but the ability to detect overall changes in large-scale systems with multiple time series deteriorates
Solution Approach 1:
The patent combines multiple individual time series into a composite time series by summing their values at each time point. This merging approach enables the detection of overall system changes while preserving the ability to detect individual change points, as the composite series reflects collective behavior across all monitored time series.
Solution Approach 2:
The patent introduces a new dimension by creating a composite time series that aggregates information from multiple individual time series. This dimensional transformation allows simultaneous detection of both individual and overall changes, as the composite series provides a macro-level view while individual series maintain micro-level details.
2Speed
If change point detection is performed for each time series independently, then detection speed for individual series is improved, but the ability to identify simultaneous changes across multiple series deteriorates
Solution Approach 1:
By merging multiple time series into a composite series, the patent enables simultaneous detection of changes across all series. The composite series captures collective behavior, allowing identification of simultaneous changes without sacrificing the speed benefits of independent analysis, as the merging operation can be performed efficiently.
Solution Approach 2:
The patent performs preliminary merging of time series into a composite series before change point detection. This preliminary action preserves simultaneous change information that would be lost in independent analysis, while the subsequent detection process maintains computational efficiency by working with the aggregated data structure.
3Reliability
If the observed time series is approximated by a smooth time series using regression analysis, then reliability of individual change point detection is improved, but the complexity of analyzing multiple time series simultaneously increases
Solution Approach 1:
The patent reduces analysis complexity by merging multiple time series into a single composite series. This approach maintains the reliability of regression-based change point detection while simplifying the overall analysis process, as a single regression model is fitted to the composite series rather than multiple separate models.
Solution Approach 2:
The composite time series serves multiple functions: it captures overall system behavior for detecting global changes, while also preserving individual series characteristics through the summation operation. This universal structure enables both individual and collective change detection using a unified analytical approach.
Data Source
AI summary
Change points of a system represented by a plurality of time series are detected more appropriately. An information processing system includes means for learning, with respect to each of a plurality of time series, models that approximate partial time series respectively and are defined by parameters of the partial time series respectively, the partial time series being obtained by dividing a corresponding time series into a plurality of segments at change point candidates; and means for detecting, with respect to each of the change point candidates for the plurality of time series, a global change point that is a change point for the plurality of time series based on a difference between a parameter of a first partial time series starting from a time point of a corresponding change point candidate and a parameter of a second partial time series before the corresponding change point candidate, and outputting the global change point.