Time-Series Anomaly Detection Using Temporal Pattern Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques fail to effectively distinguish and extract patterns related to behavior with temporal constancy from those without, leading to patterns being mixed or not detected correctly, especially in time-series data analysis.
Innovation Solution
An information processing device constructs second data by stacking first data with respect to time, extracts a constant pattern with temporal constancy, and generates a difference to isolate a random pattern without temporal constancy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If matrix decomposition techniques (PCA or SVD) are used to detect main patterns in time-series data, then the detection process can identify frequent behavior patterns, but the behavior having temporal constancy and random behavior cannot be distinguished from each other, and random behavior may be regarded as noise and not detected
Solution Approach 1:
The patent segments time-series data into multiple time periods and performs separate analysis for each period. By dividing the data chronologically, the system can identify patterns that persist across periods (temporal constancy) versus those that appear randomly. This segmentation allows distinct treatment of systematic patterns and random variations, resolving the contradiction between detecting main patterns and preserving information about temporal constancy versus randomness.
Solution Approach 2:
The patent introduces a temporal period dimension to the analysis by comparing patterns across different time periods. This additional dimensional perspective allows the system to distinguish between patterns with temporal constancy (consistent across periods) and random behavior (inconsistent across periods). The dimensional transformation from single-period to multi-period analysis enables differentiation that was not possible with traditional matrix decomposition alone.
2Quantity of substance
If there are a large number of patterns which are as frequent as the behavior having temporal constancy, then the matrix decomposition regards the random behavior as a noise, but the random behavior may not be detected as a pattern
Solution Approach 1:
By segmenting data into multiple time periods and analyzing each separately, the system can identify random behaviors that occur in specific periods without being overwhelmed by frequent patterns from other periods. This temporal segmentation allows random behaviors to stand out as anomalies within their specific time contexts, improving detection accuracy even when many frequent patterns exist.
Solution Approach 2:
The patent applies different analysis approaches to different time periods, treating each period's patterns with local consideration. Random behaviors are detected by comparing local period-specific patterns against expected temporal constancy, allowing precise identification of random events even in the presence of numerous frequent patterns elsewhere in the dataset.
3Measurement precision
If tensor decomposition is applied to tensor generated by periodic stacking to detect periodic pattern, then a periodic pattern can be detected, but the calculation has a large calculation amount and result values are not converged into one value
Solution Approach 1:
Instead of applying complex tensor decomposition to the entire dataset at once, the patent segments the analysis into multiple time periods and processes each period separately. This segmentation reduces the computational burden on each analysis step and improves convergence by focusing on smaller, more manageable time segments, while still achieving comprehensive periodic pattern detection across the full dataset.
Solution Approach 2:
The patent performs partial decomposition analyses on segmented time periods rather than attempting complete decomposition of the entire dataset. This partial action approach reduces calculation complexity and improves convergence by breaking down the excessive computational task into manageable partial analyses that can be combined to achieve the overall detection goal.
Data Source
AI summary
An information processing device according to the present invention includes: a memory; and at least one processor coupled to the memory. The processor performing operations. The operations includes: constructing second data that is acquired, based on first data containing a plurality of observation values in a plurality of times, by stacking the first data with respect to the times, and extracting a constant pattern that is a combination of the observation values having temporal constancy in the first data, based on the second data; generating a difference between the first data and the constant pattern in the time; and extracting a random pattern that is a combination of the observation values without temporal constancy, based on the difference.