Anomaly Detection in Periodic Data via Residual Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods fail to effectively detect anomalies in periodic or semi-periodic data in real-time, especially when data exhibits multiple periodicities or variations that are not constant over time, leading to missed alerts and inefficient system responses.
Innovation Solution
A computerized method using Power Spectrum or Phase Dispersion Minimalization techniques to find and fit periodic signals, subtract them from data, and generate residual points to identify anomalies, with real-time alerts triggered by significant deviations, applicable to various data types including satellite, GPS, and medical data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional anomaly detection methods are used on periodic data, then simple detection can be achieved, but anomalies in complex periodic data with multiple periodicities cannot be effectively detected
Solution Approach 1:
The patent segments the complex periodic data into multiple periodic components, each with its own period and amplitude. By decomposing the data into separate periodic signals that can be individually analyzed, the system achieves accurate anomaly detection without being overwhelmed by the complexity of the overall signal structure.
Solution Approach 2:
The patent applies partial action by focusing on detecting anomalies relative to each individual periodic component rather than attempting to analyze the entire complex signal at once. This allows the system to identify deviations from expected periodic behavior without requiring complete understanding of all periodic interactions simultaneously.
2Reliability
If real-time anomaly detection is implemented, then timely system responses can be triggered, but false alerts increase due to cumulative probability of random anomalies
Solution Approach 1:
The patent implements feedback by continuously comparing new data points against the established periodic model and using the residual analysis to adjust anomaly detection thresholds. The system learns from patterns in the residual data and adjusts its sensitivity accordingly, reducing false alerts while maintaining reliable detection of genuine anomalies.
Solution Approach 2:
The patent performs preliminary action by establishing a baseline periodic model from historical data before attempting real-time anomaly detection. This pre-characterization of normal periodic behavior allows the system to distinguish between expected variations and genuine anomalies, reducing false alerts while enabling timely responses to real issues.
3Measurement precision
If periodic signals are fitted to data, then periodic variations can be accounted for, but anomalies hidden within periodic patterns remain undetected
Solution Approach 1:
The patent extracts the periodic components from the data by fitting mathematical models to represent the expected periodic behavior. By separating and removing these predictable periodic signals from the overall data stream, the system isolates the residual variations that contain the anomalies, making them visible and detectable despite being hidden within the periodic patterns.
Data Source
Figure 1
Figure 2~3A
Figure 3B~3C
AI summary
A computerized system for monitoring physical data for anomalies, which physical data are predictable given predetermined information, the system comprising a predetermined information data point generator operative to compute a sequence of model data points which the physical data, given predetermined information, can be expected to duplicate at each of a corresponding sequence of temporal sampling points; a wayward point monitor including a processor operative for monitoring the physical data including identifying wayward points within said physical data that are incongruous with the predetermined information; and an anomalous episode-prompted alarm generator operative for identifying anomalous episodes, each including a cluster of wayward points satisfying predefined anomalous episode-defining criteria and generating an alarm for each anomalous episode identified.