Anomaly Detection Using Sparse Decomposition for Time Series
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional analytics computing systems inaccurately and inefficiently identify anomalous values in time series data, particularly when dealing with multiple seasonal trends, missing or non-real values, and varying data types, leading to false positives and wasteful use of computing resources.
Innovation Solution
The system decomposes a metrics time series into latent components, such as spike and level components, using an optimization algorithm that accounts for significance thresholds and excludes non-real values, allowing for intelligent identification of significant anomalies without separating data into training and testing periods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional anomaly-detection algorithms are uniformly applied to all values in a time series, then the system can identify potential anomalies, but it produces false positives when encountering missing or non-real values
Solution Approach 1:
The patent applies different processing rules to different types of values within the time series. Real values undergo anomaly detection analysis, while missing or non-real values are handled separately through imputation or exclusion. This localized differentiation prevents uniform algorithm application from generating false positives, thereby improving both detection accuracy and reliability.
Solution Approach 2:
The time series data is segmented into distinct categories: real values, missing values, and non-real values. Each segment is processed according to its specific characteristics, with real values subjected to anomaly detection and other values handled through appropriate imputation or filtering methods, thus eliminating false positives while maintaining detection precision.
2Measurement precision
If the entire latent-component series is analyzed to identify anomalies, then comprehensive anomaly coverage is achieved, but computing resources are wasted on non-real and insignificant values
Solution Approach 1:
The patent extracts and removes non-real and insignificant values from the latent-component series before applying anomaly detection algorithms. By taking out these problematic values that would otherwise consume computing resources without contributing to meaningful anomaly identification, the system maintains comprehensive anomaly coverage while significantly improving computational efficiency.
Solution Approach 2:
Instead of applying anomaly detection to the entire latent-component series including all problematic values, the patent applies the detection algorithm only to the relevant subset of real values. This partial action approach avoids the excessive computation wasted on non-real values while still achieving complete anomaly identification among valid data points.
3Ease of manufacture
If a time series is separated into training and testing periods, then the anomaly-detection algorithm can be tuned, but anomalies may be missed when the training period does not represent the testing period
Solution Approach 1:
The patent implements an anomaly detection system that processes the entire time series uniformly without requiring separate training and testing periods. The algorithm is designed to be self-adaptive, handling multiple seasonal patterns and data characteristics through a single unified processing framework, thereby eliminating the representation mismatch problem while maintaining algorithm tuning capability through parameter optimization on the full dataset.
4Ease of operation
If anomaly-detection algorithms are rigidly applied regardless of data variations, then the system operates consistently, but it misidentifies seasonal variations, zero values, and non-real numbers as anomalies
Solution Approach 1:
The patent transforms the rigid, static anomaly detection approach into a dynamic system that adapts to varying data characteristics. The algorithm automatically adjusts its behavior based on the type of values encountered, applying appropriate processing rules for real values, missing values, and non-real values. This dynamic adaptation maintains operational consistency while eliminating misidentification of seasonal variations and other legitimate data patterns as anomalies.
Data Source
AI summary
The present disclosure describes systems, non-transitory computer-readable media, and methods for determining latent components of a metrics time series and identifying anomalous data within the metrics time series based on one or both of spikes/dips and level changes from the latent components satisfying significance thresholds. To identify such latent components, in some cases, the disclosed systems account for a range of value types by intelligently subjecting real values to a latent-component constraint for decomposing the time series and intelligently excluding non-real values from the latent-component constraint. The disclosed systems can further identify significant anomalous data values from latent components of the metrics time series by jointly determining whether one or both of a subseries of a spike-component series and a level change from a level-component series satisfy significance thresholds.