Target System Anomaly Control via Partitioned Matrix Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Anomaly detection in complex target systems like communications networks or industrial processes is computationally heavy and memory-consuming due to the large volume of monitored variables, making efficient analysis challenging.
Innovation Solution
A method involving partitioning a matrix of monitored variables into submatrices, standardizing them with calculated constants, and performing anomaly detection and aggregation to reduce memory usage, using methods like robust principal component analysis and max function aggregation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If anomaly detection is performed on large volume of monitored variables, then detection accuracy is improved, but memory consumption and computational load increase
Solution Approach 1:
The patent divides the large matrix of monitored variables into multiple submatrices based on dimensions such as time, location, or device type. Each submatrix is processed separately through standardization and anomaly detection, then results are aggregated. This segmentation reduces the memory required to hold any single data structure and allows processing of large-scale monitoring data with limited memory resources while maintaining comprehensive anomaly detection coverage.
2Measurement precision
If anomaly detection is performed on large volume of monitored variables, then detection accuracy is improved, but computational load increases
Solution Approach 1:
By segmenting the monitoring data into submatrices, the patent reduces the computational complexity of each individual anomaly detection operation. Instead of computing anomaly scores across the entire large matrix simultaneously, the system performs standardized anomaly detection on smaller submatrices separately, then aggregates results. This division of computational work reduces the peak computational load and memory requirements while maintaining comprehensive anomaly detection capability across all monitored variables.
3Reliability
If matrix of monitored variables is processed as a whole, then anomaly detection comprehensiveness is improved, but processing efficiency decreases
Solution Approach 1:
The patent segments the comprehensive monitoring data into dimension-based submatrices (e.g., by time period, location, or device type) and processes each submatrix through standardization and anomaly detection independently. The results are then aggregated to provide comprehensive anomaly detection coverage. This approach maintains the thoroughness of whole-matrix analysis while improving processing efficiency through parallelizable, smaller-scale operations that reduce computational bottlenecks.
4Measurement precision
If standardization is performed on entire matrix, then data normalization accuracy is improved, but memory usage increases
Solution Approach 1:
Instead of standardizing the entire large matrix of monitored variables at once, the patent applies standardization separately to each submatrix after partitioning. This allows the system to compute mean and standard deviation for each submatrix independently, requiring only enough memory to hold one submatrix and its statistical parameters at a time. This segmented standardization approach maintains data normalization accuracy while dramatically reducing peak memory usage compared to whole-matrix standardization.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A computer implemented method for controlling a target system. The method is implemented by obtaining a first matrix (310) comprising a plurality of monitored variables as columns and a plurality of observations of the monitored variables measured from the target system as rows, wherein each row is associated with one or more dimensions; partitioning (402) the first matrix into a plurality of submatrices (412-416), each submatrix comprising a subset of rows of the first matrix; separately processing the submatrices by standardizing (403) values of the submatrix (412- 416) to obtain standardized submatrix (422-426); and processing (404) the standardized submatrix (422-426) by performing anomaly detection on the values of the standardized submatrix and by aggregating results of the anomaly detection by the respective one or more dimensions to a result matrix (430); and outputting the result matrix (430) or information derived from the result matrix (430) for the purpose of controlling the target system.