Anomaly Detection Model Using Singular Value Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection models implemented by microcontrollers are complex and memory-intensive, making them difficult to implement and resource-constrained for real-time monitoring in various fields.
Innovation Solution
A method using singular value decomposition to generate a robust anomaly detection model with a maximum Mahalanobis distance threshold, which reduces memory usage and allows for efficient implementation on microcontrollers by projecting data onto a new base and calculating Mahalanobis distances, thereby defining normal operation limits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms are used to generate anomaly detection models, then detection accuracy is improved, but model complexity and memory requirements increase
Solution Approach 1:
The patent segments the anomaly detection model into distinct components: a training phase executed on a computer server that generates model parameters, and an execution phase on the microcontroller that uses pre-computed matrices (mean matrix, covariance matrix, Mahalanobis distance threshold). This segmentation allows complex computations to be performed offline while the embedded device uses only simple matrix operations, resolving the contradiction between detection accuracy and device complexity
Solution Approach 2:
The patent applies preliminary action by pre-computing all complex model parameters (mean matrix, covariance matrix, and Mahalanobis distance threshold) during an offline training phase using machine learning algorithms on a powerful computer server. These pre-computed parameters are then stored in the microcontroller's memory, enabling the embedded device to perform anomaly detection using only simple matrix operations without requiring complex real-time computations, thus maintaining high detection accuracy while minimizing device complexity
2Measurement precision
If complex models with numerous parameters are used, then detection accuracy is improved, but memory space consumption increases
Solution Approach 1:
The patent extracts and pre-computes all parameter-heavy operations during the offline training phase, extracting only the essential result matrices (mean matrix, covariance matrix, Mahalanobis distance threshold) that need to be stored in the microcontroller. This extraction eliminates the need to store and process the original complex model parameters, training data, and computation algorithms in the embedded device, significantly reducing memory space requirements while preserving detection accuracy
Solution Approach 2:
The patent transforms the anomaly detection problem by changing the parameters from raw sensor data to processed statistical parameters (mean vector, covariance matrix, Mahalanobis distance threshold). This parameter transformation allows the system to work with compact numerical representations that require minimal storage space while maintaining the ability to perform accurate anomaly detection through simple distance calculations
3Reliability
If real-time anomaly detection is implemented on microcontrollers, then system reliability is improved, but computational resources are overwhelmed
Solution Approach 1:
The patent segments the anomaly detection system into an offline model generation component and an online detection component. The complex model generation using machine learning algorithms is performed offline on a powerful computer server, while the microcontroller only executes simple real-time detection using pre-computed matrices. This segmentation enables real-time detection on resource-constrained devices without overwhelming processing resources
Solution Approach 2:
The patent substitutes complex real-time computational mechanics with simpler pre-computed mathematical operations. Instead of performing complex machine learning computations in real-time on the microcontroller, the system uses pre-computed mean matrices, covariance matrices, and Mahalanobis distance thresholds that require only simple matrix multiplication and comparison operations, dramatically reducing processing complexity while maintaining real-time detection capability
Data Source
AI summary
A method implemented by computer for generating a model for anomaly detection in a system includes obtaining a learning data matrix corresponding to a normal operation of the system, decomposing the learning data matrix into singular values of the matrix of learning data, calculating a new base, defining a maximum Mahalanobis distance threshold representing a limit of the normal operation of the system, and defining an anomaly detection model from the new base and from the maximum Mahalanobis distance threshold.