Incremental PCA Anomaly Detection for Memory-Limited Microcontrollers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods for microcontrollers in physical systems are inefficient, requiring significant memory and computational resources, and are not well-suited for applications where training data is scarce, especially when implementing incremental anomaly detection.
Innovation Solution
A method involving incremental calculation of principal components and orthonormalization using a modified Gram-Schmidt algorithm, combined with a detection threshold, to develop a lightweight anomaly detection model that can be executed by a microcontroller, allowing for real-time classification of normal and abnormal data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional anomaly detection methods (Gaussian Mixture model, Z-score) are used, then anomaly detection capability is achieved, but memory usage and computational complexity increase significantly
Solution Approach 1:
The patent extracts only the essential features needed for anomaly detection by using Principal Component Analysis to identify the most significant variance directions in the data. Instead of storing or processing all training data, only the top k principal components are retained, dramatically reducing memory requirements while preserving the most important patterns for detecting anomalies.
Solution Approach 2:
The patent transforms the data representation by changing from raw feature space to principal component space. This parameter transformation allows the system to work with a reduced set of orthogonal features that capture the essential variability, reducing computational complexity for subsequent anomaly detection operations while maintaining detection accuracy.
2Speed
If incremental anomaly detection is implemented with limited memory, then real-time monitoring is enabled, but model development efficiency decreases
Solution Approach 1:
The patent performs preliminary dimensionality reduction by calculating principal components incrementally as data arrives. By pre-processing the data stream to extract dominant variance patterns before anomaly detection, the system enables real-time monitoring with reduced computational load, balancing the trade-off between real-time capability and model development efficiency.
3Measurement precision
If more training data is collected for model development, then model accuracy improves, but memory requirements and processing time increase
Solution Approach 1:
The patent extracts the most informative aspects of the training data through Principal Component Analysis, retaining only the top k principal components that capture the majority of variance. This extraction process allows the model to achieve good accuracy with significantly reduced memory requirements, as only the essential patterns are stored rather than the complete training dataset.
Solution Approach 2:
The patent changes the parameter representation from raw data features to principal component scores. This transformation compresses the information content while preserving the most significant variations, enabling accurate anomaly detection with reduced memory footprint by working in a lower-dimensional transformed space.
Data Source
Figure 1~3
Figure 4
AI summary
According to one aspect, a computer-implemented method for developing an anomaly detection model is proposed, the method comprising: - obtaining (30) a training data stream, - incremental calculation (31) of the principal components of the training data stream, - orthonormalization (32) of the principal components of the training data stream so as to obtain an orthonormal basis representing the training data stream, - development (33) of an anomaly detection model including said orthonormal basis and a detection threshold defined by the user.