Multivariate Time-Series Anomaly Detection With Corrective Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex systems with multiple sensors generate vast time series data, making it challenging to identify anomalous behavior and determine the source of anomalies, leading to difficulties in monitoring and correcting such behavior.
Innovation Solution
A method using convolutional neural networks to encode original signature matrices, convolutional long-short term memory networks to model temporal patterns, and deconvolutional neural networks to decode and detect anomalies, with a loss function determining the occurrence of anomalies and a controller performing corrective actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple sensors are deployed to monitor complex systems, then monitoring coverage is improved, but data complexity and difficulty of anomaly identification increase
Solution Approach 1:
The patent segments the complex multivariate time series data into individual univariate time series from each sensor, then processes each series independently through separate LSTM networks. This segmentation transforms the complex joint analysis problem into multiple simpler independent analysis problems, reducing the computational complexity while maintaining comprehensive monitoring coverage across all sensors.
Solution Approach 2:
The patent transforms the complex multivariate anomaly detection problem into an unsupervised learning framework that operates in a different dimensional space by using reconstruction error as the anomaly metric. Instead of directly analyzing the high-dimensional multivariate relationships, the system projects each univariate series through an autoencoder and measures deviation from the reconstructed pattern, effectively changing the problem dimension from multivariate relationship analysis to univariate reconstruction fidelity.
2Ease of manufacture
If traditional anomaly detection methods are used, then implementation simplicity is maintained, but false alarm rates increase and true detection rates decrease
Solution Approach 1:
The patent implements feedback through the autoencoder architecture where the reconstructed output is continuously compared with the original input to compute reconstruction error. This feedback mechanism allows the system to learn normal patterns during training and then use the learned patterns to detect anomalies by measuring deviation from expected behavior, significantly improving detection accuracy while maintaining implementation simplicity through the standardized autoencoder framework.
Solution Approach 2:
The unsupervised learning approach enables the system to service itself by automatically learning normal operational patterns from training data without requiring manual labeling or threshold configuration. The autoencoder autonomously identifies anomalies through reconstruction error computation, eliminating the need for external expert intervention in anomaly definition and reducing false alarms while improving true detection rates.
3Measurement precision
If unsupervised learning is implemented for anomaly detection, then false alarms are reduced and true detection rates improve, but computational complexity increases
Solution Approach 1:
The patent segments the computational workload by processing each sensor's time series independently through separate LSTM autoencoders rather than jointly processing all multivariate data together. This segmentation divides the large-scale computational problem into multiple smaller independent problems, reducing memory requirements and enabling parallel processing, thereby improving anomaly detection accuracy through dedicated univariate analysis while managing computational complexity through distributed processing.
Solution Approach 2:
The patent applies partial action by focusing computational resources on detecting anomalies in individual univariate series rather than attempting to model all multivariate relationships simultaneously. Each autoencoder performs excessive analysis on its specific sensor data by learning comprehensive temporal patterns through LSTM, ensuring high detection accuracy for that sensor while the overall system manages complexity by processing sensors independently and aggregating results.
Data Source
AI summary
Methods and systems for anomaly detection and correction include generating original signature matrices that represent a state of a system of multiple time series. The original signature matrices are encoded using convolutional neural networks. Temporal patterns in the encoded signature matrices are modeled using convolutional long-short term memory neural networks for each respective convolutional neural network. The modeled signature matrices using deconvolutional neural networks. An occurrence of an anomaly is determined using a loss function based on a difference between the decoded signature matrices and the original signature matrices. A corrective action is performed responsive to the determination of the occurrence of the anomaly.