Sensor Anomaly Detection Using Stable and Dynamic State Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection systems in IoT systems face challenges in distinguishing between normal dynamic behavior and anomalous behavior, leading to false positives and difficulty in obtaining training data for rare anomaly events, especially in systems with periods of stable and dynamic operation.
Innovation Solution
A method and system that utilizes a neural network to differentiate between stable and dynamic states using reconstruction errors from stable and dynamic models, allowing for accurate anomaly detection and automatic corrective actions during stable periods, while ignoring dynamic periods to avoid false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If anomaly detection systems continuously monitor system operation without distinguishing between stable and dynamic states, then anomaly detection coverage is maximized, but false positives increase due to inability to distinguish normal dynamic behavior from anomalies
Solution Approach 1:
The patent segments the operational state space into distinct stable and dynamic states using separate neural network models. The stable model processes data during stable states while the dynamic model processes data during dynamic states, allowing the system to distinguish between normal dynamic behavior and actual anomalies by comparing reconstruction errors from both models.
Solution Approach 2:
The patent implements dynamic adaptation by switching between different neural network models based on the current operational state. The system dynamically determines which model to use (stable model or dynamic model) based on reconstruction error comparisons, enabling it to adapt its detection behavior to the current system state and reduce false positives during dynamic transitions.
2Measurement precision
If the system uses a single unified model for anomaly detection, then model complexity is minimized, but the system cannot accurately distinguish between stable and dynamic operational states leading to false positives
Solution Approach 1:
The patent divides the anomaly detection function into separate specialized models: a stable model for detecting anomalies during stable states and a dynamic model for detecting anomalies during dynamic states. Each model is trained on specific state data, enabling precise state differentiation through comparison of reconstruction errors from both models.
Solution Approach 2:
The patent applies local quality by training different neural network models with specialized characteristics for different operational states. The stable model is optimized for stable state data while the dynamic model is optimized for dynamic state data, allowing each model to exhibit the appropriate detection sensitivity and specificity for its target state.
3Productivity
If the system performs corrective actions continuously, then system responsiveness is maximized, but unnecessary corrective actions are triggered during dynamic states causing operational disruption
Solution Approach 1:
The patent performs preliminary state classification by comparing reconstruction errors from the stable model and dynamic model before executing any corrective actions. This preliminary determination of whether the system is in a stable or dynamic state prevents premature or inappropriate corrective actions from being executed during dynamic transitions, while still maintaining rapid response during stable states.
Solution Approach 2:
The patent implements feedback control by continuously monitoring reconstruction errors from both models and using this feedback to dynamically determine whether corrective actions should be executed. The system feeds back the comparison results of reconstruction errors to control the corrective action logic, ensuring actions are only taken when the system is in a stable state and truly requires intervention.
Data Source
AI summary
Methods and systems for anomaly detection include determining whether a system is in a stable state or a dynamic state based on input data from one or more sensors in the system, using reconstruction errors from a respective stable model and dynamic model. It is determined that the input data represents anomalous operation of the system, responsive to a determination that the system is in a stable state, using the reconstruction errors. A corrective operation is performed on the system responsive to a determination that the input data represents anomalous operation of the system.