Sensor Relationship Model Evaluation for Noise-Robust Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing system model evaluation systems face challenges in achieving high accuracy for abnormality detection due to noise and limitations in selecting inter-sensor-value relationships, which affects the reliability of system models used in operation management.
Innovation Solution
A system model evaluation system that creates candidates of system models by varying the pattern of selecting inter-sensor-value relationships and evaluates these candidates using predetermined evaluation data, thereby improving prediction accuracy and reducing noise impact.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a system model is created using sensor values to represent inter-sensor-value relationships, then the model can capture system behavior patterns, but the model accuracy deteriorates due to noise in sensor data and limitations in selecting relationships
Solution Approach 1:
The patent segments the system model creation process into multiple candidate models, each representing different patterns of inter-sensor-value relationships. By dividing the single model approach into multiple segmented candidates with different relationship patterns, the system can evaluate and select the most accurate model while reducing the impact of noise in any single model.
Solution Approach 2:
The patent changes the parameters of the system model by varying the pattern of selecting inter-sensor-value relationships. Different candidate models use different selection patterns (parameters) for establishing relationships between sensor values, allowing the system to find the optimal parameter configuration that maximizes accuracy while minimizing noise impact.
2Measurement precision
If multiple candidate system models are created by varying selection patterns, then model evaluation accuracy improves, but the complexity of the system model creation process increases
Solution Approach 1:
The patent introduces dynamics into the model creation process by automatically generating multiple candidate models with varying selection patterns and dynamically evaluating them. The system adaptively creates and assesses different model configurations, selecting the best performing model based on evaluation results, which manages complexity through automated dynamic processes rather than static manual configuration.
Solution Approach 2:
The system performs self-service by automatically evaluating multiple candidate models and selecting the optimal one without requiring extensive manual intervention. The evaluation process autonomously assesses each candidate model's accuracy and reliability, reducing the operational complexity for users while maintaining high evaluation standards.
3Reliability
If sensor values are used to create inter-sensor-value relationships, then the system model can be built from available data, but the reliability deteriorates due to noise in the sensor values
Solution Approach 1:
The patent implements feedback by evaluating candidate models against evaluation data and using the evaluation results to select the most reliable model. The system feeds back the performance metrics of each candidate model and uses this information to determine which model to deploy, creating a closed-loop process that improves reliability by selecting models that have demonstrated accuracy in evaluating data.
Solution Approach 2:
The patent performs preliminary action by creating and evaluating multiple candidate models before final deployment. The system预先 (in advance) generates various model configurations, evaluates their performance on evaluation data, and selects the best model beforehand, ensuring reliability is established before the model is used for actual system monitoring and anomaly detection.
Data Source
AI summary
There is provided a system model evaluation system including a system model candidate creation part configured to create a candidate(s) of a system model by changing a pattern of selecting an inter-sensor-value relationship created by using sensor values acquired from sensors arranged in a system to which the system model is directed. This system model evaluation system further includes a system model evaluation part configured to evaluate the candidate(s) of the system model by inputting predetermined evaluation data to the created candidate(s) of the system model.