Time Synchronization Anomaly Detection Using Predicted Parameter Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing time synchronization systems face challenges in detecting abnormalities with high accuracy due to fluctuations in time synchronization parameters, which are affected by transmission periods and network delays, making it difficult to identify abnormalities on the order of microseconds.
Innovation Solution
A configuration involving a relay apparatus and a monitoring apparatus that utilize machine learning and cumulative distribution tables to calculate and compare abnormality degrees of time synchronization parameters, facilitating the detection of abnormalities by predicting future parameter values and analyzing prediction errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high accuracy time synchronization is implemented, then time synchronization accuracy is improved, but abnormality detection capability deteriorates due to parameter fluctuations masking microsecond-level abnormalities
Solution Approach 1:
The system performs preliminary learning of normal parameter fluctuation patterns during a learning phase before actual monitoring. By pre-establishing the baseline behavior of time synchronization parameters under normal conditions, the system can later detect deviations from this baseline as abnormalities, effectively separating normal fluctuations from true anomalies.
Solution Approach 2:
The system continuously monitors time synchronization parameters and compares them against learned normal patterns, providing feedback on abnormality degrees. This feedback mechanism enables real-time detection and notification of abnormalities while maintaining high-accuracy synchronization operations.
2Measurement precision
If machine learning is used to detect abnormalities, then abnormality detection accuracy is improved, but device complexity increases due to additional processing requirements
Solution Approach 1:
The patent extracts only the essential features needed for abnormality detection from the time synchronization parameters. Instead of implementing complex general-purpose machine learning systems, the invention focuses on learning and comparing specific parameter patterns (time synchronization parameters and their fluctuations) to detect abnormalities, thereby reducing system complexity while maintaining detection accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An inference unit (1033) predicts as a prediction parameter value, a future parameter value of a time synchronization parameter used for time synchronization calculation which is calculation for synchronizing times of two devices. When a measured parameter value which is a measured value corresponding to the prediction parameter value is generated with elapse of time, a cumulative distribution table generation unit (1034) calculates as a prediction error, a difference between the measured parameter value and the prediction value. A cumulative probability calculation unit (1036) calculates an abnormality degree indicating possibility that an abnormality is included in the measured parameter value, using the prediction error.