Randomized Learning Set for Outlier Detection in Communication Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional outlier detection methods rely on threshold values, which require complex determination and often filter out normal system state measurements, leading to incorrect evaluations of system states.
Innovation Solution
A randomized method is used to form a modified learning set by removing or weighting unmarked measured values, allowing for the inclusion of valuable data points and preventing incorrect evaluations by using a score function, transformation function, and weighting function to assess the probability of system states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If threshold values are used for outlier detection, then abnormal measured values can be identified, but normal system state measurements with large deviations are incorrectly filtered out
Solution Approach 1:
The patent changes the parameter used for outlier detection from fixed threshold values to dynamic percentile-based thresholds. By calculating thresholds based on the distribution of measured values (e.g., 1st percentile, 5th percentile), the system adapts to the actual data distribution and avoids filtering out normal measurements with large deviations. This resolves the contradiction by making the threshold parameter flexible rather than fixed.
Solution Approach 2:
The system performs self-service by automatically calculating thresholds and evaluating system states without requiring external intervention or complex manual threshold determination. The learning system autonomously adapts to the measured values and adjusts the thresholds accordingly, eliminating the need for complex external tests and evaluations while maintaining detection accuracy.
2Measurement precision
If complex tests and evaluations are used to determine threshold values, then outlier detection accuracy improves, but system complexity increases
Solution Approach 1:
The system eliminates the need for complex external tests and evaluations by implementing self-service threshold determination. The learning system automatically calculates thresholds based on the percentile distribution of measured values, performing the determination process internally without requiring complex external intervention. This reduces system complexity while maintaining precision.
Solution Approach 2:
The patent introduces percentile-based thresholds as an intermediary mechanism between raw measured values and outlier detection decisions. Instead of directly applying complex tests, the system uses the percentile distribution as an intermediate step to automatically determine appropriate thresholds, simplifying the overall process while maintaining accuracy.
3Reliability
If threshold-based outlier detection is used, then abnormal values can be identified, but the learning set quality deteriorates due to inclusion of faulty measurements
Solution Approach 1:
The patent implements feedback by using the learning system to evaluate system states based on measured values and threshold comparisons. The results of these evaluations feed back into the learning process, allowing the system to continuously improve its model of normal system behavior. This feedback mechanism ensures that only high-quality measurements are used to update the learning set, improving learning set quality while maintaining outlier identification capability.
Solution Approach 2:
The system changes the parameter for selecting learning set members from fixed thresholds to dynamic percentile-based thresholds. By using percentiles (e.g., including measurements above the 1st or 5th percentile), the system dynamically adjusts the criteria for inclusion based on the distribution of measured values, ensuring that faulty measurements are excluded while maintaining high learning set quality.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for evaluating measured values of a system S, which may be in a fault-free or faulty state, wherein the system S comprises at least one communication network, a network component of a communication system, and/or a service of a communication network, comprising the following steps, preferably in the following order: forming a set V of unlabeled measured values v of the system S; forming a modified learning set V' with measured values v' for a learning system L by removing and/or weighting measured values from the set V using a random method; forming a model M for evaluating measured values of the system S by the learning system L from the modified learning set V'; and evaluating measured values of the system S by an evaluation system B using the model M.Furthermore, the invention relates to a system for evaluating measured values of a system S, which may be in a fault-free or faulty state, wherein the system S comprises at least one communication network, a network component of a communication system, and/or a service of a communication network, comprising: a device for generating a set V of unlabeled measured values v of the system S; a device for generating a modified learning set V' with measured values v' for a learning system L by removing and/or weighting measured values from the set V using a random method; learning system L suitable for generating a model M for evaluating measured values of the system S from the modified learning set V'; and evaluation system B suitable for evaluating measured values of the system S using the model M.