Telecom KPI Anomaly Detection via Dynamic Statistical Profiles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for identifying anomalies in performance indicators of telecommunications systems are inefficient in accurately and reliably reporting anomalies, as what constitutes an anomaly can vary over time and are difficult to consistently detect.
Innovation Solution
A system and method that classify performance indicators using statistical profiles, such as distribution, behavioral, or unidentified profiles, to determine anomalies by correlating data series and calculating noise-to-signal ratios, allowing for real-time identification and reporting of anomalous data points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods compare data points to averaged performance indicators within a fixed threshold, then the detection process is simple, but the accuracy and reliability of anomaly reporting deteriorates because what constitutes an anomaly varies over time
Solution Approach 1:
The patent applies dynamics by transitioning from static threshold-based anomaly detection to dynamic statistical profile-based detection. The system learns and adapts statistical profiles for each performance indicator over time, allowing anomaly thresholds to dynamically adjust according to actual system behavior patterns rather than using fixed predetermined thresholds. This resolves the contradiction by making the detection method adaptive while maintaining reasonable complexity through automated statistical learning.
Solution Approach 2:
The system implements self-service through automated statistical profile learning and anomaly detection. Instead of requiring manual configuration of anomaly thresholds or expert intervention to define what constitutes an anomaly, the system automatically learns the statistical characteristics of performance indicators and autonomously identifies anomalies. This self-organizing capability improves detection accuracy without proportionally increasing operational complexity.
2Adaptability or versatility
If the system uses fixed thresholds for anomaly detection, then the implementation is straightforward, but the adaptability to changing system conditions deteriorates
Solution Approach 1:
The patent implements feedback mechanisms where the system continuously monitors performance indicators, compares actual values against learned statistical profiles, and uses the results to refine anomaly detection. The feedback loop allows the system to adapt to changing conditions by updating its understanding of normal versus anomalous behavior patterns, thereby maintaining both adaptability to new conditions and reliability through consistent statistical evaluation criteria.
Solution Approach 2:
The system applies parameter changes by transitioning from fixed threshold parameters to dynamic statistical parameters. Instead of using static anomaly thresholds, the system employs statistical profiles with parameters such as mean, standard deviation, and confidence intervals that automatically adjust based on observed system behavior. This allows the anomaly detection parameters to change adaptively while maintaining reliable detection through statistically sound methods.
3Productivity
If conventional methods use simple averaging for performance indicators, then the computational load is low, but the ability to accurately identify anomalies deteriorates due to varying anomaly definitions over time
Solution Approach 1:
The system applies preliminary action by performing statistical profile learning during normal operation before anomalies occur. The system continuously builds and updates statistical profiles of performance indicators during standard operations, so that when anomalies do occur, the detection can immediately leverage the pre-established understanding of normal behavior patterns. This preliminary learning phase enables accurate anomaly identification without adding significant computational burden during critical detection moments.
Data Source
AI summary
A system and method for identifying anomalies in indicators, such as key performance indicators (KPIs) of a telecom system are disclosed. The method can learn over time behavior of the indicator and can statistically identify what should be considered anomalous. Learning can be performed on a per indicator basis that each presents different statistical qualities. The method can associate the indicator to a profile, such as one of several statistical distributions and can operate accordingly. Association may be determined by the correlation of the indicator to statistical distribution. The method can identify correlations between indicators when identifying the statistical distribution and especially when the associated statistical distribution is an unidentified profile. The method can include comparison of actuals versus prediction and sending alerts when anomalies are found. The system can be configured to receive data points respective of indicators and implement the method while continuously determining data points constituting anomalies.