Local Network Quality Modeling for Accurate Abnormality Alarms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Network analysis devices face challenges in distinguishing between different abnormal states in communication networks, leading to incorrect alarm notifications and adjustments in abnormality determination thresholds, which can result in missed or false alarms.
Innovation Solution
The implementation of a network analysis program that performs local modeling analysis using a local model based on previous training data to estimate current network quality and determine abnormalities, while also considering the distribution of connections and feedback from operators to adjust the abnormality determination threshold dynamically.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the abnormality determination threshold is set based on average packet loss rates from aggregated network analysis data across multiple connections, then the device can perform collective analysis to reduce monitoring costs, but when the number of packets decreases during out-of-service periods, the average packet loss rate increases artificially causing exaggerated alarms
Solution Approach 1:
The patent segments the network analysis data by dividing connections into multiple connection groups based on communication node groups (e.g., bases). Instead of aggregating all connections into a single average, the device performs separate analysis for each connection group. This segmentation prevents the artificial inflation of average packet loss rates during low-traffic periods and allows for more accurate detection of actual network abnormalities in each segment.
2Reliability
If the abnormality determination threshold is increased to prevent exaggerated alarms during low-traffic periods, then false alarms are reduced, but the device cannot generate necessary alarms when actual network abnormalities occur
Solution Approach 1:
The patent implements dynamic threshold determination for each connection group based on its specific characteristics and traffic patterns. Instead of using a single static threshold for all connections, the device calculates appropriate thresholds separately for each connection group, allowing the system to maintain high sensitivity for detecting actual abnormalities while avoiding false alarms. This dynamic approach adapts to varying network conditions in different connection groups.
3Productivity
If network analysis data from multiple connection groups is aggregated and analyzed collectively, then monitoring cost is reduced, but the device cannot distinguish between different abnormal states leading to incorrect alarm notifications
Solution Approach 1:
The patent segments network analysis data by connection groups and performs separate abnormality determination for each group. This segmentation preserves the distinctive characteristics of different connection groups, allowing the system to identify and distinguish between different types of abnormal states (e.g., abnormalities in specific bases or regions) while still maintaining cost-efficient monitoring through automated analysis.
4Ease of operation
If the device determines abnormalities based on overall average packet loss rates across all connections, then simple threshold comparison can be performed, but the device generates false alarms when traffic patterns change during out-of-service periods
Solution Approach 1:
The patent divides the network into multiple connection groups and performs separate abnormality determination for each group. This segmentation maintains operational simplicity through automated processing while significantly improving reliability by preventing false alarms that occur when aggregating data across all connections. The system automatically identifies abnormalities in specific connection groups without requiring complex manual analysis.
Data Source
AI summary
A computer readable network analysis program of performing local modeling analysis of determining an estimated value of a current network quality corresponding to explanatory variable vector in current aggregated data based on a local model including local training data; determining an abnormality in the network based on whether or not a measured value of the current network quality is lower than a threshold; determining whether or not a distribution of the connections having the measured value of the network quality exceeding the threshold is present in a large size; extracting an individual-analysis-target connection group including more than predetermined proportions of connections in the distribution of the connections having the large size; and performing the local modeling analysis to the individual-analysis-target connection group and the remaining connection groups to determine the abnormality in the network.