LTE Network Anomaly Detection via Gaussian Probability Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional abnormal value detection methods in LTE networks are inefficient in detecting anomalies in large data sets, particularly those that are new or unanticipated, due to their reliance on supervised algorithms and lack of consideration for time axes, leading to delayed identification and high error rates.
Innovation Solution
A method using a latent semantic analysis model based on Gaussian probability (GPLSA) that incorporates time axes to cluster data points, calculate likelihood values, and assign them to normal, intermediate, or abnormal regions, enabling rapid detection of abnormal values by dividing data into training and testing sets and applying a clustering algorithm to identify 'strong' and 'long' abnormal values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If supervised algorithms are used for abnormal value detection, then the detection can identify known abnormal patterns, but it cannot detect new or unanticipated anomalies and requires large amounts of labeled training data
Solution Approach 1:
The patent inverts the conventional supervised learning approach by using unsupervised learning to detect anomalies. Instead of training the system to recognize known abnormal patterns with labeled data, the system learns the normal data distribution and identifies deviations from this distribution as anomalies, enabling detection of previously unseen anomaly types without requiring labeled training examples of the anomalies themselves.
Solution Approach 2:
The patent introduces an intermediary statistical model that represents the normal data distribution. This model acts as a mediator between the raw network data and the anomaly detection process, allowing the system to identify anomalies by measuring how well data points fit the learned normal distribution, rather than directly comparing against labeled anomaly examples.
2Productivity
If conventional abnormal value detection algorithms are used, then the processing can be performed, but the time length from appearance of abnormal value to correction is not reduced and error rate remains high
Solution Approach 1:
The patent applies preliminary action by continuously learning and updating the normal data distribution model in real-time as new data arrives. This allows the system to be prepared for anomaly detection at any moment, reducing the time from anomaly appearance to detection. The model is proactively maintained rather than waiting for anomalies to occur before analysis.
Solution Approach 2:
The patent replaces traditional mechanical anomaly detection methods with a statistical modeling approach. Instead of using rule-based systems or threshold comparisons, the system uses probability distributions and statistical measures to identify anomalies, enabling more efficient and accurate real-time detection with lower error rates.
3Measurement precision
If manual inspection of log files is performed by experts, then detailed analysis can be conducted, but the huge amount of data makes it difficult and time-consuming to find abnormal values
Solution Approach 1:
The patent implements self-service by automating the anomaly detection process entirely. The statistical model automatically learns from the data and identifies anomalies without human intervention, eliminating the need for experts to manually inspect log files. The system serves itself by continuously updating its understanding of normal behavior and autonomously flagging deviations.
Solution Approach 2:
The patent extracts the essential characteristics of normal network behavior into a statistical model, separating the key patterns from the overwhelming volume of raw data. This extraction process creates a compact representation that captures the essence of normal operation, allowing the system to detect anomalies efficiently without processing every individual data point manually.
Data Source
AI summary
A method and system for detecting abnormal values in an LTE network is provided: dividing measured data into a training and a testing set; defining clusters and parameters in the training set, and finding the cluster to which each point belongs using clustering algorithms; calculating a likelihood of each point based on parameters and clustering results; assigning the likelihood into an abnormal, an intermediate or a normal region according to a set warning and alarming threshold; and applying a calculated model to the testing set, the likelihood of each point is calculated and assigned to a region, thereby finding abnormal values in the testing set. The variation of data points versus time may be better understood by introducing time axes into the model, thereby multiple abnormal values may be discovered from a sequence of multiple points. The method can immediately detect abnormal values and the error rate is low.