Procedure-Level Anomaly Detection for Telecom Fault Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods in telecom networks lack the ability to capture detailed information on software performance, fail to identify subtle failures, and struggle with complex network structures, leading to inefficient troubleshooting and sub-optimal decision-making.
Innovation Solution
Collect and process procedure-level time series data to derive feature time series for use in unsupervised multivariate time series anomaly detection models, leveraging hierarchical event graphs to enhance fault localization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If KPIs and PM counters are used for anomaly detection, then common faults can be detected, but detailed software performance information is lost and subtle failures cannot be identified
Solution Approach 1:
The patent segments the monolithic KPI/PM counter system into multiple hierarchical levels: procedure-level metrics, function-level metrics, and module-level metrics. This segmentation allows detailed software performance information to be captured at lower levels while maintaining the aggregated view at higher levels, thus preventing information loss while enabling precise fault detection.
Solution Approach 2:
The patent introduces a new dimension of observation by collecting metrics at the procedure level, which is a finer granularity than traditional KPIs. This dimensional change from aggregate system-level metrics to procedure-level metrics enables the detection of subtle failures that were previously invisible, while still maintaining compatibility with existing KPI frameworks.
2Loss of information
If log files are used to capture detailed software state information, then comprehensive fault information is obtained, but storage requirements increase and computational cost rises
Solution Approach 1:
The patent extracts only the essential performance information from detailed log files at the procedure level, rather than storing complete log data. By extracting key metrics such as procedure execution time, success/failure status, and error codes, the system maintains comprehensive fault information while significantly reducing storage requirements compared to full log file retention.
Solution Approach 2:
The patent employs lightweight procedure-level metric structures that are inexpensive to store and process, replacing the need for expensive, long-term retention of detailed log files. These metric objects are designed to be compact and efficient, providing necessary fault information without the storage burden of complete software state logs.
3Device complexity
If traditional rule-based anomaly detection methods are used, then implementation is simple, but they cannot handle complex network structures and evolving anomaly patterns
Solution Approach 1:
The patent transitions from static rule-based detection to dynamic machine learning-based detection. The ML models are trained on historical procedure-level metrics and can adapt to evolving anomaly patterns in complex networks. This dynamic approach allows the system to automatically learn and adjust to new failure modes without requiring manual rule updates, significantly improving adaptability while managing complexity through automated model training.
Solution Approach 2:
The patent introduces machine learning models as an intermediary between raw procedure-level metrics and anomaly detection decisions. This intermediary layer transforms complex, high-dimensional metric data into meaningful anomaly detections, handling the complexity of network structures while maintaining interpretability and adaptability to evolving patterns.
4Measurement precision
If supervised ML methods are used for fault detection, then detection accuracy improves, but the labeling process becomes tedious and scalability is reduced
Solution Approach 1:
The patent performs preliminary unsupervised clustering of procedure-level metrics to identify potential anomaly patterns before applying supervised classification. This preliminary action reduces the burden of manual labeling by pre-organizing data into meaningful clusters, allowing annotators to focus only on ambiguous cases rather than labeling all data points, thus reducing time loss while maintaining detection accuracy.
Data Source
AI summary
A method of pre-processing data for use in training one or more machine learning, ML, models for use in detecting anomalies occurring during execution of one or more procedures at a plurality of network nodes in a network, includes: obtaining procedure level time series data relating to the execution of a first procedure at the plurality of network nodes; and deriving, from the procedure level time series data relating to the execution of the first procedure at the plurality of network nodes, one or more first feature time series of one or more respective first feature values, wherein the one or more first feature time series are for use in training a first ML model associated with the first procedure.