Telecom Network Anomaly Detection via Autoencoder Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection methods in telecommunications networks require significant domain knowledge and manual tuning to set thresholds and handle seasonal variations, failing to detect complex anomalies and correlations between Key Performance Indicators (KPIs).
Innovation Solution
A method using an Artificial Neural Network (ANN) or autoencoder to encode and decode time series data, associating performance parameter values with time representations, calculating reconstruction errors, and analyzing these errors to identify anomalous behavior without the need for expert intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If threshold-based anomaly detection is used, then the method is simple to implement, but it requires considerable domain knowledge for threshold setting and cannot detect complex anomalies
Solution Approach 1:
The system performs self-learning by automatically analyzing historical KPI data to establish dynamic thresholds and anomaly detection rules without requiring manual domain expertise. The machine learning models autonomously adapt to network patterns and seasonal variations, eliminating the need for expert intervention in threshold setting while maintaining high detection accuracy for complex anomalies.
2Device complexity
If static thresholds are used for anomaly detection, then the implementation is straightforward, but the thresholds cannot account for seasonal variations in network traffic
Solution Approach 1:
The system transitions from static thresholds to dynamic, time-adaptive thresholds that automatically adjust according to seasonal patterns, daily cycles, and network conditions. The machine learning models continuously learn from historical data to capture temporal variations, enabling the anomaly detection mechanism to adapt its sensitivity and thresholds based on the time of day, day of week, and seasonal trends without increasing implementation complexity.
3Ease of operation
If simple KPI comparison to thresholds is used, then the method is easy to manage, but abnormal combinations of KPI values cannot be detected
Solution Approach 1:
The system merges multiple individual KPI analyses into a unified multivariate anomaly detection framework. By combining information from multiple KPIs and their interrelationships through machine learning models, the system can detect abnormal combinations and correlations that would be invisible when analyzing individual KPIs in isolation, while maintaining ease of operation through automated analysis.
4Measurement precision
If sophisticated statistical approaches with decomposition are used, then seasonal variations can be handled, but delicate manual tuning is required
Solution Approach 1:
The system automatically performs the delicate manual tuning required for statistical decomposition methods through self-supervised learning and automated hyperparameter optimization. The machine learning models autonomously adjust decomposition parameters, seasonal periods, and trend components based on historical data patterns, eliminating the need for expert manual configuration while maintaining the ability to detect abnormal KPI combinations and handle complex seasonal variations.
Data Source
AI summary
A method for analysing performance of a telecommunications network is disclosed. The method comprises assembling values observed for a network performance parameter over a time window and constructing a time series by associating each performance parameter value with a representation of a time at which the value was observed. The method further comprises encoding the time series to generate a compressed representation of the time series, decoding the compressed representation of the time series to reconstruct the time series, calculating a reconstruction error between the time series and the reconstructed time series, and analysing the reconstruction error to determine if the assembled performance parameter values indicate an anomalous behaviour during the time window. Also disclosed are an apparatus and a computer program product configured to conduct methods for analysing performance of a telecommunications network.