Telecom Alarm Prediction via KPI Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current telecommunication network monitoring systems primarily focus on determining alarm conditions rather than predicting them, leading to potential instability and revenue losses, as they lack real-time predictive capabilities and incremental learning mechanisms.
Innovation Solution
A method for continuous learning and online prediction of alarms in telecommunication networks using a customized Machine Learning model that monitors key performance indicators (KPIs) at both Network Element and communication route levels, allowing for automatic prediction of alarms before they occur by considering the environment, topology, and interactions between elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing anomaly detection methods (k-NN, Cluster analysis, SVM) are used to determine alarm conditions, then alarm detection capability is provided, but real-time prediction capability and incremental learning ability are lacking
Solution Approach 1:
The system transitions from static anomaly detection methods to dynamic predictive modeling by continuously updating the machine learning model with new data. The model adapts to changing network conditions through incremental learning, enabling real-time prediction while maintaining manageable complexity through automated processes.
Solution Approach 2:
The machine learning model performs self-updating through incremental learning, automatically improving its prediction capabilities without requiring manual reconfiguration or intervention. The system serves itself by continuously learning from new data points and adjusting its parameters autonomously.
2Measurement precision
If a customized Machine Learning model is used for real-time alarm prediction, then prediction accuracy is improved, but computational requirements and system complexity increase
Solution Approach 1:
The system performs preliminary data processing and feature extraction before feeding data to the machine learning model. By pre-processing data and selecting relevant features in advance, the computational burden on the model is reduced, enabling real-time prediction with manageable complexity.
Solution Approach 2:
The patent replaces complex manual analysis and traditional detection mechanisms with automated machine learning algorithms. This substitution enables the system to handle complex prediction tasks with standardized computational processes, improving precision while keeping system management straightforward through automation.
3Ease of manufacture
If offline learning methods are used to build predictive systems with history logs, then model training is simplified, but real-time prediction capability is reduced
Solution Approach 1:
The system transitions from periodic offline batch processing to continuous online learning where the model is constantly updated with new data. This continuous action maintains prediction accuracy in real-time while keeping the system adaptable to changing network conditions without requiring complete retraining.
Solution Approach 2:
The system performs preliminary offline training to establish initial model parameters, then uses these pre-trained parameters as a foundation for rapid online adaptation. This two-stage approach combines the ease of offline training with the speed requirements of real-time prediction.
Data Source
Figure 1
Figure 2a~2b
Figure 3~6
AI summary
The present disclosure relates to a method performed by a network node (7) configured for making automatic predictions in a telecommunication network (1). The method comprises obtaining a first value of a first KPI for a first network entity, NE, (2; 3) in the telecommunication network. The method also comprises obtaining a second value of a second KPI for a communication route (5; 6) between said first NE and a second NE (3; 4). The method also comprises predicting, automatically and based on the obtained first and second values, that an alarm will be triggered at the second NE.