Abstract
This paper proposes a centralized management framework for mobile communication network alarms, introducing a classification hierarchy and a six-stage model. It explores recessive maintenance, logical correlation rules for network equipment, and the application of data mining algorithms (like Apriori and WINEPI) to identify root causes. The research concludes with a roadmap for an intelligent alarm analysis system designed to reduce network noise and improve operational efficiency.
Introduction: The Challenge of the "Alarm Storm"
As mobile networks transition to complex 3G and 4G architectures, operators face an overwhelming "alarm storm." In this environment, a single equipment failure can trigger a cascade of alerts, often described as the "click alarm, multi-click dissemination" effect. Traditional management approaches, which treat alarms as isolated incidents, are no longer sufficient. Operators are often reactive, struggling to distinguish between "garbage alarms" (e.g., from unused ports) and critical performance alarms that directly impact user experience. This paper proposes a centralized management framework that leverages logical correlation and data mining to filter noise and prioritize actionable insights.
Methods for Centralized Alarm Management
The authors outline a multi-layered approach to transition from fragmented monitoring to an intelligent, centralized system.
Classification Hierarchy
Alarms are categorized into five distinct types to better understand their impact: Equipment level, Performance level, Perceived service, Volume overload, and Network traffic anomalies. This hierarchy helps operators move beyond raw equipment status to focus on service quality.
The Six-Stage Model and Logical Correlation
The framework scales management through a six-stage model, moving from single network elements to regional, cross-specialty monitoring. Central to this are logical correlation rules:
- Intra-equipment rules: Eight rules (e.g., filtering, suppression, temporal relation) manage alerts within a single device.
- Inter-equipment rules: Five rules (e.g., topology and causal correlation) identify relationships between groups of equipment.
Data Mining and Recessive Maintenance
To evolve beyond static rules, the system integrates data mining algorithms like Apriori and WINEPI. These tools discover frequent episode patterns in historical data, allowing the system to "learn" root causes. Furthermore, the strategy emphasizes recessive maintenance—identifying "sub-healthy" states, such as optical power attenuation, before they lead to full outages.
Key Insights: Prioritizing Performance
A core tenet of this research is that Performance > Equipment. Indicators like call success rates are more valuable than raw hardware alerts because they align with customer perception. Additionally, context matters: alarms in high-traffic "hot spots" or during holidays are weighted more heavily. Effective monitoring also requires "garbage collection"—the systematic removal of alarms from decommissioned equipment that creates unnecessary noise.
Results: A 90% Reduction in Alarm Volume
The implementation of these maintenance patterns and centralized monitoring in a test district yielded significant quantitative improvements. The research reports that the number of single network equipment alarms dropped from 354 in September 2012 to just 38 by February 2013. This nearly 90% reduction in alarm volume significantly eases the burden on monitoring personnel, allowing them to focus on genuine critical failures.
Future Work
While the current system effectively uses "expert-combed logic," the authors acknowledge that real-time analysis can be further improved. Future development aims for a more autonomous "intelligent system management" phase, incorporating neural networks or fuzzy logic to handle the increasing complexity of 5G deployments where manual rule-setting may become unsustainable.
Summary updated