A Research On Key Problems In Alarm Correlation Analysis Based On Knowledge Discovery

Modern telecommunication networks are characterized with large scale, complexity, and heterogeneity, that requires we have to manage telecommunication networks effectively to maintain their high reliability and high usability. As an important problem in network fault management, alarm correlation analysis can help network administrators to delete redundant alarms, locate faults and predict faults before they happen. However, traditional alarm correlation analysis methods can hardly work well when networks are complex and changeful, while the knowledge discovery method can overcome the shortage of traditional methods. In this thesis, we apply knowledge discovery technology to alarm correlation analysis and study several key problems including incremental episode rules mining, alarm prediction patterns mining and the real-time of alarm prediction patterns mining, and the work which has been finished are as follows:1. We research the incremental episode rules mining. As a kind of important alarm correlation knowledge, episode rules can be used to delete redundant alarms and to analyze the root alarm denotes the fault. At present most of the episode rules mining methods are based on the framework of WINEPI algorithm. However, WINEPI algorithm is inefficient under the conditions of repeated mining caused by changed mining parameters. In this thesis we propose a incremental episode rules mining algorithm TWIER for the repeated mining caused by changed time window, an important parameter. By using the mining result under...