Alarm Correlation Analysis Based On Association Rules In Telecommunication Networks

In alarm correlation analysis applications, existing methods for obtaining rule knowledge depend on expert experiences. These methods lead to fast decline of fault management effi-ciency, which cannot satisfy practical requirements of telecommunication networks. Based on mining weighted correlation rules, an alarm association analysis method is proposed for the considered problem. The main contributions of this thesis are as follow:(1) An improved alarm transaction extraction method is proposed. Because these uniform s-liding time windows in traditional methods cannot effectively extract alarm transactions, a clustering algorithm is used to divide the alarm data into some relatively uniform groups. The uniform sliding time window method is used to extract alarm transactions in each group in order to improve the alarm transactions extraction efficiency.(2) An entropy method is designed to confirm the alarm weights, which is a kind of objective method to reduce manual participation in dealing with weights of alarms. It avoids the devia-tion caused by human factors and is suitable fora large number of alarm database. Compared with other alarm weight determination methods, (e.g. Analytic Hierarchy Process), the en-tropy method is more practical and effective.(3) An improved weighted association rule mining algorithm(WEclat) is proposed. Different from existing weighted association rules scanning database repeatedly, the proposed weight-ed association rules mining algorithm is based on data compression and WIT-Tree structure. Only one scan is needed. Data compression techniques are used to reduce memory con-sumption and efficiency of mining algorithms is improved.(4) An incremental mining algorithm of weighted association rules is proposed which is based on storing Tidset of 1-Itemset (TWEclat). To improve efficiency of the mining algorithm in incremental environments, IWEclat is proposed to avoid rescanning the original database and generating Tidset of 1-Itemsets.(5) Based on the alarm data in a network management system, the alarm correlation analysis module is designed and implemented to evaluate the proposed algorithm.