Researches On The Key Technologies Of Case-based Reasoning And Its Applications In Telecommunication Alarm And Fault Diagnosis

Case-based reasoning (CBR) is a kind of learning system based on experience.With the rapid development of big-data technology, the researches of case-basedreasoning technologies have important significance in theory and practice. In thisthesis, the method of case representation and similarity measure, the model ofcase-based reasoning and the case classification algorithm as well as the case basedassociation rules discovering algorithm are studied. The major work of this thesis is asfollows:First, the model based the optimization problems is put forward, case revisingwith iterative optimization is used to increase the diversity of case base. Theprocedure that artificial bee colony algorithm for solving function optimizationproblems is used to simulate the procedure of case-based reasoning.Themultidimensional greedy search artificial bee colony algorithm(mdABC) is putforward. Experimental results show that the mdABC algorithm has more fastconvergence speed and better computational precision in solving the multimodal andhigh dimensional function optimization problems.Second, based on the research of the case classification algorithm and associationrules discovering algorithm, in order to solve the problems of the parameter selectionof kernel function and the penalty factor in the support vector machine, a newclassification algorithm combining of support vector machine and improved immuneclone is proposed. The new algorithm accelerates the speed for finding the globaloptimal solution and improves the classifying accuracy.In order to reduce thecomputational cost of the Apriori alogrithm, based on horizontal identifier lists oftransactions to calculate the support count of candidate sets and using addressindexing mechanism in pruning process，a high-powered muliti-dimensional miningalgorithm is proposed. Experimental results show the new algorithm can significantlyreduce the run time and memory space.Third, the case similarity measure methods based on maximum clique problemare examined. After analysis of the procedure of local search and evolution algorithmto resolve the maximum clique problem, the penalty_BLS algorithm and phased BLSalgorithm(PBLS) are proposed. The PBLS algorithm not only weakens the dependence on case base of the algorithm, but also improves the efficiency. Aiming atthe defects of evolution algorithm for the maximum clique problem inmore complicated, long-running and poor generality, a fast genetic algorithm(FGA) isput forward to solve the maximum clique problem. A newchromosome repair method based on degree, combing with elitist selection based onrandomly paired, uniform crossover and inversion mutation operator are adopted inthe FGA, it can speed up the search, at the same time, can effectively avoid algorithmtrapped into local optimum. The algorithm is tested on DIMACS benchmark graphand experimental results show the FGA has better performance and high generality.Finally, in order to solve the problem of alarm management and fault diagnosisin mobile communication networks, combining the rule-based reasoning, model-basedreasoning and case-based resoning, a new alarm management model based on hybridresoning is presented. According to the hierarchical feature of communicationnetwork, the alarm tree representation model is introduced, alarm management andfault diagnosis based on alarm tree and hybrid resoning are implemented.