Design And Implementation Of Fault Diagnosis System Based On Parallel High Utility Pattern Mining

With the rapid development of network information technology,many industries introduce information systems into workflow.Many different information systems are widely used.However,too many information systems will sharply increase the cost of maintenance,and it is difficult for fault diagnosis.In the electric power information system,the frequent interactions during systems,and the faults generated by massive data,can not be solved manually.So making fault diagnosis process intelligent is becoming more and more important.Fault diagnosis based on high utility pattern solves the problem of indicator importance differences.However,high utility pattern mining(HUPM)faces the challenge of low efficiency and long running time,due to the newly introduced definitions.Moreover,traditional HUPM in big data environment are difficult to satisfy the requirements.In this thesis,based on the real electric power business data,an efficient HUPM algorithm is proposed,improve the performance,and using it to design the fault diagnosis core algorithm.The accuracy and running time is better than similar algorithm.The design and implementation of the fault diagnosis system is completed.The thesis addresses the balance between pattern accuracy and efficiency in fault diagnosis.The research contents of this thesis are as follows:(1)The study of HUPM algorithm based on bit operation.The efficiency improvement of HUPM algorithm is still a key problem.In this thesis,to improve the construction process of utility list in HUPM algorithm,combined with bit operations,a statecompressed utility list structure called SCUL(State-Compressed Utility List)is proposed,and a HUPM algorithm called SCUL-Miner is designed.The experiments show that the running time of SCUL-Miner is 1/6 of state-of-art,although the memory usage is little large.(2)The study of parallel,adaptive and pruning optimization of HUPM algorithms.Under the data background of electric powder fault information,it is very important to select algorithm parameters quickly and reasonably,and improve data throughput.The existing HUPM algorithms mainly focus on the design of data structure and the pruning method,but lack of research on the parallelization and the data structure selection.Based on the analysis of SCUL-Miner,this thesis parallelizes core functions,adaptively selects data structures,and prunes the search space.Finally achieving a paralleled,selfadaptive and pruning algorithm called MSCUL-Miner(Modified SCUL-Miner).Experiments show that the performance of the proposed parallelization strategy is 3 times faster than SCUL-Miner; The adaptive data structure selection is more reasonable,reducing the memory consumption by 50% without affecting the running time significantly; The speedup of pruning strategy is up to 5 times SCUL-Miner.(3)Based on MSCUL-Miner,design the fault diagnosis core algorithm and implement the system.Combined with the pattern knowledge base and MSCUL-Miner algorithm,a complete fault diagnosis algorithm is proposed.Compared with the experimental results,the accuracy is 12 % higher than best algorithm,and the running time is 18% less than the fastest algorithm.Then the system is implemented.First,the fault alarm information is preprocessed,and then the fault diagnosis core algorithm is run to obtain the fault classification of the new fault information,then complete the fault diagnosis,to satisfied the work needs of the power information system operation and maintenance personnel.