The Application Of Association Rule Mining Based On Rough Set In Substation Equipment Fault Diagnosis

Fault diagnosis of power device is an information analysis technology by using intelligent methods to detect potential fault of the power device, which is helpful to improve the stability of the power system. With the development of sensor technology,Growth of the power industy data is tend to be exponential. The change of monitoring precision or frequency will cause exponential change in data volume. To cater the future big data era, the application of association rule mining in fault diagnosis of power device is studied in this thesis. Traditional fault diagnosis methods, such as neural networks has the problem of unalbe to explain itselfs’ output and reflect the objective laws between features and performance accurately. Parameter Configuration and calculation for artificial immune are Extremely complex. Because these methods are not proposed on the principle of how fault generated, they are difficult to reflect the objective laws of fault and performance between features. On the background of big power data, fault diagnosis rules minig from the data is easy to understand and more readily to converted to expertise experience.Through analysis of rough set theory and association rule mining method proposed substation fault diagnosis model based on rough sets and association rule mining. The use of rough set theory in the preprocessing operation such as discretization, filled and reduction of original data set, are in order to improve the quality of data mining algorithm for association rule mining. In order to solve the problem of the high false positive rate in substation equipment fault diagnosis when the information are incomplete,a weighted similarity data filled method is proposed based on the importance and dependence of attributes to achieve the efficient reduction of missing data. In order to solve the problem that the efficiency of the attribute reduction algorithm’s exponential decay with the increasing of amount of data, the genetic algorithm was introduced to solve the attribute reduction problem in complex space. Defined based on attribute importance and dependence of the fitness function, can move in two directions simultaneously chromosome high importance and high dependence of convergence, for association rule mining to provide a more streamlined and higher information content of the data. A fitness function based on both attribute importance and dependence is defined in this paper, which can lead chronmosome simultaneously converge in two directions with high importance and high dependence. Proving a more streamlined and higherinformation content data for association rule mining.The major task of association rules mining is to find out all frequent itemsets that satisfy support threshold form database. The generation of candidate frequent itemsets is a NP-hard problem. To solve the above problem, existing association rules mining algorithms mostly use pruning method to find frequent itemsets, which in turn bring about the need for multiple traversal of database. To solve the multiple traversal problem,with the deep study of FP-Growth and DHP algorithm, combining the advantages of the both, a tree structure is defined to storage one traversal result, which reduce by half the number of database scans. With experimental verification and comparative analysis against BP neural network method, results show that the method of association rules mining based on rough set is better in both data processing and diagnosis efficiency aspects. The method given in this paper has a strong application value in the field of power transformer device fault diagnosis.