Study On Data Mining Method And Its Applications In Power System Fault Diagnosis

Data mining is a multi-discipline domain, which gains alimentation from many fields, such as DB technology, artificial intelligence, machine-learning, NN (neural network), statistics, pattern recognition, and knowledge base system and so on. Since the emerges of the Data mining, it has attracted many attention in the computer field, artificial intelligence field etc. at home and abroad. Meanwhile, application study in every field is springing up.Current data mining method is to mine association relationship between ID-set in database. Mined information implies connection between ID-set, but does not imply quantity relationship. In power system applications this quantity relationship exists widespread. It plays an important role in decision system. In this paper a concept and an algorithm about the mining of the more frequent itemsets are proposed, which can mine quantity relationship among each ID-set in database by utilizing information data of fault ID-set, and reflect influence and relationship between frequency of itemsets elements and other itemsets. For instance, each transaction in a transaction DB includes itemsets' category and quantity. But what we concern most is whether the association itemsets with a sharp quantity proportion are related or not. Maybe they are not related at all, just a coincidence. This can interpret some accidental phenomena in power system.When the data-recorder dynamically increases, the current methods of data mining will face repeated scan of the database, thus causing a waste in the computer system resource. Aiming at a solution to such a problem, this thesis proposes a new dynamic growing mining method for large-scaled databases and more frequency itemsets. The new method makes use of the results of the preceding mining and specified data of ID of the newly emerged faults, and it only does mining of the incremental part. When this mining is over, a supplementary scan will be conducted to the part of primitive database. This new method overcomes the defects such as unnecessary repeated scan of primitive database, excessively large space of search, a prolonged process of search and low efficiency of calculation. It is a full optimization of the system resource of the computer.As for the minine method of association regulations with candidate itemsets. thisthesis proposes a new bi-segment correlative algorithm. This method conducts only one respective scan of the primitive database DB in its calculation of 1—item candidate itemsets and 2 —item candidate itemsets. While in its calculation of frequency itemset at higher levels, there is no needs of the scan of the primitive database but only that of the candidate frequency closed itemsets at the preceding level. This will greatly decrease the field of the search, raises the calculation efficiency and saves system resource.As for the mining of the association regulations with no candidate itemsets, this thesis introduces the method of FP-T calculation into the "tree" based mining of the multi-level association. It makes improvements to the method of FP-T calculation as follows: it realizes the data mining in multi-level frequency itemsets by constructing an index form with a special node-link pointer and making use of the filiations between the multi-concept levels, making it possible to mine the regulations of association between different concept levels. This calculation method is capable of reducing the repeated scans needed for the database and mining more related information.There are so many attribute parameters in power systems. If all these parameters join to the calculation and mining they must greatly affect execution efficiency of data mining program. How to pretreated data and optimize parameters in calculation process is an important means to improve efficiency in data mining in power system. This paper studies a binary mining algorithm, which makes logically "and" arithmetic through the use of binary status values "1" and "0". It shields useless property value, extracts useful information, extracts key information from many property value field, and forms decide rule. This method has the characteristics of simple principle, efficient mining, also strong fitting, plays an important role in power system data mining and is also suitable to mining and analysis of the huge data houses.The waveforms of power system's parameters are related closely to its operation status. It can reflect and foresee that certain status and faults will occur in power system. In this paper, fractal theory is used to analyze parameter waveform. This method can overcome the defects in tradition methods which processing the high frequency instantaneous waveform of power system as a noise data, and make full...