The Data Mining Approach For Islanding Detection In Distributed Generation

Data mining technique is capable of extracting useful informationfrom the mass of information. During the networking era, application ofdata mining technique covers all aspects of society, including financial,industry, economic and social life. Among the systems of smart gridconsidering vast amounts of information, the technique can well performits advantages. Combined with specialized knowledge of power system,this kind of technique raises new solutions to many traditional problems.Further, the distributed generation systems possess many differentcharacteristics compared with traditional power systems. Thus, theunderstanding of targets’ features is important. Today, the distributedgeneration systems require optimization of threshold setting in islandingdetection and concept drift are required to be taken into consideration,which are the suitable for data mining techniques to cope with.Firstly, we discuss a framework of smart grid system of massiveinformation. Secondly, the paper introduces the mass storage and samplingtechnology, describing the three-dimensional model, which consists of datapreprocessing, data window and processing results. According to thefeatures of smart grid framework, the preprocessing of three dimensionmethods are outlined. After that, algorithms of data mining are reviewed,covering clustering, classification, etc. In this chapter, characteristics ofeach method, including its appropriate targets, are described.Then, the key issue of distributed generation system is analyzed, thatis islanding detection. Recalling the previous research work, the thresholdsetting of islanding achieves little concern while it becomes more popular recently. Specifically, the key feature recognition and single classifier’sinductive bias problem are analyzed in this paper. Correspondingly,solutions like RELIEF algorithm to help evaluating critical features and theuse of multiple classifiers have been put forward. Finally, meta-learningstrategy is proposed for the optimization of islanding detection, exhibitinggood classification accuracy.In a subsequent chapter, it proposes a micro-clustering algorithm andweighted and ensemble sampling method to cope with islanding detectionconsidering concept drift. Through the micro-clustering resampling, it caneliminate the randomness of sampling and resolve the problem of memoryoverflow. Combined with SCADA system, it helps to achieve an organiccombination of asynchronous online and real-time processes. The strategyof weighted and ensemble sampling, considering advantages of instancebased and ensemble based method, can improve the accuracy of islandingdetection. To verify the effectiveness of methods proposed above, suddenand gradual concept drift are taken into account. Comparison experimentalresults show that the proposed method has a high precision and strongrobustness, possessing good practical potential.