Bad-data Detection And Identification Of Power Systems Using Fuzzy Clustering Methods

The security and stability of power systems are dependent on the quality of real-time data in power systems. But because of unreasonable layout, unfavorable communication and relaxed management for real-time measurement systems, the bad-data in real-time data must be detected and identified. Bad-data detection and identification is one important part of power system state estimation, the purpose of that is to remove a few bad-data from the measurement data and then enhance the reliability of state estimation, which is very important to the security and stability of power systems. In this thesis, according to the fuzzy and uncertainty of the real-time data in increasingly complex power systems, fuzzy mathematics theory has been used to detect and identify the bad-data in power systems naturally.Firstly, the basic conception and research status of bad-data detection and identification in power systems have been imported. And the common methods of bad-data detection and identification have been put forward detailedly.Secondly, the conception of fuzzy mathematics, the theory and basic methods of fuzzy cluster analysis have been provided. And based on the analytic comparison between the fuzzy equivalence and fuzzy ISODATA methods, one integrated fuzzy clustering method is presented.Finally, with Fortran 6.5 and VC++, the bad-data identification system is realized separately based on the fuzzy equivalence relation, fuzzy equivalence relation and integrated fuzzy clustering to identify bad-data. Meanwhile, the features of the three cluster methods have been analyzed by comparison. The three methods have been proved to be efficiency by using the case simulation test, and the merits and flaws of the three methods have been compared through the case simulation. Meanwhile, the clustering method of fuzzy equivalence matrix to bad-data detection and identification has been the key point to research. The method of searching best threshold value has been discussed by lucubrating. The clustering method has been proved to have some advantages, relatively to some traditional methods for bad-data detection and identification. Besides, measured samples from SCADA system are compressed by data preprocessing, and then a new bad-data identification system has been deduced. In the last of this thesis, it is proposed tentatively that the optimal clustering centers computed have certain instructive significance to improve the distribution of measuring points in WAMS.