Application of Gustafson-Kessel clustering algorithm in the pattern recognition for GIS

This paper simulates four typical defects in GIS for PD detection, and uses the pulse, amplitude, phase and number of PD to form the three-dimensional PQN matrix. Based on the PQN, three two-dimensional distributions of Hqmax similar to Phi, Hqmean similar to Phi and Hn similar to Phi can be achieved. Then the new G-K clustering method is introduced to separate the four different defects according to the parameters of Sk, Ku, Pe, Q and CC. At last the comparison with different clustering methods is achieved based on the evaluation criterions which are PC, CE, PI and SI.