| With the expansion of China’s power system,the grid structure is becoming more and more compact,and the capacity of single generators is increasing,making the impact of loss of excitation faults become more serious;and as the oscillation period becomes longer,the delay of loss of excitation protection increases;At the same time,the parallel operation of the units and the change of the system operation mode also affect the accurate action of the loss-of-excitation protection to a certain extent.In order to improve the selectivity and quick action of generator loss-of-excitation protection,this paper proposes a related loss-of-excitation protection scheme based on support vector machine(SVM).First,a generator loss-of-excitation protection method based on SVM for intelligent trajectory identification is proposed.The measured impedance trajectory contains lots of generator operating information,and its motion characteristics can essentially reflect the operating state of the generator.Perform global and local feature extraction on the measured impedance trajectory,and calculate the statistical parameters of the extracted motion feature sequence to form a total of 84-dimensional features;then perform Principal Component Analysis(PCA)dimensionality reduction on the feature space to form the corresponding training input In the feature space,the genetic simulated annealing algorithm(GSAA)is used to optimize the parameters of the SVM;a loss-of-excitation protection model based on the SVM is formed.Compared with the traditional loss-of-excitation protection,this method improves the quick action and selectivity of the generator loss-of-excitation protection action.Secondly,Secondly,based on the detailed description of the SVM classification mechanism,the SVM classification accuracy of the impedance trajectory segment with different time windows and the characteristics of insufficient SVM classification accuracy of the samples at low time windows are analyzed;take the SVM classification function distance value under the low time window trajectory segment samples as the output value,and set the classification function The output distance confidence threshold is used to determine whether to use the SVM classification distance value as the basis for judgment,and finally an adaptive loss-of-excitation protection based on the distance of the SVM classification function is formed.To a certain extent,the SVM adaptive classification judgment further shortens the time required for partial loss of magnetization fault judgmentt.Finally,the simulation samples verify that the above two schemes can accurately identify the loss of excitation fault,which effectively improves the accuracy and quickness of the generator loss of excitation protection.At the same time,when the method in this paper is used as a loss-of-excitation protection scheme,it can also accurately identify oscillation and short circuit faults. |
