Excitation surge and fault current identification method based on kernel function limit learning machine

Abstract

The invention discloses an excitation surge and fault current identification method based on a kernel function limit learning machine. The method comprises a step of collecting the high and low voltage side current of each phase to make a difference to obtain the differential current of each phase, a step of carrying out experience modal decomposition on the differential current of each phase to obtain the IMF sequence corresponding to each phase, a step of forming the IMF component of each phase into a trajectory matrix, carrying out singular value decomposition on the trajectory matrix to obtain the singular value corresponding to each phase, and obtaining the singular value spectrum entropy corresponding to each phase based on the information entropy, a step of collecting the current data of a transformer fault condition, and dividing the current data into a training sample set and a test sample set, a step of establishing the kernel function limit learning machine of each phase with the singular value spectrum entropy of each phase as an input amount and the current type as an output amount, and using the training sample set to train the kernel function limit learning machine, and a step of using the test sample set to train the trained kernel function limit learning machine to carry out testing and evaluating. The method has the advantages of advantages of fast convergence speed, stable generalization performance and high prediction accuracy.

Description

Technical field [0001] The invention relates to a method for identifying inrush current and fault current based on a kernel function extreme learning machine, and belongs to the technical field of intelligent substation equipment diagnosis. Background technique [0002] With the continuous expansion of my country's power grid capacity, more and more large-capacity power transformers are put into the operation of the power system. Once they fail, they will have serious adverse effects on industry, agriculture and people’s lives, and will also cause A lot of economic losses. Therefore, in order for the entire power grid to operate more safely and stably, the protection of power transformers has attracted the attention of relevant power system departments. [0003] At present, one of the main problems of power transformer protection is how to correctly identify the inrush current and fault current. Commonly used methods are mostly based on the waveform characteristics of the inrush c...

AI Technical Summary

Problems solved by technology

[0004] The common method is to use the wavelet transform to preprocess the differential current, and extract the energy spectrum as the characteristic input of the neural network. The trained network can correctly distinguish the excitation inrush current and the fault current, but the selection of the wavelet base and the decomposition scale in the wavelet transform is still not unified. The principle followed, and there is energy leakage in wavelet decomposition; later, an adaptive wavelet neural network excitation inrush recognition method was proposed. After testing, this method has a high recognition rate, but it requires a high number of training samples. When the actual number of samples is limited , the accuracy of this method will be limited by

Images