A Method of Using Virtual Samples to Train Neural Networks to Diagnose Transformer Faults

Abstract

The invention discloses a method of utilizing a virtual sample to train a neural network to diagnose a transformer fault. The method comprises the steps of establishing characteristic areas of faults of a transformer, selecting virtual fault samples, adding real fault samples, constructing a BP neural network, training the BP neural network and performing fault diagnosis. The method of the invention utilizes an orthogonal table to construct the virtual sample uniformly distributed and capable of reflecting all the data spatial characteristics to construct the training sample set, and adds samples diagnosed to be erroneous by an improved three-ratio-method in practical engineering into the sample set which is used to train the neural network to avoid nonuniform distribution of the training sample and make the trained neural network stronger in generalization ability.

Description

technical field [0001] The invention relates to a method for diagnosing transformer faults, in particular to a method for using virtual samples to train a neural network to diagnose transformer faults, and belongs to the technical field of fault diagnosis of power main equipment. Background technique [0002] As one of the most important and critical electrical equipment in the power system, the safety and reliability of the power transformer is directly related to the safety and stability of the power system. In order to ensure the safe and economical operation of the transformer, it is very necessary to diagnose the latent faults and their types in the transformer timely and accurately. [0003] In recent years, experts, scholars and technical personnel from various countries have done a lot of research work on transformer fault diagnosis, especially the theory of using neural network technology to diagnose transformer faults has achieved many fruitful research results. H...

AI Technical Summary

Problems solved by technology

However, the failure rate of transformers is relatively low, the fault samples collected from actual engineering are limited and the distribution of samples in the data space is extremely uneven, and the generalization ability of the neural network is very dependent on the training samples. If the training samples cannot meet the The requirements of feature representation and distribution uniformity will seriously affect the accuracy of its fault diagnosis. Therefore, most of the researches on transformer fault diagnosis based on neural network remain in the scope of theoretical discussion, and are rarely applied in engineering practice.

Images