Distribution transformer fault detection method and system

Abstract

The invention relates to a distribution transformer fault detection method and system. According to the distribution transformer fault detection method, a distribution transformer fault detection model based on a robust multi-core extreme learning machine-auto-encoder is adopted. According to the method, a detection result is obtained based on monitoring data obtained in real time in the operation process of the distribution transformer, so that the problems of data imbalance and noise interference are solved, and therefore, the fault detection rate is improved and the fault false alarm rate is reduced.

Description

technical field [0001] The present invention relates to the field of transformer fault detection, in particular to a method and system for fault detection of low-voltage transformers in distribution networks based on Robust multi-kernel extreme learning machine-autoencoder (RMKELM-AE) . Background technique [0002] Distribution transformers (referred to as distribution transformers, 35KV and below, mainly 10KV) are distributed in every corner of towns and villages, with a large number and complex distribution. As the core equipment in the power system to realize electric energy conversion, the distribution transformer, once it fails, will cause power outages, affecting people's normal life, and cause serious accidents such as fires and explosions, causing major economic losses. [0003] In order to reduce the incidence of distribution transformer failures, power grid companies often use manual irregular or regular inspections to check the operating status of transformers. ...

AI Technical Summary

Problems solved by technology

[0004]During the operation of distribution transformers, the occurrence of faults is still a small probability event, which leads to the fact that the actual fault state samples obtained are far smaller than the normal state samples, that is, there are data imbalance problem

Some traditional pattern recognition algorithms such as support vector machines and neural networks regard fault detection as a binary classification problem, but there are disadvantages of low fault detection rate in the data imbalance environment

The main reason for this disadvantage is that the traditional pattern recognition algorithm takes the overall minimization of the training error as the learning goal, and then ignores the learning of fault states with less data

In addition, in the actual operating environment, the normal state data of distribution transformers are easily disturbed by noise, which affects the robustness of the model, resulting in the disadvantage of high fault false alarm rate. The main reason for this disadvantage is that the traditional fault detection model does not have the ability to suppress Ability to interfere with noise

Images