Fault arc neural network optimization training method

Abstract

The invention relates to a fault arc neural network optimization training method, and the method comprises the following steps: extracting a plurality of characteristic quantities of arc faults and anormal circuit current signal time-frequency domain, performing training by adopting a plurality of neural network optimization training strategies, and then finding out a model with the best test setperformance, wherein the final training model has higher recognition accuracy in fault arc recognition; employing 1GHz AD for collecting current signals in a line, extracting multiple time-frequencydomain characteristic values from each half-wave, extracted and splicing characteristic vectors into a characteristic matrix, processing the characteristic matrix by a convolutional neural network, and in a neural network training mode, and adopting seven different training means to seek a model with the best generalization ability; and finally, judging whether the arc exists according to an output result of the neural network.

Description

technical field [0001] The invention belongs to the field of fault arc detection, and mainly relates to a neural network training strategy for fault arc characteristic values ​​to ensure that the model has good generalization ability. Background technique [0002] Electrical fires account for a high proportion of fire accidents in today's society, and arc faults are one of the important causes of electrical fires. Arc faults are usually caused by the aging and damage of wiring, equipment insulation, or poor electrical connections. The occurrence of arcs will release high temperature, which is extremely easy to cause fires. The arcs that occur can be divided into series fault arcs and parallel fault arcs according to their fault types. Due to the relatively large current when parallel fault arcs occur, existing overcurrent protection devices and short circuit protection devices can play a partial protective role. When a series arc occurs, although the current is abnormal, th...

AI Technical Summary

Problems solved by technology

[0004] The traditional arc detection method is mainly to set the threshold value for the extracted feature value for discrimination. Due to the variety of load conditions in the actual power consumption environment, the way of setting the threshold value has different threshold values ​​for different loads, so it is difficult to adapt There is still a lot of room for improvement in performance under different load environments

At the same time, the traditional arc detection methods focus on signal frequencies mostly within a few MHz, and the signals collected at low sampling rates will appear to be indistinguishable from the arc signal and the signal of some common household appliances, resulting in misjudgment.

Images