VGGNet检测矿井供电漏电应用研究

Research and Application of VGGNet in Mine Power Supply Leakage Detection

在煤炭安全生产过程中,漏电故障作为矿井主要故障之一,具有严重的危害性.目的:为了能够准确稳定识别故障并保护井下设备,提出一种基于深度学习的自动识别方法.方法:为模拟矿井下生产作业环境,首先搭建附加直流电源矿井漏电仿真模型,并利用瞬时对称分量法进行分析,然后根据仿真模型的故障与正常波形的不同特征,提出面向矿井漏电波形图的数据集扩展方法,最后基于深度学习卷积神经网络VGG-Net模型,构建浅层VGG4-Net、VGG7-Net进行故障波形数据自动识别.结果及结论:实验结果表明,文章提出的VGG7-Net模型分类效果较好,表现出较高的准确性与稳定性,其Acc、Pre、Roc、F-1分别达到0.9768、0.9908、0.9665、0.9785,验证了深度学习模型在矿井漏电检测识别中具有一定的可靠性和可行性.

In the process of coal safety production,the leakage fault is one of the main faults in the mine,which has serious harm.Objectives:In order to accurately and stably identify faults and protect underground equipment,a recognition method based on deep learning was proposed.Methods:To simulate the underground production operation environment.Firstly,a simulation model was built on additional DC power supply mine leakage and analysed with using the instantaneous symmetrical component method.Then,data set expansion was proposed for mine leakage waveform graphs according to both the fault of the simulation model and different characteristics of normal waveforms.Finally,shallow VGGNet-4 and VGGNet-7 were constructed based on the deep learning convolutional neural network VGGNet model to perform two-classification processing of waveform data.Results and Conclusions:The experimental results showed that the VGG7-Net model proposed in this paper had a better classification effect,showing high accuracy and stability,and its Acc,Pre,Roc and F-1 reached 0.9768,0.9908,0.9665,0.9785,respectively.The deep learning model was verified to have certain reliability and feasibility in mine leakage detection and identification.