本质安全型防爆电气设备故障电弧识别技术

Arc Fault Identification Technology for Intrinsic Safety Apparatus

为提高爆炸性环境防爆电气设备故障电弧检测的可靠性,依据标准搭建基于IEC火花发生装置的本质安全电路故障电弧试验平台,实时采集电路正常和故障情况下的电流电压波形图像,提出基于卷积神经网络图像识别方法与故障电弧诊断理论结合的电弧检测方法,通过提取波形图像特征向量,判断电路是否产生故障电弧,并通过梯度下降算法、反向传播理论、损失函数优化,不断提高检测模型识别准确度。经测试,该模型对本质安全型防爆电气设备故障电弧的检测准确率达到92%,交叉熵误差值为0.1,说明卷积神经网络模型能准确检测故障电弧,可为爆炸性环境本质安全型防爆电气设备故障电弧检测识别提供技术支持。

This paper aims to enhance the detection infallibility of explosion-proof electrical arc fault in explosive environment.An arc fault detection platform of the intrinsically safe circuit was constructed to collect the current and voltage waveform images under normal and fault conditions of the circuit based on an IEC spark generator accord⁃ing to the standards.Then,the arc was detected based on the image recognition of convolutional neural network(CNN)combining with the arc fault diagnosis theory.Arc fault was detected by extracting the eigenvector of waveform images.The detection accuracy was improved through Gradient-Descent algorithm,Back Propagation theory,and Loss Function.The test results show that the model detection accuracy reaches 92%,and the loss is 0.1.It is proved that the designed CNN model can detect the intrinsic safety apparatus arc fault accurately and provide technical support for detecting the arc fault of intrinsically safe explosion-proof electrical apparatus in explosive environment.