基于弹性振动和深度学习的变压器状态识别

Research on transformer state recognition based on elastic vibration and deep leaning

针对当前传统的变压器状态识别算法需要人工干预的问题,研究了一种能够自动提取特征并分类的一维卷积神经网络算法。该算法通过3层卷积池化层自动提取信号特征,并通过全连接层展为一维矢量,最终通过Softmax层进行分类。鉴于弹性振动信号抗干扰能力较强,选择弹性振动信号作为信号处理研究对象,运用基于一维卷积神经网络和弹性振动的方法对变压器状态进行识别,并通过采集500kV变压器的弹性振动信号获取的数据集进行验证,结果表明该算法的准确率优于BP、SVM和SAE算法,能对变压器的不同状态实现自动有效识别。

In order to overcome the shortcomings of traditional transformer state recognition algorithms that require human intervention,this paper proposes a new state recognition method,which uses the One-Dimensional Convolutional Neural Network(the 1D-CNN algorithm)to automatically extract the features and classify.The algorithm could automatically extract the signal features through three convolutional and pooling layer,expands the extracted features into a 1D vector through a fully connected layer,and finally classifies the signals through a Softmax layer.Since the strong anti-interference capability of the elastic vibration,a dataset of elastic vibration signals of a 500kV transformer is used to verify the proposed algorithm,and shows that the accuracy of the 1D-CNN algorithm is higher than BP,SVM and SAE in transformer state recognition.