基于深度学习与不平衡样本集的输电线路故障分类

Transmission Line Fault Classification Based on Deep Learning and Imbalanced Sample Set

针对输电线路各类型故障样本间的数量不平衡会造成人工智能算法对故障中的少数类样本识别精度不足的问题,提出了一种基于Borderline-SMOTE(BSMOTE)算法与卷积神经网络(CNN)相结合的输电线路故障分类方法。该方法首先利用BSMOTE算法对位于分类边界上的少数类样本进行过采样合成处理,改善样本间的不平衡度,然后将所提取的一维故障电流信号样本重构成二维灰度图像数据形式,并在Pytorch深度学习框架下搭建了CNN网络模型,利用模型的自主学习能力对灰度图像进行特征自提取与辨识,减少传统人工设计特征提取的工序,完成对输电线路故障类型的分类。实验结果表明该模型能够提高对少数类故障样本的识别能力,准确地判断故障类型,并对噪音具有较强的抗干扰能力。

Targeting the low accuracy in identifying minority fault samples with artificial intelligence algorithm caused by an imbalance between the number of samples of various types of transmission line faults,the paper proposes a transmission line fault classification method based on the combination of Borderline-SMOTE(BSMOTE)algorithm and convolutional neural networks(CNN).Firstly,BSMOTE algorithm is used to oversample and synthesize the minority samples located on classification boundary to improve the imbalance between the samples.Then the extracted one-dimensional fault current signal samples are reconstituted into twodimensional gray-scale image data format,and CNN model is built under a Pytorch-based deep learning framework,using the model’s self-learning ability to perform the feature self-extraction and identification of grayscale images,reducing the feature extraction processes for traditional manual design,and completing the classification of transmission line fault types.The experimental results show that the model can improve the identification of the minority samples,accurately judge the fault types,and has strong anti-noise interference ability.