基于神经网络的输电线路故障识别方法

A Transmission Line Fault Identification Method Based on Neural Network

针对长期暴露在露天环境下输电线路容易发生开股、断股情况,提出了一种基于神经网络的输电线路故障识别方法,选择轻量级网络MobileNet训练模型与单发多盒探测器(single multibox detector,SSD)目标检测算法。建立导线故障样本缺陷库,对导线故障图片进行处理,在已有模型上对故障特征进行训练,调整超参数优化模型,对模型进行测试,最终将模型在嵌入式设备上进行部署。结果表明,与传统的Faster-RCNN结合VGG16网络模型相比较,大小为传统模型的1/23.78,测试速度快了28倍,精确度为92.60%。该系统不仅有较好的识别效果,而且满足实时性。

Aiming at the long-term exposure of open transmission and transmission lines,it is easy to open and break stocks.A neural network-based fault identification method for transmission lines was proposed.The lightweight network MobileNet training model and SSD target detection algorithm were selected.A wiring fault sample defect library was established,the wiring fault picture was processed,the fault feature on the existing model was trained,the hyperparameter optimization model was adjusted and tested and finally the model on the embedded device was deployed.The results show that compared with the traditional Faster-RCNN combined with the VGG16 network model,the size is 1/27.78 of the traditional model,the test speed is 28 times faster,and the accuracy is 92.60%.The system not only has a good recognition effect but also meets real-time performance.