Research On Fault Detection Method Of Power Insulator Based On Deep Learning

As an important component in the power network,the insulator plays a role of support and electrical insulation in the operation of the transmission line.Due to its large quantity and long-term exposure to nature,it is easily affected by natural factors such as climate and temperature,and it is prone to failures such as falling off and damage.Once the insulator in the transmission line fails,it is easy to cause power interruption of the entire line,and even lead to grid failure in the entire area,causing huge economic losses to industrial and agricultural production.At present,the use of aerial imagery for insulator fault detection is a research hotspot because it saves time and effort,can accurately and efficiently determine the operating state of the insulator,and has high detection efficiency.Because the artificial feature extraction of the traditional insulator image fault detection method is limited,the accuracy of the insulator image fault detection of complex background is low.Therefore,this thesis first introduces the deep learning target detection method into the insulator fault detection.Secondly,for the problem that it is difficult to quantitatively analyze the detection results,the quantitative segmentation detection network in the deep learning semantic segmentation method is applied to the insulator fault detection.Aiming at the problem that FNC and Mask R-CNN networks are subject to complex background interference,which causes some backgrounds to be misdetected as foreground insulator regions and affect the accuracy of fault detection,a novel second-order FCN network structure is proposed for insulator fault detection.The work is as follows.(1)For the acquired aerial insulator image,the deep-learning FCN network is used to automatically perform the feature extraction of the insulator image layer by layer,and the insulator region is preliminarily segmented;(2)For the background of the insulator background of the complex background,the background of the insulator is misdetected as the foreground insulator region affects the accuracy of fault detection.This thesis uses the morphological reconstruction filter algorithm to filter the false detection area in the output image of the first-order FCN network to accurately segment the insulator region.(3)Multiplying the accurately segmented insulator region image by the original image to remove the background,and obtaining an image of the region insulator containing only the foreground.As a training set for fault detection,a second-order FCN network is constructed on the basis of the regional insulator for final insulator fault detection.In order to verify the effectiveness of the method,this thesis compares the second-order FCN fault detection results with the FCN and Mask R-CNN test results.The experimental results show that the second-order FCN-based insulator fault detection method can reduce the interference of complex and variable background regions and improve the accuracy of insulator fault identification.It is an effective insulator fault detection method.