Depth Identification Of Underground Power Cables Of LSTM Based On Self-Attention Mechanism And ResNet

Underground power cables are an important part of the power grid system and an important way to reliably and economically transmit electricity.Underground power cables are installed in complex underground environments for a long time,and it is easy to produce safety hazards under long-term environmental erosion,such as short circuit and breaking of power cables,aging and falling off of the insulation layer on the surface of the cable.Therefore,it is the focus of ground detection to accurately and conveniently find out the direction,underground burial depth and fault point of underground power cables without damaging the ground.In view of the problems of slow acquisition speed,few elements and low accuracy of traditional ground detection methods commonly used in China,and in view of the advantages of automatic feature extraction and classification and recognition of Residual Network(ResNet)and Long Short-Term Memory(LSTM)in deep learning,this paper is based on transient electromagnetic method.Based on the hybrid model of self-attention LSTM and ResNet,this paper improves the calculation accuracy of underground cable depth recognition,and realizes the imaging display of underground cable more accurately and scientifically,which has practical significance for underground space detection in the field of ground exploration.The main research contents and conclusions of the paper are as follows:(1)The noise reduction process is performed on the timing electromagnetic data signal collected by the transient electromagnetic detector,and white noise is added to the input data signal by the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN),and the noise in the original signal is canceled by multiple superposition averages.(2)Deep identification of single-emission frequency underground power cables based on SA-LSTM neural network.The induced voltage at different sampling time points is collected for a single reflection frequency to obtain its normalized value for current,which is used as the input of the SA-LSTM network together with the transient electromagnetic device parameters and sampling time,and the network output is the apparent resistivity.It can avoid the problem that the unique resistivity value cannot be determined due to the duality characteristics of the correspondence between induced electromotive force and resistivity,and at the same time,the feature extraction of the input dataset of the network is carried out by using the self-attention mechanism,which reduces the dimension of the input data vector,improves the computational accuracy and computational efficiency of the network,and optimizes the weight and accuracy of the network.(3)Construction of in-depth identification of multi-frequency underground power cables based on generative adversarial network and ResNet network.The output data of the SA-LSTM neural network model will be plotted at the equivalent depth,and the depth identification and imaging of underground power cables under multi-frequency conditions can be obtained by adjusting different transmission frequencies.These images are then input into the generative adversarial network and ResNet network for image fusion,and finally output the generated image,which can make the path direction and depth of underground power cables more accurate and clear.(4)The proposed method was experimentally verified by the path detector developed by the research group,and the effectiveness of the proposed method was verified.