Wireless Channel Modeling In Complex Underground En-Vironment Based On Deep Learning

With the development of coal mine industry,the intelligentization of underground wireless communication system has become a research hotspot in recent years.Wireless channel is an important part of the communication system.However,due to the complex environment in the mine roadway,the transmission of electromagnetic wave is severely limited,which leads to the development of underground wireless communication system blocked.In this thsis,the propagation characteristics of underground electromagnetic waves are analyzed,and the fading model of mine wireless channel is established by deep learning method and the receiving field intensity is predicted.In this thsis,the complex underground tunnel environment is abstracted into an ideal empty straight rectangular tunnel.Based on the theory of waveguide,the propagation characteristics of electromagnetic wave in underground tunnel are analyzed.The factors causing electromagnetic wave fading are mainly studied,such as carrier frequency,unsmooth degree of underground roadway,incline degree of underground roadway,coal dust scattering in the channel,humidity in the roadway and sending and receiving distance.Based on the similarity analysis of neural network and wireless channel structure,this thsis proposes a modeling method of large scale fading and small scale fading based on deep learning.Firstly,the propagation process of electromagnetic wave in the well is analyzed,and the fading simulation data is generated as the training set of the neural network.Secondly,the neural network structure is designed according to the training set,and compared with the channel model established by support vector machine.Finally,the model is verifi ed by experiments with the actual measured data of a mine tunnel in our school.The experimental results show that the small-scale fading model established by deep learning method is superior to SVM,but the large-scale fading model established by deep learning method is slightly inferior to the model established by SVM.In order to solve the problem that the prediction performance of large-scale fading SVM is better than that of LSTM,and the prediction performance of small-scale fading LSTM is better than that of SVM,as well as the demand for prediction accuracy,this thsis designed a hybrid network model of SVM-LSTM,which allocated the weight of the prediction results of SVM and LSTM,and optimized the allocation of weight through genetic algorithm.The experimental results show that the hybrid network designed in this thsis can combine the advantages of two single models to predict the fading,and improve the prediction accuracy of the channel fading.