Research On High-Speed Railway Radio Propagation Scenario Identification And Channel Prediction Based On Deep Learning

In recent years,with the rapid development of high-speed railway in China,digitalization and intellectualization have become an important development direction of high-speed railway in the future.As an important part of intelligent high-speed railway system,high-speed railway mobile communication system is responsible for the highspeed railway wireless communication services.Wireless channel propagation characteristic is one of the most important factors which influence the performance of the communication system.To ensure the safety and efficient transmission for the information of high-speed train operation and passengers’ services,the communication system need to identify the scenarios precisely and to make predictions for the future wireless channel state,and then the transmission mode can be adjusted according to the current scene channel state by intelligent decision-making and adaptive transmission technology to improve high-speed communication system performance.Instead of adopting deep learning algorithms which is more powerful in nonlinear mapping and feature extraction,most current researches on scenario identification adopt the traditional machine learning algorithms to identify traditional scenarios such as indoor and vehicle network,which is difficult to deal with the fast time-varying and non-stationary highspeed railway scenarios.In addition,the deep learning based channel prediction methods are still lack of the research on high-speed wireless channel.Therefore,to fill the aforementioned research gaps,this paper carries out research on high-speed railway propagation scenario identification and wireless channel prediction with deep learning algorithms.The main research contents of this paper are as follows:(1)Based on the previous high speed railway channel measurements,we obtained a large amount of wireless channel data and preprocessed the data.Then a variety of highspeed wireless channel characteristic parameters were extracted and the high-speed wireless channel datasets were constructed and divided into training datasets and testing datasets to provide the basis for later research work.(2)According to deep learning algorithms in the application of propagation scenario identification studies,a propagation scenario model based on long short-term neural network and the weighted score fusion was proposed.The weight initialization scheme and optimizer used in the training process of the model were determined.The hyperparameters of the proposed model were determined by autocorrelation analysis and K-fold cross-validation strategy.A variety of evaluation indicators such as accuracy,confusion matrix and F1-Score were used to evaluate the influence of different feature groups and feature fusion methods for the performance of identification models.The proposed model was compared with models based on support vector machine,random forest,deep neural network and recurrent neural network.Finally,the time and space complexity of the identification models were evaluated by parameters and floating computing operations of models.(3)The research on high-speed railway wireless channel prediction was developed and a wireless channel prediction model based on Transformer for high-speed railway was proposed.The training schemes and prediction schemes used in the realization of the model were discussed,and the hyper-parameters of the model were determined based on the K-fold cross-validation strategy and autocorrelation analysis.Multiple evaluation indicators such as root mean square error and error vector amplitude were used to evaluate the prediction performance of the proposed model.The proposed model was compared with models based on deep neural network,recurrent neural network and long short-term neural network.The time and space complexity of different prediction models were evaluated.