The Research And Design Of Cab Signal Denoising Process Based On Deep Learning

The cab signal is an important part of the train operation control system.While the train is running,the transmission of cab signals is often interfered by a lot of noise,which in turn affects signal demodulation and affects driving safety.Since the existing cab signal noise reduction methods have poor effect on noise reduction in the effective frequency band,most of the signal misinterpretation caused by noise interference on the current railway adopts the signal degradation processing method,which greatly reduces the efficiency of railway transportation.Based on the deep learning method,this thesis studies the noise reduction of cab signals from the end-to-end extension of single time domain to multi-channel end-to-end,and develops the experimental simulation platform of cab signals to realize the simulation of code sending and decoding,and the denoising model was applied to achieve noise reduction.The main research and application of the thesis are as follows:(1)Through the research on the cab signal system,combined with the time domain and frequency domain characteristics of the cab signal,analyze the way and principle of noise interference,and summarize three typical noise interference,including unbalanced traction current interference,adjacent line interference and insulation interference,which are common in cab signal interference.(2)The noise-free frequency shift keying signal is generated by software modulation and then the corresponding noise is superimposed to establish a simulation data set.A long-short-term memory fully convolutional neural network(LSTM-FCN)model was built,and the model was trained with the training set.After the test set was tested,before and after model processing,the signal-to-noise ratio of the sampling rate of 8000 Hz and12500Hz increased by an average of 17.43 d B and 17.22 d B,and the root mean square error is reduced by 1.231 and 1.226 on average.(3)The LSTM-FCN model is improved based on the encoder-decoder framework,and the optimized network is tested.For the data sets with sampling rates of 8000 Hz and12500Hz,the signal-to-noise ratio before and after denoising is increased by an average of 17.66 d B and 17.29 d B,the root mean square error decreased by 1.245 and 1.253 on average.Compared with the commonly used time series prediction model Long Short Term Memory Neural Network(LSTM),the signal-to-noise ratio is improved by 5.18%and 2.98%,and the root mean square error is reduced by 5.560% and 5.330%,respectively.The improved encoder-decoder LSTM-FCN network model is verified with railway site misinterpreted signals,which shows that the noise reduction model is feasible and generalizable.(4)Based on the construction and training of the cab signal denoising model,the cab signal experimental simulation platform is developed in C# programming language.The simulation platform is divided into six modules to realize realize signal sending function,demodulation function,noise adding function,model loading function,data storage function,interface display function.Import the denoising model built with Python,and finally complete the development of the cab signal experimental simulation platform.The functions of the platform are tested one by one,and it is proved that it can be used for cab signal simulation signal sending,demodulation and denoising.The platform can be used not only for experimental research on cab signal noise reduction,but also for practice,teaching and other scenarios.There are 71 figures,11 tables and 45 references.