Research On Estimation Of Track Longitudinal Level Irregularity Of Rail Transportation In Long Tunnel Construction

With the advancement of railway construction towards mountainous areas in the central and western regions,the construction of long tunnels is increasing.Improving the speed of track transportation for TBM construction materials in long tunnels will have higher requirements for track smoothness,therefore it is necessary to strengthen the detection of track geometric irregularities.However,in the construction of long tunnels,manual detection of track irregularities during rail transportation is not only time-consuming and labor-intensive,but also increases economic costs.Therefore,it is of great significance to detect track irregularities through the dynamic response of vehicles.This thesis obtains data on axle box acceleration and track irregularity through dynamic simulation.On the basis of analyzing the correlation between axle box acceleration and track irregularity,a high and low track irregularity estimation model for rapid rail transportation in long tunnel construction was established by combining deep learning and intelligent algorithms.By inputting the axle box acceleration of rail transportation vehicles during the construction of long tunnels into the estimation model,the amplitude of track unevenness can be obtained in real-time.The main work and research results of this article are as follows:(1)Based on the vibration state of high-speed rail transportation vehicles in the construction of long tunnels under the excitation of track irregularity at a certain mileage position,and the characteristics determined by the track irregularity conditions and previous vehicle vibration state of the mileage position and the previous section,a recurrent neural network is selected.Comparing the principles and characteristics of RNN,LSTM,and GRU,taking into account accuracy and training time,an estimation model is constructed by combining LSTM and GRU networks.Based on the characteristics of intelligent algorithms,choose particle swarm optimization algorithm to optimize the network.(2)Build a training dataset for estimating models.Using inverse Fourier transform and dynamic simulation methods to obtain track irregularity time-domain and vehicle axle box acceleration data,respectively.In view of the complex relationship between vehicle response and track geometric irregularity,combined with coherence analysis,the axle box vertical acceleration is determined as the input of the model,and the track height irregularity is determined as the output of the model.(3)The hyperparameter of the neural network is optimized by empirical manual parameter adjustment and PSO particle swarm optimization algorithm respectively.Compared with manually optimized models,the PSO particle swarm optimization algorithm reduces RMSE by 75% and training time by 4.2%.Finally,the optimal parameters found by the PSO particle swarm optimization were selected,and an LSTM layer with 1 neurons in the input layer,43 neurons in the first layer of the hidden layer,61 neurons in the GRU layer in the second layer of the hidden layer,RMSProp as the update gradient algorithm,epoch 40,regularization coefficient 0.001,learning rate0.005 2,and time step 468 were built,The LSTM-GRU instantaneous track height irregularity estimation model for high-speed rail transport in the construction of long and large tunnels with a batch size of 82 for training hyperparameter.The prediction results of the model are basically the same as the original data trend,and can also predict well in areas with larger amplitudes,with a prediction error of no more than 0.4mm.(4)Based on the estimation model of the height unevenness of the rail transportation track in the construction of a long tunnel,the predictive performance of the model was studied under different axle box positions,different routes,different input variables,and noise scenarios with track unevenness.Research has shown that the vertical acceleration at different axle box positions has little impact on the prediction accuracy of the estimation model.The same power spectrum has good prediction performance for different lines,but the prediction effect for track height irregularities generated by different power spectra is very poor.When the train is running at a constant speed,considering the train speed or lateral acceleration of the axle box as input variables does not significantly improve the prediction accuracy of the model;In the case of train inertia operation,considering the train speed or lateral acceleration of the axle box as input variables significantly improves the prediction accuracy of the model.When the signal-to-noise ratio is negative,that is,when the noise power is greater than the signal power,the prediction results are poor and the trend is severely inconsistent;When the signal-to-noise ratio is positive,the trend of the predicted results is basically the same as the actual data,and as the signal-to-noise ratio continues to increase,the prediction accuracy of the model becomes higher and higher.