| The deformation of tunnel is a complex process.Due to the combined effects of many factors,the monitoring data are often nonlinear and non-stationary.How to extract and analyze the real deformation of the tunnel from the above nonlinear and non-stationary monitoring data is a very important part of the tunnel construction and operation stage.Traditional methods,such as time-frequency analysis,wavelet transform and empirical mode decomposition,have some problems in deformation data processing,such as lack of adaptability and end effect,which have a great impact on the results of monitoring data processing and are not conducive to the effective analysis and prediction of deformation data in the future.Therefore,based on the signal processing method of local mean decomposition(Local Mean Decomposition,LMD),this thesis studies a more effective denoising model to optimize the shortcomings of LMD method for denoising analysis of deformation data,and proposes a multi-scale combination model on this basis.The reliability of the above methods is analyzed and discussed according to the simulation experiments and engineering examples.(1)The principle and characteristics of LMD and SSA methods in deformation data denoising are analyzed,and SSA method is used to optimize the shortcomings of LMD single method in denoising process.In the process of SSA denoising,the setting of parameters has a great influence on the denoising results.Based on the existing research,the window length of one of the SSA parameters is fixed to half of the sequence,and the Hurst index and the contribution rate of eigenvalues are used to determine the second parameter of SSA,that is,the reconstruction order.Combining LMD with SSA,a denoising method based on LMD-SSA is studied.(2)According to the characteristics of tunnel surrounding rock deformation,the exponential function and sine signal are used to simulate and reconstruct the tunnel surrounding rock deformation signal.The denoising ability of LMD-SSA is studied in different SNR cases,and compared with the single denoising method,when the SNR is 15 d B,the RMSE of LMD-SSA is reduced by 7.1% and 5.0% respectively,when the SNR is 25 d B,At this time,RMSE decreased by 77.4% and 53.8%.When the signal to noise ratio is 5 d B,the accuracy of LMD-SSA is slightly lower than that of LMD because of the serious influence of noise on the signal.By comparing the wavelet threshold denoising,it is found that the proposed method is more adaptive in complex situations.Compared with the single denoising method LMD and SSA,LMD-SSA has higher accuracy,which is basically consistent with the conclusion of the simulation denoising experiment.Although part of the wavelet threshold denoising is better than the method in this thesis,the adaptability of wavelet threshold denoising is poor in different data.The feasibility and effectiveness of LMD-SSA for deformation data denoising are further verified.(3)Analyze the principle of the existing prediction model GM(1,1)and ARIMA,and optimize the cumulative reduction formula of GM(1,1).The LMD-SSA-GM-ARIMA combination forecasting model is constructed,which is based on denoising and further uses the characteristics of LMD decomposition to reconstruct the sum of the denoised PF component and the remaining PF component,thus obtaining two time-scale components with different physical meanings.The optimized GM(1,1)is used for modeling and prediction,and ARIMA is used for modeling and prediction of the volatility of the reconstructed PF component.The LMD-SSA-GM-ARIMA model is used to predict the vault settlement and surrounding rock convergence of Haitianbao Tunnel and Yumo Tunnel,and the results are compared with those of single GM(1,1),grey Verhulst and ARIMA.The results show that the LMD-SSA-GM-ARIMA model has high prediction accuracy and is a reliable and effective method. |
PDF Full Text Request