Data Processing And Forecast Analysis Of Surface Deformation By TBM Tunneling

Construction of subway tunnel destroyed the balance of the original ground stress, followed by the inevitable deformation of the surface. For the reason that the subway goes through most urban area, around where the surrounding buildings are relatively dense. Hence, it is quite necessary to monitor and predict the ground surface deformation.Based on the subway construction interval of TBM tunnel in one certain city as the engineering background, FLAC3D is used for numerical simulation in the early stage of the construction. And construction monitoring site is arranged, the monitoring data are collected, of which with the wavelet method to detect and denoise. The last step was to establish the prediction model with time series and BP neural network for model optimization, then the optimal prediction model is applied to prediction work of the surface deformation.FLAC3D software model is established according to the lithology of this interval, construction factors, with the aim to simulate the ground subsidence after excavation of tunnel. It finds that ground subsidence right above the tunnel axis is largest, which is -17.9 mm. Analyzing record point settlement curve, the result is that subsider is roughly normally distributed, which reflects that the ground subsidence right above the tunnel axis is largest, and the more far away from the settlement, the smaller the settlement value. Compared with maximum subsidence value of FLAC3D stimulation, it is consistent with the actual measured maximum subsidence value, therefore, it is obvious that the numerical simulation method is suitable for the surface subsidence prediction work.Using wavelet method in the detection of monitoring data and eliminating gross error in data, and adopting the method of control variates to compare the influence of different threshold selection, selection of wavelet base, decomposed layers and scal selection to data denoising mean square error (mse) and signal-to-noise ratio of this data group. It is finalized that, denoising effect of sym5 wavelet, soft-threshold rules, rigrsure threshold rules, I layer decomposition, and scal-sln to this group data are best.Creating a single time series and BP neural network model after denoising of the data, and comparing the prediction accuracy of data before and after denoising, it proves the good effect of wavelet denoising method. Combining the advantages of these two kinds of single model, the optimal weighted model and residual error correction ARMA - BP model are established. Comparing the SSE, MSE and MAPE data of prediction models, it discovered that optimal weighted model, on the basis of wavelet denoising, is superior to other forecasting models, and it can be applied to the the surface subsidence prediction work of this subway interval.