Abstract: In order to improve the accuracy of tunnel deformation prediction, firstly,we discussed the denoising results by different wavelet transformation parameters,and divided the tunnel deformation data into trend term and error term. Secondly,we established individual and combinatorial forecasting models to predict the trend term and error term, and compared the results so as to verify the proposed prediction model in this paper. The results show that the denoising results of Sym8 wavelet function are the optimum by using the soft threshold selection method,heuristic threshold criteria and eight-layer wavelet decomposition. By removing the maximum error and then determine the weights of the rest predicted values in reciprocal order, the prediction accuracy of trend term and error term by combinatorial forecasting model has improved by 2.5-3.5 times and 4.0-5.4 times respectively than that by individual forecasting model, hence the reliability of the prediction results are enhanced. Through example analysis of the prediction model,we verified the feasibility and validity of this prediction method,and the results of high feasibility could meet the requirements of large deformation prediction.

Key words: tunneling engineering, wavelet denoising, large deformation, combination forecasting, error reciprocal after removing maximum error, comparative analysis