Abstract: There are many random errors in the monitoring data of tunnel surrounding rock. The monitoring data is usually denoised for reducing or eliminating the disturbance of the random errors. Based on the theory of wavelet transform, as an example, the monitoring data of a tunnel surrounding rock is processed by a technique of wavelet denoising with db3 wavelet function and heursure soft threshold. The result of pressure prediction is given by using 5-15-1 BP neural network for the original data and the de-noised data, and the training steps are 2 448 and 450 respectively. The error of the pressure prediction for the original data is larger than that for the de-noised data. The results show that the method of wavelet denoising is efficient and reliable, is sensitive to distinguish noise and useful information, is particularly suitable to analyze monitoring data of surrounding rock.

Key words: monitoring of surrounding rock, wavelet transform, noise reduction, Mallat algorithm