Wavelet Denoising And Roughness Penalty Smoothing For Slope Deformation Monitoring Data

Accurate monitoring data are the precondition of accurate information of slope deformation and stability, as well as a basis for slope engineering and construction. Since monitoring instruments are always influenced by many factors such as manual operation, their own defects, climate and environment and so on, the original data always contain noise which will definitely affect the credibility of the results once being used to forecast and backstep. Therefore, it is of great significance for data utilization to efficiently eliminate the noise, improve the quality of the monitoring data and obtaine the real characteristics of slope deformation.At present, there are many de-noising approaches, from the traditional moving average method, least squares with a basis system, kernel smoothing, Fourier transform to the latest methods wavelet analysis and roughness penalty smoothing, all of which are widely applied in all kinds of fields. However, the traditional de-noising approaches cannot be well applied in slope deformation monitoring because of theory defects of their own and disadvantages in the process of the application. Thus, based on the purpose of probing the application in slope monitoring project by the latest two methods:wavelet analysis and roughness penalty smoothing, the study goes as follows:1. Make use of the wavelet thresholding de-noising experiment to study some factors which affect the result of de nosing such as wavelet function,thresholding rule,decomposition level and thresholding methods. Results indicate that compared with lower-order wavelet, the higer-order wavelet is generally better. When the signal-to-noise ratio of noised-signal is bigger, it is better to use minimax and rigrsure. Otherwise, sqtwolog and heursure is better. The optimal de-nosing affect, at least 90% of which, can be reached by adopting 5 decomposition levels. Compared with hard threshold, soft threshold is better.2. Since wavelet de-noising cannot deal with unequally spaced data, put forward Hermit interpolation which can transfer unequally spaced data into equally spaced data sequence. The feasibility has been approved in practice with sufficient data.3. Study the efficiency of roughness penalty smoothing through de-noising experiments. Results shows that roughness penalty smoothing can effectively filter the noise, eliminate noise mutation and strengthen signal quality, whose de-noising effect is as good as that of the wavelet threshold de-noising.4. The monitoring data of two slope constructions in practice are de-noised separately by wavelet analysis and roughness penalty smoothing which obviously eliminate the mutation in original data caused by the noise. The characteristics of slope deformation are clearer and the data quality is also improved. This successful application proves the efficiency of wavelet analysis and roughness penalty smoothing in de-noising monitoring data of slope.