Simulation and prediction of engineering deformation through comparisons of time serial and wavelet decomposition

Deformation on earth's surface covers crust displacement due to tectonism and land slide or subsidence caused by human activities or natural evolvement. The former is explained by tectonician through geologic investigation and researches. The local creep deformation arisen from engineering activities on earth's surface is complex and is often related with local engineering safety. So this kind of deformation attracts wide attention of many scholars and engineers, and is therefore discussed in this paper in way of simulation and prediction. In our work, we compared Auto Regressive (AR), Moving Average (MA) and ARMA models and used AR model to replace other time serial models based on their equivalence. It is found that AR or ARMA are fit for modeling or prediction, but it is hard to obtain deformation mechanism. Wavelet transform (WT) has shown great potential in information extraction and identification. It is also used as a tool to deal with deformation extraction and analysis in this paper. Tests have shown that it can be applied to distinguish different components from mixed observation serials. It is known that an observed serial in deformation monitoring is composed of sophisticated components and each represents different contents and is attributed to some acting factors. In this research, regional and engineering deformation observation is employed as inputs for wavelet decomposition; contents from different frequency scales are obtained at different layers. Deformation trend and rapid deformation changes are found from this multiple inspection transformation. Practical examples are given to reveal the feasibility of wavelet decomposition as a useful inspection tool for deformation analysis. From these work, we come to see that AR approach of time serial is fit for modeling and prediction, while wavelet transform is more flexible in deep inspection of deformation details and can exhibit subtle variation in observation serials. So some comprehensive comparisons are made in terms of time domain and spectrum domain to summarize merits and demerits of both methods. Practical data from field using some instruments are used for analysis and validation to verify the flexibility of suggested models.