Research On Geotechnical Deformation Problem Baed On Time Series Prediction Method

Geotechnical engineering deformation is the macroscopic reaction of a complex mechanics within system and it contains the information of mechanics evolution. If the evolution rule can be determined, the existing deformation data can be ued to model and predict future deformation. According to the predict result, engineers can adjust construction plan and take corresponding measures in time, which can effectively reduce the possibility of engineering accidents. This method successfully avoids the complex deformation internal mechanism and it can be used as an effective way of informatization construction and dynamic control for engineering. Therefore, the research of geotechnical engineering deformation prediction and control is of great significance.The engineering material and environment of geotechnical engineering is rock soil mass, which is an inhomogeneous and anisotropic elastic-plastic sticky body. Because of the complexity of geological conditions, the mechanics parameter and mechanics phenomenon of rock soil mass is random and uncertain, which results in the deformation prediction and control of geotechnical engineering is difficult. In addition, geotechnical engineering deformation is influenced by geological conditions, site conditions, ground load, construction method, construction schedule, time and temperature and other factors. Because of these factors, the deformation series contains intrinsic regularities of rock soil mechanics variation and certain randomness at the same time. That is, the measured deformation series can be decomposed into trend series and random series. Wherein, trend series reflects the inherent law of geotechnical deformation and it is the main basis for deformation prediction. Random series belongs to noise series and with certain stability. If choosing to remove random series, the precision and authenticity of prediction results will be reduced. Therefore, in the process of deformation prediction, the prediction model of trend series and random series should be established respectively according to their respective characteristics.Based on the theory of time series prediction, the wavelet transform, least squares support vector machine of particle swarm optimization (PSO-LSSVM) and autoregressive moving average model (ARMA) were combined together and a united prediction method and model of geotechnical engineering deformation was put forward in this paper. The basic idea was as follows:For the measured deformation data of early stage in construction, firstly using Db4orthogonal wavelet to decompose and reconstruct it into trend time series and random time series. The trend time series was pretreated by phase space reconstruction theory and then setting up a PSO-LSSVM model to predict. The random time series was predicted by ARMA model in EViews software. Finally, predicted values of two series were summed up as final prediction results. By this means, the prediction and control of deformation in later stage or operation stage could be realized. This method was successfully used to predict the deformation of a pit engineering example and a foundation engineering example. The result is satisfied.