Deformation Analysis On Geotechnical Construction Baed On Chaotic Time Series Prediction Method

Measured datas from geotechnical construction deformation contain the dynamics evolution rule within the engineering system, having similar characteristics of the information between local with overall and history with future. The suitable model can be established based on the dynamic evolution information of deformation datas from monitoring been fully exploited and the development rule of deformation tendency been determined to predicts the future deformation. According to guiding construction by the predict result, the possibility of engineering accidents can be effectively reduced, which is of great significance to the achievement of the informationization construction and dynamic control of engineering.Geotechnical construction belongs to the open complex system and its deformation show a complex nonlinear dynamic characteristics because of the various influencing factors, such as geological condition, site and the surrounding environment, construction conditions and load, the change of the underground water level and climate, etc, which is difficult to accurately predict the deformation on account of its internal characteristics and evolution rule are difficult of using the traditional analysis method. Therefore, the deformation and prediction of nonlinear characteristics of geotechnical construction has become an important research direction in the field of geotechnical engineering.Based on the actual monitoring data of geotechnical construction deformation and the theory of chaotic time series prediction, Firstly, using autocorrelation function method and mutual information method to determine the delay time τ, respectively using saturated correlation dimension method(G-P algorithm) and Cao method to determin embedding dimension m, of the two important parameters of phase space reconstruction on deformation time series be pretreat by two sub-spline interpolation metheod. Then a chaotic characteristics determination method of the improved small data sets combines with improved surrogate data be proposed, the reconstruction step of original small data sets and the phase generator of surrogate data method have been improved, the reliability of the calculated maximal Lyapunov exponent can be tested using this method. Finally, adding-weight one-rank local-region and maximal Lyapunov exponent method model of chaotic forecasting method have been established, simultaneously the least squares support vector machine model(LSSVM) based on chaotic theory has been established, application of the three models predict and comparative anylysis the future deformation. Among the LSSVM modle, its two parameters (penalty coefficient γ and kernel parameter σ) were determinated by a method of using uniform experimental design to optimize select the best combination of parameters of particle swarm optimization(PSO).