The Prediction Research Of Foundation Pit Deformation Rolling Based On LSSVM-ARMA

The development of city scale promotes the construction of urban underground engineering,especially the construction of the subway.However,the stability of the deep foundation pit is difficult to be guaranteed due to the complexity of the surrounding environment,geological conditions and process.The problem of foundation pit deformation becomes more and more serious and the prediction of the deformation of the foundation pit has become an indispensable part in the construction of deep foundation pit.In order to precisely predict the deformation of foundation pit,the LSSVM-ARMA(least squares support vector machine-auto regressive moving average model)based on wavelet transform is proposed to realize the rolling prediction of foundation pit deformation time series.Firstly,due to the different frequency of deformation monitoring of foundation pit,uses three-order spline interpolation method to interpolate the deformation data of foundation pit of subway station.Secondly,due to the highly nonlinear deformation data of the foundation pit,the deformation data includes the inherent trend of soil mechanics and the data is randomness.Wavelet transformation is used to decompose and reconstruct the deformation data of foundation pit,so the trend term and random term of deformation data of foundation pit are effectively distinguish.Finally,the LSSVM-ARMA model is used to rolling predict the time series of trend term and random term.The sum of the predicted value of the trend term and the random item is taken as the final prediction value.The predicted values of LSSVM-ARMA model and the predicticted values of PSO-LSSVM(Particle Swarm-least squares support vector machine)model and SVM(support vector machine)model are compared and analyzed.The result shows that the prediction accuracy of the LSSVM-ARMA model is high.It is of great significance to improve the safety of foundation pit and similar projects in Guangzhou.At last,in the paper,the GUI(graphical user interface)prediction system that is based on MATLAB is designed,the visualization of LSSVM-ARMA model prediction process is realized.The prediction process is displayed dynamically by graphical results,and it has important practical value.