Applying One Fuzzy Neural Network Model Based on Multivariate Phase Space Reconstruction to Predict Deformation of Deep Foundation Pit XI Xue-feng, FU Bao-chuan,

The ability to accurately and consistently predict deformation of deep foundation pit in civil engineering is required by the designer and manager in planning and conducting construction activities. Since deformation of deep foundation pit that belongs to a multi-variable nonlinear system evolution with time-varying is vague and uncertain, it is prone to a certain degree of prediction errors. Thus a prediction model which adopts a fuzzy inference method provides a solution to fit the uncertain and vague properties of deep foundation pit’s deformation. Furthermore, for its multi-variable and nonlinear characteristics with time-varying, a neural network based on multivariate phase space reconstruction is a feasible method. According to combination of these technologies above, multivariate time series predicting method for the deformation of deep foundation pit is studied, and an improved fuzzy neural network model is proposed based on multi-variable phase-space reconstruction technology for it. By the various time series delay and embedding dimension determined respectively in this model, the multi-variable series of excavation deformation for deep foundation pit have been done in the first phase space reconstruction. The neural network input extraction by the use of partial least squares regression method can be the strongest impact components. Finally non-linear fitting between the various components has been completed via the artificial neural network, which is also utilized to determine the significant fuzzy rules in fuzzy inference processes. With practical application for deformation prediction of deep foundation pit, the method's effectiveness has been verified. In addition, compared with the traditional BP network, fuzzy neural network has better convergence and accuracy.