The Diaphragm Wall's Deformation Forecasting Based On Neural Networks

From the point view of deformation forecast in construction of the concrete diaphragm wall pit, artificial neural network (ANN) and finite element numerical method (FEM) are adopted separately to solve this problem. The main work can be summarized as follows.First, based on the analysis on different types of ANN, the network applications of BP neural network and RBF neural network are researched respectively, from which a new improved BP network construction method is proposed.Second, the decisive parameters governing the diaphragm wall's deformation are brought forward from the model analysis. Two types of diaphragm wall's deformation forecast model are established based on the BP neural network and RBF network respectively, in the first of which the diaphragm wall is taken as an integer, in the second the diaphragm wall is discretized according to the digging phases and survey points. Furthermore a hybrid method to predict the diaphragm wall's deformation based on the above two neural networks BP and RBF is put forward.Third, the finite element numerical analysis model is established considering the pit's spatial and different soil strata effects and many construction factors, in which the nonlinear hyperbola soil spring constitutive model is introduced and the stiffness matrix of plates on nonlinear elastic foundation is derived. The pit's whole excavation course can be effectively simulated using this model and the effects of different boundary conditions on the calculation results are discussed.It can be concluded from practice that the results of the diaphragm wall's deformation forecast ANN model and FEM numerical calculation model can satisfy the requirements of practical engineering projects and supply good theoretical basis for structure design and construction.