Research Of Pile-anchor Supporting Pit’s Deformation Based On Finite Element And Neural Network

With China’s urbanization process accelerated, super high-rise buildings, subway project growing gradually, deep foundation pit engineering develops towards larger, deeper in scale, and is subject to complex venue and sensitive surroundings et al. Therefore, we should not only ensure its strength and stability, but also strictly control the deformation of the supporting structure in excavation process. Because of uncertainty and regional characteristics in deep excavation engineering, there are many problems in this field worth studying.In this paper, deformation properties of pile anchor support structure in deep pits were studied. The current research state of composite support structure, deep foundation deformation, influencing factors and the conventional design methods of pile anchor supporting pit were described. Researches showed that the composite supporting structure of upper anchor pile and lower soil nailing wall analyses was more suitable for finite element analyses. Based on finite element software PLAXIS2D, two kinds of deep foundation pit model were set up:the first was for pile anchor supporting pit in a single kind of soil, by using grey correlation method, deformation had been researched under different ground overload, prestressed anchor and soil stiffness modulus, pile embedded depth and flexural rigidity; the second was the composite suppoting pit as described before, the deformation of the pit had been analyzed under different rock shoulder width, anchor prestress and rock-socketed depth. Then an project instance in Guangzhou was calculated by the use of the second model,2methods were used to predict the engineer’s deformation, including FEM and time series forecast based on BP neural network. Finally, the results of the2methods and actual monitoring results were compared. The main results of the paper:1) From the results of gray relational analysis of maximum pile’s horizontal displacement, maximum surface subsidence and uplift under different influencing factors, the horizontal displacement of pile and the surface subsidence were most affected by ground overloaded, the pit bottom bulge was most affected by prestressed anchor. 2) Through the deformation of the second model calculated by FEM, the deformation curve were closely related to the first row of prestressed anchor’s position; pit deformations were affected most by pile’s rock-socketed depth. But the prestressed anchor at the bottom of pile and rock shoulder width affected less.3) The displacement at the top of pile and surface subsidence were predicted by BP network. Prediction effect was different when using different training functions (trainbr, trainlm, trainscg, trainbfg), the issues raised in this article was more suitable for trainlm function.4) Through the comparison between FEM(finite element method) and BP neural network method for calculating the deformation of the engineering instance, we saw that both methods can reflect the overall trends of the deformation in the pit, and the BP network’s computing speed and prediction accuracy was much higher than the FEM. But BP network can’t made it clear that the deformation mechanism, so the neural network cannot replace the FEM.