Research On Deformations In Excavations Based On BPN

In recent years, speedy development of high-rise buildings and full use of subterranean space accelerate the development of deep excavations, which brings a mass of geotechnical engineering problems during the period of construction. Due to the complex stratum conformation and the defect in design and construction management, deep excavation accidents occur frequently, which result in huge economic losing and time delay, and make negative effects in society. The high accident rate is mainly caused by the super complicacy of foundation pit engineering itself. Problems in available theoretical calculation methods for the supporting system of foundation pit are first discussed, and then this article analyzed the mechanism of BP neural network and the failure mechanism of foundation pit supporting system. Based on the analysis of sample data, accident forms and their reasons in foundation pit supporting systems, the displacement models of foundation pit supporting patterns are established with the example data of case-in-site supporting system as training and prediction sample data through BPN. Trying use the more reliable improved BPN, this article put forward the forecast research on the level displacement of supporting structure with the various factors data influencing the supporting as input samples. The contrast between of prediction output and case samples shows that the algorithm of BPN which has strong nonlinear mapping capability can provide a feasible means to forecast the displacement of supporting structures in deep excavations. From the research, we could come to the following knowledge and conclusions: (1) ANN has the advantage of solve kinds of nonlinear problems with characteristics of stronger adjustability, nonlinearity, learning function and fault-tolerance. It is difficult to establish the revealing relation between input and output with traditional mathematic methods for the reason that problems of foundation pit engineering are vague and nonlinear, etc. And factors influencing the deformation of foundation pit are too many. Though the result of impact can be caught in the monitor of deformation, the impact degree of each factor can not be expressed accurately. Therefore, some methods based on mathematics including regression analysis, gray forecast have larger prediction errors. As for ANN, kinds of relation information exists in weights of the distributed memory network, which derives from knowledge learning , and can correctly reflect the relation between input and output. So ANN has a better prediction precision than other algorithms. (2) Under the circumstances of supporting pattern and brace scheme are determined, construction parameters are determinants impact the control of supporting wall deformation. As far as the complexity and diversity of impact factors are concerned, ANN should fully consider the impacts of various factors to improve the prediction accuracy. (3) As for some geotechnical engineering problems such as foundation pit deformation, etc. ,the prediction accuracy will be enhanced as the amount of ANN training data increases. Therefore, such a ANN prediction model is especially feasible for the district which...