基于BP神经网络的复合地基承载力预估研究

Study of Prediction of the Composite Foundation Bearing Capacity Based on BP Neural Network

鉴于目前单桩复合地基、双桩复合地基和四桩复合地基载荷试验确定承载力的缺陷,利用MATLAB神经网络工具箱的BP网络的理论,通过影响复合地基的7个主要指标来对单桩、双桩和四桩复合地基承载力进行预估。从预估的结果来看,双桩复合地基的承载力与载荷试验值最接近,误差较小;单桩复合地基次之,双桩复合地基最大,均满足工程上的要求。分析了单桩复合地基和四桩复合地基误差较大的原因:1.单桩复合地基不能够很好地反映桩与桩之间,桩与桩间土之间的相互作用机理;2.四桩复合地基的7个指标与实际存在较大的差异。

Taking into consideration the definitive bearing capacity defect on the foundation loading test of the single-pile, double-pile and four-pile composite foundation; Using BP network theory of the toolbox in the MATLAB neural network, the bearing capacity of the single-pile, double-pile and four-pile composite foundation is evaluated through seven pieces of chief impact index mark of composite foundation. Judging from the forecasted result, the bearing capacity of double-pile composite foundation has the closest value to the load test values, and the error value between the two is the least. And the single-pile value is larger while the four-pile value is at the most. But they can satisfy with the requirement of engineer. The cause to a larger error value on the single-pile and four-pile composite foundation is analyzed as follows： 1. The interacting mechanism of pile-and-pile and pile-and-earth is not able to reflect in the single-pile composite foundation. 2. The bigger difference between seven pieces of indexing mark and the reality still exists in the four-pile composite foundation.