单桩荷载与沉降关系曲线的神经网络预测方法

A Neural Network Algorithm for Predicting Load-Settlement Curve of Single Piles

本文提出了一个预测单桩荷载—沉降关系曲线的 BP神经网络模型。该网络模型包含两级网络 ,第一级网络以桩长、桩径、桩侧土层的分布情况 ,桩侧与桩端土层的物理力学指标为输入元 ,来先预测出单桩的极限承载力及其对应的极限沉降 ;第二级网络以第一级网络的输入元和输出元作为其输入元 ,进一步预测各分级荷载下桩的沉降。实际预测精度一般能达到 15 %以内 ,且稳定性较好 。

A neural network model for predicting the vertical load-settlement behaviour of single piles is proposed in the paper. The proposed network model is a two level network. The primary network, taking the pile length, pile diameter, distribution of soils along the pile,and the physico-mechanical characteristics of soils at the side and the end of the pile as the input data, is to predict the ultimate bearing capacity of the single piles and the corresponding ultimate settlement.The secondary network is to take the input data and the output values of the primary network as its input data to further predict the settlement of the pile under various stages of loads. The accuracy of this model from practical computation is generally within 15% with a good stability, indicating that neural network model is slightly superior to the conventional empirical formula methods and the numerical methods.