广义回归神经网络预测加筋土支挡结构高度

General ized regression neural networks for predicting design height of GRW

土工合成材料加筋支挡结构(Geosythetics-ReinforcedRetainingWall,简称GRW)设计方法主要是建立在似粘聚力理论基础之上的半经验设计法。由于土性及加筋机理的复杂性,常常要对它们进行人为假定,导致计算结果差强人意。神经网络方法与传统方法的不同之处在于不需要主观假定,而是模拟人脑思维,通过数据样本的学习来获得预测结果。引入神经网络技术来预测加筋土支挡结构的设计高度是一种新尝试。由于本问题具有样本容量非常有限、影响因素复杂多样的特点。因此,采用适用于稀少样本数据的广义回归网络(GeneralRegressionNeuralNetwork)来预测加筋土支挡结构设计高度。基于MATLAB神经网络工具箱及文献犤1犦的挡墙离心模型试验结果,建立了一个可用于加筋支挡结构设计高度预测的GRNN网络。通过对足尺试验、实际工程及模型试验结果的检验,表明网络的学习是成功的,具有一定指导意义。

Current design methods for Geosythetics-Reinforced Retaining Wall(GRW)are mainly semi-empirical method based on homo-cohesion theory.Due to the complexity of soil property and reinforcement mechanism,many subjective assumptions have to be made,which often lead to error results in the actual cases.An artificial neural network(ANN)model is fundamentally different from the conventional calculation medel.One of its distinctive features is that it is based on experimental data rather than on assumptions made in developing a mathematical model.The ANN model learns from experimental data and forms neural connection stimuli from the learning process functioning somewhat like a human brain.Because of its unique learning,training,and prediction characteristics,the ANN model has great potential in soil engineering application,particularly for situations where good experimental data are available and where conventional constitutive modeling may be difficult and time consuming.Based on MATLAB neural network toolbox and centrifuge test data,a generalized regression neural network(GRNN)was designed for predicting the design height of reinforced retaining walls,which can give better prediction results with fewer training specimen data.Checked by some full scale test data and case history measured values,the network proved to be working successfully and provided satisfactory predictions for the design height,implying that the GRNN is applicable for predicting design height of reinforced retaining walls.