摘要
通过采用神经网络工具,探讨沥青混合料的抗剪强度预估方法.通过对比,采用不同输入参数,选定沥青类型、集料类型、空隙率、级配类型、公称粒径、油石比等六个影响因素作为输入参数.引入了误差分级迭代法进行网络学习训练,通过对比常规BP算法和误差分级迭代法,发现后者能有效减轻初始权值和阈值对训练和样本预测的影响,也能较好控制样本预测的误差.因此,建议采用基于误差分级迭代法的BP神经网络方法,预测沥青混合料的抗剪强度.
Neural networks tool was adopted as a forecast method of asphalt mixture shearing strength. By comparing different input parameters, six influence factors were determined as the input parametern, including the asphalt type, gravel type, void percent, grading type, particle size and asphaltaggregate-ratio. Then, error grade iterative method was introduced for the neural networks training. A comparison between the results of general BP algorithm and the error grade iterative method show that the latter was able to reduce the influence of original weight value and the threshold value on the sample training and forecasting, and the latter is better in the forecast error control. Thus it is proposed that the error grade iterative BP neural networks method be applied to the asphalt mixtures shearing strength forecast.
出处
《同济大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2008年第9期1191-1194,1204,共5页
Journal of Tongji University:Natural Science
基金
国家杰出青年科学基金资助项目(50325852)
关键词
沥青混合料
强度预估
BP神经网络
误差分级迭代
抗剪强度
asphalt mixture
strength forecast
BP neural networks
error grade iterative
shearing strength