基于GA-BP神经网络的泡沫轻质土强度预测  

Strength Prediction of Foam Light Soil Based on GA-BP Neural Network

作　　者：周中[1] 邓卓湘 陈云 胡江锋 ZHOU Zhong;DENG Zhuoxiang;CHEN Yun;HU Jiangfeng(School of Civil Engineering,Central South University,Changsha 410075,Hunan,China)

机构地区：[1]中南大学土木工程学院,湖南长沙410075

出　　处：《华南理工大学学报（自然科学版）》2022年第11期125-132,共8页Journal of South China University of Technology(Natural Science Edition)

基　　金：湖南省自然科学基金资助项目(2020JJ4743)。

摘　　要：泡沫轻质土的抗压强度是其重要的力学性能。精准地预测和调整泡沫轻质土的抗压强度,对于提高施工效率有重要的现实意义。为实现对泡沫轻质土抗压强度的智能控制和优化,设计了包含4节点输入层、8节点隐层、1节点输出层的拓扑结构,输入层采用遗传算法(GA)对BP神经网络的权重和阈值进行改进。以水固比、粉灰比、细集料掺合比以及气泡率4个参数作为输入参数,28天抗压强度为输出参数,以室内实验数据作为样本,使用均方差(MSE)、决定系数(R2)和相对误差等对优化前后两种模型进行验证和对比分析,并以此为基础建立了基于不同性能需求的配合比设计方法。结果表明:相比BP神经网络,GA-BP神经网络训练的适应度函数值更大、均方差更小,预测值与实际值的拟合度可达到0.946,具有更强的预测精度和泛化能力,同时遗传算法的全局搜索能力也弥补了BP神经网络容易陷入局部最优的缺陷,且能更好地指导粉煤灰泡沫轻质土强度预测配合比设计。基于GA-BP神经网络的泡沫轻质土强度增长预测模型可实现对泡沫轻质土抗压强度的灵活调整,对于工程施工具有重要的参考价值。Compressive strength is an important mechanical property of foamed lightweight soil. Accurately pre-dicting and adjusting the compressive strength of lightweight foam soil is of great practical significance for improv-ing construction efficiency. For intelligent control and optimization of foam light soil, this study designed a topologystructure including 4 node input layer, 8 node hidden layer and 1 node output layer. The weight and threshold ofBP neural network were improved by genetic algorithm(GA) in input layer. Using the four parameters of water-solidratio, fly-ash ratio, fine aggregate mixing ratio and bubble rate as input parameters and 28-day compressive strengthas output parameters, the two models before and after optimisation were validated and compared using meansquared error(MSE), coefficient of determination(R2) and relative error as samples. Based on this, a method for de-signing the mix ratio based on different performance requirements was established. The results show that comparedwith BP neural network, the GA-BP neural network has a larger fitness function value and smaller mean square de-viation;the fit between the predicted and actual values can reach 0. 946, with stronger prediction accuracy and generalization ability;the global search ability of the genetic algorithm also makes up for the defect that BP neural net-work can easily fall into local optimum, and can better guide the fitting ratio design of the strength prediction of flyash foam lightweight soil. The GA-BP neural network based strength growth prediction model for foam lightweightsoils enables flexible adjustment of the compressive strength of foam lightweight soils, and it is of important reference value for engineering construction.

关 键 词：泡沫轻质土 抗压强度 BP神经网络 遗传算法 GA-BP神经网络 

分 类 号：U414[交通运输工程—道路与铁道工程]