基于GAAA-RBF神经网络的工程估价方法  

Novel Project Evaluation Model based on GAAA-RBF Neural Network

作　　者：叶青[1] 刘婧[1] 

机构地区：[1]华侨大学土木工程学院,福建厦门361021

出　　处：《数学的实践与认识》2013年第13期28-33,共6页Mathematics in Practice and Theory

基　　金：中央高校基本科研业务费专项基金(JB-ZR1162);华侨大学高层次人才科研启动项目(12BS131)

摘　　要：利用遗传-蚁群混合算法(GAAA),对RBF神经网络的主要结构参数中心矢量、基宽向量和网络权重进行组合优化,建立了GAAA-RBF神经网络组合算法的工程估价模型.将55个工程造价案例,随机抽取10个作为预测样本,剩下的45个作为训练样本.通过与相同结构的RBF神经网络相比较,结果表明算法克服了RBF神经网络易陷于局部极值、搜索质量差和精度不高的缺点,改善了RBF神经网络的泛化能力,收敛速度快,输出稳定性好,提高了工程造价的预测精度.In this paper, the Genetic Algorithm-Ant colony Algorithm （GAAA） was utilized to optimize the center of structural parameters, base-flat vector and the weight of neural network, finally establishing a novel project evaluation model based on GAAA-RBF. The paper chose 55 examples of construction engineering to build up the model of project cost estimation, among them, 10 samples were randomly chosen for forecast and the rest 45 for training samples. The comparison between RBF neural network with the same structure shows that the problem for trapping into partial extremes, poor searching and low precision can be solved by using GAAA-RBF neural network. Moreover, the generalization ability of RBF neural network was improved with highly speed of convergence and good stability, eventually increasing the precision of forecasting the project cost estimation.

关 键 词：工程估价模型 GAAA-RBF神经网络 最优参数组合 RBF神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]