基于RBF人工神经网络的房屋建筑工程造价预测研究  

Research on building project cost estimation based on RBF artificial neural network

作　　者：莫俊文[1] 吴海龙 Mo Junwen;Wu Hailong(School of Civil Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China;Gansu Gannan Housing and Urban Construction Bureau,Hezuo 747000,China)

机构地区：[1]兰州交通大学土木工程学院,甘肃兰州730070 [2]甘肃省甘南州住建局,甘肃合作747000

出　　处：《工程造价管理》2018年第6期15-23,共9页Engineering Cost Management

基　　金：国家自然科学基金项目(项目号:70961004)

摘　　要：将设计阶段的消耗量限额指标作为影响因素,寻求消耗量指标与房屋建筑工程造价之间的非线性关系,采用RBF人工神经网络模型预测工程造价。文章调查收集了兰州市房屋建筑领域比较典型的20幢建筑物的单方造价指标和消耗量指标,运用主成分分析法进行KMO和Bartlett检验,计算变量的初始特征值和提取平方和载入,对样本数据进行筛选,挑选出影响工程造价的主要指标。在MATLAB环境下,将所筛选的造价影响因素作为模型训练样本集的输入向量,将主要造价指标作为输出向量,对样本集数据归一化后,调用RBF人工神经网络对数据进行训练,构建预测工程造价的广义RBF人工神经网络模型。最后通过四个实际样本对模型进行验证,造价指标预测值和实际值的对比分析表明,所建模型的预测精度是可靠的。Consumption quota indicator in des consumption indicator and cost of housing bu ign din stage is regarded as an influence factor to seek the non-linear relationship between g construction project. RBF artificial neural network model is adopted to forecast the project cost. Unilateral cost indicator and consumption indicator of more typical 20 buildings in Lanzhou building field are investigated and collected in the paper. Principal component analysis is applied for KMO and Bartlett test to calculate variable initial characteristic value and abstract square and load. Sample data is screened to select main indicators affecting project cost. The screened cost influence factors are regarded as input vector of model training sample set under MATLAB environment. Main cost indicator is regarded as nput vector. After the sample set data is normalized, RBF artificial neural network is called for training data. Generalized RBF artificia neural network model for predicating project cost is constructed. The model is verified through four actual samples. Comparative analysis on cost indicator predicted value and actual value shows that the constructed model has reliable prediction precision.

关 键 词：工程造价预测 主成分分析 RBF人工神经网络 

分 类 号：TU-9[建筑科学]