基于SVR的工程建设项目快速投资估算方法研究  

Study on the Fast Investment Estimation Method of Construction Projects Based on SVR

作　　者：陈小波[1,2] 张媛媛 崔平 CHEN Xiao-bo;ZHANG Yuan-yuan;CUI Ping(School of Investment&Construction Management,Dongbei University of Finance&Economics,Dalian 116025,China;Construction Management Research Center,School of Investment&Construction Management,Dongbei University of Finance&Economics,Dalian 116025,China;XIONG BING Model Worker Innovation Studio of China Construction Second Engineering Bureau,Beijing 100160,China)

机构地区：[1]东北财经大学投资工程管理学院,辽宁大连116025 [2]东北财经大学投资工程管理学院工程管理研究中心,辽宁大连116025 [3]中建二局熊兵劳模创新工作室,北京100160

出　　处：《工程管理学报》2020年第1期143-148,共6页Journal of Engineering Management

基　　金：国家自然科学基金青年科学基金项目(71701033);大连市青年科技之星项目(2017RQ005).

摘　　要：在建设项目前期,如何快速而准确地估算工程项目的造价,对项目的投资决策具有很大的意义。针对传统造价估算方法的不足之处,采用SPSS统计分析软件进行工程造价指标的相关性分析及指标体系选取,将之作为输入变量,使用真实案例训练集样本训练SVR模型并进行仿真模拟预测。为了验证提出的SVR模型的有效性,引入BP人工神经网络来进行预测结果的对比验证。结果表明,SVR模型得到的预测值平均绝对百分比误差约为5%,拟合优度R2高达0.97,远小于BPNN模型的预测误差14%,即提出的SVR估算模型要比BP人工神经网络预测模型具有更良好的泛化能力,预测精度更高,因此其在工程项目前期投资估算实践中具有一定的现实意义。At the early stage of a construction project,how to estimate project cost in a quick and accurate way is of great significance to the investment decision of the project.In response to the shortcomings of traditional cost estimation methods,SPSS statistical analysis software was used to conduct correlation analysis of engineering cost indicators and establish an index system,which was taken as input variables.Real case training set samples were used to train the SVR model and conduct simulation prediction.Finally,to verify the effectiveness of the SVR model proposed in this paper,BP artificial neural network was employed to carry out the comparative verification of prediction results.The results show that the SVR model's predictive value of mean absolute percentage error is about 5%,goodness-of-fit R2 is as high as 0.97,which is far less than the prediction error of 14%in the BPNN model,indicating that the proposed SVR estimation model has better generalization ability and forecasting accuracy,which is better than the BP artificial neural network prediction model.Consequently,the findings of this study has a certain practical significance in the prophase of the project investment estimation practice.

关 键 词：SVR:BP神经网络 成本估算 仿真模拟 

分 类 号：TU723.3[建筑科学—建筑技术科学]