【作者】 马辉；

【导师】 王雪青；

【作者基本信息】 天津大学 ， 管理科学与工程， 2006， 硕士

【摘要】 在建设工程项目生命周期的多次计价中,投资估算是最粗略的,但却是对整个项目影响最大的。它是项目决策的依据,也是制定投资计划和控制投资的有效工具。然而,由于建设项目投资估算工程所处的阶段比较靠前,许多有关工程费用的信息无法确定,并且在工程的执行过程中还会发生许多不可预见的影响工程造价的事情,使得项目投资估算的编制显得十分困难。所以找到一种切实可行的建设工程项目快速投资估算方法有着非常深远的意义。本文在大量查阅和深入研究国内外建设工程项目投资估算方法和估算模型的基础上,鉴于影响建设工程项目投资估算的影响因素和工程咨询公司的数据基础条件,提出了将多元回归分析和径向基函数神经网络相结合的投资估算模型。首先,借用SPSS软件对历史数据进行多元回归分析,确定回归模型,通过对模型的假设检验确认模型的有效性。然后,根据径向基神经网络原理构造了三层径向基函数神经网络模型,详细阐述了径向基神经网络的学习算法和训练仿真过程,并结合历史数据,在MATLAB语言环境下,对神经网络进行训练、仿真以及校验,从而对模型的可靠性加以验证。最后,本文在基于多元回归估算模型和基于径向基神经网络的投资估算模型的建模过程中,通过对两种模型的比较和局限性分析,总结了两种模型的优缺点,提出将两者相结合运用于建设工程项目投资估算的一套新的思路和方法。实践证明这种组合后的方法克服了原有两种估算模型的局限性,具有更高的估算精度。同时,本文提出了构建建设工程项目投资估算信息系统的思想,通过分析系统的结构和数据流程,对主要模块进行了概要设计,为系统的开发提供了理论指导。更多还原

【Abstract】 The investment estimation is the most unshaped one as well as the most influential one among all the evaluations during the long period time of construction project. It is the foundation of project decision and the effective tool of making investment plan as well as controlling it. But it is difficult to estimate the investment because of its preliminary position and the situation that many unpredictable factors which influence the construction cost might appear in the operating of the whole project. So it is vital to find a feasible method that could be used for investment estimation of construction projects.Based on great amount of investigation and deep analyze of investment estimation methods and models of construction project both domestic and abroad, according to the influential factors of investment estimation and data of construction consulting companies, this paper developed a investment estimation model based on multiple statistics analysis and RBF neural network. Regarding to the model based on multiple statistics analysis, the effectiveness has been proved by setting up regressive model through SPSS and virtual examination; Regarding to the model based on RBF neural network, 3 layers RBF neural network has been built, particularly expatiate the calculation of network and neural network has been trained and examined by testing figures under the circumstance of MATLAB language. A new system of thought and method is invented after comparison and limitation analysis on both the above mentioned models which were finished in the course of establishing them and this system of using the two models comprehensively has been proved to be more specific compared to the traditional systems of using them individually. In the last, the information system of investment estimation was introduced. Some theoretical support has been offered by the analysis of several aspects such as functional demand, natural demand, data flow etc, the program of divisions of the system and the brief design of respective divisions.更多还原