【作者】 杨润莲；

【导师】 李红昌；

【作者基本信息】 北京交通大学 ， 项目管理， 2009， 硕士

【摘要】 招标是我国现行的工程项目、服务和物质采购的重要方式,是社会主义市场经济的重要组成部分,其本质是一种交易方式。现行的招标评标方式主要是基于专家等评标委员会成员打分,这或多或少的存在个人主观因素的影响,这相应地减少了评标工作的科学性和公正性的成分,就产生了进一步解决评标过程中的科学性和公正性的需要,而评标模型是首要解决的问题。各种不同的招标方式和模型有其特殊性和适用的范围,在实际应用中必须考虑项目的特点。评标是招标过程中的重要组成部分,其决定了标的实现,其本质是一个高维非线性空间的最优化数学问题。SVR(Support Vector Regression:支持向量回归)算法是基于统计学习理论的小样本学习机器,与人工神经网络的经验风险最小化原理不同,其建立在结构风险最小化原则基础上,通过非线形内积函数将线性不可分的低维空间数据映射到一个线性可分的高维特征空间,在这个特征空间中进行分类和回归拟合的方法,从而有效克服维数灾难和过拟合问题。较之于人工神经元网络,该算法具有小样本、泛化性能好以及所得解是全局最优解的优势,这恰好弥补神经网络的不足。理论已经证明,支持向量机在参数选择合适的前提下可以任意精度逼近任意非线性函数。经典的SVR算法只能解决一维输出变量的回归问题,本文将一种改进的SVR算法引入建设工程项目招标评标之中,可以有效解决多维输出变量的回归问题。GA(Genetic Algorithm:遗传算法)作为一种群体进化的仿生全局最优化算法,具有许多传统梯度优化算法不具备的优点,已被广泛应用于各个领域。将GA与改进的SVR算法相耦合,采用遗传算法在SVR算法训练过程中自动搜索能使训练效果最优的SVR网络参数,以提高SVR算法的泛化性能,SVR算法运于招投标的文献资料可检索到的约20余个,而形成改进的GA-SVR算法,并将之应用到建设工程项目评标中则是一个全新的尝试。首先,论文从招标投标的历史出发,讨论了招标投标的沿革,分析了我国招标投标的历史,探讨了影响评标结果的因素,分析了常用的招标方法和模型,指出了它们的特点和适用范围,不存在包罗万象的普适的评标方法和模型,而完善招标投标的数量化和模型化是具有实际意义的事情。在总结20多个项目实际评标结果的基础上设计了基于改进GA-SVR算法的评标模型,并将设计好的改进GA-SVR算法的评标模型在A公司的B项目评标中进行了实际应用,并结合了实际工程建设中的效果,证明了这个模型具有良好的实用性。本文的主要结论有:第一,将改进GA-SVR算法的评标模型创新性地应用于招投标领域,B项目的成功证明了该模型的可靠性和可实施性。第二,训练好的改进GA-SVR模型能够在一定程度上替代评标专家用于建设项目的招标评标工作,减少人为因素的影响,提高了评标结果的客观性和公正性;同样这个评标模型也可以作为评标的辅助工具,来检验评标结果是否合理。第三,招投标中的评标工作,实质上是对经济标、技术标的各项指标综合评价,房地产项目的招投标有其基本特性,对工期、质量等指标有着较为特殊的要求,在此将评标因素进行优化控制,通过改进GA—SVR算法模型映射评标因素间的非线性关系,为评标因素的权重的设定提供决策依据。更多还原

【Abstract】 Invitation to tender is an important way in project, service and material purchasing, and it is also an important component of Socialist Marketing Economics .It is a way of trade essentially. Because of the members of the bid evaluation committee are more or less impacted by subjective factor in the process of bidding evaluation, which decrease scientificity and fairness of the evaluation. The first problem of evaluation model is to increase scientificity and fairness of bidding evaluation. Different modes of invitation to tender have their own particularity and range of convenience and we must consider the trait of the given project. Bidding evaluation is an important component of bid inviting, which decide the achievement of bid and it is a decision problem and decision set. Transfer a dimensional vector to duality is complex mathematic process.Improved algorithm of GA-SVR ( Support Vector Regression) is the developing of support vector machines and the core of the statistical learning theory. It is a method based on the principle of structural risk minimization which maps the data from an inseparable low-dimensional space to a high-dimensional linear separable space through the non-linear function of the inner product, and then analyses the data by classification and regression analysis. This method has been applied to the construction field, the financial and economic field, but its application to the evaluation is still an innovation.Firstly the paper starts from the history of bid inviting and tender, discusses the evolution of bid inviting and tender, analyses the history of bid inviting and tender in China, probes into the factors which affect the result of evaluation, analyses usual methods of bid inviting, points out their characteristics and application scope. There is no universal evaluation approach, perfecting the evolution model of bidding is a meaningful thing. Secondly, the paper designs the improved GA-SVR algorithm evaluation model which translates the elements of the multi-vector represented tender program into a binary vector and compares these binary vector in two-dimensional space, derived the optimal plane, calculate the optimal vector. Finally author applies the improved GA-SVR algorithm evaluation model to a company’s B project evaluation and combines with the results of the actual construction, proves that this model has good practicability.The conclusions: Firstly, the GA-SVR algorithm evaluation model of innovation applied to the bidding, the success of Project B proved that the reliability of the model and enforceability. Secondly, the trained GA-SVR model can be improved to some extent, experts in alternative evaluation of the tender evaluation for construction projects, work to reduce the influence of human factors to improve the results of the evaluation objectivity and impartiality; the same in this evaluation model can be used as complementary evaluation tools to test evaluation results are reasonable. Thirdly, the bidding of the evaluation work, in essence, marked on the economy, technology underlying the comprehensive evaluation index, real estate bidding for the project has its own basic characteristics, schedule, quality and other indicators have a more specific request. It will optimize the evaluation factor control, by improving the GA-SVR model mapping algorithm is non-linear relationship between the evaluation factors for the evaluation factors set the weights to provide basis for decision making.更多还原