基于改进SVM的创新项目评价模型研究

发布时间：2018-01-08 16:09

本文关键词：基于改进SVM的创新项目评价模型研究　出处：《重庆理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：科技创新是促进国家经济增长的重要源泉,也是我国创新体系中的重要组成部分。目前全国各地均已建立了不同类型的项目创新平台,为研究者、生产者以及管理者等参与主体开展深度合作、资源共享、进行合力创新攻关带来了方便。对项目创新平台上的项目数据进行提取分析,进行项目的准确评价有着重要意义。由于创新过程是一个极其复杂和不确定性的动态社会过程,整个项目中涉及到的数据非常复杂尤其是今后随着各领域的交叉合作,平台上的项目数据会越来越多,如何对项目进行客观评价,为管理者提供准确的决策信息成为了一个亟需解决的重要问题。为解决以上问题,论文从数据挖掘的角度出发,结合机器学习方法建立评价模型。论文所涉及的主要工作包括以下几个方面:第一,介绍了关于科技创新的背景和意义,概述了当前国内外学者对解决创新项目评价问题的研究现状,以及解决问题所使用到的相关理论知识,并分析了其不足之处,最后展望了建立模型所使用的新方法。第二,论述了建立项目评价模型所涉及到的理论方法和技术,主要包括降维算法理论、遗传算法、支持向量机理论等知识。第三,依据确立的项目评价指标体系采集相应的数据并预处理获得实验数据,选用不同的核函数利用支持向量机分类器进行学习训练,通过比较预测准确率得出了最佳的核函数。为了削减数据的冗余信息,减少分类器的学习训练时间提升模型的性能,使用不同的降维算法对实验数据进行特征提取,将处理后的样本输入分类器进行分类任务,实验结果表明LLE算法在分类中应用良好。第四,通过对构建的项目评价模型LLE+SVM进行分析,指出了不足之处并进行了两点相关改进。主要是从前端降维和后端分类入手,利用样本数据自身带有的类别标签信息将传统的局部线性嵌入算法改进为带有监督功能的降维方法,增强了对数据的降维效果;针对支持向量机中核函数参数以及惩罚因子最优问题,使用改进遗传算法对支持向量机进行参数寻优,得到整体性能最佳的支持向量机。通过上述改进最终确立了高效的创新项目评价模型。准确评价是对创新项目进行有效管理的首要条件,基于本文所构建的项目评价模型能够高效准确地评估项目,对于以后在创新项目平台上进行快速客观地管理决策项目,具有重要的现实意义。
[Abstract]:Scientific and technological innovation is an important source of promoting national economic growth and an important part of China's innovation system. At present, different types of project innovation platforms have been established all over the country for researchers. Producers as well as managers and other participants to carry out in-depth cooperation, resource sharing, joint efforts to solve the key innovation brought convenience. The project data on the project innovation platform for extraction and analysis. It is very important to evaluate the project accurately because the innovation process is a very complex and uncertain dynamic social process. The data involved in the whole project is very complex, especially with the cross-cooperation of various fields in the future, the project data on the platform will be more and more, how to evaluate the project objectively. Providing accurate decision information for managers has become an important problem that needs to be solved. In order to solve the above problems, this paper starts from the angle of data mining. The main work of this paper includes the following aspects: first, the background and significance of scientific and technological innovation are introduced. This paper summarizes the current research status of domestic and foreign scholars to solve the problem of evaluation of innovative projects, as well as the relevant theoretical knowledge used to solve the problem, and analyzes its shortcomings. Finally, the new methods used to build the model are prospected. Secondly, the theoretical methods and techniques involved in the establishment of the project evaluation model are discussed, including dimensionality reduction algorithm theory and genetic algorithm. Support vector machine theory and other knowledge. Third, according to the established project evaluation index system to collect the corresponding data and pre-processing to obtain experimental data, select different kernel functions to use support vector machine classifier for learning and training. The best kernel function is obtained by comparing the prediction accuracy. In order to reduce the redundant information of the data, the learning and training time of the classifier is reduced to improve the performance of the model. Different dimensionality reduction algorithms are used to extract the features of experimental data and the processed samples are input into the classifier for classification task. The experimental results show that the LLE algorithm is well applied in the classification. 4th. Through the analysis of the project evaluation model LLE SVM, this paper points out the shortcomings and makes two related improvements, mainly starting with the front-end reduction and back-end classification. The traditional local linear embedding algorithm is improved to dimensionality reduction method with supervisory function by using the class label information of the sample data itself, which enhances the dimension reduction effect of the data. Aiming at the optimization of kernel function parameters and penalty factors in support vector machines, an improved genetic algorithm is used to optimize the parameters of support vector machines. Get the best overall performance of the support vector machine. Through the above improvements finally established an efficient innovation project evaluation model. Accurate evaluation is the most important condition for the effective management of innovation projects. The project evaluation model constructed in this paper can evaluate the project efficiently and accurately. It is of great practical significance to manage the project quickly and objectively on the platform of innovative project.
【学位授予单位】：重庆理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP18;TP311.13

【参考文献】