一种改进的类比估算方法及案例研究
文献类型：学位论文
| 作者 | 李效云 |
| 学位类别 | 硕士 |
| 答辩日期 | 2010-06-03 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 北京 |
| 导师 | 王永吉 |
| 关键词 | 软件成本估算 类比估算 相似度函数 第一类曲线积分 成本估算工具 ImpAnalogy |
| 学位专业 | 计算机软件与理论 |
| 中文摘要 | 软件成本估算是软件工程领域的重要活动之一。类比估算由于具有易于理解、不受数据规模的影响、不需要本地校准、降低异常点对估算精度的影响等优势，所以是软件成本估算常用的方法之一。项目属性集的选择、相似度函数定义和修正求解是类比估算三个比较重要的方面。 通过对比分析目前类比估算相似度的定义，得出其存在的问题是：没有充分利用模型和历史项目集信息从而在一定的数据集上可能降低估算精度。针对这种问题，本文提出一种改进的类比估算方法，该方法充分利用了模型和历史项目集的相关信息，定义模型属性信息空间上由第一类曲线积分计算的弧长为相似度。 项目属性集选择相关程度比较低的因子以解决由于因子之间相关程度高对估算精度产生影响的问题；修正求解采用基于相似度倒数或平方倒数加权法以解决相似项目数难以确定的问题。 在成本驱动因子的定义和管理、成本估算模型的支持与集成、模型校准、模型精度分析这四个方面，本文分析了现行的成本估算软件存在的问题及发展趋势。根据分析结果，并结合改进的类比估算方法开发了改进的类比估算工具ImpAnalogy。该工具基于Eclipse RCP框架，采用MVC架构实现了数据管理、因子管理、模型精度分析、模型共享等模块。 本文使用工具ImpAnalogy进行了三个案例研究。 案例一：为一个没有使用任何估算方法和估算工具的企业提供一个估算精度相对较高的方法和工具。结果表明改进后的类比方法的估算精度高于基于欧氏距离和AMH的类比估算方法、基于模型的指数加权调整方法、COCOMO类型方法和回归方法。 案例二：为一个准备申请CMMI 4级的企业提供一个估算精度较高的方法。结果表明改进后类比方法的估算精度高于采用欧氏距离和AMH的类比估算方法、COQUALMO类型方法，但是低于回归方法。 案例三：采用NASA93数据集验证方法的有效性，增强说服力。结果表明改进后的类比方法的估算精度高于采用欧氏距离的类比估算方法和回归方法；采用基于相似度倒数或平方倒数加权调整算法的估算精度高于单项目调整算法和回归方法。 |
| 英文摘要 | Software cost estimation is one of the most important activities in software engineering. Because Analogy-based estimation has lots of advantages such as easy to understand, not affecting by the size of data, not needing calibration and reducing the affecting on estimation precision by outlier, it often be used in software cost estimation. Selecting item attribute set, the defection of similarity and molding to solve are the three important aspects of analogy-based estimation. By comparing and analyzing the current methods which are used to define similarity, a conclusion that the exiting problem of the current similarity definition is that they have not make full use of model information to the definition of similarity has been made, and on a certain set of data this may affect the estimation precision. To resolve this problem, an improved method which defines the similarity which is the length of arc computed by line integrals of the first type in the space of model information has been proposed. To solve the problem that the drivers whose relevance is high may affect the estimation precision, these drivers whose relevance is low have been selected. To solve the problem that the number of similarity item set is difficult to determine, the weighted method which based on the reciprocal of similarity or the square of similarity has been used. This thesis analyzes the problems and trends of the existing software estimation tools in the definition and management of cost drivers, the supporting and integration of estimation model, model calibration and model precision analysis. Based on the conclusion and improved method, ImpAnalogy has been developed. Based on RCP framework and using MVC model, this tool realize data management, driver management, model precision analysis and model sharing modules. Using the ImpAnalogy, three case studies have been executed. Case study one: the purpose is to provide a relatively higher estimation precision method and tool for a company which does not use any estimation method and tool. The result is that the estimation precision of the improved analogy-based estimation is higher than that of these analogy-based methods which are based on Euclidean distance and AMH, weighted index based on model method, COCOMO model type method and regression. Case study two: the purpose is to provide a relatively higher estimation precision method for a company which is applying for CMMI 4. The result is that the estimation precision of the improved analogy-based estimation is higher than that of these analogy-based methods which based on Euclidean distance and AMH and COQUALMO model type method, but is lower than that of regression. Case study three: the purpose is to prove that the method is effective and convincing by using the data set of NASA93. The result is that the estimation precision of the improved analogy-based estimation is higher than that of that analogy-based method which based on Euclidean distance and regression; the estimation precision of the methods which is based on the reciprocal of similarity and the square of similarity is higher than that of single-item linear adjustment and regression. |
| 学科主题 | 计算机工程 |
| 语种 | 中文 |
| 公开日期 | 2010-06-16 |
| 源URL | [http://124.16.136.157/handle/311060/2384]  |
| 专题 | 软件研究所_互联网软件技术实验室 _学位论文 |
| 推荐引用方式 GB/T 7714 | 李效云. 一种改进的类比估算方法及案例研究[D]. 北京. 中国科学院研究生院. 2010. |
入库方式： OAI收割
来源：软件研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。