【作者】 冯楠；

【导师】 李敏强；

【作者基本信息】 天津大学 ， 信息管理与信息系统， 2007， 博士

【摘要】 在软件项目的投资、开发以及实施过程中,风险是时时存在的。而且随着软件系统复杂度和需求的不断增长,风险管理的作用变得越来越突出。因此,如何辨识风险、分析风险以及控制风险,降低风险发生的概率或造成的损失,是一个亟待解决的问题。本文首先回顾和分析了国内外研究现状,指出了当前要解决的主要问题;然后,从风险这一概念入手,对风险管理以及软件风险管理进行了简要的分析,详细说明了风险管理的理论基础以及软件项目风险管理相对于其它项目风险管理的特殊性。介绍了4套典型的风险管理体系,分析了各个体系的特点,并对这些风险管理体系进行了比较;第三,在构建软件项目风险评价体系的基础上,提出一种基于人工神经网络的软件项目投资风险评价模型。并通过实际IT项目评价数据的验证,说明该模型能够利用已知的样本数据进行有效的软件项目投资风险评价。在构建风险评价模型过程中,采用因子分析的方法对样本数据进行降维处理,同时利用了一种基于黄金分割原理的优化算法确定隐含层节点数;第四,提出一种动态软件项目开发风险管理模型。该模型通过风险跟踪模块实时监控风险状态,并根据风险状态的变化不断调整风险列表,使整个风险管理活动在一个具有持续反馈功能的风险管理流程当中进行,从而实现了动态的风险管理过程;第五,针对风险分析模块提出了一种基于贝叶斯网络的风险分析方法及其实现过程,并利用实际案例说明贝叶斯网络在软件项目风险分析过程中的应用过程。该方法能够根据不断更新的风险数据改变各个风险节点的状态,是支持软件开发过程中实现动态风险管理的基础。详细分析了贝叶斯网络的建模过程,即贝叶斯网络的结构学习过程和贝叶斯网络的参数学习过程。最后,总结全文,并指出该领域可能的发展方向。更多还原

【Abstract】 Risk is ubiquitous in the process of investment or development and implementation of software project. Risk management becomes more important with the increase of software complexity and requirement. It is an urgent problem to study the strategies to analyze and control risk.The main contents of this paper are as follows: Firstly, the history and state-of-the-art of the risk management are reviewed both domestically and abroad. Then, the concepts about risk management and software risk management are defined based on the concept of risk. The theoretical foundation of risk management and the particularity of software project risk management are discussed in detail. After four styles of risk management systems are analyzed, their advantages and disadvantages are compared. Thirdly, based on establishing a risk evaluation system of software projects, a risk evaluation model of software project investment based on Artificial Neural Network (ANN) is proposed. The model is capable to evaluate risks effectively in the process of software project investment by verifying the data of software projects. And in the process of modeling, the factor analysis is utilized to decrease dimension of sample data and an optimization algorithm based on the principle of golden section is designed to find the optimal number of hidden layer nodes. Fourthly, a dynamic risk management model of software project development is proposed. The model can monitor risk states continually by using risk track module and then adjust risk list according to the change of risk states so that the activities of risk management run in the flow with feedback. Fifthly, A risk analysis implementation flow based on Bayesian Networks(BNs) for risk analysis module is presented. In addition, the BNs is applied to risk analysis module of a real case. The method can update risk data and change the states of risk nodes online. It is fundamental to implement dynamic risk management in the process of software development. In the process of modeling BNs, the structure learning and parameter learning of BNs are analyzed detailedly. At last, after summarizing main content discussed in this paper, potential research directions are pointed out.更多还原