【作者】 王淑静；

【导师】 倪志伟；

【作者基本信息】 安徽大学 ， 计算机应用技术， 2006， 硕士

【摘要】 范例推理是人工智能领域中较新崛起的一种重要的基于知识的问题求解和学习的方法，它是根据过去的成功或失败的事例来推导出新问题的解，它是一种知识库同推理机融为一体的新的推理技术。近年来，关于范例推理的研究及其系统的开发受到人们的普遍关注。范例推理是由目标范例的提示而得到记忆中最相似的源范例，并由源案例来指导目标范例求解的一种策略。范例推理不仅是关于人类认知的心理学理论，而且将成为智能计算机系统技术新的基石之一。范例推理技术在许多领域都可以使用，尤其在不好总结出专家知识的领域效果很好。 然而在范例推理中也存在一些问题，主要体现在范例工程过程的自动化，即范例知识的自动生成，如范例结构及其内容、相似性评估知识、现有范例库的自动更新、修正知识库的获取、索引模式等。而这些知识的获取也存在一定程度的瓶颈问题，这些知识是由领域专家与知识工程师通过不断积累慢慢取得的。 对于给定的目标范例，如何从范例库中检索和选择出最为相似的范例决定了范例推理系统的学习与推理性能。范例间的相似性度量是关键。其中范例的特征项权重对检索的质量与速度都起到了重要作用。对范例库特征项权重的提取也就是来发现范例的不同特征具有不同的重要性。 遗传算法作为一种借鉴生物界自然选择和自然遗传机制的高度并行、自组织、自适应的搜索算法，由于其隐含并行性和收敛的全局性两大显著特点，使其尤其适用于处理传统搜索方法难于解决的复杂问题。禁忌搜索算法是一种局部搜索修正算法，模拟人类具有记忆功能的寻优特征。即人们常常在对已经搜索过的地方暂时不会再次去搜索，先对尚未搜索过的地方进行搜索，如果没有更为合适的解，则返回原来已经搜索过的地方。禁忌搜索算法通过局部邻域搜索机制和相应的禁忌准则，来避免重复迂回搜索，并通过期望标准来释放一些被禁忌的优良个体，进而保证多样化的有效搜索，以此来最终实现全局优化。 遗传禁忌算法结合了遗传算法的随机搜索能力、并行性和禁忌搜索算法的记忆功能，有效地解决了遗传算法的爬山能力差、早熟的问题。该算法把禁忌搜索的“多样化”引入到遗传算法的交叉算子和变异算子中，生成禁忌交叉算子和禁更多还原

【Abstract】 Case-Based Reasoning is a kind of important methods for question answering and learning based on knowledge which grows up recently in artificial intelligence field. It is to reason answer for new problem based on the past successful or failed cases. It is a new reasoning technology combing knowledge base with reasoning. In recent years, Case-Based Reasoning research and its system development has received people’s universal attention. Case-Based Reasoning is a strategy which gets the most similar source case in the memory by the hints of the target case and gets the answers by the instruction of the source case. Case-Based Reasoning is not only about human cognition psychology theory, but also will become the foundation stone of the intelligent computer system technology. Case-Based Reasoning can be used in many different fields, especially in the fields which are difficult to conclude experts’ knowledge. But there are some problems in Case-Based Reasoning as well, mainly embodies in Case-Based project automation, that is automatic production of case knowledge, such as case structure and its contents, similar evaluation knowledge, automatic updating of case-base, achievement of amendment knowledge database, index pattern etc. But there are also some choke point problems in attaining this knowledge which is accumulated slowly by experts in specified fields and knowledge engineers.As to target case given, how to check and choose the most similar case from case-base decides learning and reasoning functions of case-based reasoning system. The similarity between cases is the key. The feature weight of case plays an important role in checking quality and speed of index. Extract for feature weight of case-base is also to discover that different characters of case have different importance.Generic Algorithm is a parallel highly, organically itself, adapt itself searching algorithm referring to mechanism of nature selection and nature inheritance in biology field. Because of its two prominent features which are connotative parallel and convergent globality, Generic Algorithm is typically suitable to process complex更多还原