【作者】 郝国生；

【导师】 巩敦卫；

【作者基本信息】 中国矿业大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 交互式遗传算法把人的智慧和遗传算法结合起来,主要用于解决无法建立显式函数的隐式性能指标优化问题。交互式遗传算法在发挥人类智慧的同时,也需要面对人自身的局限性。人的认知局限性和易疲劳特点,使得交互式遗传算法的种群规模较小和进化代数较少,这限制了交互式遗传算法的优化性能。许多学者研究了改进交互式遗传算法性能的方法,这些方法几乎都与用户偏好信息相关。由于用户偏好信息往往综合了多种用户认知规律,因此,为了更好地获取用户偏好信息,必须深入研究交互式遗传算法中用户的认知规律。但是,已有研究成果中对用户认知规律的研究却很少。本文通过研究交互式遗传算法中用户的认知规律,进而研究交互式遗传算法收敛理论和性能改进方法。本文内容主要从以下5个方面展开:(1)研究交互式遗传算法中用户的参照认知规律,分别考虑理论参照认知和实际参照认知的算法收敛理论,提出交互式遗传算法全局收敛的强条件和弱条件;(2)研究交互式遗传算法中用户的理性认知规律,提出用户保持理性是交互式遗传算法全局收敛的充分条件,并针对赋予适应值的不同方法给出用户保持理性的最大进化代数估计;(3)研究交互式遗传算法中用户的不确定性认知规律,给出用户偏好知识提取、表示及更新方法,并结合定向变异,提出了改进算法性能的方法;(4)研究交互式遗传算法中用户的选择性注意认知规律,提出获取用户选择性注意的种群初始化方法和跟踪用户选择性注意的个体生成方法,并给合用户选择性注意知识,提出算法性能改进的方法;(5)研究交互式遗传算法系统的实现,给出交互式遗传算法的系统实现框架、模块划分,并给出基于交互式遗传算法的三维动漫人物造型系统。本文的研究成果不仅丰富了交互式遗传算法的基础理论,而且为把交互式遗传算法应用于工程实践提供了理论指导。更多还原

【Abstract】 Interactive genetic algorithm (IGA) combines human’s intelligence with genetic algorithm (GA) together to solve problems in which their performance indices are implicit or difficult to be expressed by explicit functions. When IGA makes use of human’s intelligence, it has to consider human’s limitations. For example, the population size and the evolutionary generation should not be more than 20 for human’s fatigue and limitation of cognition ability. Therefore, the performances of IGA are often restricted by small population size and a few evolutionary generations. Many researchers have studied methods to improve IGA’s performances. Almost all of the methods are based on the information of the user’s preference. In fact, a user’s preference is the synthesis of different kinds of cognitions. So it is important to study on the principles of the user’s cognition, which will be helpful not only to get the information of the user’s preference, but also to study the methods to improve IGA’s performance. But it is regret that there have been few researches on the principles of the user’s cognition.This dissertation mainly focused on the principles of the user’s cognition in IGA. Firstly, the principle of the user’s reference cognition in IGA is studied. Also, this dissertation addresses its influence on the convergence of IGA. The strong condition and weak condition of convergence of IGA with fitness noise are given. Secondly, the principles of the user’s rationality cognition are studied. Based on the principles, we find that the ability for the user to keep rational state is a sufficient condition for the convergence of IGA. In order to help the user to keep rational state, the maximum generations should be different for different methods of fitness assignment. Based on this viewpoint, the maximum generation problem was studied. Thirdly, the principle of users’uncertainty cognition is studied. In order to identify the uncertainty information, the method of quantities identification is given. Then the method to abstract users’preference knowledge from certainty information is given and the method to express and update the user’s preference knowledge is studied. Fourthly, the principle of the user’s selection attention cognition is studied. In order to get the knowledge of the user’s attention, we consider two optimization problems: (1) the maximum number of gene sense units that attract the user’s attention with small population size and (2) the minimum size of population in which the knowledge of the user’s attention to all the gene sense units can be deduced. In order to make use of the above knowledge, the special method to initialize population and the method to track the attention fluctuation are given. Finally, we address the realization of IGA and the realization of 3 dimension cartoon characteristics design which is based on IGA is given.The studies on the principles of the user’s cognition in IGA not only enrich the basic theory of IGA, but also provide necessary instruction for IGA application.更多还原