【作者】 吴晓琴；

【导师】 李龙澍；

【作者基本信息】 安徽大学 ， 计算机应用， 2007， 硕士

【摘要】 近几年，智能优化算法倍受人们关注，如人工神经网络、遗传算法，为解决复杂问题提供了新的方法，并在诸多领域取得了成功。组卷问题是一个在一定约束条件下的多目标参数优化问题，针对传统的组卷算法具有组卷速度慢、成功率较低、试卷质量不高等缺点。本文着重研究遗传算法在组合优化中的应用，为了避免简单遗传算法收敛速度慢以及局部收敛的问题，引入了一种改进遗传算法。该算法利用不断淘汰相似个体，并不断补充新个体的方法，达到了扩大搜索空间，稳定群体的个体多样性目的。通过详细分析试卷的各项约束条件如知识点、难度系数、区分度，建立了一个智能组卷数学模型，利用改进的遗传算法实现了智能组卷。改进后的遗传算法采用分段实数编码，把同一题型的试题放在同一段，组成试卷的各道试题的题号直接映射为基因，用实数编码避免解码过程，提高了运算效率，而且交叉和变异操作都在各段内部进行，因此可以保证组卷过程中各题型题量的正确匹配，还要保证同一题型知识点不重复。对适应度函数设计，调整了强度约束的权值。根据需求分析，对系统进行了四个功能模块的设计。这四个功能模块分别是题库管理模块、试卷生成模块、成绩分析模块和系统维护模块。组卷模块是系统的核心，在组卷模块中组卷方式有三种：人工组卷、自动组卷、向导组卷。可以直接生成的Word形式试卷以及试卷答案，在Word中可以对试卷以及试卷答案进行编辑修改并打印出来。成绩分析模块除具有学生成绩进行统计分析功能外，还能通过成绩分析结果对题库中试题属性进行更新。实验结果表明，新方法的组卷成功率和收敛速度都得到明显提高，并且较好地克服了未成熟收敛现象，只要试题库中的试题数量适中，试题类型完备，分布合理，由该算法产生的试卷就能满足用户的各项需求指标。更多还原

【Abstract】 The optimization algorithm becomes popular in the recent years, such as theartificial neural network and genetic algorithms etc. It provides as a new method toresolve complicated problems and gains success in lots of fields, auto-generatingexamination paper is a constrained multi-object optimization problem. Traditionalalgorithms of composing test paper have the disadvantages of slow convergence,lowsuccess rate and quality. This paper mainly studies the use of GA in the Combinationoptimization. In order to avoid slow convergence and local convergence of simplegenetic algorithm (SGA)for intelligent test paper generation, a kind of improvedgenetic algorithm (IGA) has been proposed in this paper. This algorithm usesunceasing elimination of similar individual method to quickly enlarge the searchspace and to stabilize the individual diversity of the group, n this article a new methodof composing test paper based on the improved genetic algorithm is given.. After acareful Analysis of each binding condition in the test paper, we have set up amathematical model for automatic test paper-making based on knowledge point,difficulty factor, distinguishing degree, etc. and have realized automatic testpaper-making with improved genetic algorithm. Improved genetic algorithm adoptssegment real number code, putting the question of the same type on the same section,and then the question number maps gene directly. Real number code avoid decodingprocedure, it may enhance operation efficiently. In addition, crossover and mutationoperation conduct in the interior of each section, it may guarantee the quantity of eachtype correct matching and different knowledge point of the question of the same typein the process of test paper-making. The fitness function is designed, the weights ofstronger Constraint conditions are enlarged.According to the functional demand of intelligent Test Paper system, we havedesigned four functional modules: examination database, test paper-generation, gradeanalysis and system setup. Test paper generation module is the core of the system, itincludes three ways of test paper-making: manual, guide and automatic. Test paper aswell as its answers can be sent directly into Microsoft Office Word,in which testpaper as well as its answers are edited, revised and printed out. Grade analysis modulecan analyze the student grades statistically and update each binding condition thequestion in examination database by analyzing grade. The new method is moreefficient and easier to get over premature convergence than the traditional algorithms. It is proved by a number of experiments provided by this article, the test paper formed by the algorithm meets all the users’ requirements if the quantity of test questionsis moderate and reasonable.更多还原