【作者】 刘永芳；

【导师】 金菊良；

【作者基本信息】 合肥工业大学 ， 市政工程， 2004， 硕士

【摘要】 城市系统是个典型的巨系统。因此，分析、研究、规划和管理城市系统应从巨系统的观点出发。城市系统评价是认识和研究城市系统的一种科学方法，为城市的规划、建设、管理等提供科学的依据。 文中对常用的评价方法进行了分析研究，指出了常用评价方法的优缺点，指出不同的评价方法有不同的适用环境，应用时应当结合实际的情况。详细阐述和分析了投影寻踪评价方法，并对其用于城市系统评价进行了尝试，结果表明投影寻踪评价方法应用于城市系统评价是可行且有效的。采用遗传算法对于投影寻踪方法在评价过程中涉及到的模型优化问题进行优化，遗传算法是模拟生物“优胜劣汰”进化过程而形成的一种高度并行、随机和自适应的通用性全局搜索算法，能够处理非线性较强的优化问题。对于标准遗传算法存在的问题，文中提出了一种改进的遗传算法——扰动式遗传算法，并对其运行效果进行了分析，改进后的算法在提高精度的同时能够达到全局收敛，并能有效地处理多极值问题。人工神经网络具有记忆功能，根据专家知识和源数据样本对网络进行训练，建立网络模型，就可以对同类评价问题进行评价，这样可以节省大量的财力物力，并且操作简单易行。论文在最后还提出了基于遗传算法的逼近理想点评价模型，并对其应用进行了尝试，评价结果合理且直观。更多还原

【Abstract】 City system is a typical immense system, and analysis, research, planning and management of city system should be performed in view of huge system. Evaluation of city system, which provides a scientific basis of planning, construction and management of city system, is a scientific method to understand and research city system.Regular evaluation methods are analyzed and studied in this paper. After their advantages, disadvantages and different application conditions are presented, it is concluded that these methods should be applied according to actual condition. First of all, projection pursuit algorithm is elaborated and analyzed, and is applied to evaluation of city system. Research results show that application of projection pursuit to evaluation of city system is feasible and effective. Genetic algorithm is a highly collateral, random, self-adaptive, general and globe search algorithm, which simulates biologic evolution process. In this paper, genetic algorithm is applied to optimizing the model optimum in what is evaluated by projection pursuit algorithm. Due to some problems of standard genetic algorithm, an improved genetic algorithm called disturbance genetic algorithm (DGA) is presented. According to the effect of DGA, this improved algorithm can deal with multiple hump function efficiently and achieve the global convergence with higher precision. Then artificial neural network, which is of memory function, can form the network model by training network according to expert knowledge and source data samples. What is mentioned above make it economical and easy to evaluate the same kind of problems by artificial neural network. Finally, the technique for order preference by similarity to solution based on genetic algorithm is also presented and applied, and its evaluation result is reasonable and practical.更多还原