【作者】 尹雄锐；

【导师】 张翔； 夏军；

【作者基本信息】 武汉大学 ， 水文学及水资源， 2004， 硕士

【摘要】 降雨径流模拟是水文模拟的重要组成部分。到目前为止，人们对水文过程规律的认识还不十分清楚。近二十年人类活动的剧烈影响，土地利用／土地覆被变化日益加剧，使得对机理的认识更加困难，再加上数据有限这一瓶颈问题，使得传统的水文模拟方法面临更加严峻的挑战。因此，水文模拟的研究迫切需要引入新的理论和方法。在这种形势下，基于智能计算的水文模拟方法应运而生。作为该方法的探索，本文将主要工作放在基于智能计算方法的降雨径流的模拟上，大致开展了如下研究： (1) 分析我国目前水文水资源面临的新的形式以及水文模拟所面临的困难，介绍人的智能作用和人工智能的兴起，总结了智能计算在水文模拟中的应用。 (2) 针对土地利用／土地覆被变化对水文过程的影响问题，结合潮河流域的实际情况，建立了加入土地覆被因子的BP网络模型模拟流域降雨径流过程，结果表明，加入土地覆被因子的BP网络比没有加入土地覆被因子的网络模型表现出更高的精度。 (3) 针对降雨径流模拟中，由于大流量样本少而使模拟精度不高的问题，提出了基于模糊C均值和基于自组织映射两种流量分类方法的神经网络预报模型。针对分类过程中一些类与类之间的边缘样本在选择局部神经网络模型时出现的误判问题，在上面已经建立的模型的基础上，建立了加入专家经验的模糊逻辑选择模型。构成Fuzzy+ANN+专家经验的综合智能模型。以王家厂流域为例进行了方法的实现，计算结果表明，使用基于分类的网络模型比没有分类的总体模型有更高的模拟精度，而应用模糊逻辑选择的基于自组织映射分类的BP网络模型有最高的精度。 (4) 遗传编程是新的智能计算方法，在国内尚未见到在模拟降雨-径流关系上的应用，本文尝试应用该方法建立了王家厂水库洪水预报，结果令人满意。更多还原

【Abstract】 The simulation of rainfall-runoff is an important part of hydrology simulation. So far, the law of the hydrological cycle is still unclear. However, with the strong influence of human actions and land use and land cover change increasing rapidly, it has being becoming more difficult for us to understand it. Furthermore, the data sacristy makes the traditional hydrologic simulation facing the more rigorous challenge. So new theory and methods are being urgently needed for the study on hydrology process simulation. Under these conditions, the hydrologic simulation method based on the intelligent calculation emerges, as the times require. This thesis focuses on employing the intelligent calculation method to simulate the rainfall-runoff simulation and the following four parts are included:(1) At the first of this paper the new situation that the hydrology and water resource is facing and the difficulty that the hydrologic simulation are encountering in our country are analyzed, and the intelligent functions of the humans and the springs up of the artificial intelligence are introduced, then the intelligent calculation’s application in hydrologic simulation is summarized.(2) Aimed to the influence that land use and land cover change has had on the hydrologic process, and taking Chao river basin for example, an BP networks model is established to simulate the basin rainfall-runoff process with the land cover factors. Compared with the simulation that didn’t consider the factor of land use/land cover, the more accurate simulating result is obtained by the model considering the influence of land use/land cover.(3) Owing to the lack of the big runoff samples in hydrology series, good precision is seldom obtained in the simulation of big flow. To improve this problem, a kind of neural network forecasting model is brought forward. This model rank the runoffs firstly and is based on two runoff taxonomies: the fuzzy C mean and the self-organization mapping. However, when the samples on the border of two sorts choose the local neural network models, they tend to make mistakes. According to this problem, a fuzzy logic selection model based on expert experiences is established.Taking the Wangjia Chang basin for example, the calculation results shows that the simulation precision of neural network model with classification is higher than that of integrated model without classification, and the fuzzy logical selection based on self-organization mapping classification-BP network model has the highest accuracy of all.(4) The hydrologic development requires the continuous introduce of new methods. So this paper attempts to use a newer intelligent technology-genetic programming to establish the rainfall-runoff experiential formula. The data is from the Wangjia Chang basin and results are satisfying, then the new method should be recommended.更多还原