【作者】 杨正祥；

【导师】 姜铁兵； 康玲；

【作者基本信息】 华中科技大学 ， 系统分析与集成， 2011， 博士

【摘要】 水是生命的源泉,是生态环境系统中最活跃、影响最广泛的因素,又是工农业生产过程中不可取代的重要资源。今天,水资源系统已演变为一个多目标、多属性、多层次、多功能和多阶段的复杂巨系统。随着所研究系统广度与深度的扩大,传统方法对于现代水资源系统的高维、非凸、非线性等复杂问题的处理已日显掣肘。近年来,随着现代应用数学和计算机技术的迅猛发展,人们提出了人工智能计算理论与分析方法以解决复杂系统问题,这些方法的引入极大地促进了系统分析技术和水科学信息分析计算的研究和发展,使这个领域的深入研究具有相当广阔的空间。本文在汲取多种新的理论和分析技术的基础上,采用了一些新思路、新方法,在水文水资源研究方面取得了一些进展。其主要研究成果和创新总结如下:(1)分析了复Morlet小波变换的时域和频域分析特性,研究了小波变换过程中伸缩因子和平移因子对时频域分辨率的影响,使用了复Morlet小波进行周期性识别的原理和公式。(2)从Parseval定理和卷积定理的角度上分别推导了的快速且实用的非正交小波变换系数的计算方法。该方法把非正交小波变换系数的计算转换为频域的计算,充分利用了快速傅里叶变换运算速度快、编程效率高的优点,为复Morlet小波的周期识别提供了算法基础。(3)结合小波理论和神经网络各自的优势,建立基于小波理论的神经网络预测模型,利用小波神经网络的非线性模拟能力进行水库年径流量的预测研究,取得了较好的模拟和预测结果。并把预测结果和影响径流的时空因素进行多方位的比较分析。(4)针对所建立的水库优化调度模型,提出了结合模拟退火和逐步优化算法的求解方法,证明了此方法处理复杂约束条件优化问题的有效性,改善了智能算法中常见的“局部最优”难题,对水库优化调度决策理论的探索有一定成效。(5)因为传统马斯京根河道洪水演进模型参数率定中存在线性化、计算繁琐、精度差等问题,分别引入遗传算法、模拟退火算法和粒子群算法用于马斯京根模型参数率定。对多种算法在马斯京根问题中的应用进行了综合性比较,对实践有较好的指导作用。更多还原

【Abstract】 Water is the source of life and the most active factor and most extensive affect in the eco-environmental system. Meanwhile, it is an important resource and can not be replaced in industrial and agricultural production process. Today, the water system has evolved into a multi-objective, multi-attribute, multi-level, multi-functional and multi-stage complex giant system. With the breadth and depth of the system expansion, it is difficult for the traditional methods of water resource systems to deal with modern high-dimensional, non-convex and nonlinear complex issues. Recent years, with rapid development of the modern applied mathematics and computer technology, artificial intelligence computation theory and analysis methods have been proposed to solve complex system problems. The introduction of these methods has greatly facilitated the development of systems analysis techniques and water scientific information analysis and calculation. So there is a very wide space for further research in this area.Based on learning of a variety of new theories and analysis techniques, some new ideas and new methods are used, and some progress has been made in analysis of hydrology and water resources. In this paper, the research works and innovations are as follows:(1) The characteristics analyses of time and frequency domain of the Morlet wavelet transform are studied; The effect of the change of wavelet dilation and translation factor on the time and frequency domain resolution is analyzed; The principle and formula of the Morlet wavelet period recognition are derived.(2)A quick, simple and practical algorithm of coefficients of non-orthogonal wavelet transform is derived form the Parseval theorem and convolution theorem.The method turns non-orthogonal wavelet transform coefficient calculation to the frequency domain calculation, takes advantage of the FFT algorithm, has the advantages of high operation speed and high programming efficiency and provides a algorithm basis for identification the period by the Morlet wavelet.(3)The method integrating simulated annealing algorithm and progressive optimization algorithm is proposed for the optimal operation model for hydropower system. The effectiveness is proved to be suitable for solving optimization problem with complex constraints by this algorithm; The common "local optimization" problem in intelligent algorithm is solved; The theory of optimization scheduling decisions of hydropower systems is developed.(4)The prediction model of neural network is established based on wavelet theory which combines advantages of wavelet theory and neural network. The runoff of river is predicted with the nonlinear simulation capabilities of wavelet neural networks, achieving good results of the simulation and prediction. The multi-faceted comparison and analysis of runoff prediction results and impact of spatial and temporal factors are carried out.(5)In order to solve the problem of linearization, complexity and poor accuracy for parameter estimate of Muskingum Routing Model, this paper introduces three modern intelligent algorithms - Genetic Algorithm (GA), Simulated Annealing Algorithm (SA) and Particle Swarm Optimization Algorithm (PSO) for the parameter calibration of Muskingum model. Through specific simulation, the results of five methods are produced. Then according to the calculation, comparison and analysis of five methods comprehensively, it is found that the results of three modern intelligent algorithms are fit significantly and better than traditional methods.更多还原