【作者】 周激流；

【导师】 丁晶；

【作者基本信息】 四川大学 ， 水文学及水资源， 2000， 博士

【摘要】 遗传算法是人工智能的关键技术之一，世界各国都将其作为一个重要的研究课题。本文的主要研究内容有： 1．系统综述了国内外遗传算法的研究进展，分析了部分模型的思想和技术处理特点，概述了遗传算法在水问题中的应用情况，提出了需要注意的动向。 2．根据参考文献分析的典型遗传算法(CGA)不成熟收敛的起因，本文提出了一种可克服CGA不成熟收敛缺陷的，进而寻得全局最优解的新遗传算法(NGA)，从理论和实验两方面证明了其改进后的遗传算法能有效地克服不成熟收敛、进而搜索到全局最优解。 3．根据自适应遗传算法(AGA)的原理，本文提出了一种新的MAGA(MODIFIED ADAPTIVE GENETIC ALGORITHM)——增强的自适应遗传算法，该方法不仅能够加快普通遗传算法的收敛速度，而且能够有效地保证种群的多样性，通过求解具有多个极值点的函数优化问题，计算机仿真实验结果表明该方法是非常有效地。 4．开展了NGA和MAGA在水问题中的一系列应用的研究，它们是：用SGA和NGA优化马斯京根模型参数，用SGA和MAGA优化暴雨强度公式中的参数，实例计算表明了它们在水问题的优化问题中是有一定工程实用价值的。 5．近年来，人工神经网络(ANN)和遗传算法(GA)相结合的研究已引起了人们极大的关注。本文首先系统综述了该学科领域的发展现状，然后提出将增强性自适应遗传算法(MAGA)和BP算法相结合，利用二进制编码来同时优化多层神经网络的网络结构和权值，通过对洪水灾害评估建模和岷江紫坪埔洪水预报模型的实验，证明了这种方法能有效地避免BP算法陷入局部极小和遗传算法过早收敛，结果是满意地。更多还原

【Abstract】 Genetic algorithm (GA) is one of the key technologies for artificial intelligence and is deemed as an important subject of investigations in various countries. Main works in the paper are illustrated as follows.1. This paper presents a systemic review about the state-of -art progresses of genetic algorithm(GA) both at home and abroad, gives an analysis of the concepts adopted and the characteristics of technological processing for some models, summarizes the application of GA on water, and points out the tendencies that deserve attention.2. Based on the study of the reason of premature convergence in canonical genetic algorithms, a new genetic algorithm is proposed in this paper. The experiment results and theory analysis show that such an improved genetic algorithm can find global optimal beyond premature convergence efficiently.3. Proposed in this paper is a novel genetic algorithm (MAGA) , which not only can keep the population diversity but also has quicker convergence speed. It is applied to optimizing functions with multi-model. Computer simulation results prove its validity.4. A series of applications of NGA and MAGA are made, which include optimizing the parameters of Muskingum routing model with SGA and NGA, optimizing the parameters of the Formula of Storm Intensity with SGA and MAGA. The results indicate that these algorithms are practical and efficient on water.II5. Research involving some sort of combination of genetic algorithms (GAs) and Artificial Neural Networks (ANNs) has attracted a lot of attention recently. Firstly, this article presents a brief review of the state of the art and research prospects in this area. Secondly, the MAGA algorithm coding in binary is engaged to optimize the weights and topology of multi-layer neural network. Furthermore, the experiment result, which establishing the model of evaluation of flood disaster effect, shows that by combining the binary-coded GA with BP algorithm, the entrapment in local optical optimum of BP and the premature of GA can be prevented efficiently and satisfactorily results are obtained.更多还原