【作者】 杨晓华；

【导师】 陆桂华； 张建云；

【作者基本信息】 河海大学 ， 水文学及水资源， 2002， 博士

【摘要】 本文对单纯形算法、Rosenbrock算法和模式搜索算法等多种常规优化方法，模拟退火算法、混沌算法和最大熵优化等多种革新优化算法以及具有很大发展潜力的遗传算法(GA)进行了系统的分析，并针对不同水文模型所出现的大量高维、多峰、非线性、不连续、非凸性等复杂的参数优选问题，建立了改进的加速遗传算法、改进的混合加速遗传算法并对上述改进的加速遗传算法进行数值模拟、系统的理论研究、应用研究并与常规优化方法进行比较，提高了水文模型参数优选的稳健性和解的精度以及遗传算法的速度。主要研究内容如下： 1．对二进制编码加速遗传算法的杂交算子、变异算子进行改进，建立了改进的加速遗传算法，提高了GA的全局优化能力。 2．利用混沌变量特定的内在随机性和遍历性，建立了简单混沌优化算法。 3．在进化过程中，根据种群的实际情况，随时调整杂交、变异算子大小，把这一思想应用于加速遗传算法中，提出基于动态杂交、变异操作的自适应加速遗传算法，克服了算法的早熟收敛，在一定程度上提高了GA的全局优化能力。 4．二进制编码有时不便于反映所求问题的结构特征，对于一些连续函数的优化问题，GA的局部搜索能力也较差。相邻整数的二进制编码可能具有较大的Hamming距离，这种缺陷将降低遗传算子的搜索效率。格雷码遗传算法虽然可以克服二进制编码的Hamming悬崖这一缺点，但也存在着收敛速度慢等问题。为此，本文提出基于格雷码编码的加速遗传算法，并将格雷码加速遗传算法应用于河流横向扩散系数、结构最优设计和非线性极大极小问题等参数识别问题中，大大提高了格雷码遗传算法解的精度和搜索效率。 5．二进制编码需频繁的编码和解码，计算量大。实数编码遗传算法虽然不需频繁的编码和解码，但局部搜索能力有时也较差。本文在实编码遗传算法中加入单纯形法、模拟退火法和模式搜索法，提出了单纯形混合加速遗传算法、模拟退火混合遗传算法和模式搜索混合遗传算法，在一定程度上，减少了GA的计算量，提高了GA的搜索效率、全局优化能力和解的精度。 6．建立了上述两点杂交、两点变异格雷码加速遗传算法的模式定理，给本文所建立的格雷码加速遗传算法提供了理论基础。 7．对模拟退火算法进行了改进，建立了改进的模拟退火算法，在一定程度上提高了模拟退火算法的搜索效率。 8．和用基于最大熵原理的变尺度算法(DFP)利实编码加速遗传算法(RAGA)，解决了带约束的环境优化问题。 9．本文对10种优化方法的全局收敛性进行了数值实验和分析比较，结果表明混合加速遗传算法SAGA、JHGA的全局优化性能较好。 10．给出了12种参数优选方法在水文模型中的应用结果，建立了适合于水文模型尤其是流域水文模型的混合加速遗传算法。更多还原

【Abstract】 The usual optimization methods of simplex’ s algorithm, Rosen Brock algorithm, Hooke-Jeeveshybrid algorithm, the reformation optimization methods of simulated annealing algorithm, chaos algorithm, maximum entropy method and the potential genetic algorithm are systematically analyzed. Improved accelerating genetic algorithm and hybrid accelerating genetic algorithm are proposed for lots of multidimension, multiapex, nonlinear, discontinuity, nonprotruding about complex parameter optimization problem in Hydrological model. Comparisons with some usual optimization methods were made through the application of Hydrological model in practical. Results showed that the above improved accelerating genetic algorithm have the features of convenient, fast, constringency and steady. They could be applied extensively in the optimal problems. The main achievements are as follows:1. Crossover operator and mutation operator are modified for binary accelerating genetic algorithm. Improved accelerating genetic algorithm is presented. The total optimization ability is improved.2. An chaos algorithm is established by use of the intrinsic stochastic property and ergodicity of chaos movement.3. According to population’ s fact, crossover operator and mutation operator are adjusted for accelerating genetic algorithm in evolution in time. Adaptive accelerating genetic algorithm for dynamic crossover and mutation operator are presented. Early constringency is conquered. The total optimization ability is improved.4. The structure in problem can be difficultly reflected by use of binary code. GA local searching ability is quite worse for continuum function. Near binary code integers could be of large Hamming distance, that will debase genetic algorithm’ s searching efficiency. Although gray code genetic algorithm can get over the Hamming distance problem, the convergence speed of gray code genetic algorithm is also slow. So gray code accelerating genetic algorithm is presented for continuum function. For the parameter identification of determining the transverse difussion coefficient of river, the optimum design of structures and minimax problems, gray code accelerating genetic algorithm gets good solution. Searching efficiency and solution’ s precision are greatly improved for gray code genetic algorithm.5. Binary coding need frequent coding and decoding, and the amount ofcalculation is big. Al though real coding genetic algorithm needn’ t frequent coding and decoding, local searching ability of real coding genetic algorithm is sometime difference. Simplex’ s algorithm, simulated annealing algorithm or Hooke-Jeeves algorithm is added in real coding genetic algorithm, Simplex hybrid accelerating genetic algorithm, simulated annealing hybrid accelerating genetic algorithm and Hooke-Jeeves hybrid accelerating genetic algorithm are presented, at a certain extent the calculation steps of algorithm is reduced, searching efficiency, global optimization ability and solution’ s precision are improved.6. Two-point crossover and two-point mutation’ s scheme theorem is established for above gray code accelerating genetic algorithm. It provides a theory base for above gray code accelerating genetic algorithm.7. Simulated annealing algorithm is improved, an improved simulated annealing algorithm is established and searching efficiency is improved for simulated annealing algorithm.8. DFP and RAGA on maximum entropy theory are proposed, an environment optimization problem is good solved.9 . The global convergence is numerically tested and analized for ten optimization methods. Result indicating Hybrid accelerating genetic algorithms---SAGA, JI1GA are good for the global convergence.10. Calculation result of twelve parameter optimization methods is given for Hydrological model. Hybrid accelerating genetic algorithm are established for Hydrological model especially for basin Hydrological model.更多还原