【作者】 陈钢；

【导师】 曹炬；

【作者基本信息】 华中科技大学 ， 运筹学与控制论， 2014， 硕士

【摘要】 最优化问题广泛存在于航空航天、工业工程、经济管理、交通运输等领域，这些优化问题促使最优化理论和方法不断进步。随着计算机技术的迅速发展，优化问题的复杂程度也越来越高，单纯形法、最速下降法、共轭梯度法等传统优化算法难以求解这些具有大规模、非线性、不连续、多极值等特点的优化问题。于是，智能优化算法应运而生，常见的有遗传算法、粒子群优化算法、蚁群优化算法等。然而，大多数智能优化算法存在早熟等问题，因此，智能优化算法的改进工作具有一定的实际意义。针对粒子群优化算法的早熟问题，本文提出了一种将粒子群优化算法和最速下降法相混合的智能优化算法——多策略粒子群优化算法。在粒子群寻优过程中，最优粒子采用差商最速下降策略和矫正下降策略进行局部搜索，非最优粒子采用聚集策略进行全局搜索。当整个种群陷入局部极值点时，最优粒子和非最优粒子分别采用随机移动策略和扩散策略进行移动，以跳出局部极值点。4个典型benchmark测试函数的数值实验结果表明，本文提出的算法比其他两种类粒子群优化算法具有更强、更稳定的全局搜索能力。针对云搜索优化算法的参数控制难、算子较复杂等问题，本文采用单因素实验法对算法中涉及的参数与算子进行了分析和优化，并提出了一种将云搜索优化算法和模式搜索法相混合的智能优化算法——一种改进的云搜索优化算法。在云团优化过程中，云会不断地生成、飘动、降雨、收缩和扩张，最终，整个云团聚集在低气压区。采用蒙特卡罗法生成云的方式能减少算法的计算复杂度，采用模式搜索法抖动云层内中心水滴的方式能提高算法的收敛速度和跳出局部极值点的能力。4个典型benchmark测试函数的数值实验结果表明，本文提出的算法比标准粒子群优化算法和云搜索优化算法具有更强、更稳定的全局搜索能力。更多还原

【Abstract】 There are a large number of optimization problems which widely exsit in theaerospace, industrial engineering, economic management, transportation and other fields.These optimization problems prompt the continuous improvement of optimization theoryand methods. However, the complexity of optimization problems become more and moresophisticated with the rapid development of computer technology. Furthermore, it isdifficult for traditional algorithms, such as the simplex method, steepest descent method,conjugate gradient method and so on, to solve these problems with the characteristics oflarge-scale, nonlinear, discontinuous, multi extreme values, etc. Thus, some intelligentoptimization algorithms come into being, such as Genetic algorithm, Particle SwarmOptimization algorithm, Ant Colony Optimization algorithm, etc. Nevertheless, themajority of intelligent optimization algorithms suffer from premature convergence andother issues. Therefore, this paper proposed two inproved hybrid intelligent optimizationalgorithms, which may have some practical significance.For the premature problem of Particle Swarm Optimization, a new hybrid intelligentoptimization algorithm called Mult-strategy Particle Swarm Optimization algorithm isproposed, which combines Particle Swarm Optimization algorithm with steepest descentmethod. In the process of particle swarm optimization, the optimal particle performs localsearch by using the steepest descent strategy with difference quotient and correctivedecline strategy, non optimal particles perform global search by using aggregation strategy.While the entire population is trapped in local minima, the optimal particle and nonoptimal particles use random mobile strategy and diffusion strategy to escape from thelocal extremum point respectively. In the end, the performance of Multi-strategy ParticleSwarm Optimization algorithm is tested with four typical benchmark functions and iscompared with other two improved Particle Swarm Optimization algorithms’. Numericalresults indicate that the proposed algorithm has better performance including stability, theability of global search, etc.In order to solve problems on parameter settings and the computational complexity of operators in Clouds Search Optimization algorithm, this paper analyses and optimizesparameters and operators involved in the algorithm by using the method of single-factorexperiments. In addition, a hybrid intelligent optimization algorithm with Clouds SearchOptimization algorithm and Pattern Search mehod is proposed, which is called anImproved Clouds Search Optimization algorithm. In the process of clouds searchoptimization, clouds will be constantly formed, flowed, rainfall, shrinked and expanded.Ultimately, the whole cloud cluster gathers in the area of the lowest pressure. By the way,The formation of clouds by using Monte Carlo method can reduce the complexity ofalgorithm, and the jitter of the central droplet for each cloud by using Pattern Searchmethod can improve the convergence speed and the ability to jump out of local minima. Inthe end, the performance of an Improved Clouds Search Optimization algorithm is testedwith four typical benchmark functions and it is compared with the standard ParticleSwarm Optimization algorithm and basic Clouds Search Optimization algorithmrespectively. Numerical results indicate that the proposed algorithm has betterperformance including stability, the ability of global search, etc.更多还原