【作者】 马松涛；

【导师】 梁静；

【作者基本信息】 郑州大学 ， 检测技术与自动化装置， 2013， 硕士

【摘要】 在现实社会的生产实践中,大量实际问题的解决最终可转化为对问题的优化。而这些待优化的问题往往非常复杂,集中表现在多模值、维数高、多目标和动态性等方面。传统的进化算法受限于自身的机理和结构单一,收敛精度低、对初值敏感、易陷入局部最优,对高维复杂问题的处理比较吃力。由Eberhart和Kennedy教授在1995年共同提出了粒子群优化算法。它源于鱼群和鸟群的觅食行为,进而提出来的一种具有代表性的群体性智能进化算法。由于粒子群优化算法具有搜索速度快、全局搜索能力强、鲁棒性强等突出优点,所以短短几年时间,粒子群优化算法已经成为计算智能领域的新的研究热点,并被应用到许多领域之中。随着大量科研人员对粒子群优化算法的深入研究,粒子群优化算法已被成功地用来解决静态单模优化问题。但是实际生产中的许多优化问题需转化为多模优化问题和动态优化问题。面对这些复杂的优化问题,不但要求优化算法能够迅速的、准确的找到全局最优极值点,而且要找出所有的局部最优极值点并能够及时的跟踪变化的全局最优极值点。这对于粒子群优化算法则是一种新的挑战。现从以下三方面对本文所做的工作进行阐述：(1)粒子群优化算法的发展经历了惯性权重线性递减、全信息、单维搜索、拓扑结构、多种群等策略改进的阶段。本文对粒子群优化算法发展过程和各种改进版本在第2章作了详细的描述和分析。(2)多模优化问题在现实生活中非常常见。例如：路径优化、数据分析、蛋白质结构预测等。这类问题不仅要寻找一个全局最优极值点,有些场合需要同时找出其余的局部最优极值点。对于多模优化问题,经典的优化算法往往易陷入局部最优点而难以找到全局最优解,更难于找到所有的局部最优极值点。但是,基于物种形成原理的小生境技术的引入使多模优化问题的求解决逐步实现。本文将在第3、4章详细介绍小生境技术的精华之处,并测试验证基于局部搜索的小生境粒子群优化算法的优势。(3)现实世界的许多问题中存在很多动态元素。某些变量的状态常常随着时间的变化而变化。例如：股票市场、路径规划、物流配送、投资分配等。所谓动态优化问题,优化一类问题时不仅为了获得问题的全局最优解,还要能及时的检测到环境的变化,精确的跟踪最优解随时间变化的轨迹。鉴于上述改进的小生境粒子群优化算法在多模函数成功案例,我们在第5章对其做了更深入的研究和改进,并在动态优化测试函数上进行实验分析,结果证实改进算法的有效性。更多还原

【Abstract】 Nowadays, in the real society, a large number of practical problems can be treated as optimization problem. The problem which is needed to be optimized is always complex. The complexity mainly consists of multi-modal, high dimension, multi-objective or dynamic, etc. The traditional evolutionary algorithm is limited by its own mechanism and single structure which results in low convergence precision, sensitivity to initial value, easy falling into local optimum, bad capacity of handling problems of high-dimensional complex problems.Particle swarm optimization algorithm was proposed by Eberhart and Kennedy in1995. It is a typical swarm intelligence optimization algorithm which originated in the fish and bird flock foraging behavior. Because of the outstanding advantages of particle swarm optimization algorithm such as fast search speed, good global search ability and strong robustness, particle swarm optimization algorithm has become the new research spot in the field of computational intelligence and has been applied to many fields.Along with more deep researches about particle swarm optimization algorithm, the algorithm has been successfully applied to solve the static single modal optimization problem. Many optimization problems in the actual production need to be transferred to multi-modal optimization and dynamic optimization problem. These complex optimization problems require optimization algorithm to find the global optimal extreme value point quickly and accurately, to find out all the local optimal extreme value points and track the changes of the global optimal extreme value point in time. This is a new challenge for particle swarm optimization algorithm.This thesis expounds the work from the following three aspects:(1) The development of particle swarm optimization algorithm consists of the linear decreasing inertia weight, full information, one-dimensional search, topology structure, and multi-swarm strategy, etc. In this thesis, the development process and improved versions of particle swarm optimization algorithm are described and analyzed in chapter2.(2) Multi-modal optimization problems are very common in real life such as the path planning, data analysis and prediction of protein structure, etc. This kind of problem not only needs a global optimal extreme value point, but also needs the rest of the local optimum extreme points in some cases. For multi-modal optimization problem, a classical optimization algorithm often falls into local optimal point, so it is difficult to find the global optimal solution, not to mention all local optimal extreme value points. Then, the niche technology which is based on the principle of speciation is introduced to solve the multi-modal optimization problem gradually. The essence of the niche technology is introduced in chapter3and4in detail. And the advantages of niching particle swarm optimization algorithm based on local search was tested and verified.(3) In fact, many problems consist of a lot of dynamic elements. Some variables often vary over time such as the stock market, path planning, logistics allocation, investment allocation, etc. Dynamic optimization problem is the class of problems which are not only intended to obtain the global optimal solution of a problem, but also has the ability to detect changes in the environment timely and track the trajectory of optimal solution which changes over time precisely. Considering that the improved niche particle swarm optimization algorithm discussed above has succeeded in multi-modal function, we did some further research and improvement in chapter5. Some experiments are conducted on dynamic optimization test functions in order to analyze the performance of the algorithm. The results prove the efficiency of the improved algorithm.更多还原