动态搜索空间策略下的粒子群算法改进及其拓展研究

发布时间：2018-01-04 19:39

本文关键词：动态搜索空间策略下的粒子群算法改进及其拓展研究　出处：《江西理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着以粒子群为例的群智能算法在各领域内愈发广泛的使用,其算法后期早熟以及最终解精度不高等现象成了务须重视并尝试解决的问题。本文以粒子群算法为切入点,通过观察粒子在搜索过程中具体空间特性,逐步改进并扩展优化策略,最终构建出具有一定广泛适用性的优化策略。具体主要包括以下三方面:(1)为进一步研究和优化粒子群算法,在采用非线性学习因子的同时,提出了一种新的牵引策略来共同优化粒子群算法(Particle Swarm Optimization Algorithm based on Homing HMPSO)。该策略通过使粒子发生偏移于最优解的位移,增加粒子活性,从而提升算法后期的寻优能力。依实验需求将各基准函数进行调整变换并通过仿真实验进行寻优测试。结果表明,在算法后期的寻优能力有明显提升,且具有较好的鲁棒性。最后,估算出算法寻优结果精度高于指定阀值精度的概率区间,证明该策略具有良好可信度。(2)为进一步缓解粒子群优化算法在其后期收敛速度慢、早熟等问题,提出了一种挂载式的、依赖自适应阀值和已知全局最优解的压缩搜索空间策略。并在此基础上对粒子重新分配初始位置、调整速度权值来提升算法的后期探索能力。实验表明,在使用相同的权重和学习因子策略时,比之原粒子群优化算法具有较好的表现,在对量子粒子群算法进行嵌入时依然具有一定效果。该策略可以有效避免早熟问题,提升算法在后期的寻优效果,具有较好的鲁棒性。(3)群体智能算法的主要任务便是在有限的时间内尽可能的获得精度更高的解。但由于早熟等常见问题,使得一个精度更高的解需要通过提供额外的迭代次数来取得。为能彻底解决早熟问题的同时保持原算法主体不变且可与现有优化理论协同优化,在前期仿真实验和理论证明的基础上提出了一种逐层演化的改进策略。利用在原算法中构建基于搜索空间压缩理论的自适应系统,通过逐层的压缩、选择、再初始化的操作,以包括压缩后搜索空间在内的社会信息作为遗传知识,指导寻优过程,从而实现最终解精度的提升、避免早熟问题的出现。对基准函数进行仿真实验可以看出该策略在提升算法精度,增强后期个体活性方面具有良好的表现。上述三个策略,依次证实了:提升种群多样性有助于提升粒子群算法最终表现;在同等情况下,压缩搜索空间可以使得算法最终表现得到提升;逐层的演化策略作为在种群多样性与搜索空间二者的基础上构建的优化策略较之于前者具有更好的普适性。
[Abstract]:With particle swarm optimization as an example, swarm intelligence algorithm is more and more widely used in various fields. In this paper, particle swarm optimization (PSO) is taken as the starting point, and the specific spatial characteristics of particles in the search process are observed. Gradually improve and expand the optimization strategy, and finally build an optimization strategy with a wide range of applicability, including the following three aspects: 1) for further research and optimization of particle swarm optimization algorithm. The nonlinear learning factor is used at the same time. A new traction strategy is proposed to optimize particle swarm optimization (PSO). Particle Swarm Optimization Algorithm based on Homing HMPSO). . this strategy shifts the particle to the optimal solution by causing the particle to shift to the optimal solution. Increase particle activity, thus improve the ability of optimization in the later stage of the algorithm. According to the requirements of the experiment, the benchmark functions are adjusted and transformed, and the optimization tests are carried out through simulation experiments. The results show that. In the later stage of the algorithm, the optimization ability is obviously improved, and the algorithm has good robustness. Finally, the probability interval of the accuracy of the algorithm is estimated to be higher than the specified threshold precision. It is proved that the strategy has good reliability.) in order to further alleviate the problems of slow convergence rate and premature convergence of particle swarm optimization algorithm, a mount formula is proposed. The search space strategy depends on adaptive threshold and known global optimal solution. On this basis, the initial position of particle is reassigned and the velocity weight is adjusted to improve the ability of the algorithm to explore in the later stage. When using the same weight and learning factor strategy, it has better performance than the original particle swarm optimization algorithm. This strategy can effectively avoid the premature problem and improve the optimization effect of the algorithm in the later stage. The main task of swarm intelligence algorithm is to get more accurate solution in limited time. However, due to the common problems such as precocity and so on. In order to solve the precocious problem completely and keep the main body of the original algorithm unchanged and cooperate with the existing optimization theory, a more accurate solution needs to be obtained by providing additional iterations. On the basis of previous simulation experiments and theoretical proof, an improved strategy of hierarchical evolution is proposed. An adaptive system based on search space compression theory is constructed in the original algorithm. Reinitialize the operation, including the compressed search space, including social information as genetic knowledge, to guide the optimization process, so as to achieve the final solution accuracy. To avoid the problem of precocity. The simulation of benchmark function shows that the strategy has a good performance in improving the algorithm accuracy and enhancing the individual activity in the later stage. The three strategies mentioned above. It is proved in turn that improving the diversity of population is helpful to the final performance of PSO. In the same situation, the algorithm can be improved by compressing the search space. As an optimization strategy based on population diversity and search space, the evolutionary strategy of layer by layer has better universality than the former.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】