基于蝙蝠算法的启发式智能优化研究与应用

发布时间：2018-07-10 11:41

本文选题：群体智能 + 蝙蝠算法　；参考：《北京工业大学》2016年博士论文

【摘要】：群智能优化算法是一类模拟动物群体行为的随机优化算法,该类算法利用动物群体所突显的智能来求解问题。目前为止,已经提出了很多群智能优化算法,这些算法从动物的视觉、听觉、嗅觉等角度出发来求解问题,但它们普遍存在局部搜索性能较差,以及算法过早收敛等问题。而蝙蝠则具有一种完全不同的觅食方式—回声定位。因此,模拟这种觅食方式的蝙蝠算法对上述问题提供了一种完全不同的解决方式。本论文以蝙蝠算法为基本算法,研究高效的改进蝙蝠算法,以期进一步改善蝙蝠算法性能使其应用范围更广阔。与已有蝙蝠算法的研究成果相比较,本论文系统地从性能分析、理论分析、算子设计、算法融合、知识学习、多目标优化等方面进行研究,并将蝙蝠算法应用在生物信息学、无线传感器网络等多个工程问题中。本文研究成果和创新点如下:(1)针对蝙蝠算法中速度震荡问题,引入速度权重概念,设计了不同的速度权重,使蝙蝠个体能根据其适应值自适应的调整飞行速度,更好的在搜索空间内进行搜索;针对标准蝙蝠算法搜索方式发散和搜索区域不完整的缺点,通过扩大频率范围,使蝙蝠的搜索区域覆盖整个搜索空间,提高了蝙蝠算法的全局搜索能力。(2)针对蝙蝠算法局部搜索能力差的问题,从算法融合角度去优化标准蝙蝠算法,分析不同策略与智能算法的优化机理,从纯数学理论和其它智能优化算法中汲取优点,将其与蝙蝠算法相结合以进一步提高算法性能。数学理论方面,利用Powell法增强标准蝙蝠算法的局部搜索能力。智能优化算法融合方面,本文分析并采纳了遗传算法、模拟退火算法和分布估计算法的优异算子,将其引进到蝙蝠算法中进行融合,从而进一步优化蝙蝠算法。(3)针对蝙蝠算法个体信息利用率低的缺点,将知识学习引入到蝙蝠算法中。首先,蝙蝠个体在寻优过程中不断利用其历史知识和群体知识调整优化,有利于群体成员向更好的方向移动,加快算法收敛速度。其次,针对高维多峰问题,利用相似度聚类函数将蝙蝠群体分为不同的蝙蝠簇,使蝙蝠有针对性的进行区域知识学习。最后,引入偏好知识维度概念,提高蝙蝠个体学习能力。(4)在多目标蝙蝠算法的基础上引入非支配快速排序策略,该策略不仅可以筛选出距离真实前沿较近的个体,而且可以使得所求个体均匀的分布在真实前沿的边缘。偏好多面体策略的引入进一步降低了决策者在选择非劣解时的困难,该策略利用非减拟凹函数代替决策者参与非劣解的筛选,所得的非劣解可以有效的反映决策者的偏好。(5)将蝙蝠算法应用到多个不同领域,其中包括生物信息学中的RNA二级结构预测问题、蛋白质折叠预测问题,无线传感器网络中的覆盖问题、定位问题以及团簇优化问题。针对不同问题,对蝙蝠算法离散化,并设计了不同目标优化模型。
[Abstract]:Swarm intelligence optimization algorithm is a kind of stochastic optimization algorithm for simulating animal group behavior. This kind of algorithm uses the intelligence of animal groups to solve the problem. So far, many swarm intelligence optimization algorithms have been proposed. These algorithms solve the problem from the angle of animal vision, hearing and smell, but they are generally local. The bats have a completely different foraging method - echolocation. Therefore, the bat algorithm that simulates this way of foraging provides a completely different solution to the above problem. This paper uses bat algorithm as the basic algorithm to study the efficient bat algorithm. In order to further improve the performance of bat algorithm to make its application wider, compared with the achievements of the existing bat algorithm, this paper systematically studies the performance analysis, theoretical analysis, operator design, algorithm fusion, knowledge learning, multi-objective optimization and so on, and applies the bat algorithm to bioinformatics and wireless sensor networks. The research results and innovation points of this paper are as follows: (1) according to the speed shock problem in bat algorithm, the speed weight concept is introduced to design different speed weights, so that the bat individual can adjust the flight speed adaptively according to its adaptive value, search in the search space better, search for the standard bat algorithm. By expanding the frequency range, the search area of bats is covered with the whole search space, and the global search ability of the bat algorithm is improved by expanding the frequency range. (2) in view of the problem of the poor local search ability of the bat algorithm, the standard bat algorithm is optimized from the angle of algorithm fusion, and the different strategies and intelligent calculation are analyzed. The optimization mechanism of the method is derived from the pure mathematical theory and other intelligent optimization algorithms. The algorithm is combined with the bat algorithm to further improve the performance of the algorithm. In mathematical theory, the Powell method is used to enhance the local search ability of the standard bat algorithm. The intelligent optimization algorithm combines the square surface. This paper analyzes and adopts the genetic algorithm, and simulated the regression. The excellent operators of fire algorithm and distribution estimation algorithm are introduced into the bat algorithm to further optimize the bat algorithm. (3) the knowledge learning is introduced into bat algorithm for the disadvantage of low utilization rate of individual information in bat algorithm. First, the bat individuals continue to use their historical knowledge and group knowledge in the process of optimization. The adjustment and optimization will help the group members to move in a better direction and speed up the convergence speed of the algorithm. Secondly, according to the Gao Weiduo peak problem, the bat group is divided into different bat clusters by similarity clustering function, so that the bat can learn the regional knowledge pertinence. Finally, the concept of preference knowledge dimension is introduced to improve the individual learning ability of the bat. (4) on the basis of the multi target bat algorithm, the non dominated fast sorting strategy is introduced. This strategy can not only screen out individuals near the real front of the real distance, but also make the individual distributed evenly on the edge of the real frontier. In this strategy, the non substandard concave function is used instead of the decision maker to select the non inferior solutions. The non inferior solutions can effectively reflect the preference of the decision-makers. (5) the bat algorithm is applied to a number of different fields, including the RNA two structure prediction in bioinformatics, the problem of protein folding prediction and the coverage in Wireless Sensor Networks For different problems, the bat algorithm is discretized and different target optimization models are designed.
【学位授予单位】：北京工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP18

【参考文献】