基于探索能力和开发能力的智能算法设计

发布时间：2018-07-05 03:58

本文选题：智能算法 + 猴群算法　；参考：《天津大学》2016年博士论文

【摘要】：智能算法是基于自然现象运行机制的随机优化算法,具有结构简单、易于操作和全局优化能力强等优点,在决策优化、系统优化、工程设计等诸多领域都具有广泛应用.然而在处理复杂优化问题时,现有的智能算法依然会出现早熟收敛和停滞问题.为了从算法的运行机理上探索早熟收敛和停滞问题的解决方案,基于全局探索能力和局部开发能力的有效平衡,本文分别设计了猴群算法和差分进化算法的改善机制,并提出了一种新型智能算法.主要工作包括:(1)设计了基于自组织分层结构和时变参数的改进方案,用于提高猴群算法的优化性能.在改进方案中,利用个体的适应值信息和优化空间的边界信息,同时融合提出的选择算子、基于适应值的替换算子和排斥算子重新设计了原始猴群算法的爬、望和跳操作;采用了分层结构组织其核心操作,并利用设计的自组织机制协调核心操作的执行;利用单个时变参数替代了原始猴群算法中的多个固定参数,提高了算法应用的便捷性.大量比较实验表明改进方案明显优于原始猴群算法和7种表现优异的智能算法.(2)设计了基于高斯变异和动态参数的改进方案,用于提高差分进化算法的优化性能.在改进方案中,利用随机选择个体的适应值信息设计了新型高斯变异算子和改进了一种典型变异算子,并利用累计分值信息提出了两种变异算子之间的协作规则;分别采用余弦函数和高斯函数实现了缩放因子的周期性变化和交叉概率的波动性变化.大量比较实验表明改进方案明显优于5种差分进化算法变型和两种表现优异的群智能算法.(3)设计了基于再初始化策略和优化空间调整策略的改善机制,用于提高差分进化算法的优化性能.在改善机制中,结合种群的优化状态和交叉算子提出了再初始化策略,用于恢复算法的全局探索能力;利用最优个体信息和具有波动性的动态参数设计了优化空间调整策略,用于防止再初始化策略引发的过度探索.所设计的改善机制具有算法独立性,可以便捷的移植到各种差分进化算法中.大量的比较实验表明改善机制可以有效提高多种差分进化算法的优化性能.(4)设计了基于军事理论中联合作战策略的新型智能算法联合作战算法,用于处理大规模复杂优化问题.在联合作战算法中,利用精英个体信息和优化空间的动态调整信息设计了攻击操作,用于探索新区域;利用正态分布和交叉算子设计了防御操作,用于开发局部区域;利用随机排序技术提出了整编操作,用于恢复种群多样性.采用多个大规模复杂优化问题进行了全面系统的比较实验,结果表明联合作战算法明显优于6种表现优异的智能算法.
[Abstract]:Intelligent algorithm is a random optimization algorithm based on natural phenomena, which has the advantages of simple structure, easy operation and strong ability of global optimization. It is widely used in many fields, such as decision optimization, system optimization, engineering design and so on. However, in dealing with complex optimization problems, the existing intelligent algorithms will still have premature convergence and stagnation problems. In order to explore the solution of premature convergence and stagnation problem from the operation mechanism of the algorithm, based on the effective balance of global exploration ability and local development ability, this paper designs the improvement mechanism of monkey swarm algorithm and differential evolution algorithm, respectively. A new intelligent algorithm is proposed. The main works are as follows: (1) an improved scheme based on self-organizing hierarchical structure and time-varying parameters is designed to improve the optimization performance of monkey swarm algorithm. In the improved scheme, using the fitness information of individuals and the boundary information of the optimization space, the proposed selection operator is fused at the same time, and the crawling, lookout and jump operation of the original monkey group algorithm is redesigned based on the fitness replacement operator and the exclusion operator. The core operation is organized with layered structure, and the implementation of the core operation is coordinated by the self-organizing mechanism designed. A single time-varying parameter is used to replace many fixed parameters in the original monkey swarm algorithm, which improves the convenience of the algorithm application. A large number of comparative experiments show that the improved scheme is obviously superior to the original monkey swarm algorithm and seven intelligent algorithms with excellent performance. (2) an improved scheme based on Gao Si mutation and dynamic parameters is designed to improve the optimization performance of differential evolution algorithm. In the improved scheme, a new Gao Si mutation operator and a typical mutation operator are designed by using the fitness information of randomly selected individuals, and the cooperative rules between the two mutation operators are proposed by using the cumulative score information. CoSine function and Gao Si function are used to realize the periodic variation of scaling factor and the fluctuation change of crossover probability. A large number of comparative experiments show that the improved scheme is obviously superior to five differential evolution algorithms and two excellent swarm intelligence algorithms. (3) an improved mechanism based on reinitialization strategy and optimized spatial adjustment strategy is designed. It is used to improve the optimization performance of differential evolution algorithm. In the improvement mechanism, combining the optimal state of the population and the crossover operator, the reinitialization strategy is proposed to recover the global exploration ability of the algorithm, and the optimal spatial adjustment strategy is designed by using the optimal individual information and the dynamic parameters with volatility. Used to prevent over-exploration caused by reinitialization policies. The improved mechanism is algorithmic independent and can be easily transplanted to various differential evolution algorithms. A large number of comparative experiments show that the improved mechanism can effectively improve the optimization performance of various differential evolution algorithms. (4) A new intelligent algorithm based on the joint operational strategy in military theory is designed. Used to deal with large-scale complex optimization problems. In the joint operation algorithm, the attack operation is designed by using the information of the elite individual and the dynamic adjustment information of the optimized space to explore the new area, the defense operation is designed by using the normal distribution and the crossover operator to develop the local area. In this paper, an integration operation is proposed to restore population diversity using random sorting technique. A comprehensive and systematic comparison experiment is carried out with several large-scale and complex optimization problems. The results show that the joint operations algorithm is superior to the six intelligent algorithms with excellent performance.
【学位授予单位】：天津大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP18

【参考文献】