【作者】 何泳；

【导师】 韩敏；

【作者基本信息】 大连理工大学 ， 测量计量技术及仪器， 2016， 硕士

【摘要】 现实中大量的科学研究与工程实践问题都可归结为多目标优化问题。粒子群优化作为一种群体智能计算模型,易于实现且收敛速度快,适合于求解多目标优化问题,吸引了众多学者进行广泛而深入地研究。目前国内外已有部分多目标粒子群优化算法的研究成果,但仍然存在一些不足：方面,大部分算法缺乏监测种群进化环境的机制,无法获得实时的反馈信息,难以决定在何时调节何种进化策略到何种程度；另一方面,粒子群优化算法在求解高维多目标优化问题时优化能力急剧下降。针对以上不足,本文主要进行了以下研究：(1)针对缺乏监测进化环境机制的问题,设计了相应的监测机制,并在平衡全局搜索和局部搜索、平衡解集收敛性和多样性两方面分别提出了两种改进的多目标粒子群优化算法。在基于高斯混沌变异和精英学习的自适应多目标粒子群算法中,通过监测种群的收敛状态来自适应调节惯性权重和学习因子；而且,提出精英学习策略和改进的高斯混沌变异算子来调节局部搜索和全局搜索能力。而在基于档案解集状态的自适应多目标粒子群算法中,混合了两种全局向导选择策略,并通过监测档案解集所处的状态来自适应调整这两种策略的选择概率；此外,分别对粒子和档案解集中的个体进行扰动,根据档案解集状态和迭代次数动态调整这两种扰动的概率,使算法能兼顾解集的收敛性和多样性。(2)针对多目标粒子群优化算法在求解高维多目标优化问题上的不足,提出一种基于参考点的高维多目标粒子群算法。在目标空间中引入一系列的参考点,根据参考点来筛选出兼顾收敛性和多样性的非支配解作为粒子的全局向导,并提出了基于参考点的档案维护方法,维持解集的多样性。(3)将多目标粒子群优化算法应用于转炉炼钢铁合金加入量计算问题中。在转炉炼钢生产过程中,如何保证钢水中各成分含量达标的同时降低生产成本是影响钢铁企业生产效益的一个重要问题。本文在回声状态网络进行软测量的基础上,将该问题转化为一个多目标优化问题,并采用改进的多目标粒子群优化算法对其进行求解,在实际炼钢数据上的仿真实验表明,所提方法能在保证钢水成分含量达标的同时有效降低铁合金投入的成本。更多还原

【Abstract】 Many problems in scientific studies and engineering practices could be transformed into multi-objective problems. As one of the swarm intelligence based models, particle swarm optimization(PSO) is is easy to implement and has fast convergence speed. PSO is suitable for solving multi-objective problems and a great number of scholars have been researching multi-objective particle swarm optimization algorithms.The research of the multi-objective particle swarm optimization has achieved lots of results, but there are still some deficiencies to be improved. On the one hand, most multi-objective particle swarm optimization algorithms lack the mechanisms of monitoring the swarm evolutionary environment. These algorithms cannot obtain real-time information feedback and are hard to decide when to adjust which evolutionary strategy to what extent. On the other hand, when solving the many objective problems, the optimization performance of particle swarm optimization algorithms degrades severely. The main contents of this paper are as follows:(1). To deal with the lack of mechanisms of monitoring the evolutionary environment, we designed the mechanisms and proposed two improved multi-objective particle swarm optimization algorithms to balance the global and local researching abilities, the convergence and diversity respectively. In the adaptive multi-objective particle swarm optimization algorithm with gaussian chaotic mutation and elite learning, we adjust the inertia weight and acceleration coefficients adaptively by monitoring the convergence state of swarm. Further more, an elite learning strategy and an improved mutation operator with the features of Gaussian function and chaotic sequence are proposed to adjust the local and global researching abilities. In the adaptive multi-objective particle swarm optimization algorithm based on the status of external archive, we combined two global best selection strategies and adjust the execution probability of the two strategies adaptively by monitoring the state of external archive. Further more, perturbation operators are applied to the particles in swarm and solutions in archive, and the perturbation probabilities are adjusted adaptively by the status of external archive to balance the convergence and diversity.(2). We proposed a reference-point-based particle swarm optimization algorithm for many-objective optimization. Aiming at the insufficiency of performance when solving many-objective optimization problems, we introduced a structured set of reference points in the objective space. After that, we can select the solutions which have both good convergence and diversity as the global best. Further more, we proposed a truncation method of external archive based on the reference points to maintain the diversity of the solution set.(3). We applied our multi-objective particle swarm optimization algorithm to the calculation of adding alloy in basic oxygen furnace steelmaking process, how to reduce the costs while ensure the contents of the liquid steel meet the requirement is an important issue which affect the steel plant’s benefits. On the basis of the soft measurement by echo state network, we transformed the the issue to solve a multi-objective optimization problem. After that, our improved multi-objective particle swarm optimization algorithm is applied to solve the problem. Simulation results on real data of basic oxygen furnace steelmaking showed that our method could reduce the costs of alloys and ensure the contents of the liquid steel meet the requirement.更多还原