【作者】 戴朝华；

【导师】 陈维荣；

【作者基本信息】 西南交通大学 ， 电气系统控制与信息技术， 2009， 博士

【摘要】 优化是科学研究、工程实践和经济管理等诸多领域十分关心的课题,其目的是建立目标函数并找到使目标函数最小或最大的解。随着人类生存空间的扩大以及认识与改造世界的不断深入与发展,各个领域,特别是人工智能与控制领域,存在大量具有多模态、非线性以及建模困难等特点的复杂系统;因此,人们对高效优化技术和智能计算显示出与日俱增的热情和关注,寻求适合大规模并行且具有智能特征的算法已成为许多学科的一个重要研究方向。社会性动物的个体遵循简单的规则,通过个体与个体之间、个体与环境之间的交流,往往能产生惊人的自组织行为。在过去的20年里,社会性动物的这些行为受到越来越多的研究人员的关注,并受其启发,提出了用于解决分布式优化问题的群体智能算法。优化问题出现在人类生活的方方面面,针对每个问题搜索优化解成了人类的一种基本行为。本文围绕对人类搜索行为的模拟,用于求解优化问题。本文的主要研究成果归纳如下。(1)提出了云自适应遗传算法(CAGA)。将AGA的作用机理转化为人类特有的自然语言描述:高于种群平均适应度的个体,随着适应度的增加,交叉、变异概率逐渐减小;而低于种群适应度的个体采用最大交叉、变异概率。然后,引入正态云模型,对上述语言描述进行建模,用于自适应确定交叉、变异概率。由于正态云模型的随机性和稳定倾向性特点,使交叉、变异概率既具有传统AGA的趋势性,满足快速寻优能力;又具有随机性,提高了算法避免陷入局部极值的能力。典型函数优化和TSP问题验证了CAGA的有效性。(2)提出了云进化算法(CEA)。基于连续函数“介值定理”思想和模拟人类的聚焦搜索行为,利用云运算实现交叉、变异操作,提出了CEA。由于正态云模型具有随机性和稳定倾向性的特点,随机性可以保持种群多样性从而避免搜索陷入局部极值,稳定倾向性又可以很好地保护较优个体并对全局最优值进行自适应定位,从而较大程度克服了遗传算法局部搜索能力差、收敛速度慢和进化无记忆性等问题。典型函数优化和FIR数字滤波器设计验证了CEA的有效性。(3)提出了搜寻者优化算法(SOA)。对人类搜索行为进行深入研究,将优化视为搜寻队伍在搜索空间对最优解的搜索,以搜寻队伍为种群,以搜寻者所处位置为优化问题的候选解,提出了SOA。SOA利用“经验梯度”确定搜索方向、不确定性推理确定步长,完成搜寻者在搜索空间中的位置更新,实现解的优化。(4)对本文提出的SOA作为一种新的群体智能算法的有效性进行了理论分析,阐述了SOA相对于其它智能优化算法的相同点和不同点,研究了SOA参数对性能的影响;将SOA应用于CEC05的benchmark函数优化,验证了SOA的有效性。(5)将本文提出的SOA应用于神经网络(ANN)训练。由于ANN性能对网络结构敏感,为了保持网络复杂度和泛化能力问的平衡,同时使用了权值训练、结构进化和优化的正则化性能函数。典型的模式识别和函数逼近问题验证了算法的有效性和优越性。(6)将本文提出的SOA应用于IIR数字滤波器优化设计。将滤波器的优化设计等价于系统辨识问题,利用SOA调整IIR数字滤波器的系数,使系统的输出与IIR数字滤波器的输出问的均方误差最小,从而将IIR数字滤波器设计的任务转化为一个最小化优化问题。典型实例验证了算法的有效性。(7)将本文提出的SOA应用于质子交换膜燃料电池(PEMFC)的优化建模。根据PEMFC的极化曲线模型,采用SOA对该模型的参数进行优化,实现了将SOA应用于燃料电池模型优化。仿真结果表明,SOA可有效用于PEMFC建模。(8)将本文提出的SOA应用于电力系统无功优化。通过调整变压器变比、补偿电容器容量和发电机端电压等控制变量,在尽可能保证各节点电压幅值和发电机无功输出、支路无功功率满足电力系统运行安全和电能质量所需的同时,使有功网损最小。以IEEE 57节点系统为例进行仿真实验,验证了新算法的有效性和优越性。更多还原

【Abstract】 Optimization has been widely used in many fields such as scientific research, engineering practice and economic management, etc. The aim is to create an objective function and find a solution for minimizing or maximizing it. With the extension of human knowledge and activities, optimization technique and algorithm has become the target of increasing interest due to the increasing demands for improved optimization performance in many complex systems where the involved objective functions are non-linear, multi-modal, and even cannot be expressed in explicit mathematical forms and their derivatives cannot be easily computed. In many research fields, pursuing an effective optimization method has become one of the main objectives for the scientific researchers.The wonderful and constructive self-organization behavior usually emerges from the individuals of social animals who follow some simple rules and communicate with each other and their environments. In the past 20 years, these behaviors of social animals have been attracting more and more attention of researchers from which swarm intelligence computation was proposed. Swarm intelligence is an algorithm or a device and illumined by the social behavior of gregarious insects and other animals, which is designed for solving distributed problems. Optimization tasks are often encountered in many areas of human life, and the search for a solution to a problem is one of the basic behaviors to all mankind. The dissertation focuses on learning from the advanced social animal, human, and simulating their behaviors for solving optimization problems. The main contributions given in this dissertation are as follows.(1) A novel cloud-based adaptive genetic algorithm (CAGA) was proposed. The rule for the probability of crossover (mutation) can be described by natural linguistic variables as follows: the probabilities of the individuals with hyper-average fitnesses decrease with the fitness increasing while the probabilities of the individuals with sub-average fitnesses are set at a fixed maximum value. Then, a normal cloud model is introduced to model the rule and adaptively give the probabilities of crossover and mutation. Because cloud model has the properties of randomness and stable tendency, the first property is able to help GA avoid a local optimum, and the second property can improve its convergence speed. The benchmark function optimization and TSP problems proved the effectiveness of CAGA.(2) A novel cloud-based evolutionary algorithm (CEA) was proposed. Based on the intermediate value theorem of the continuous function and the simulation of human focusing search, CEA is implemented using cloud models as crossover and mutation operators. Unlike GA with the property of "absence of memory", CEA searches the optimal solutions around the current generations until to converge to the optimum for as few generations as possible. Owning to the stable tendency of cloud model, CEA does not easily get lost and is able to locate the region in which the global optimum exists. On the other hand, the randomness of cloud model can maintain the diversity of the population and make CEA enough robust not to get stuck at a local optimum. Benchmark function optimization and digital FIR filter design proved the effectiveness of CEA.(3) A novel stochastic search algorithm called as seeker optimization algorithm (SOA) was proposed. In SOA, optimization is viewed as a search of search population in the search space. After the detailed study of human searching behaviors, SOA is established with search group as population and the position of a seeker as a candidate solution. In the algorithm, the choice of search direction is based on the empirical gradient by evaluating the response to the position changes, and the decision of step length is based on uncertainty reasoning by using a simple Fuzzy rule. Then, according to the given search direction and step length, the position update is conducted so as to implement the solution evolution.(4) It is proved that SOA is a new swarm intelligence algorithm. Further theoretical analysis on the validity of SOA was conducted. The difference and sameness between SOA and other intelligent optimization algorithms were presented. The influence of the parameters on the performance was studied. Then, SOA was applied to the benchmark functions of CEC05 and presented the better optimization outputs for some functions. The results of the experiments proved the effectiveness of SOA.(5) A SOA-based training algorithm was proposed for the evolution of weight values and structure. Since the MLP’s performance is sensitive to network architecture, to achieve a proper balance in ANN’s network complexity and generalization capability, two approaches are used in this study, namely: structure evolution and the use of the mean squared error with regularization performance function (MSEREG). Hence, the individual in SOA consists of three parts: the link switch bits, the connection weight values and the regularization parameter. The benchmark problems from pattern classification and function approximation were used to evaluate the new algorithm.(6) A SOA-based evolutionary method is proposed for digital IIR filter design. In this study, IIR filters are designed for the system identification purpose. In this case, the parameters of the IIR filter are successively adjusted by SOA until the error between the outputs of the filter and the system is minimized. Several widely used design examples were used to exhibit the effectiveness of the proposed method.(7) A new optimized model of proton exchange membrane fuel cell (PEMFC) was proposed by using the proposed SOA. In this case, SOA was used to automatically tune the parameters of the polarization curve model of a PEMFC to minimize the mean square error between the optimized model and the previously sampled experimental data. The simulation results have shown that SOA is suitable for modeling a PEMFC.(8) The proposed SOA was applied to reactive power optimization. The aim is to minimize the active power loss in the transmission network by tuning the generator voltage, the transformer tap and the shunt capacitor/inductor and simultaneously keeping the load-bus voltage and the generator reactive power within the suitable limits to maintain the voltage quality. The new method is tested on IEEE-57 bus power systems, and the experimental results showed that SOA is suitable for reactive power optimization.更多还原