【作者】 张伟；

【导师】 王宁；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2014， 博士

【摘要】 人工蜂群(简称ABC)算法是一类新的仿生智能优化算法。由于其特有的分工协作、信息交流等机制,自其问世以来就受到了人们广泛的关注。ABC算法结构简单、概念清晰,而且容易实现、全局优化性能好,有广阔的应用前景。尽管如此,基本的人工蜂群算法仍有诸多不足,例如后期收敛速度慢、局部搜索能力弱、不能处理离散和约束问题等。本文对人工蜂群算法进行深入研究,结合其他生物智能算法的机制,提出了多种人工蜂群混合算法并用于求解工程背景的优化问题。本文的主要研究工作如下：(1)针对具有离散和连续变量、多重约束的机组组合调度优化问题,结合遗传算法(简称GA)在求解离散问题方面的优势和ABC算法较强的实数优化能力,提出一种基于约束分解和修复的ICGA-ABC算法。该算法使用整型编码(简称1C)的染色体更直观地表达调度解并减小编码长度；通过在遗传进化中嵌入修复操作提高搜索可行解的能力,避免大规模时的“组合爆炸”问题；使用ABC算法快速求解简化后的实数域“经济负载分配”问题。将ICGA-ABC用于多个不同规模机组组合调度问题的求解,结果表明与LR、 BCGA、 ICBF等方法相比,本文所提出的算法可获得更高质量的调度解,有效减小了总发电成本。(2)针对人工蜂群算法在搜索后期局部优化能力差、收敛速度慢的问题,受生物界中大肠杆菌觅食行为的启发,将细菌趋药性作为局部搜索策略嵌入到人工蜂群算法的“观察蜂相”中,提出了一种混合ABC算法(称为HABC)。在选择操作上,该算法引入自适应Boltzmann概率,有利于种群维持多样性、避免陷入局部最优、加快收敛速度。通过对8个典型测试函数的求解,结果表明所提算法在求解精度及可靠性方面都有显著提高。将该算法用于求解质子交换膜燃料电池模型参数估计问题,对两个电池实例的实验结果表明所建模型的预测值与实验值的误差更小,能更好地反映实际系统的非线性特性。(3)受RNA分子操作的启发,将RNA分子操作算子引入人工蜂群算法中以提高种群多样性、避免重复搜索和早熟收敛,提出了一种RNA-ABC算法。结合使用Oracle罚函数法和通用的离散量转换技术,可以将复杂约束转化为无约束问题。通过对8个典型非线性约束优化问题的寻优测试,验证了所提方法的有效性。使用RNA-ABC算法求解炼油厂短期汽油调合调度优化问题,所得的调合配方能获得比其他方法更高的调合利润。(4)为提高人工蜂群算法的搜索效率,特别是增强对高维问题的寻优能力,将差分进化算法(简称DE)中的几种变异操作和交叉算子引入到“雇佣蜂相”,提出了一种具有竞争型差分操作的混合ABC算法(称为DABC)。该算法使用新的组合式向量产生策略并引入竞争,以增强雇佣蜂对搜索的引领作用,提高算法性能。通过对测试函数的寻优,验证了所提出的算法求解较高维问题时在搜索速度上的提升。将该算法用于桥式吊车系统的RBF神经网络优化建模,实验结果表明,使用DABC算法可以获得具有良好拟合精度和泛化能力的RBF网络模型。更多还原

【Abstract】 Artificial bee colony (ABC) algorithm is a kind of relatively new bionic intelligent algorithm. Due to its unique mechanisms, such as division of labor, information exchange, it has attracted widespread attention since it proposed. Artificial bee colony algorithm has simple structure and clear concept, furthermore, it is easy to implement and possess excellent global optimization performance. Therefore, it has broad application prospects. However, the classical ABC algorithm also has some drawbacks and deficiencies, such as slow convergence speed at later stage, weak local search capability, and it could not deal with discrete variables and problems with constraints. Inspired by the behavior of biological intelligence, the ABC algorithm is studied and improved in this dissertation, and then the proposed algorithms are applied to solve some complex engineering optimization problems.The main contributions of this dissatation are summarized as follows:(1) To solve complex unit-commitment scheduling problems with mixed variables and heavy constraints, the superiorities of Genetic Algorithm (GA) in discrete problems and ABC algorithm in continuous problems are combined and the ICGA-ABC algorithm based on repair operation is proposed. This algorithm adopted integer-coded (IC) chromosomes to represent the scheduling solutions intuitively, with shorter string lengths. The proposed repair operation improves the search capabilities of feasible solutions, and avoided the potential "combination explosion" The UC problem is then simplified to an Economic Load Distribution (ELD) problem and ABC is adopted to solve it efficiently. The ICGA-ABC is used to solve more UC problems with different numbers of generators, and the results show that it can obtain better schedulings than that of LR, BCGA and ICBF, which could result in the reduction of total power production cost.(2) The convergence speed of ABC algorithm will slow down much at later stage. Inspired by the foraging behavior of E.coli, chemotaxis effect is embedded into the ABC algorithm as local search strategy and the HABC algorithm is proposed. For selection operation, adaptive Boltzmann probability is adopted to adjust selective pressures, which could improve population diversity, avoid premature convergence, and accelerate the convergence velocity. The performances of proposed HABC are validated by8typical benchmark functions. The HABC is firstly applied to solve the parameter estimation problems of the PEM Fuel Cell model, and the results show that the optimized model by HABC can predict experimental data points more accurately, and can reflect the nonlinear property of the system better.(3) Inspired by RNA molecular, the three RNA operators are introduced into the ABC algorithm to improve population diversity and avoid trapping into local minima. The RNA-ABC algorithm is combined with Oracle penalty function technology to solve optimization problems with complex constraints. Discrete variables are changed to continuous ones and equality constraints are transformed to inequality constraints, which results in that Oracle penalty method can adaptively handle all kinds of constraints. The performance of the proposed algorithm has been validated by8typical benchmark functions with complex nonlinear constraints. The proposed algorithm is applied to solve short-term gasoline-blending scheduling problem. The experimental results show that the proposed approach can obtain better recipes than two other methods, and can reach a higher profit.(4) In order to improve the efficiency of artificial bee colony algorithm, especially for high-dimensional optimization problems, a variety of mutation and crossover operators from Differential Evolution (DE) algorithm are introduced into the employed bee phase of ABC, and the DABC algorithm based on the competition is proposed. Combined with the new vector generation policy and competitive mechanism, the leading role of employed bee phase has been enhanced, and the performance is improved. Numerical experiment on some typical benchmark functions demonstrate that the hybrid algorithm has speed up the search process of solving high-dimensional problems. Then, the proposed algorithm is used to train and optimize the RBF neural network model for modeling the overhead cranes system. The experimental results show that the DABC algorithm can obtain satisfactory RBF network model with excellent fitting accuracy and generalization ability.更多还原