【作者】 王康泰；

【导师】 王宁；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2011， 博士

【摘要】 作为求解复杂优化问题的有效方法,遗传算法从问世之日就受到了人们的关注。遗传算法是模拟达尔文的生物进化理论而产生的一类随机优化算法,具有较强的全局寻优能力和较好的通用性。但是,传统的遗传算法也存在着一些缺点和不足,例如算法局部搜索能力差、容易陷入局部极值等。受生物RNA分子的编码和操作启发,把RNA分子编码和操作方式与遗传算法相结合,本文对RNA遗传算法展开深入研究,并将所提出的RNA遗传算法用于化工过程的建模问题。本文的主要研究成果如下：(1)针对传统遗传算法的个体随进化不断趋同,易导致早熟收敛的现象,受生物界相似个体竞争排挤现象的启发,提出了一种基于相似剔除策略的RNA遗传算法。该算法采用了基于碱基的四进制编码方式,并用RNA置换操作和RNA颈环操作代替传统的交叉操作。在选择操作中,通过增加相似剔除规则来剔除相似个体,以达到提高种群多样性,避免早熟收敛的目的。通过5个典型无约束测试函数的寻优实验,验证了该算法在避免早熟收敛上的有效性。将该算法用于三个动态系统建模的参数估计问题,结果表明所建模型能较好地反映实际系统的动态特性。(2)受生物遗传信息表达过程的启发,提出了一种基于蛋白质特性RNA遗传算法。.模拟生物遗传信息表达过程中DNA分子先转录成RNA分子再翻译成蛋白质分子的过程,设计了RNA再编码操作和蛋白质折叠操作。通过对典型测试函数的计算结果表明,所设计的新的交叉操作使RNA遗传算法在搜索精度和稳定性方面都有较大的提高。将该算法用在重油加氢裂解建模的参数估计中,仿真结果验证了该算法的有效性。(3)针对传统遗传算法的变异概率取值的盲目性,提出了一种基于信息熵动态变异概率RNA遗传算法。该算法中每一代个体的每一位的变异概率是根据当前种群中个体该位碱基的分布情况来设定,可克服传统变异概率的设定与当前种群个体特性无关的弊端。通过对4个典型测试函数寻优验证了该算法的有效性。用该算法求解短期汽油调合优化调度问题,仿真结果表明,该算法能获得更大的调合利润。(4)受生物膜结构的启发,提出了一种基于膜结构的分层RNA遗传算法。该算法将RNA遗传算法嵌入膜系统中,2个膜子系统的RNA遗传算法的计算结果作为外部表层膜系统的RNA遗传算法的部分初始种群。针对具有约束的非线性测试函数的寻优结果表明,该算法具有满意的寻优精度。将该算法用于求解短期汽油调合优化调度问题,仿真结果表明,该算法得到的调合利润高于其它2种优化算法得到的结果。(5)提出了一种变搜索空间的RNA遗传算法。该算法根据进化过程中种群个体的分布情况,对搜索空间做出适当的调整以减小对无效空间的搜索。通过5个典型测试函数寻优实验验证了该算法的有效性。将该算法用于CSTR神经网络建模问题,仿真实验结果表明,用该算法优化得到的神经网络具有较高的拟合精度和较小的网络复杂度,较好的反映了实际系统特性。更多还原

【Abstract】 As an effective method to solve complex optimization problems, genetic algorithm (GA) has being attracted people’s attention since it proposed. Being global search techniques, GAs simulate the processes of natural evolution and own some remarkable advantages, such as excellent global exploration ability and good applicability. However, traditional genetic algorithms have some drawbacks, such as poor local searching ability and premature convergence etc. Inspired by RNA molecular coding and operations, RNA genetic algorithms and their applications are studied in this dissertation. The main contents are as follows:(1) These individuals in population are becoming more and more similar with evolution, which makes GA be prone to premature convergence. In order to overcome this drawback, the RNA genetic algorithm with similar individuals rejected (srRNA-GA) is proposed. The algorithm adopts encoding of the nucleotide bases and RNA permutation operation and RNA stem-loop operation are designed for the proposed genetic algorithm. In selecting operation, the strategy of rejecting similar individuals is applied to improve population diversity and avoid premature convergence. The performances of the proposed RNA-GA are validated by five benchmark functions. The efficiency and accuracy of this proposed algorithm are demonstrated in the three practical parameter estimation problems.(2) Inspired by the expression of bio-genetic information, a RNA genetic algorithm based on protein characteristics is proposed. In this algorithm, the RNA recoding operation and protein folding operation are designed to replace conversional crossover operation by simulating the process from DNA molecular to protein ones in biology. Numerical experiments with some benchmark functions and the parameter estimation problem in hydrocracking of heavy oil demonstrate the effectiveness of this proposed algorithm.(3) In order to overcome the drawback setting GA’s parameter blindly, the RNA genetic algorithm with dynamic mutation probability according to entropy is proposed. In this algorithm, the values of mutation probability are decided by nucleotide bases distribution of the current bits of population. The numerical results on four benchmark functions show the effectiveness of this proposed algorithm. The solution of the short-time gasoline blending scheduling problem shows that the proposed algorithm gain a higher profit.(4) The membrane structure based hierarchical RNA genetic algorithm is proposed. In this algorithm, the RNA genetic algorithms are imbedded into the membrane system and the final results of the RNA-GAs in the two membrane sub-systems are used to determine a part of initial population of the RNA-GA in the skin membrane system. Numerical results on six benchmark functions demonstrate excellent search performance of this proposed algorithm. The algorithm is applied to solve short-time gasoline blending scheduling problem. The experimental results show that this proposed algorithm can obtain a higher profit.(5) The RNA genetic algorithm with a changeable problem search space is presented for constrained nonlinear optimization problems. Along with running, the changeable problem search space is re-built for promoting the search efficiency. The performances of this proposed algorithm are validated by five benchmark functions. It is also used to model the CSTR process by a neural network. The simulation results show that higher precision and lower complexity are reached.更多还原