【作者】 陈霄；

【导师】 王宁；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2010， 博士

【摘要】 遗传算法是模拟生物进化过程的一类随机性全局优化算法,广泛应用于化工过程的建模与优化中。然而遗传算法的局部搜索能力较弱、易早熟收敛,而且常用的二进制编码方法不能表达丰富的遗传信息,因此在其计算模型中没有反映出遗传信息对生物体的调控作用,尤其是起关键作用的DNA编码机制的调控作用。近年来,随着DNA计算的发展,人们发现基于DNA的智能系统能反映生物体的遗传信息,有利于发展功能更强大、能解决更复杂问题的智能方法。受DNA生物特性的启发,本文对DNA遗传算法及应用进行了深入的研究,主要研究工作如下：(1)受DNA分子操作启发,提出了多种新型交叉算子。利用Markov链模型分析讨论了具有新型交叉算子的DNA遗传算法的收敛性。测试函数计算结果表明所提新型交叉算子可以有效改善种群多样性,减少了寻优代数。将该算法用于催化裂化主分馏塔的参数估计问题中,实验结果表明所建立的模型反映了复杂系统的动态性能。(2)受DNA和遗传信息表达过程的启发,基于碱基编码方式,提出了多种新型变异算子。利用Markov链模型分析了具有新型变异算子的DNA遗传算法的收敛性。测试函数计算结果表明所提新型变异算子可以显著提高DNA遗传算法的收敛速度,增强算法克服问题欺骗的能力。将该算法用于渣油加氢过程的参数估计问题中,比较结果表明了该算法的有效性。(3)将所提出的新型交叉和变异算子在DNA遗传算法中配合使用。测试函数的检验结果表明配合使用新型操作算子可以进一步提高算法性能。使用具有新型交叉算子和变异算子的DNA遗传算法对重油热解过程建模,仿真实验结果表明了所建模型的误差小。(4)针对具有不等式约束的非线性规划问题,提出了一种混合DNA遗传算法。该算法将DNA遗传算法的全局搜索能力和SQP算法的局部搜索能力相结合。测试函数比较结果证明了该混合算法的有效性。利用该混合算法优化汽油调合问题的配方,结果表明该混合算法可以实现调合产品的质量指标卡边控制,增加生产利润。(5)提出了一种双链DNA遗传算法的广义回归神经网络建模方法,用来解决非线性系统建模问题。使用该方法对一个非线性系统进行建模,仿真结果表明所提方法的建模精度优于其它神经网络方法,并将该建模方法用于延迟焦化过程的建模,仿真结果表明所建模型精度高,其误差标准差和AIC指标较小(6)针对复杂非线性系统,提出了一种混沌DNA遗传算法的T-S模糊递归神经网络建模方法。在碱基编码和新型操作算子的基础上,混沌DNA遗传算法通过对劣质个体进行混沌细搜索,来提高个体品质。将该方法用于pH中和过程建模,仿真实验和比较结果表明所建立的模型的拟合精度高。(7)针对多目标优化问题,提出了一种多目标DNA遗传算法。测试函数的仿真研究表明,该算法可以更好的逼近Pareto前沿,解的分布更均匀,搜索速度更快。将该算法用于设计基于T-S模糊递归神经网络的广义预测控制器,对一个pH中和过程进行控制,仿真实验结果表明所建模型的精度更高,控制效果更好。更多还原

【Abstract】 A genetic algorithm (GA) is a stochastic global optimization technique which simulates natural evolution process, and it has been widely employed in the modeling and optimization problems of chemical engineering processes. But GA has weak local searching ability and tends to premature. Furthermore, the binary encoding method cannot represent the abundant genetic information and reflect the organism control function of genetic information in its computing model, especially the function of the DNA encoding method. Recently, with the development of DNA computing, researchers find that the intelligent systems based on DNA can reflect the genetic information of organism and develop more powerful intelligent methods to solve complex optimization problems.Inspired by the biological characteristics of DNA, DNA genetic algorithms (DNA-GA) and their applications are studied in this dissertation. The main contents are as follows.(1) Inspired by DNA molecular operations, several novel crossover operators are proposed. The convergence of the DNA genetic algorithm with the crossover operators is analyzed in terms of Markov chain model, and the results of the test experiments show that the algorithm can effectively maintain the diversity of the population and reduce the required evolution generations. Then, the algorithm is used to estimate the parameters of a fluid catalytic cracking unit model, and the solution of typical test functions show the model can reflect the dynamic property of the complex process.(2) Inspired by DNA and the expression of genetic information, some novel mutation operators are desigened based on the nucleotide bases encoding method. The convergence of the DNA genetic algorithm with the mutation operators is analyzed in terms of Markov chain model. The results with some typical test functions show the proposed operators can largely increase the comvergence speed of the DNA genetic algorithm, and improve the capability of overcoming the fraudulence. The algorithm is applied to model a hydrogenation reaction, and the results illustrate the effectiveness of the algorithm.(3) Both the novel crossover and mutation operators are adopted in the DNA genetic algorithm. The results of several test functions show that with the help of a proper combination of two kinds of operators, the performance of the DNA genetic algorithm can be improved. This algorithm is applied in the parameter estimation of the heavy oil thermal cracking model, and the results show that a smaller modeling error is reached.(4) A hybrid DNA genetic algorithm is presented for the nonlinear optimization problems with inequality constraints. The hybrid algorithm integrates the global searching ability of DNA genetic algorithm and the local searching ability of SQP. The comparison with typical test functions shows the effectiveness of the proposed hybrid algorithm. The optimal solution for the gasoline-blending scheduling problem gained through the hybrid algorithm shows that the higher profit and the product quality constraints are achieved.(5) A double-chain DNA genetic algorithm (dcDNA-GA) based general regression neural network (GRNN) method is proposed for nonlinear systems. In this method, GRNN optimized by dcDNA-GA is used to the modeling of a delayed coking process. The comparison and simulation results show that the smaller error and AIC indicators are gained.(6) A Chaos DNA genetic algorithm (CDNA-GA) based T-S fuzzy recurrent neural network method for complex nonlinear systems is suggested. In this method, the T-S neural network optimized by the CDNA-GA is used to the modeling of a pH neutralization process. The comparision and simulation results show the feasibility and advantage of this method.(7) A DNA multiobjective genetic algorithm is proposed for multi-objective optimization problems. The results of typical test functions show that the proposed algorithm can converge nearer to the Pareto front, and the spread and the precision of the solution are both improved. This algorithm is used to design a generalized predictive controller for a pH neutralization process, and the simulation results show the higher precision and better performance is obtained.更多还原