Research On DNA Computing Based Genetic Algorithms And Its Applications

DNA computing is a novel research field and still in its infancy. There are many theoretical and practical problems in DNA computing itself. Genetic algorithm (GA) is similar as DNA computing in certain aspects, which is the simulation of the process of biological evolution at the molecular level. Therefore, GA can be utilized as a bridge for DNA computing to solve complex optimization problems. Since the sophisticated knowledge can be expressed as DNA molecular encoding and the information can be detected and processed by simulating DNA operations, accessing and updating the information in the evolution not only adequately employ the pioneering idea of DNA computing, but also can solve the complicated optimization problems existing in automatic control, pattern recognition, decision, machine learning, etc.. With these points in mind, this dissertation focuses on the combination of DNA and GA to solve the modeling and optimization problems of control systems. The main contents are as follows.(1) The Genetic algorithm based on RNA computing is proposed according to RNA operations, DNA sequence selection and mutation model. This method uses A, T, G, C to encode the variables, and combines RNA genetic operations and DNA sequence model to design the crossover and mutation operators of GA. The convergence analysis in terms of Markov model shows that RNA-GA with elitist strategy can converge in probability 1 to the global optimum. Comparisons of RNA-GA with SGA for typical test functions show the advantages and efficiency of the proposed algorithm. The applications of chemical process parameter estimation show the practicability of the proposed algorihm.(2) The DNA double chromosome and SQP based hybrid genetic algorithm is presented for nonlinear programming problems with inequality constraints. In the global exploration phase, the DNA double chromosome structure is used to overcome the "Hamming cliff" problem and keep the diversity of the population. RNA operators are applied to improve the global searching capability and the sequential quadratic programming (SQP) method is implemented to quickly find the local optimum and raise the solution accuracy. The convergence speed analysis and simulation results on typical test functions demonstrate the reliability and efficiency of the proposed algorithm. The applications of gasoline blending recipe optimization show the advantages of the proposed algorithm.(3) The DNA computing based non-dominated sorting genetic algorithm is suggested for multi-objective optimization problems. First, the gene level operators of RNA computing areadopted to enhance the global searching capability of GA. Then, the inconsistent multi-objective fitness functions are converted into a single objective function by Pareto sorting and individual crowding measuring. Finally, external population is introduced to keep the Pareto front individuals and the maintenance algorithm is proposed to maintain the evenness of individual distribution. The convergence analysis and simulation results on typical test problems show the improvement of the proposed algorithm, in the spread of solutions and the convergence near the true Pareto-optimal front.(4) How to select the basis function center the network structure, has not yet given an effective method. The splicing system based genetic algorithm is put forward to optimize the RBF network, which is used to extract valuable process information from input/output data. The novel RBF network training technique includes the network structure into the set of function centers by compromising between the conflicting requirements of reducing prediction error and simultaneously decreasing model complexity. The effectiveness of the proposed method is illustrated through the development of dynamic models for a continuous stirred tank reactor (CSTR) by comparing with different RBF network training methods.(5) Based on the operators of DNA computing, the multi-objective genetic algorithm is proposed to optimize neural networks. Both the structure complexity and the approximation performances of a RBF network are optimized. Once a group of Pareto optimal solutions are derived, the appropriate RBF network can be chosen in terms of the sum of absolute value of the testing error (SAE). Simulations on a CSTR show fairly good fitting ability of the proposed method and the multi-objective GA can obtain better results than single objective GA.(6) DNA-GA algorithms are used for the optimization design of control systems. A stabilizing space based multi-objective GA is proposed for the optimization of PID controllers of the unstable plants of first order plus time delay. For the application of generalized predictive control (GPC) to pH processes, the splicing based GA algorithm is applied to optimize fuzzy neural networks for modeling of a nonlinear pH neutralization process. A fuzzy neuron hybrid control method is proposed for mold level control of the continuous steel casting with big uncertainties and grave nonlinearities. Since there exist the problem of optimized selection of several parameters in the model-free fuzzy neuron hybrid controller, RNA-GA is utilized to determine the controller parameters. Simulation results demonstrate the efficiency of the DNA-GA methods.