DNA Computing Based Genetic Algorithm

This chapter first reviews the research status of genetic algorithm theory, encoding problem, constrained optimization, and multi-objective optimization. Secondly, it briefly introduces the biological basis, the problems of DNA biocomputing, and the significance of the combination of genetic algorithm and DNA computing. Finally, the main work and organizational structure of this book are introduced.

Based on the biological RNA operations, DNA sequence selection, and mutation model, a RNA genetic algorithm (RNA-GA) algorithm is described in detail in this chapter. RNA molecules A, T, U, and C are utilized to encode the chromosome, and RNA molecular operations and DNA mutation model are combined to improve the crossover and mutation operators of SGA. The convergence of RNA-GA is analyzed using the Markov chain model. Five benchmark functions are applied to demonstrate the application process of the RNA-GA algorithm, and compare with SGA to effectively show the results by alleviating the premature convergence and improving the exploitation capacity of SGA.

By utilizing the global exploration of GA and local exploitation characteristics of sequential quadratic programming (SQP), a hybrid genetic algorithm (HGA) is proposed in this chapter for the highly nonlinear constrained functions. Thereafter, the theoretical analysis for the convergence of the HGA is then made. In the global exploration phase, the Hamming cliff problem is solved by DNA double-helix structure, and DNA computing inspired operators are introduced to improve the global searching capability of GA.

In this chapter, DNA computing based non-dominated sorting genetic algorithm is described for solving the multi-objective optimization problems. First, the inconsistent multi-objective functions are converted into Pareto rank value and density information of solution distribution. Then, the archive is introduced to keep the Pareto front individuals by Pareto sorting, and the maintaining scheme is executed to maintain the evenness of individual distribution in terms of individual crowding measuring.

Because of the complex nonlinear characteristics of chemical processes, traditional numerical optimization algorithms generally cannot be used to solve the modeling and optimization problems. In this chapter, the estimation of model parameters for heavy oil thermal cracking is firstly solved by RNA-GA. Then, we use DNA-DHGA to solve the recipe optimization problem of gasoline blending with heavy nonlinear inequality constraints. DNA computing based GAs are efficient in solving the optimization problems in chemical processes.

Radial basis function (RBF) neural network is efficient to model nonlinear systems with its simpler network structure and faster learning capability. The temperature and pressure modeling of the coke furnace in an industrial coke equipment is not very easy due to disturbances, nonlinearity, and switches of coke towers. To construct the temperature and pressure models in a coke furnace, RBF neural network is utilized to improve the modeling precision. Moreover, the shortcoming of RBF neural network, such as over-fitting is overcome.

Fuzzy neural networks are quite useful for nonlinear system identification with only input/output data information available. A fuzzy neural network and its improved framework are proposed and the improved genetic algorithms are designed for the structure and parameter optimization to catch the unknown plant dynamics. The hybrid encoding/decoding, neighborhood search operator and maintaining operator are presented to optimize the structure of the input layer, fuzzy rule layer and the parameters of the membership functions together. The liquid level and oxygen content modeling problems in the industrial coke furnace described in Ch. 6 are utilized to compare the performance of several methods. Simulation results show that GA optimized fuzzy neural network is superior in modeling precision and generalization capability. Fuzzy neural networks are quite useful for nonlinear system identification with only input/output data information available. A fuzzy neural network and its improved framework are proposed and the improved genetic algorithms are designed for the structure and parameter optimization to catch the unknown plant dynamics. The hybrid encoding/decoding, neighborhood search operator and maintaining operator are presented to optimize the structure of the input layer, fuzzy rule layer and the parameters of the membership functions together. The liquid level and oxygen content modeling problems in the industrial coke furnace described in Ch. 6 are utilized to compare the performance of several methods. Simulation results show that GA optimized fuzzy neural network is superior in modeling precision and generalization capability.

Distributed parameter systems (DPSs) are difficult to model due to their nonlinearity and infinite-dimension characteristics. This chapter adopts principal component analysis (PCA) to derive a hybrid modeling strategy for modeling such systems. The strategy consists of a decoupled linear autoregressive exogenous (ARX) model and a nonlinear Radial Basis Function (RBF) neural network model.

In this chapter, GA is used to optimize the controller design. First, a new PID controller is designed by using a non-minimal state-space model through predictive function control. The weighting matrix in the predictive function controller is optimized through GA so as to achieve a relatively desired closed-loop control performance. Secondly, a fuzzy neuron non-model controller is designed for a continuous casting process with strong nonlinearity and severe uncertainty, and its parameters are optimized through RNA-GA. Finally, a MOGA based on parameter stabilization space of the PID controller is used to control the first-order lag unstable process. The simulation results confirm the effectiveness of GA and its improved format in the optimization of the control system design problem.

With the development of intelligent algorithms, the learning-based algorithm has been considered as viable solutions to various optimization and control problems. GA can also be efficient to optimize the new emerging intelligent algorithm. Here, an adaptive fuzzy energy management control strategy (EMS) based on Q-Learning algorithm is presented for the real-time power split between the fuel cell and supercapacitor in the hybrid electric vehicle (HEV) in order to adapt the dynamic driving pattern and decrease the fuel consumption. Different from the driving pattern recognition based method, Q-Learning controller observes the driving states, takes actions, and obtains the effects of these actions. By processing the accumulated experience, the Q-Learning controller progressively learns an appropriate fuzzy EMS output tuning policy that associates suitable actions to the different driving patterns. The environment adaptation capability of fuzzy EMS is then improved needless of driving pattern recognition. To enhance the learning capability and decrease the effect on the initial values of Q-table, GA can also be utilized to optimize the initial values of Q-Learning based fuzzy energy management.