Hybrid Evolutionary Algorithms For Two Classes Of Optimization Problem And Their Applications

Many optimization problems arise in almost every field such as science, business and engineering applications. A global optimal solution to these problems rather than a local optimal solution is desired. Evolutionary algorithms have been shown to be one kind of global and robust methods for solving highly nonlinear and nondifferentiable optimization problems. Furthermore, they are even suitable for problems whose derivatives are not available or whose objective functions can not be expressed in explicit mathematical forms. These problems can not be solved by traditional gradient-based optimization techniques. This dissertation is focused on evolutionary algorithms for unconstrained optimization and constrained optimization problems. The hybrid evolutionary algorithms are applied to solve the parameter optimization problem of nonlinear model and the set point optimization problem for oil production process. The main contributions on these subjects can be summarized as follows.(1) Considering the drawback of randomly aimless search of basic genetic algorithm, a novel adaptive gradient guiding crossover genetic algorithm (AGCGA) is proposed to solve unconstrained optimization problems. It executed crossover operation via choosing special indivi-duals from the set range of the negative gradient of the optimal indivi-dual got from the current population, and made the offspring closer to the optimal solution. It guaranteed the purpose and feasibility of the crossover operation. The convergence of AGCGA is analyzed. The simulations on 12 benchmark functions indicate that AGCGA can greatly improve precision in searching the optimum value.(2) A dynamic hierarchical hybrid particle swarm optimization (DHHPSO) algorithm is proposed for unconstrained optimization problems. In the DHHPSO algorithm, by using parallel PSO algorithm, hierarchical parallel variables are employed for global and local search respectively. Hierarchical ways of parallel variables are dynamically adapted according to the search phases. In the global search, chaotic mechanism is introduced to the algorithm to enhance the global search capability. In the local search, simplex method is used to search local optimization solution. Simulations show that DHHPSO has better optimization performance than other global algorithms.(3) A hybrid genetic algorithm based on clustering good point set crossover is proposed to solve constrained optimization problems. The constrained optimization genetic algorithm (COGA) is improved from two basic aspects of COGA (i.e., genetic algorithm and constraint-handling techniques). The clustering good point set crossover operation can effectively make use of the information carried by the parents and generate representation offspring in order to maintain and increase the diversity of population. In addition, a local search scheme is introduced to enhance the local search capability and speed up the convergence of the proposed algorithm. As for constraint-handling technique, a new individual comparison criterion is proposed, which can adaptively select different individual comparison criterion according to the proportion of feasible solution in current population. The proposed algorithm is tested on 18 well-known benchmark functions and 3 engineering constrained optimization applications, and the empirical evidence show that the algorithm proposed is effective.(4) A hybrid particle swarm optimization based on modified augmented Lagrange function is proposed for solving constrained optimization problems. The basic steps of the proposed hybrid algorithm comprise an outer iteration and an inner iteration. The inner iteration, in which a nonlinear bound constrained minimization sub-problem of the modified augmented Lagrange function, is solved by improved PSO algorithm. The outer iteration is performanced which updates the Lagrange multipliers and penalty parameters using a first-order update scheme. It is proven that the new algorithm can guarantee the convergence towards the optimum of the problem. The proposed algorithm is tested on 18 well-known benchmark functions and 3 engineering constrained optimization applications, and the experimental results show that it is very suitable and steadier than other algorithms from the literature for different formal of constrained optimization problems. (5) Parameters optimization method of nonlinear model is key problem in the area of modeling and control. A hybrid genetic algorithm based on steepest descent method is proposed to determine the structure of RBF neural network and optimize its parameters. Case studies on various time series and different actual data series show that the HGA-RBF neural network modeling approach exhibits much better prediction accuracy compared with some other existing methods.(6) Two constrained optimization evolutionary algorithms are proposed to solve the set point optimization problem for oil production process. One is multi-objective constrained optimization evolutionary algorithm based on dynamical selection and replacement strategy; the other is constrained optimization evolutionary algorithm based on hybrid crossover scheme. Finally, the numerical simulations are made to verify the proposed model and optimization method, using well-known benchmark functions and a set of data obtained by a commercial software Eclipse on a heterogeneous reservoir Synfield.