The Research Of Using Soft Computing Methods For Intelligent Optimization

Intelligent optimization methods are especially effective to solve problems that are difficult to solve by traditional methods. In this dissertation, the research of intelligent optimization is based on integration soft computing (SC). The involved works are presented as following:1. By combining genetic algorithm (GA) and heuristic methods, a hybrid approach is proposed to solve and optimize a general schedule problem of constrained resources and shortest time. The approach is characterized by its tractability and acceptable computation and is especially effective to solve large problem instances.2. A classification method based on GA integrating decision function is provided. The method solves such problems of classification as selection of appropriate function representation and parameters, which are critical but not easily determined by traditional methods.3. A dynamic fitness based GA is put forward to solve 0-1 integer programming problem. The approach finds a good balance between search space and efficiency. Its performance is exhibited by solving a quadratic 0-1 problem and the influence of its parameters is also analyzed by comparison.4. To avoid the complexity and limitation of BP-MLP network, we refer to the functional link artificial neural network (FLANN) and conduct research of control simulation of a nonlinear CSTR model. The simulation result shows its performance and the influential result of parameters given by comparison is useful.5. By analyzing relation and equivalency between radial basis function network (RBF) and fuzzy logic (FL), a new method of designing fuzzy RBF network is provided, which synthesizes the advantages of fuzzy reasoning and fast learning speed of local network. Its effectiveness is validated by the application results of function approximation and control simulation.6. Based on competitive learning, a supervised classification method is presented. Compared with traditional ones, the method relies less on experiential knowledge and determines parameters more easily. The method also finds good initial parameters that make computation fast to obtain correct classification result.To sum up, some novel use and combination of SC applications are provided in this dissertation. We expect that these research works will improve and expand problem-solving capability of intelligent optimization to a large spectrum.