Research On Gene Expression Programming Algorithm And Its Application

Gene Expression Programming (GEP) is a new evolutionary computing model, proposed by Candida Ferreira in 2001, which is a new technology based on gene expressing rules. It has remarkable differences with the traditional GA (Genetic Algorithms) and GP (Genetic Programming) in form of individual expressing and processing and so on. Based on the foundation of GA, GP develops fast and it surpasses the GA in performance. By contrast with GP, GEP has advantage in processing symbols and expressions.This thesis introduces several key technologies of GEP, including the construction of genes and chromosomes in GEP and the design of fitness function and genetic operations. By contrast with GA & GP, GEP has characteristics of simple, linear and compact chromosomes, easy genetic operations and that the expression trees are exclusively the expression of their respective chromosomes. A difficult problem to solve in GEP is analyzed that is the contradiction between algorithm convergence and population diversity and the premature phenomena. In order to solve this problem, an improved GEP algorithm is proposed based on the idea of symbol-weight changing, classification and new individuals generating style. An heuristic accelerating searching strategy and diversity operators are designed during new individuals generation. Several experiments are done using the improved GEP algorithm, which indicates that the improved GEP algorithm is better than the traditional GEP and it sovle the contradiction between algorithm convergence in a good way. The successful rate of the improved GEP algorithm is enhanced by 14.8% and 21.0% respectively on average.In practice, the improved GEP algorithm is applied to function mining and image registration problems. The function mining problem is analyzed. The experiments applying the improved GEP algorithm to function mining indicate that the improved GEP algorithm is efficient. The image registration problem is summarized which introduces the significance of image registration, the transformation type, the registration process based on the control points and so on. The advantage and disadvantage of the current model is analyzed. The experiments applying the improved GEP algorithm to image registration are performed in detail and we obtain good results.A software for the improved GEP algorithm is implemented. The system infrastructure, the interface and the corresponding operations of the software are introduced in the thesis.