Research And Application Of Improved Methods On Genetic Programming

Genetic Programming(GP)is a typical intelligent optimization algorithm,which can automatically generate computer programs and other complex structures to solve problems,so as to solve various optimization problems.GP simulates the evolution process in the population,and uses genetic operators such as selection,crossover,and mutation to optimize the individuals in the population,and finds the optimal solution based on the fitness.However,GP suffers from premature convergence and code bloat,which have a significant impact on the quality of the solution.Therefore,in view of the above problems of GP,the following research work is carried out in this paper:(1)The premature convergence problem will cause the algorithm to fall into a local optimum,reduce the search performance and affect the optimization effect of the solution.For premature convergence,this paper improves the selection mechanism of the algorithm from the perspective of diversity,and proposes a Genetic Programming Algorithm Based on Cluster Tournament Selection and Parent Matching(CMGP).CMGP divides the population into multiple subpopulations through clustering.By automatically adjusting the selection pressure,the algorithm effectively maintains the diversity of the population and improves the search ability of the algorithm.For the problem of the single fitness evaluation,CMGP improves the Binary String Fitness Characterization(BSFC),which further clarifies the behavior within the individual.In addition,CMGP utilizes BSFC to achieve targeted crossover operations.Through parent matching,the algorithm achieves a better balance between exploration and development from the perspective of pairwise diversity.CMGP uses different benchmark problems to conduct multiple comparative experiments,and the results show that the algorithm can effectively improve the problem of low GP population diversity.By maintaining population diversity and avoiding premature convergence,the optimization ability and convergence speed of the algorithm have been greatly improved.(2)Code bloat is an inevitable result of the evolution of GP,which causes the algorithm to consume a lot of computing resources and reduces the interpretability of the solution.Aiming at the bloat problem,this paper proposes a Genetic Programming Bloat Control Algorithm Based on Clustering(CBGP)based on CMGP.CBGP proposes a new clustering method on the basis of maintaining population diversity.The new clustering method helps avoid further evolution of redundant nodes.In order to prevent abnormal individuals from affecting the performance of GP optimization,CBGP proposes Fitness Library Replace(FLR).Through the dynamically updated adaptive library,the algorithm optimizes the individual structure of the population.In addition,for destructive crossover,CBGP improves the matching method of CMGP,reducing the possibility of producing destructive offspring.By conducting multiple comparative experiments on benchmark problems,the results show that CBGP can effectively alleviate the bloat problem and reduce the consumption of computing resources while maintaining the optimization performance and convergence speed.To sum up,for the problems of premature convergence and code bloat,this paper proposes different GP improved algorithms based on population diversity,and verifies its optimal performance.By maintaining the diversity of the population,the algorithm improves the possibility of producing good offspring,while different individual selection strategies effectively avoid invalid crossover and optimize the individual structure.In addition,the good performance in classification experiments shows that the proposed algorithms have good practical value.