Research On The Structured Gene Expression Programming And Its Applications

Gene Expression Programming (GEP) is a genetic programming algorithm designed by the inspiration from structure and function of genes. It absorbed the genetic algorithms and genetic programming’s advantages, and solved complex problems by simple coding. Recently scholars have done some meaningful exploration and research on gene express program function, particularly in function mining, classification rule discovery, time series forecasting, etc. have made some impact in these areas, while the study of gene expression programming aspects still has many problems to be solved, such as large-scale gene expression population, the big number of iterations, the huge time and space resources needed by the release of the generation of expression trees have become bottleneck of improving gene expression programming algorithm performance.We carefully studied the entire framework of gene expression programming algorithm for the existence of these problems and proposed Structured Gene Expression Programming (SGEP) algorithm. In processing the gene expression, the gene expression is further abstracted to form the genetic structure of the template, in the course of evolution those gene structure template who qualified survive and eventually produce high-quality genome consists of high-quality structure template, making the algorithm convenient, high efficiency, required less space and improve the performance advantages in operation, and get additional high-quality genetic template can provide better domain knowledge for other similar problems, help improve the understanding of the problem dealing, which will apply the successful experience of evolution to other fields.The main work and contribution of this paper is as follows:(1) Proposed structural gene expression programming algorithm. First, by evaluating the genome randomly generated, selected high-quality genes, and then get the abstract quality structural gene expression by removing function symbols, then these high quality structural gene expression were separately to generate new genome, and these new gene expression mutate, inverted string, string interpolation to calculate the fitness value, to elect better quality structural gene expression for solving the problem of evolution. Structural gene expression programming algorithm extract useful structure from gene expression to evolve expression which is better in expressing, abstracting and keeping qualify DNA, solving problem than the gene expression programming algorithm, experimental results show that the structural gene expression programming algorithm can more accurately mining functions, and in earthquake prediction can better accurately predict earthquakes of magnitude than gene expression.(2)Apply the structural gene expression programming algorithm combined with RBF neural network to handle the discovery of complex functions. The optimized RBF neural network based on structural gene expression programming algorithm has more powerful search capabilities than the structural gene expression formation programming, as well as in the evolutionary process without considering the function of specific forms of versatility. By mining complex functions experimental results show that SGEP-RBF algorithm has the ability to dig out the function more accurate and find internal links between the data than the RBF algorithm, gene expression programming algorithm, and the accuracy of SGEP-RBF is better than other two algorithms. And SGEP-RBF algorithm was also applied to the weather forecast, the results show that the SGEP-RBF algorithm can accurately predict the weather information.(3)Apply the structural gene expression programming algorithm to processing text classification problems. First, structural gene expression programming text classification establish VSM feature space vector for each document from training text collection, then the expression of the programming document classifier constructed initialized population by the text categorization which based on the structural gene expression programing, and then use structured gene expression programming algorithm to process groups in order to identify the intrinsic function relationship between the text attribute and text class, and finally use relational model found during training to judge categories operations on new text, and get the final classification result. Through test over text data sets, as well as Sina Weibo, which showed that structural gene expression programming is better for text classification than gene expression programming.