Fuzzy Adaptive GEP Algorithm And Its Application

Gene Expression Programming(GEP)is a new evolutionary algorithm derived from the combination of genetic algorithm and genetic Programming.This algorithm is very flexible and simple in coding and decoding,has strong expressive ability and is easy to carry out genetic operation,and has high efficiency in solving complex problems.Compared with traditional mathematical statistical method,GEP only need to select an appropriate fitness function to evaluate chromosomes,don't need to have deep mathematical foundation,don't even need to understand in detail the search space or other auxiliary information,knowledge,data form can accurately depict complex relationships between data,this is also one of the main reasons for its wide application.Compared with the current popular neural network method,GEP does not need to prepare a large number of experimental data as training data,and the phenomenon of overfitting is rare.In practical application,the final solution of GEP can be converted into a specific calculation formula,which is more conducive to the specific implementation and implementation of the application field.However,as an evolutionary algorithm,it is difficult to avoid GEP algorithm falling into local optimization,and many researchers have provided solutions to this problem.The main reason for this phenomenon is the gradual convergence of the algorithm in the iterative process,which is also accompanied by the loss of population diversity.The probability of genetic manipulation is the key factor affecting convergence and diffusion.Fuzzy control is a kind of intelligent control technology developed from fuzzy mathematics.In order to enhance the ability of GEP to jump out of local optimization,this paper proposed a Multicellular GEP Algorithm Based On Fuzzy Control(MGEP-FC)by combining Multicellular GEP(MC_GEP)with Fuzzy Control method,and adjusted the genetic operation probability of MC_GEP in the process of iteration.In addition,combining MGEP-FC algorithm with wavelet analysis,a precipitation modeling and prediction algorithm(WT_FMC-GEP)based on fuzzy Multicellular gene expression programming and wavelet analysis was proposed to model and predict the real precipitation data with non-linear and non-stationary characteristics.The main work of this paper is as follows:(1)This paper analyzes the problem that the population diversity of GEP is easy to be lost and get into local optimum in the process of iterative optimization,and proposes a feasible solution--adaptive adjustment of genetic operation probability,so that the algorithm can reach the balance between convergence and decentralized optimization in the process of iteration.(2)This paper proposes a strategy of multiple parallel genetic manipulation,which does not use the traditional fixed mode of "crossover before mutation",effectively avoids the large and irreversible destruction of excellent gene sequence by crossover manipulation,and keeps the opportunity for poor individuals to get evolution through crossover manipulation.(3)According to the fuzzy control method and multiple parallel genetic operation strategy,a fuzzy adaptive Multicellular GEP algorithm(MGEP-FC)was proposed.The algorithm adaptively adjusts the probability of genetic manipulation by fuzzy control and optimizes it by multiple parallel genetic manipulation iteration.Experimental results of 15 Benchmark function optimization experiments show that compared with traditional GEP,MC_GEP and other function optimization algorithms,this algorithm has significantly improved the stability,global convergence ability and optimization speed.Moreover,the symbolic regression experiments of 10 Benchmark functions show that MGEP-FC is very suitable for function discovery and data modeling problems with complex structures.(4)Combined with wavelet analysis,MGEP-FC proposed a precipitation modeling prediction algorithm based on fuzzy Multicellular gene expression programming and wavelet analysis(WT_FMC-GEP)for the prediction modeling of precipitation data,and conducted precipitation prediction modeling for the real precipitation data sets of three regions with large differences in longitude,latitude and climate.Experimental results show that the WT_FMC-GEP algorithm not only performs better than BP neural network,support vector regression machine,gene expression programming and other time series prediction algorithms,but also performs better than the Multicellular gene expression programming and wavelet analysis based precipitation prediction algorithm,with better application prospects.