Research And Improvement Of Evolutionary Programming Algorithms

Evolutionary Programming (EP) is a kind of stochastic optimization algorithm. The goal of EP is to achieve intelligent behavior through simulated evolution. EP was inactive for many years until being reborn in the 1990's in a new form, and has been widely used as a branch of evolutionary computation. EP algorithms are based on an arbitrarily initialized population of search points which evolves towards better and better regions in the search space by means of randomized process of mutation and selection. Similar to the genetic algorithm, premature convergence is also a big problem of evolutionary programming. So avoiding premature convergence and balancing the ability of exploration and exploitation has become one of the important aspects of EP's study.The major works and initiative points in this paper are as follows:1. This paper presents a brief overview of evolutionary computation and evolutionary programming.Evolutionary computation (EC) is a kind of heuristic optimization algorithms. It's the study of computational systems which use ideas and get inspirations from nature evolution and adaptation. EC has been applied with success to many numerical and combinatorial optimization problems, and has been accepted as a robust global optimization algorithm. There are many advantages of EC, such as robustness, effectiveness and simple operation.Evolutionary programming is a branch of evolutionary computation. It was first proposed in 1960's. EP focuses on the process of the evolution, and does not need coding for solutions. Mutation is the unique driving force for individual evolution. EP algorithms are tested by analyzing the simulation result. The theoretical analyzing of EP algorithms should be enriched.2. Several EP algorithms are introduced in detail, and a new algorithm using a mixed mutation strategy is proposed.The classic evolutionary programming (CEP) is the basic algorithm of evolutionary programming. FEP and SPMEP improved CEP by replacing the Gaussian mutation in CEP by Cauchy mutation and Single-Point mutation. MSEP inspired from evolutionary game theory is an algorithm using a mixed mutation strategy. In MSEP each individual chooses one mutation operator for a set of mutation operators according to its strategy parameters.In this paper, we propose an improved EP algorithm named SPCEP. It uses a mixed strategy based on Gaussian mutation and single-point mutation. In SPCEP, each of the two mutation operators will generate one offspring. Then a compare will be conducted between them, and the best one will be chosen. Simulation results show that SPCEP is superior to CEP and SPMEP for some functions.3. A new mutation operator named Bagging is proposed based on the study of search step and the exploration and exploitation capability of EP algorithms. And a new Bagging algorithm is proposed by combing the Bagging mutation and CEP together.The capability of exploration and exploitation is very important in EP algorithms. Sometimes it impacts the convergence speed and the accuracy of algorithms. Generally, adjusting the search step size can regulate this ability. Bagging mutation operator adjusts the step size by two control parameters. Then the step size will become smaller and smaller with the time going on. By this way, the EP algorithm can regulate the ability of exploration and exploitation.In this paper, we present a new EP algorithm by combining the Bagging and Gaussian mutation operator together. We use the algorithm to solve optimization problems, and eight benchmarks are chosen in our experiments. Because of the limitation of CEP, Bagging algorithm has bad results in solving high-dimensional multimodal functions. We modify the algorithm by changing the Gaussian mutation operator into Single-Point mutation operator. This improves the accuracy of the algorithm.