Research On Cultural Algorithms And Their Applications

This paper first gives a deep analysis on characteristics of existing evolutionary algorithms, then uses the concept of social\cultural evolution theory in social science for reference, and finally describes a series of novel algorithms, namely cultural algorithms. Cultural algorithms are dual inheritance systems based on knowledge that consist of a population space and a belief space. The population space consists of individuals.The problem solving experience of individuals selected from the population space is used to generate problem solving knowledge that resides in the belief space. Cultural algorithms utilize some characteristics and knowledge from population in the pending problems for restraining the degenerative phenomena during evolution process, so as to improve the algorithmic efficiency. Cultural algorithms are optimal algorithms in essence, therefore, they can be used in some fields, such as cybernation, pattern recognition, optimal design, meshing learning, network security, etc. There are also some examples of these attempts described in this paper.This paper first researches the status quo of evolutionary algorithms theories, main branches of evolutionary algorithms and the similarities and differences between them. Then based on analyzing the disadvantages of evolutionary algorithms, many new algorithms and strategies are proposed, and can be summarized as follows:Firstly, this paper describes the computational framework of cultural algorithms and designs its population space, belief space and all functions. Most problems can be converted to nonlinear unconstraint/constraint optimization problems in practice. So this paper researches on how to solve these kinds of problems by cultural algorithms and apply standard nonlinear unconstraint/ constraint functions sets to test the performance of cultural algorithms. The computational experiment shows that cultural algorithms can produce substantial performance improvements. The nature of these improvements and the type of knowledge that is most effective in producing them will depend on the structure of the problem. Generally speaking, the best performance is produced using normative/situational knowledge to decide both step size and direction, while in most cases.Secondly, based on evolutionary epistemology proposed by Sir Karl Raimund Popper, a computational framework of knowledge evolution strategies is proposed at first. Then define the new concepts about knowledge evolution strategies and design all functions. The key idea behind knowledge evolution strategies is that both of hypotheses set and knowledge set, which are connected with conjecture and refutation method, evolve into the truth cooperatively, i.e. the best solutions of the problem to be solved. Then a specific implementation of Knowledge Evolution Strategies for nonlinear unconstraint optimization problems is produced. The computational experiment shows that knowledge evolution strategies can produce substantial performance improvements as expressed in terms of high speed for convergence and avoid the premature convergence to some degree.Thirdly, attempt to apply cultural algorithms into cluster analysis problems. After analyzing the disadvantages of the classical clustering algorithm, a series of novel hybrid clustering algorithm based on cultural algorithms are proposed, which uses the merits of the global optimization and higher convergent speed of cultural algorithms and combines with fuzzy c-mean/spherical shell/line and hard c-mean clustering algorithms. It includes hybrid fuzzy c-Mean clustering algorithm based on cultural algorithms, hybrid fuzzy c-spherical shell clustering algorithm based on cultural algorithms, hybrid fuzzy c-line clustering algorithm based on cultural algorithms and hybrid hard c-mean clustering algorithm based on cultural algorithms. Comparing with the classical Fuzzy c-means/spherical shell/line and hard c-mean clustering algorithms, the experimental results show that the algorithms proposed in this paper not only avoid the disadvantages of the classical Fuzzy c-means/spherical shell/line and hard c-mean clustering algorithm, but also have higher precision and higher convergent speed.Finally, pattern synthesis of antenna array is first converted into a simple target problem, which is a nonlinear optimization problem. Then objective functions of pattern synthesis with low side-lobe and with broad nulls are proposed separately, and then cultural algorithms, which are suitable to solve these problems, are produced. The results of computer simulation verify the efficiency of the proposed methods, which can satisfy requirements about broad nulls and low side lobe with high convergence rate.