Rule-centred genetic programming (RCGP): an imperialist competitive approach

摘要

Automatic programming is one of the challenging fields of AI to generate solutions for high-level programming problems. There are variant methodologies attempting to introduce an efficient technique which address problems of this domain. In this paper, a novel Rule-Centred Genetic Programming (RCGP) is proposed. RCGP benefits from a series of evolutionary rules to help the algorithm choose intelligent alterations in the chromosome of individuals during the evolution yet preserves its stochastic evolutionary nature. Further, a modified search strategy based on Imperialist Competitive Algorithm (ICA) is employed in RCGP that shows to be significantly effective to deal with various problems which differ in degree of complexity. The proposed method features competitive convergence both in the case of speed and accuracy as well as a simpler mechanism than the several existing GP methods. RCGP is tested on nine benchmark problems which are synthesis and real world. The obtained results indicate that RCGP outperforms recent GP methods and is capable of hybridizing with other types of evolutionary algorithms. The method shows to be competent enough to enhance the quality of automatic programming solutions in both aspects of accuracy and efficiency compared to existing methods.