The Convergence Mechanism and Strategies for Non-Elitist Genetic Programming

Abstract

Genetic programming is an evolutionary algorithm that proposed to solve the automatic computer program design problem by J.R.Koza in the 1990s. It has good universality and intelligence, and has been widely applied in the field of computer engineering. But genetic programming is essentially a stochastic optimization algorithm, lack theoretic basis on the convergence of algorithm, which limit the scope of its application in some extent. The convergence mechanism of non-elitist genetic programming was studied in this paper. A recursive estimation of the probability of population contains satisfactory solution with the evolution algebra was established by the analysis of operators characteristic parameters, then a sufficient condition of population converge in probability was derived from this estimation, and thereby some operational convergence strategies for many common evolution modes were provided.