Probabilistic model building Genetic Network Programming using multiple probability vectors

Abstract

As an extension of GA and GP, a new evolutionary algorithm named Genetic Network Programming (GNP) has been proposed. GNP uses the directed graph structure to represent its solutions, which can express the dynamic environment efficiently. The reusable nodes of GNP can construct compact structures, leading to a good performance in complex problems. In addition, a probabilistic model building GNP named GNP with Estimation of Distribution Algorithm (GNP-EDA) has been proposed to improve the evolution efficiency. GNP-EDA outperforms the conventional GNP by constructing a probabilistic model by estimating the probability distribution from the selected elite individuals of the previous generation. In this paper, a probabilistic model building GNP with multiple probability vectors (PMBGNP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">M</sub> ) is proposed. In the proposed algorithm, multiple probability vectors are used in order to escape from premature convergence, and genetic operations like crossover and mutation are carried out to the probability vectors to maintain the diversities of the populations. The proposed algorithm is applied to the controller of autonomous robots and its performance is evaluated.