Cartesian Genetic Programming Based Optimization and Prediction

This paper introduces a CGP (Cartesian Genetic Programming) based optimization and prediction techniques. In order to provide a superior search for optimization and a robust model for prediction, a nonlinear and symbolic regression method using CGP is suggested. CGP uses as genotype a linear string of integers that are mapped to a directed graph. Therefore, some evolved modules for regression polynomials in CGP network can be shared and reused among multiple outputs for prediction of neighborhood precipitation. To investigate the effectiveness of the proposed approach, experiments on gait generation for quadruped robots and prediction of heavy precipitation for local area of Korean Peninsular were executed.