Mining Projection Transformation Based on Gene Expression Programming of Multi-Variable Niches

Abstract

Map projection transformation is a basic operation for topographic and spatial data transformation in geographic information system. Existing methods need projection type and corresponding parameters, and manually select regression model. The transformation formulas are complex with operators based on cartology. This paper applies gene expression programming technique to projection transformation. The contributions include: (1)Formalizing the concepts of projection gene and generation gap, etc.; (2)Designing the fitness function with penalty; (3)Proposing a novel method of projection transformation-GEP based on multi-variable niches(MVN-GEP); The method automatically evolves the constants and constructs the easy formulas; proposing the algorithms of partitioning multi-variable niches(PMVN) and replacing individuals(RI); (4)Experiments show that new method is effective and the output formulas are easy. The average top fitness of geodetic abscissa is 97.1324 and that of geodetic ordinate is 97.7351; The average generation of geodetic abscissa is 238 and that of geodetic ordinate is 216.