Genetic algorithm-based method for solving an overdetermined system of equations corrupted by measurement noise

摘要

This paper describes a novel method for prediction of atmospheric temperature profiles. The profiles are described by an overdetermined system of equations corrupted by measurement noise. The goal is to predict the temperatures in four range bins with reasonable accuracy. Solving the equations by a classical linear least squares fit results in predicted temperature error over 30K for small amount of noise in the measurements themselves. Since this result is physically implausible, we developed a geneetic algorithm based mehtod and we incorporated into the fitness function correction terms based on oru experience with the atmospheric temperature profiles. The results improved dramatically. The technique devised is tested on a simulated data set and the resutls are described.