Abstract:

Large-scale optimization has become an important research theme in the evolutionary computing field. To take advantage of this progress, sophisticated algorithms have been merged in commercially inexpensive software, like python, to increase the aptitude of large data handling. In this paper, we propose an advanced algorithm, to compare Augmented and Penalty methods for resolving large-scale constrained problems of optimization. We highlight the advantages of an Augmented method in terms of faster convergence, better numerical stability and more robust performance while noting the behavior of Penalty Methods like parameter sensitivity issues, degraded performance near optimal and numerical instabilities. Using numerical experiments, our algorithm is quickly convergent and is more reliable.