TWO ALGORITHMS FOR GLOBAL OPTIMIZATION OF GENERAL NLP PROBLEMS

After a brief overview of the methods from the literature, two new algorithms (zooming and domain elimination) for global optimization of general NLP problems are introduced. Operations analysis and stopping criteria for the methods are discussed. Numerical evaluation of the methods is carried out using a set of mathematical programming test problems. Performance of the methods is compared with the Controlled Random Search (CRS) and the Simulated Annealing (SA) methods. The methods are superior to SA for the test problems, as they are more robust, efficient and accurate. The CRS is more efficient than the new methods; however, it is applicable to unconstrained problems only. Therefore, it is concluded that the new methods are useful for engineering optimization applications