Rolling algorithm with multiple runs — a non-linear discrete optimization algorithm

Abstract

Discrete Optimization (DO) problems are often encountered in real world applications. Many researchers believe that DO problems are more difficult to solve than continuous optimization problems, especially when the problems are nonlinear. In addition, although global search heuristics such as the Simulated Annealing (SA) have been successfully applied to many nonlinear DO problems, it is commonly recognized that no single algorithm works well for all problems. Therefore, new alternative algorithms are desired by researchers. In this paper we propose a simple Roiling Algorithm (RA) for solving DO problems. Because RA is easy to implement and takes little resources, we recommend multiple runs of RA for a particular application. We use simulations to show that the RA can be used to solve both discrete and continuous optimization problems. Finally, we observe in our examples that the performance of RA is comparable to that of the SA.