A Hybrid Approach for Solving Dynamic Bi-level Optimization Problems

Abstract

Several real-life decision scenarios are hierarchical, which are commonly modeled as bi-level optimization problems (BOPs). As other decision scenarios, these problems can be dynamic, that is, some elements of their mathematical model can change over time. This kind of uncertainty imposes an extra level of complexity on the model, since the algorithm needs to find the best bi-level solution over time. Despite the importance of studying these problems, the literature reflects just a few works on dynamic bi-level optimization problems (DBOPs). In this context, this work addresses the solution of DBOPs from the viewpoint of metaheuristic methods. Our hypothesis is that, by hybridizing successful solving approaches from both bi-level and dynamic optimization fields, an effective method for DBOPs can be obtained. In this regard, we propose a hybrid method that combines a coevolutionary approach and a self-adaptive, multipopulation algorithm. Experimental results assert our hypothesis, specially for certain information exchange mechanisms.