Hybrid Techniques for Detecting Changes in Less Detectable Dynamic Multiobjective Optimization Problems

Abstract

Detecting the environmental changes in dynamic optimization problems is an essential phase for a dynamic evolutionary algorithm. By determining the time points of change in the problem, the evolutionary algorithm is capable of adapting and responding to these changes efficiently. It might be more crucial for multiobjective optimization problems, since lack of efficient change detectors may not prevent evolutionary process utilizing invalid nondominated solutions due to the occurrence of changes. The change detection becomes a challenge when dealing with problems that expose less detectable environmental changes, which is a common characteristic of some real-world problems. In this paper, we investigate the performance of sensor-based and population-based change detection schemes on less detectable environmental changes. Additionally, a hybrid scheme is proposed that incorporates sensor based schemes with the population-based ones. We validate the performance of all three schemes on four different less detectable environment problems by considering different characteristics of dynamism, where hybrid techniques significantly outperform the other alternatives.