Hybridizing Change Detection Schemes for Dynamic Optimization Problems

Abstract

Detecting the points in time where a change occurs in the landscape can have an important role for a number of evolutionary dynamic optimization techniques presented in the literature. The two common ways for change detection are the population-based scheme and the sensor-based scheme. The former one requires statistical hypothesis testing, which periodically checks whether two consecutive populations are derived from different distributions or not. On the other hand, the latter one utilizes re-evaluation of a set of sensors, throughout the search process. The population-based change detectors may cause false positives and the sensor-based detectors may lack of distinction between changes and noise in fitness functions. In this paper, we propose a hybrid technique to overcome the limitations of the change detection schemes and validate it by using Moving Peaks Benchmark (MPB).