A New Prediction-Based Algorithm for Dynamic Multi-objective Optimization Problems

Karkazan K., TOPCUOĞLU H. R., Sahmoud S.

26th International Conference on Applications of Evolutionary Computation, EvoApplications 2023, held as part of EvoStar 2023, Brno, Çek Cumhuriyeti, 12 - 14 Nisan 2023, cilt.13989 LNCS, ss.194-209 identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 13989 LNCS
  • Doi Numarası: 10.1007/978-3-031-30229-9_13
  • Basıldığı Şehir: Brno
  • Basıldığı Ülke: Çek Cumhuriyeti
  • Sayfa Sayıları: ss.194-209
  • Anahtar Kelimeler: change detection, dynamic multi-objective optimization, prediction-based optimization, severity of change
  • Marmara Üniversitesi Adresli: Evet

Özet

The mechanism for reacting to the changes in an environment when detected is the key issue that distinguishes various algorithms proposed for dynamic multi-objective optimization problems (DMOPs). The severity of change is a significant approach to identify the dynamic characteristics of DMOPs. In this paper, a prediction-based strategy based on utilizing the degree of the changes is presented to address environmental changes. In case of a change detection in the given DMOP, the severity of change is evaluated and an appropriate reaction mechanism is followed based on the degree of the observed change. To accelerate the convergence process, the algorithm may respond multiple times for the same change. The performance of our algorithm is evaluated by comparing it with dynamic multi-objective evolutionary algorithms using six benchmarks. The effectiveness of our algorithm is demonstrated in the experimental study where it outperforms other compared algorithms in most of the tested instances considered.