The Effect of Objective Function Choice in Hyper-Heuristic LQR Tuning of a 3-DOF Hover
2026 10th International Artificial Intelligence and Data Processing Symposium (IDAP), Malatya, Türkiye, 6 - 07 Eylül 2026, ss.1-6, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/idap71869.2026.11727968
- Basıldığı Şehir: Malatya
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.1-6
- İnönü Üniversitesi Adresli: Evet
Özet
In this study, the tuning of the Q and R matrices of the LQR controller for the Quanser 3-DOF Hover system is treated as an optimization problem and solved automatically with a selection hyper-heuristic guided by reinforce-ment learning. The main research question is how objective-function choice affects controller performance. Five objective functions (ISE, IAE, ITAE, ITSE, and the proposed composite criterion COMP) and two selection mechanisms (Choice Function and Simple Random) were compared under a fixed Great Deluge acceptance criterion, across ten configurations, each over 15 independent runs. The results show that the main factor determining controller quality is the choice of objective function. The absolute error based IAE and ITAE produced the lowest tracking error on the yaw and pitch axes, with an RMSE about 40% lower than the squared metrics on the pitch axis. The gap between ITAE and ITSE indicates that the decisive fac-tor is not time-weighting but whether the error is treated in absolute or squared form. The selection mechanism produced a significant difference in only one of the five objective functions. Learning-based selection gained no clear advantage over random selection, owing to the similar heuristics and a single dominant basin of attraction. For square-wave tracking prob-lems where oscillation is not critical, traditional absolute-error metrics are recommended.