DEPENDENCE OF NSGA-II HYPERPARAMETERS ON MODEL PARAMETER VARIATIONS IN LQR CONTROLLER TUNING

WoS Article

en

This paper investigates the relationship between the physical properties of a controlled dynamic system and the optimal configuration of hyperparameters in the multi-objective evolutionary algorithm NSGA-II (Non-dominated Sorting Genetic Algorithm II). The aim is to determine whether, and to what extent, the optimal hyperparameter settings of NSGA-II change in response to physical modifications of the system. The test model is a rotary inverted pendulum, whose physical characteristics are altered by shifting the position of its center of mass. For each tested center-of-mass position, NSGA-II generates a Pareto front of optimal settings of a linear quadratic regulator with respect to maximum overshoot and settling time of the controlled system. The search for optimal hyperparameters for each configuration is performed using Bayesian optimization. Experimental results show that some hyperparameters exhibit weak dependence on the physical changes in the model, while others remain stable. These findings may serve as a basis for establishing general guidelines for effective tuning of evolutionary algorithms depending on the characteristics of the controlled dynamic system.

hyperparameters, evolutionary algorithm, Non-dominated Sorting Genetic Algorithm II, NSGA-II, Bayesian optimization, linear quadratic regulator, LQR, Furuta pendulum

2026-06-01

MM Science Journal

2026

June

9262–9268

7