Training of the Dynamic Systems Control: A Neural Network or a Learning Algorithm

Abstract

The dynamic system control problem under conditions of a priori uncertainty regarding the parameters of the controlled object is considered. The properties of controllers typically used in control systems are studied. Among them are a neural network, a proportional–integral–derivative (PID) controller, and a learning algorithm. The sign-changing input signal is considered in dynamic systems using the minimum time criterion. A dynamic system is represented by the first-order differential equations system, which allows using the state space method in the analysis. A feature of the research is the study of the quality of the system tuning under conditions of parametric uncertainty and the presence of homogeneous non-Gaussian noise in the phase coordinate measurement channels. The system’s reaction results for the studied approaches for the proposed mathematical model are compared. The learning algorithm showed an improvement over conventional methods by at least 40% in the evaluated indicators, in which the influence of interference is leveled by introducing a unique function of the “hysteresis” type. The modeling results are given in support of the conclusions made.

Received: 17 February 2025 | Revised: 26 May 2025 | Accepted: 26 June 2025

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available at https://drive.google.com/drive/u/0/folders/1gqzZ6Tla1RDaJAl86rp7yHHbs6Ggj_wp.

Author Contribution Statement

Dmytro Kucherov: Conceptualization, Methodology, Formal analysis, Investigation, Supervision, Project administration, Funding acquisition. Natalia Khalimon: Resources, Data curation, Writing - original draft, Writing - review & editing, Visualization. Ihnat Myroshnychenko: Software, Validation. Valerii Tkachenko: Validation.