Adaptive Neural Network-Based Feedforward-Feedback Controller for Nonlinear Dynamic System

Abstract

This study investigates the deployment of adaptive neural network-based control strategies for nonlinear dynamic systems, emphasizing the integration of Echo State Networks (ESNs) into a feedforward-feedback control architecture. Traditional controllers relying on precise mathematical modeling often fail to cope with the complexity of systems exhibiting high nonlinearity, time-varying parameters, and external disturbances. The proposed ESN-based approach harnesses reservoir computing to construct a lightweight, data-driven model capable of accurately capturing system dynamics in real time. The feedforward module provides anticipatory control actions, while the feedback loop compensates for deviations, enabling rapid convergence and robustness against parametric drift. Comparative analysis with conventional PID and LQR controllers reveals superior performance in terms of tracking accuracy, stability, and noise resilience. Preliminary simulations predict reduced steady-state error and improved dynamic response even under uncertain operating conditions. This architecture presents a scalable and efficient alternative for advanced applications in robotics, aerospace, and industrial process control. The findings affirm the viability of ESNs in redefining adaptive control paradigms by combining interpretability, computational efficiency, and real-world adaptability. Reference to this paper should be made as follows:MCE 2025, MCE825. (2025) ‘Adaptive neural network-based feedforward-feedback controller for nonlinear dynamic systems.