Learning-Assisted Model Reference Adaptive Control for Uncertain First-Order Systems: A Five-Scenario Comparative Validation Study

Abstract

In this paper, a detailed computational analysis of a Learning-Assisted Model Reference Adaptive Controller (LA-MRAC) for uncertain first-order dynamic systems is presented. The proposed controller is the classical MRAC structure with an online learning compensation term u_ML(t) to enhance the tracking performance under uncertainty, external disturbances, measurement noise and abrupt plant changes.Five increasingly difficult scenarios are developed to avoid depending on a single experiment: nominal operating, time varying parameters, external disturbances, measurement noise, and abrupt parameter changeThe proposed LA-MRAC is compared to the PID controller and classical MRAC and NN-Only learning controller using a number of quantitative measures such as RMSE, MAE, IAE, ISE, ITSE, control effort, dynamic response indices, disturbance indices, and computational execution time. The revised presentation is organized in a journal-form experimental form: before introducing each table and figure, and after using them, they are interpreted. The results show that LA-MRAC consistently achieves the lowest tracking error across all scenarios. Average RMSE is reduced by 4.79% compared with classical MRAC, 37.78% compared with PID, and 62.62% compared with NN-Only control. The improvement is obtained with control effort close to classical MRAC, while the computational time increases due to the online learning component. These results support the conclusion that the proposed hybrid adaptive-learning structure provides an effective accuracy-robustness trade-off for uncertain computer-controlled dynamic systems.