This paper presents a neural control strategy for output-voltage tracking in a Buck converter. The proposed architecture combines low computational complexity and implementation simplicity with online adaptability to system variations. Two neural networks are employed: a primary network performs real-time control, while a secondary adaptive network is updated online when significant tracking deviations are detected. Once retraining is completed, the updated network parameters replace those of the active controller without interrupting the real-time control routine. Due to its limited computational burden, the complete control scheme is implemented on a low-cost microcontroller, demonstrating its feasibility for embedded power-electronics applications. Experimental results confirm accurate voltage regulation, fast disturbance rejection, and adaptation to substantial parameter variations, including transitions between continuous and discontinuous conduction modes.
Embedded neural control: An online learning strategy for a buck converter / Corti F., Intravaia M., Bindi M., Casaucao I., Lozito G.M., Trivino A., Reatti A.. - In: RESULTS IN ENGINEERING. - ISSN 2590-1230. - ELETTRONICO. - 32:(2026), pp. 113730.0-113730.0. [10.1016/j.rineng.2026.113730]
Embedded neural control: An online learning strategy for a buck converter
Corti F.;Intravaia M.;Bindi M.;Lozito G. M.;Reatti A.
2026
Abstract
This paper presents a neural control strategy for output-voltage tracking in a Buck converter. The proposed architecture combines low computational complexity and implementation simplicity with online adaptability to system variations. Two neural networks are employed: a primary network performs real-time control, while a secondary adaptive network is updated online when significant tracking deviations are detected. Once retraining is completed, the updated network parameters replace those of the active controller without interrupting the real-time control routine. Due to its limited computational burden, the complete control scheme is implemented on a low-cost microcontroller, demonstrating its feasibility for embedded power-electronics applications. Experimental results confirm accurate voltage regulation, fast disturbance rejection, and adaptation to substantial parameter variations, including transitions between continuous and discontinuous conduction modes.I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.



