<![CDATA[Adaptive pitch control built on the quasi-ARX neural network (QARXNN) has so far been reported for wind energy conversion systems (WECS) purely in simulation, driven by artificially generated wind sequences and parameterised for utility-scale machines. The present work carries the scheme over to a small rotor of 1 m radius, typical of distributed installations in low-wind classes, and tests it against two field records of wind speed, temperature and pressure acquired with a low-cost Arduino-based weather station at Gili Raja and Mandangin Island, Madura, East Java, Indonesia. Three departures from the published formulation are documented: a sign convention in the drivetrain torque balance is corrected, the model order is lowered to match the limited length of the field records, and the minimum-variance pitch law is re-derived for the reduced-order structure. Two alternative nonlinear cores for the quasi-ARX regression, a radial-basis-function network and an Elman recurrent network, are also set out and weighed against the multilayer-perceptron core adopted here. In closed-loop simulation driven by the measured wind, the QARXNN minimum-variance controller lowers rotor-speed tracking RMSE by 9.1% and 5.2% against a carefully tuned fixed-gain PI baseline on the two records, with the nonlinear branch of the model active for roughly a quarter of the samples. Feeding the raw, coarsely quantised anemometer signal into the loop shrinks this margin considerably, an outcome traced to the mechanical bandwidth of the rotor rather than to either control law. The study is therefore offered as a proof-of-concept field validation and not as a long-term resource assessment.]]>
Copyrights © 2026
