Data-model fusion-based dual-RBFNN adaptive backstepping control for constrained power tracking in PMSG wind energy conversion systems

Power-operating-point tracking in permanent-magnet synchronous generator (PMSG) wind energy conversion systems is challenged by coupled mechanical/aerodynamic and electromagnetic uncertainties together with converter current and voltage limits. This study develops a data-model fusion-based dual radial-basis-function neural network (RBFNN) adaptive backstepping controller for below-rated operation. A dimensional turbine-PMSG model supplies wind-dependent references, while two channel-specific RBFNNs estimate lumped residual effects in the mechanical and q-axis voltage dynamics. The design combines constraint-aware adaptation, current and voltage limiters, and offline Harris hawks optimization of five gains. A complete closed-loop boundedness analysis gives explicit ultimate bounds and conditional damping margins for the adopted model. Five controller configurations (M0-M4) were compared using three optimizer seeds, five held-out simulation seeds, and nine parameter-drift and measurement-noise conditions. Under the main condition, M3 achieved a power root-mean-square error of 382.320 ± 28.187 kW, representing reductions of 28.07%, 16.13%, 18.20%, and 19.02% relative to M0, M1, M2, and M4, respectively. M3 also attained the lowest power error in all 45 seed-condition cells. The observer-assisted M4 yielded a slightly lower speed error than M3 (0.023193 ± 0.004554 versus 0.024266 ± 0.004714 p.u.). A source-derived adaptive sliding-mode comparator further supports the power-tracking benefit under matched implementation conditions. Here, ± denotes the sample standard deviation across five held-out seeds. These repeated simulations show that the proposed controller provides a consistent power-tracking advantage and balanced constrained performance within the modeled operating envelope.