A novel Data-Model Hybrid-Driven strategy for voltage control and noise resistance in wind farm

• Data-model hybrid strategy enhances wind-farm voltage control and noise robustness. • Graph-based decoding captures turbine coupling and refines power predictions. • Predictive control with prior knowledge yields feasible active/reactive setpoints. • Neural fusion adaptively combines two strategies to produce optimal commands. This paper studies voltage regulation in wind farms under volatile operating conditions, where wind uncertainty and measurement noise can induce bus-voltage fluctuations and degrade control performance. To address this challenge, we propose a data–model hybrid-driven voltage control strategy that combines data-based forecasting with physics-informed optimization. First, a data-driven encoder–decoder architecture is developed to predict the pre-output active and reactive power of each wind turbine from historical operating measurements. A graph convolutional network is then used in the decoder to capture the electrical coupling among turbines and refine the predicted power outputs. Second, a model predictive control–based model-driven strategy is designed to generate feasible active and reactive power references by incorporating prior system knowledge and operational constraints. Finally, an artificial neural network–based fusion module is trained to adaptively weight and integrate the outputs of the data-driven and model-driven strategies, producing the final optimal control commands. Simulation studies on a wind farm with 32 × 5 MW wind turbines in MATLAB demonstrate substantial improvements in voltage tracking accuracy and fluctuation suppression. Simulation results show that the proposed strategy substantially reduces voltage-control errors across the wind farm and preserves voltage regulation performance under measurement noise, thereby enhancing voltage stability and robustness.