Efficient and accurate acquisition of aerodynamic data for aircraft is a critical technological bottleneck that restricts the development of new aircraft. To address the challenges of high costs and low efficiency faced by traditional methods under limited resource constraints, this study establishes an intelligent data generation framework based on multi-fidelity models. It proposes an optimal correlated hybrid sampling strategy driven by low-fidelity model responses, enabling efficient exploration of high-dimensional aerodynamic parameter space and rapid model convergence. In the validation phase, low-fidelity viscous correction models, multi-fidelity Gaussian process models, and multi-fidelity neural network models, combined with the hybrid sampling strategy, are employed to predict aerodynamic coefficients for an X-51A-like aircraft. Experimental results show that, with only 5% high-fidelity samples, the proposed multi-fidelity modeling framework improves the average prediction accuracy by over 94% compared to a single high-fidelity model. Through efficient sampling strategies and intelligent data fusion methods, this framework constructs a high-confidence surrogate model system, providing reliable data support for the development of new aircraft and significantly shortening the development cycle.