To address the real-time and accuracy requirements for aerodynamic data modeling in online trajectory optimization of aircraft, this paper proposes a high-precision intelligent aerodynamic data fitting method based on a Multi-layer Perceptron (MLP). Frequent aerodynamic force calculations in online optimization expose limitations of traditional methods: interpolation suffers from inefficiency due to high-dimensional data processing, while polynomial fitting fails to capture nonlinear transonic aerodynamic characteristics due to order and functional constraints, leading to extrapolation errors and trajectory divergence. By leveraging nonlinear activation functions and a light-weight network structure, the MLP achieves high-precision modeling of lift and drag coefficients. Experimental results demonstrate that the MLP reduces mean squared error (MSE) by 97.7% for the lift coefficient and 91.9% for the drag coefficient compared to fifth-order polynomial fitting. Additionally, the transonic residual peaks of and are reduced to 51.8% and 39.2% of those obtained by polynomial fitting, respectively. For a typical air-to-air missile trajectory optimization problem, simulations using pseudospectral trajectory optimization software show that the MLP method requires only 0.895 s, achieving a 6.6× speedup over interpolation and reducing terminal range error by 80.5% compared to polynomial fitting.