The strong nonlinearity, tight parameter coupling, and noise interference inherent in fixed-wing aircraft dynamic models seriously degrade the accuracy of control law inverse modeling. To address this issue, a reverse reconstruction method for flight control laws based on Stacking ensemble learning was proposed. First, Savitzky–Golay (S-G) filtering combined with outlier removal was applied to preprocess raw Flight Data Recorder (FDR) data, by which the data quality was effectively enhanced. Second, a Stacking-based reverse reconstruction architecture was constructed, in which Random Forest (RF), Long Short-Term Memory (LSTM), and Transformer were employed as base learners, and XGBoost was adopted as the meta-learner. Finally, validation was conducted using real flight data, with the deflection angles of the left aileron, left elevator, and rudder selected as the inverse modeling targets. Experimental results demonstrate that the proposed model outperforms both single and conventional models in terms of prediction accuracy and robustness across all control surfaces, providing technical support for fixed-wing aircraft design, fault diagnosis, and safety analysis.