Abstract: The aerodynamic force measurement technology during the acceleration phase of the motion model is one of the key technical challenges in the maglev flight wind tunnel. The aerodynamic force measurement balance is subjected to various strong interferences such as inertial forces in the acceleration section, which masks the measured aerodynamic force values. In order to restore the aerodynamic force signal under the harsh conditions of strong interference and low signal-to-noise ratio, an aerodynamic force interference stripping algorithm based on deep learning was proposed. Firstly, the short-time Fourier transform was performed on the balance signal to determine the spectral characteristics of each interference, which was used as the input of the algorithm model. Then, an "encoder-decoder" architecture model was constructed to extract features from the complex signals measured by the balance and accurately reconstruct the desired aerodynamic force signal. After a comprehensive evaluation on the test set, this algorithm performed excellently in aerodynamic force interference stripping. The reconstruction accuracy of the aerodynamic force signals in the three components of drag, lift, and pitching moment reached approximately 93%. It provides strong support for the aerodynamic force measurement in the future test environment of the maglev flight wind tunnel.