Abstract: The emergence of digital twin technology has greatly promoted the development of mechanical condition monitoring, fault diagnosis and other technologies. The construction of twin models in structural digital twins is extremely complex, and a large amount of data is needed as a basis to ensure the accuracy of the twin models. A finite element mesh reduction method based on K-nearest neighbor classification inverse distance weighted interpolation is proposed to address the issues of large grid data, long computation time, and decreased system performance involved in the construction process of complex structured digital twin proxy models. Taking the wing lower wall panel long truss structure as the research object, a finite element mesh model of the long truss is established, and the finite element simulation results are analyzed using this method. The results show that the absolute error of the stress concentration area is 0.7 mm, and the relative error of the maximum principal stress is within 2%, providing a new approach for grid order reduction, data dimensionality reduction, and compression.