Application of Constrain Autoencoder Linear Dimensionality Reduction of Dummy Curve Data

In this paper, the linear dimension reduction of constrain autoencoder is proposed to solve the problem that the dummy curve data cannot be directly related to the discrete design parameters. This paper introduces the basic principle and restriction conditions of the constrain autoencoder, and applies the method to solve the dimensionality reduction problem of multi-type dummy response curve data. The results show that the reconstruction errors of the four types of mechanical response data with linear dimensionality reduction are less than 3%, and the covariance values of all low-dimensional data are close to 0. Therefore, the constrain autoencoder method can achieve linear dimensionality reduction and dimensionality increase reconstruction of dummy curve data, and the obtained low-dimensional data is highly independent.