Subspace clustering method for high-dimensional sparse sensor data

The invention belongs to the technical field of data processing, and relates to a subspace clustering method for high-dimensional sparse sensor data. The method comprises the following steps: preprocessing initial sensor data, including data cleaning and data interpolation, and performing high-information-content feature screening and feature subset division; aiming at the characteristics of high information content, learning by using a characteristic noise reduction variational auto-encoder to obtain hidden Gaussian variables; for a plurality of feature subsets, different subspace feature confidence coefficients are dynamically learned, specific distribution of each subset is learned on the basis of considering a topological structure of data features, and then specific structure distribution with topological information is obtained; and a mutual supervision collaborative deep clustering strategy is used to carry out full fusion between the structural characteristics and the characteristic information and mu