High-dimensional data differential privacy publishing method adopting principal component analysis

The invention discloses a high-dimensional data differential privacy publishing method adopting principal component analysis, which comprises the following steps of: 1, carrying out dimension reduction processing on original high-dimensional data through a PCA (Principal Component Analysis) method to obtain a projection matrix Z; 2, dividing the sensitive attributes into c clusters through SOM neural network clustering, calculating the sensitivities of the c clusters, and setting the sensitive levels of the clusters according to different sensitivities; and step 3, sequentially adding a Laplace mechanism according to the sensitivity level, and adding noise to different types of attributes. According to the method, dimension reduction is carried out on high-dimensional data through a PCA algorithm, then attributes are divided into different classes through SOM neural network clustering, corresponding noise is added to the different classes of attributes through a Laplace mechanism according to sensitivity, noi