Research on multi-source heterogeneous data fusion method based on cross-modal transformer

Abstract

There are a large number of heterogeneous data in manufacturing industry, which are characterized by diversity of sources, low quality and complex types, and there are also some problems such as poor data consistency, weak availability and low sharing rate. Therefore, a multi-source heterogeneous information fusion method based on Transformer is proposed in this paper. The method can fuse multi-source heterogeneous data of multiple modes of enterprises, and make use of complementary information of cross-modal heterogeneous data to form the combination and unified expression of data features. The method focuses on the interaction between multi-mode sequences across different time steps and completes the fusion from one mode to another. The empirical analysis shows that the proposed multi-mode fusion method can well fuse different types of heterogeneous data, and a good fusion data set can effectively improve the model classification effect and provide data support for subsequent data analysis and decision-making.