The social network analysis method was adopted to dynamically describe the structural characteristics of industrial linkage networks using input-output table data of 42 industrial sectors in China from 2012 to 2020, and to establish a network dynamics model based on stochastic actors to identify and analyze the evolution mechanism of industrial linkage networks. The results show that there is no significant difference in the holistic indicators of the 42 industrial networks, and the networks as a whole tend to be stable. Chemical products as well as transportation, warehousing and postal services have been at the center, and the secondary industry, especially the manufacturing industry, is at the core of the network. From 2012 to 2015, China's industrial linkage network could be divided into four segments (bidirectional spillover segments Ⅰ and Ⅱ, broker segment, and net beneficiary segment), and in the period of 2017-2020, the broker segment disappeared, and the net beneficiary segment disappeared. In 2020, the broker segment disappeared and the main beneficiary segment appeared, and the industrial linkage effect within the segment became more obvious. Industry-linked activities are concentrated in the manufacturing industries of metal products, chemical products, communication equipment, computer and other electronic equipment, and electrical machinery and equipment, and the service industries of wholesale and retail, transportation, storage and postal services, and information transmission, software and information technology services. The relative importance of endogenous structural factors is 53.11%, which plays an important role in the evolution of network relationships, the negative out-degree effect indicates that industries with more interconnected relationships are difficult to develop new relationships, and the transmission effect indicates that there is a trend of continuous refinement of the industry, and the relative importance of exogenous attribute factors is 46.89%, of which the influence of labor is the most significant, accounting for 20.52%.