Railway Data Security Governance System And Privacy Computing Technology Research

The development of emerging technologies such as cloud computing,big data,artificial intelligence and the Internet of Things has driven the digital transformation of various industries built on top of data,but the convergence of massive data is facing serious security risks while bringing great value.At present,data security governance is rebuilding from chaos to order,and from personal data protection to industry data security in depth.The railroad industry has a huge amount of heterogeneous and heterogeneous data,and data applications such as data pooling,sharing and joint reasoning of railroad business departments will bring higher level of security risks.The traditional "systembased" network security protection system can no longer meet the dual needs of railroad data flow and data security sharing,and must break the traditional concept and build"physical network-based,data-centric,data security technology-supported" on its basis."of the new all-round fine-grained data security protection system.In this context,this paper takes railroad data security governance as the research object,takes the identification of sensitive attributes of railroad structured data and the security protection of data in the use and sharing stages as the entry point,and adopts correlation analysis,differential privacy,federal learning,deep learning and other related theories and technologies to study the railroad data security governance system architecture and the application of privacy computing technology.The main research contents are as follows:(1)For the problem of identifying and grading sensitive attributes of railroad structured datasets,this paper introduces clustering analysis and association rule algorithm to realize the automatic identification and grading of sensitive attributes.The sensitivity measure of data table attributes is carried out by information entropy and maximum discrete entropy,and the association degree between attributes is mined using association analysis to achieve railroad sensitive attribute identification and grading.(2)To address the privacy leakage problem when using machine learning models for data analysis and mining in the railway data usage phase,this paper introduces differential privacy techniques to perturb the model stochastic gradient optimization to protect data privacy.The railroad image classification tasks are performed on Res Net18 and Shallow Network models and their corresponding variants,respectively,and empirically analyzed by membership inference attacks.The experimental results show that the differential privacy machine learning approach can balance the utility and privacy of the railroad business model to a certain extent and provide privacy protection for the use of railroad data.(3)To address the problems of insufficient training data samples,data noncirculation and privacy protection in the railroad data sharing phase,this paper introduces federal learning techniques to achieve secure collaboration among multiple parties of data holders.The federated average aggregation is used to perform the railway image classification task under the Res Net18 model with the added attention mechanism.The experimental results show that the federal learning technique can provide a solution to the problem of data non-circulation faced when sharing railway data.