基于联邦学习和多方安全计算的海铁联运数据安全共享方法研究

Research on Secure Data Sharing Methods for Sea-rail Intermodal Transportation Based on Federated Learning and Multi-Party Secure Computation

我国海铁联运占港口集疏运比例仍然过低,关键原因之一在于铁路对于港口适运货源的动态信息不明、营销组织不力。铁路货运营销部门缺乏在保证港口、铁路、海关三方数据隐私安全的前提下,根据铁路运力动态主动挖掘港口和海关数据中潜在适运货源的技术方法和手段,难以推出适销对路的运输产品和动态营销手段,也难以为海铁联运基础设施的建设提供有效决策依据。构建基于联邦学习和多方安全计算的铁路-港口-海关数据安全共享方法,使用结合同态加密等多方安全计算技术的梯度提升决策树作为模型训练算法,铁路、港口、海关三方地位对等协作,训练出海铁联运潜在货源识别策略;在该策略的正式运行实现过程中,铁路方能够获得路网各流向潜在适运货源的数量级,各方均看不见、带不走其余参与方的任何原始数据。

Sea-rail intermodal transportation in China still accounts for a low percentage of port throughput,and one of the key reasons is the lack of dynamic information regarding suitable cargo sources from ports and ineffective marketing organization by the railway.The railway freight marketing department lacks technological methods of proactively exploring potential suitable cargo sources from port and customs data dynamically according to railway capacity,while ensuring the privacy and security of data among the port,railway,and customs.This hampers the development of suitable transportation products and dynamic marketing strategies.Consequently,it becomes challenging to provide effective decision-making support for the construction of sea-rail intermodal transportation infrastructure.This paper constructed a secure data-sharing method among railways,ports,and customs based on federated learning and multi-party secure computation.The paper utilized gradient boosting decision trees,combined with multi-party secure computation techniques such as homomorphic encryption as the model training algorithm.The collaborative effort among the railway,port,and customs with equal status was used to train a strategy for identifying potential cargo sources of sea-rail intermodal transportation.During the formal implementation of this strategy,the railway can gain insights into the volume of potential suitable cargo sources along various routes within the network,while ensuring that each party does not have visibility or access to the raw data of other participants.