Abstract: The railway data service platform has realize centralized aggregation and asset management of railway data. To fully mine the value of railway data and gain beneficial results of comprehensive utilization of railway data asset, it is required to integrate the multi-source data from different business application systems. To cope with the problems resulting from the complexity and diversity of multi-source data and improve the results of multi-source data fusion, it is indispensable to build a unified data fusion architecture. This paper brings froward the multi-source data fusion architecture corresponding to the function modules of the railway data service platform, describeds the input/output items, data processing tasks, related techniques and features of the three levels of data fusion, explores and compares the characteristics and applicable scopes of several data fusion algorithms, and makes a brief disscussion on data storage and sharing of data fusion results, providing reference for the research and development of data fusion based on the railway data service platform.