Abstract: China's railway network contains many railway sections of construction periods and operation periods, which produce a large number of business data. However, the traditional single-node big data storage method has limitations such as slow access speed and low timeliness, which cannot effectively alleviate the pressure of data storage. Based on the idea of data hierarchical storage, this paper designed a distributed hierarchical storage architecture of big data, comprehensively considered the business attributes of railway big data in the construction period and the inherent attributes of storage database, and established a set of data value evaluation system, calculated the value of each data table under different evaluation dimensions based on expert evaluation method, and determined the corresponding storage level of each data table through K-means clustering algorithm. The paper took the railway big data in a construction period as the experimental sample for verification. The experimental results show that the value evaluation system proposed in this paper can effectively judge the storage level of railway big data in the construction period, and implement the hierarchical storage of railway big data oriented to the construction period.