To achieve the scientific management and efficient utilization of light material corrosion data ,the assessment , standardization and classification of corrosion data are carried on ,data structure design is based on the Oracle ,and the nine types of light materials corrosion data are archived and warehoused .A Corrosion data application system with lean structure and high availability is developed ,mainly based on J2EE platform and combined with Struts ,Spring and Hibernate frame-work techniques ,which can provide users with a variety of consulting services via Web to shared materials corrosion test re-sults for integrating a variety of functions such as data storage ,management ,query ,corrosion prediction and protection sys-tem recommend .

Corrosion prediction is a technology of finding the corrosion law based on material corrosion data. Due to corrosion data has the characteristics of high dimensional nonlinearity, randomness and limited sizes, many data modeling methods based on large samples are not applicable. In the process of corrosion prediction, we have to deal with missing data values, outlier detection, feature selection and regression. However, feature selection and regression would be the focus of our research in this paper. This paper adopts a modeling method combining of Grey Relational Analysis and Support Vector Regression, referred to as GRA-SVR, the former is used to select feature and the latter is used for regression. The experimental results show that, GRA-SVR method achieves higher precision than other methods such as BP Neural Network.