The SI-F method allows diagnosis of structure from data that will be acquired after set up. Presented method detects structural failure from statistical method that judges the equivalence of two data groups. One of these is called standard data group that will be acquired when structure is normal. Another is called diagnosis data group that will be acquired when structure is operated. However, there are many conditions in actual structure. Therefore, it is difficult to determine the only standard data group for long-term diagnosis. Therefore, it is necessary to prepare two or more standard data groups and to choose suitable group for the present condition. Therefore, a huge standard data groups cause a real-time decrease in the diagnosis. Optimal diagnosis needs optimization of the number of standard data groups. This report shows that the method of optimization of the number of standard data groups.