Instead of using a default 1-year history for training data, select 5+ years at 6- to 24-hour sample intervals to analyze the normal process variation. Next, based on analysis of variance, pull in one or more data sets at a higher sample rate (e.g. 5-minutes) for training data. Overlay the data sets on an X-Y plot to visualize the “coverage” of training data against history data.
Comments or questions? Write me at firstname.lastname@example.org.