Discovering Closed Periodic-Frequent Patterns in Very Large Temporal Databases