We see a temporal data warehouse as a set of temporal views defined in the past fragment of the temporal relational algebra extended with set-valued attributes and aggregation. This paper proposes an incremental maintenance method for temporal views that allows improvements over the re-computation from scratch. We introduce a formalism for temporal data warehouse specification that summarizes information needed for its incremental maintenance. According to this formalism, a temporal data warehouse W is a pair of two sets of views : the materialized component and the virtual component. The materialized component of W represents the set of views physically stored in the warehouse. The virtual component of W is a set of non-temporal expressions involving only relations kept in the materialized component. Several features of our approach make it especially attractive as a maintenance method for warehouses: (a) there is no need for storing the entire history of source databases, (b) maintenance of the temporal data warehouse is reduced to maintaining the (non-temporal) materialized component, and (c) the materialized component is self-maintainable. We build a uniform algorithm by combining two previously unrelated techniques based on auxiliary views. Our method is sufficiently general so that it can be easily adapted to treating databases with complex-valued attributes.

temporal data warehouse, temporal databases, temporal relational algebra, temporal logic, self-maintenance