Database modeling has been proposed as a nonlinear modeling technique in which historical data during normal operations are accumulated in the database and used directly for the predictions of normal states of a dynamical system. In the previous paper, the scope of this database modeling technique was demonstrated by experiments using a tank-pipeline system and the results of the predictions were almost very good but there was one bad case. This reason is considered that the database model in the previous paper was a first order model. In this paper, a database model is extended to a multiple-order model and compared with the results in the previous paper.