Development of a knowledge-based system for supervision of a continuous process requires, on the one hand, efficient and flexible knowledge structuring and, on the other hand, overall system control, which provides coherence between the diagnosis task (deduction) and the prediction task (simulation). The development of a reasoning method that combines qualitative and quantitative analysis approaches is described. This method is integrated into an overall computational system for a knowledge-based supervisor. The prototype was tested by simulating the transient behavior of the auxiliary feedwater system of a pressurized water reactor. The preliminary results indicate the feasibility of the methods and their potential for industrial applications.