Experience with large-scale dynamic model validation – application to industrial plant

Abstract

Whereas static model validation is well developed, dynamic models remain difficult to satisfactorily validate especially when it concerns large scale systems common in chemical engineering. In this paper, we focus on demarcating the domain where a model is considered validated, whereby we extend an existing methodology for static systems to dynamic systems. Secondly, the comparison between model and data is not done in the time-domain but in the wavelet domain, which has the potential to better account for the dynamic character of the system. This paper gives a short conceptual description of how to aggregate the points in the state-input space to recognize validated, non-validated and invalidated domains. This is all illustrated on real data from a condenser described by 6 state variables and where 300 sensors were involved. This proved tractable, but showed also that larger problems need further computational or conceptual development.