Black- and white-box approaches for cascaded tanks benchmark system identification

• Published in: Mechanical systems and signal processing Vol. 108; pp. 387 - 397
• Main Authors: Giordano, G., Sjöberg, J.
• Format: Journal Article
• Language: English
• Published: Berlin Elsevier Ltd 01.08.2018; Elsevier BV
• Subjects: Benchmarks; Black-box; Cascaded Tanks; Computer simulation; Containers; Descriptive labeling; Hydraulics; Identification; Mathematical models; Mechanical engineering; Non-linear system; Nonlinear system; Nonlinear systems; Pumps; System identification; Tanks; White-box; Cascaded tanks; System identification; Non-linear system; Black-box; White-box
• ISSN: 0888-3270; 1096-1216; 1096-1216
• DOI: 10.1016/j.ymssp.2018.01.008

•Identification and comparison of models for a nonlinear system using benchmark data.•Model structures with different levels of flexibility and prior knowledge compared.•White-box models built up step-wise with increasing complexity.•Best model in simulation is white-box model, RMSE is 0.25 V on validation data.•Black-box models perform good in prediction, RMSE is 0.051 V for ARMAX. This contribution consists of the identification and comparison of different models for a non-linear system: the Cascaded Tanks system. The identification of this system is challenging due to the combination of soft and hard non-linearities. Model structures with different levels of flexibility and prior knowledge are compared. The most simple ones are linear black-box models. They are extended to become non-linear black-box models, whose performances are compared with the linear ones. A second track is the investigation of a series of models with increasing complexity based on physical prior knowledge. Results show that while linear black-box models perform good in prediction, a fairly precise description of the non-linear effects is needed to achieve good performances in simulation. All models have been estimated and validated using benchmark data from a real cascaded tanks system. The contribution represents also an overview on how standard modelling techniques perform on a real identification problem.