Gaussian process model identification: a process engineering case study

• Autorzy: Kocijan J. , Azman K.
• Języki publikacji: EN

Abstrakty

EN

Gaussian process prior models, although well known in the Bayesian statistics community, are a relatively new approach for the modelling of dynamic systems and consequently a novelty in systems and control community. The Gaussian process prior model is a probabilistic nonparametric model. The complexity of such a model depends on the amount of input data used for identification and contained in the model. This complexity adds to the computational load necessary for multiple-step-ahead prediction and for model simulation. The combination of local linear models and sparse data in off-equilibrium regions can be utilised to reduce the amount of data in Gaussian process model, which is in accord with engineering practice where the situation with lots of measured data in equilibrium regions and sparse measured data in off-equilibrium regions is frequently faced. This approach is used in the paper for the modelling of a gas-liquid separator process plant, where substantial amount of measurement data in equilibrium region and sparse measurement data in distance from equilibrium region are combined in one Gaussian process model.

Słowa kluczowe

EN

nonlinear systems; model identification; Gaussian process model

• Wydawca: Oficyna Wydawnicza Politechniki Wrocławskiej
• Czasopismo: Systems Science
• Rocznik: 2008
• Tom: Vol. 34, no 3
• Strony: 31--38
• Opis fizyczny: Bibliogr. 8 poz., rys., wykr.

Twórcy

autor

Kocijan J.

autor

Azman K.

• Jozef Stefan Institute, Jamova 39, SI-1000 Ljubljana, Slovenia,

Bibliografia

• [1] Azman K., Kocijan J., Comprising Prior Knowledge in Dynamic Gaussian Process Models, [in:] B. Rachev, A. Smrikarov (eds.), Proc. Int. Conf. Computer Systems and Technologies - CompSysTech, 2005, 16.-17.6., Varna, Bułgaria, sec IIIB.2, 2005.
• [2] Gregorčič G., Lightbody G., An affine Gaussian process approach for nonlinear system identification, System Science, 29, No. 2, 2003.
• [3] Johansen T.A., Shorten R., Murray-Smith R., On the Interpretation and Identification of Dynamic Takagi-Sugeno Fuzzy Models, IEEE Transactions on Fuzzy Systems, 8, 2002, pp. 297-313.
• [4] Kocijan J., Girard A., Incorporating linear local models in Gaussian process model, Proc. IFAC World Congress, 2005, Prague, 2005.
• [5] Kocijan J., Girard A., Banko B., Murray-Smith R., Dynamic Systems Identification with Gaussian Processes, Mathematical and Computer Modelling of Dynamical Systems, 11, 2005, pp. 411-424.
• [6] Kocijan J., Likar B., Gas-liquid separator modelling and simulation with Gaussian process models, Proc. 6th EUROSIM Congress on Modelling and Simulation, 2, Full papers, Ljubljana, 2007.
• [7] Murray-Smith R., Johansen T.A., Shorten R., On transient dynamics, off-equilibrium behaviour and identification in blended multiple model structures, European Control Conference, Karlsruhe, BA-14, 1999.
• [8] Rasmussen C.E., Williams C.K.I., Gaussian Processes for Machine Learning, Cambridge, MA: MIT Press, 2005.