Dynamic Modelling of Technical Processes using Jump Markov Linear Gaussian Models

R. Morales-Menéndez, F.J. Cantú Ortiz, and A.R. Favela Contreras (Mexico)


Modelling, Jump Markov Linear Gaussian, Identification, Nonlinear Systems


Most realistic processes show a stochastic component and a finite-dimensional internal state. Jump Markov Linear Gaussian (JMLG) models are linear systems whose pa rameters evolve with time according to a Markov chain. We tested the JMLG model for both simulated nonlinear systems and real level-tank processes. Experimental tests were implemented and the observations were registered. The model parameters were learned in two stages. First, parametric identification was computed based on the Least Squares Estimation algorithm. Then, a refining procedure was carried out based on the Expectation-Maximization al gorithm. The resulting JMLG model was found to success fully represent the dynamic behavior of the process.

Important Links:

Go Back