Formalizing Candidate Model Construction with Hidden Markov Models

D. Van Welden and E. Noldus (Belgium)


Modeling, Identification, SystemApproach Problem Solver, Hidden Markov Models


In this paper a pattern recognition approach for system identification, based on a system-theoretic methodology, called GSPS (General System Problem Solving) is used to search a model for dynamic directed black-box systems. The trick is to “flatten” the input-output data into the state-observation space. The states are made up of past/present inputs and/or past outputs. The process that leads to a predictive model can be seen as a newly defined problem type in the theory of hidden Markov models (HMM). The result is a more rigid formulation in the way the model is constructed and the use of proper terminology that is more consistent with HMM. Additionally, existing model order determination problems match similar problems in the theory of HMM.

Important Links:

Go Back