Lumpability in compartmental models (Q922423)

In order to reduce the number of states for a system, lumping of states into classes represents a natural and useful procedure. The present paper discusses an alternative to the classical statistical approach: the use of fuzzy set theory in lumping methodologies A fuzzy entropy compares successive lumpings, as a measure of the intrinsic loss of information by the lumping process. Section 2 exposes the stochastic (Markov chain-based) approach to lumping for compartmental models of systems: the probabilistic structure of the random system with complete connections (RSCC) model is exposed. Section 3 analyses the particular type of lumping corresponding to systems with a limited number of observed compartments. Section 4 contains the numerical results obtained from applying the RSCC probabilistic model to such a system. Section 5 exposes the (new) proposed fuzzy model of lumping. Defining the fuzzy relations of similitude and the fuzzy entropies, the probabilities of the stochastic model are replaced by the (fuzzy) similitude degrees. The fuzzy lumping algorithm is exposed, its correctness is discussed. A numerical example, in Section 6, provides evidence on the usefulness of the proposed approach. Since the study of entropy for lumping in stochastic and fuzzy models represents a difficult problem, in practical situations it is recommended to choose that lumping technique which gives smaller loss of entropy. The fuzzy approach seems to offer better results when the available data are insufficient or unreasonable, i.e. having a noise-tolerant better behaviour.