Modeling with discrete-time recurrent fuzzy systems via mixed-integer optimization

In this paper, we present a new approach for modeling of dynamical systems with discrete-time recurrent fuzzy systems (DTRFS). The modeling problem is divided into two subproblems. The first subproblem considers the identification of the rule base of the DTRFS which can be interpreted as structural modeling. This problem is recast into integer and mixed-integer optimization problems respectively dependent on the chosen cost function and system structure. The second subproblem considers the optimization of the parameters of the system, i.e., quantitative modeling. As a result, we obtain a linguistically interpretable model of the dynamical system which additionally can be interpreted as a linguistic automaton. The applicability of the approach is shown by modeling a thermofluidic process.