Ensembles of Fuzzy Linear Model Trees for the Identification of Multioutput Systems

摘要: We address the task of discrete-time modeling nonlinear dynamic systems with multiple outputs using measured data. In area control engineering, this is typically converted into a set classical regression problems, one for each output, which can then be solved any approach. Fuzzy models, in Takagi–Sugeno form, are popular context. use Lolimot, tree learning method, to build fuzzy linear model trees. paper, we propose, implement, and empirically evaluate three extensions First, consider multioutput models. Second, propose ensembles such Third, investigate search heuristic based on simulation error (as opposed one-step-ahead prediction error), specific context systems. Finally, perform an empirical evaluation compare these approaches six case studies, both simulated data, noise: The studies include inverse dynamics robot arm, as well five additional process-industry Ensembles improve performance single- trees, while only improves very slightly. Multioutput trees exhibit comparable or worse predictive single-output providing more compact model. Overall, recommend bagging Lolimot learned by heuristic.