Results of computer-experimental performance analysis of a type of fuzzy controllers for the inverted pendulum system
Dragan Z. Šaletić Fuzzy control techniques are of interest for practical applications. Hence, it is important to understand their features. In this paper some results of computer-experimental analysis of behavior of a fuzzy controller for the cart-ball variant of inverted pendulum system are given. The impacts of changes in uncertainty in the controller are analyzed experimentally: it has been found that the performances of the controller deteriorate mildly towards a certain percentage of the rules that are removed, and that is opposite relative to the process with probabilistic uncertainty model. Also, the robustness of the controller is analyzed experimentally: it has been found that the fuzzy controller is more sensible to wrong, than to missing rules. Further, the results of the experiments with different shapes of primary fuzzy sets membership functions are given: it has been concluded that the fuzzy system is sensitive to the changes of shapes of those functions and that the optimal shapes exists. This result is a new one; it has not been given in the references. To conclude, characteristics of the discussed PD fuzzy controller type are given. The conclusion, based on the results of the computer experiments, is that the behavior of the fuzzy PD controller in the considered task is at least as good as the behavior of the crisp (traditional) PD controller. The subjects for further research activities in the considered paradigm are also pointed out.
Key words: expert system, fuzzy logic, fuzzy system, fuzzy controllers, controllers, system robustness
|