S.A. Shahdi and S.B. Shouraki (Iran)
Machine learning, Supervised learning, Fuzzy modeling, fuzzy controller
In our previous works we have introduced a novel general learning method, which could treat modeling, controlling and prediction problems in a way similar to what human being does [1,2,3,4]. We showed its advantageous by comparing with some other learning and modeling methods [2,4]. We also showed that it is implementable by simple devices [5,6]. This method includes two basic cores. One is Active Learning Method (ALM), which expresses any multi-input single-output system as a fuzzy combination of some single-input single-output systems. The other one is Ink Drop Spread (IDS), which not only serves as a fuzzy interpolating algorithm but also extracts the importance degree of each single-input single-output system in total system behavior. The proposed method was used as an unsupervised learning method. We will show in this paper that this method can also be used in supervised learning. It means that the method is modified in order to considering negative examples in its procedure. We also show a hardware implementation for this supervised learning method. Then a new approach to design a fuzzy controller automatically will be represented. Fuzzy modeling for the plant will be extracted by ALM method. By applying this controller to a practical controlling problem we will show that using negative examples in learning have a great effect in improving control parameters such as overshoot, rise time and number of fuzzy rules.
Important Links:
Go Back
