| Summary: | 碩士 === 國立虎尾科技大學 === 電機工程研究所 === 101 === This dissertation proposes a novel supervised and reinforcement evolutionary learning for TSK-type compensatory neuro-fuzzy controller. This dissertation consists of the two major parts. In the first part, we proposed a supervised evolutionary learning for TSK-type compensatory neuro-fuzzy controller. This supervised evolutionary learning composed by two parts which contain structure learning and parameter learning. The structure learning uses the data characteristics measure to determine the fuzzy rules number and initial parameters. The parameter learning used the cooperatively coevolving differential evolution to adjust the neuro-fuzzy controller’s parameters. The cooperatively coevolving differential evolution adopts the belief space, population space and cooperative coevolution into the differential evolution to increase the capability of algorithm. In the second part, we proposed a reinforcement evolutionary learning for TSK-type compensatory neuro-fuzzy controller. The controller parameters adjust by a modified cooperatively coevolving differential evolution, and used reinforcement signal as fitness to evaluate which controller parameters can use for the control problems. Finally, the proposed two evolutionary learning for TSK-type compensatory neuro-fuzzy controller are applied in various nonliear system control problems. Results of this dissertation demonstrate the effectiveness of the proposed evolutionary learning.
|
|---|
