Summary: | 碩士 === 中原大學 === 電機工程研究所 === 91 ===
he cerebellar model articulation controller (CMAC) with rectangular function used in the receptive field owns the property of evenly weighting to access ,store and renew the data of memory units . But the output of this kind of CMAC is stairwise and not differentiable .To improve this disadvantage ,the differentiable CMAC(DCMAC) with gaussiam function instead can effectively make the output differentiable . In this thesis ,it is shown that the performance of a DCMAC is better than that a CMAC. Furthermore ,the DCMAC two with modified gaussiam function for weighting was testified proposed and the one with a “blunt” gaussiam function is called the blunt DCMAC and ”aculeate” gaussiam function is called the aculeate DCMAC .The simulation results show that the blunt DCMAC works better than others in function approximation ,and the aculeate DCMAC in control performance.
|