| Summary: | 碩士 === 國立臺北科技大學 === 電機工程研究所 === 105 === The study uses a cerebellar model articulation controller (CMAC), a self-organizing neural network, recurrent network neural architecture, and a functional link neural network (FLNN) to design an adaptive function coupling self-organizing recurrent CMAC (AFCSORC). A novel design approach is adopted to implement this controller by using a self-organizing CMAC, recurrent CMAC structures, and adaptive laws, enabling the resulting static CMAC with fixed association memory layers to achieve dynamic memory capability and parameter-tuning functions. Furthermore, association memory layers are generated or eliminated on the basis of the decision-making mechanism of the layers, consequently, achieving a high-control performance. The FLNN structure is used to improve the learning speed of the CMAC. The proposed controller comprises a function coupling SORC (FCSORC) and an improved compensating controller. An integrated error function is used as the input to the AFCSORC. In addition, the improved compensating controller is designed to eliminate the errors between an ideal controller and the FCSORC. The adaptive laws of function coupling neural network weights, recurrent weights, the Gaussian function mean parameter, and the Gaussian function standard deviation are determined using a Lyapunov function-based analytical method to ensure the stability of the control system.
To verify the performance and effectiveness of the proposed AFCSORC system, we use the proposed AFCSORC scheme for controlling the direct torque control drive system of a switched reluctance motor (SRM). According to the simulation and experimental results, when the drive system is operated at 100, 300, 600, 800, 1000, 1600, ±800, and 800 rpm with an additional 1-Nm load at the third second, the load torque of the motor is 1 Nm. The maximum speed error in the transient response is lower than 11 rpm, during which the SRM exhibites a desirable torque response across a wide range of speeds. When the speed command is set to 1000 rpm, the root mean square error (RMSE) is used for a performance comparison of the speed responses among the conventional CMAC, FCMAC, and AFCSORC; the RMSE values are 2.05, 1.65, and 1.02 rpm, respectively. When the speed command is set to 800 rpm, which is not accompanied by a starting-load operation or an additional 1-Nm load at the third second, the RMSE values were 3.67, 2.36, and 1.59 rpm, respectively. In addition, the AFCSORC recovered quickly to the speed command. These results demonstrate that the RMSE value of AFCSORC is lower than that of either CMAC or FCMAC at each speed. The proposed controller exhibites enhanced robustness to external disturbances.
|
|---|
