本論文的研究目的是研製以數位訊號處理器為基礎使用模型預測控制,並利用遞迴式神經網路當作不確定性觀測器應用於永磁同步馬達定位控制系統。 本研究以傳統磁場導向控制法為基礎,使用設計好的模型預測控制器取代傳統PI控制器用於馬達轉子定位系統上,但模型預測控制應用於馬達驅動定位系統存在的參數不確定項是難以事先得知的,所以實際上要設計一個有效益的模型預測控制器是很困難的。 有鑑於此本論文使用遞迴式神經網路利用推導出的適應律估測馬達參數不確定項所產生的誤差並用補償的方式補償模型預測控制器來消除原本參數不確定所產生的誤差。 最後本研究已32位元浮點運算數位訊號處理器TMS320F28335完成所提出之永磁同步馬達定位驅動系統,並利用實驗比較所提出以遞迴式神經網路補償前後的結果。 ;The purpose of this research is to drive Permanent Magnet Synchronous Motor(PMSM) position control system by Model Predictive Control (MPC) and Recurrent Neural Network (RNN) as an uncertainty observer base on Digital Signal Processing (DSP). Base on traditional Field Oriented Control(FOC) structure and then design model predictive controller to replace PI controller apply to PMSM position control system . But model predictive control in PMSM position control system should know parameters of PMSM.So actually it is difficult to design an efficiency model predictive controller in real system. This dissertation uses recurrent neural network to derive adaptive laws to estimate parameter uncertainties of PMSM and compensate MPC to eliminate the errors that caused by parameter uncertainties. Fially, the proposed PMSM position control system is implemented in a 32-bits floating-point DSP, TMS320F28335.And compare the experimental results with before and after compensation of recurrent neural network.
