A Hybrid Optimization Study of Integrating Simulated Annealing, Linear Quadratic Regulator, and Fuzzy Control

• Main Authors: Ke-mu Wu, 吳科穆
• Other Authors: Chung-Neng Huang
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/74145188457390329323

碩士 === 國立臺南大學 === 機電系統工程研究所碩士班 === 98 === The purpose of this study presents the optimal method to solve system parameters set up by integrating Simulated Annealing (SA) combined with Linear Quadratic Regulator (LQR). This study designs the on-line control system for permanent magnet synchronous motors (PMSM). The speed controller uses the PI controller. Most of the PI controllers tune the gain manually by trial-and-error. This method not only takes time, and the gain can not guarantee the optimal value. Many scholars want to improve the accuracy of gain. They present Fuzzy-PI[1], GA-PI[2] and NN-PI[3] are used to solve above problem. Since the FUZZY-PI controller has lower accuracy and more offset, and the GA-PI needs longer computing time, while the NN-PI needs a lot of time to training. This study hence presents the SA method to improve above problem. In this study, the gains of PI controller are determined by SA-LQR. The proposed method can solve the optimal problem in need to adjust parameters. Most of the traditional optimization methods uses single optimal to adjust parameters. This study integrates SA and LQR to ensure the optimal value. The parameters of PMSM would be small amount of change when the input changed, but Fuzzy Logic Control (FLC) has good output response for this problem. For the reason, FLC can be used to device the gain in different situation. In addition, the experimental set-up is built for the PMSM and implemented by a DSP-based fully digital controller. This method can effectively improve the dynamic response and on-line correction steady state error to achieve satisfaction performance. The experimental results have shown the robust for the proposed method when the system has severe input change.