Design and Analysis of Parallel GA-fuzzy Controller

• Main Authors: Hung Lin Hou, 侯宏霖
• Other Authors: Lin Hong
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/51486092788969101878

碩士 === 國立高雄應用科技大學 === 電機工程系 === 97 === With the trends of technology change nowadays, the scale and complexity of industrial control systems are increasing gradually. Therefore, to design a controller with fine performance for large scale and complex controlled system is more difficult and time-consuming. And to develop a method of designing a controller suitable for large scale and complex system is becoming more and more important. The thesis utilizes genetic algorithm to adjust parameters of fuzzy controller; the genetic algorithm is a method to find optimal value from multi points. We only need the defined fitness function to do the evolution process, thus, it’s not limited by the condifims like the existence of function’s continuity or derivative. The thesis use methods of mix encoding, increasing mutation rate, elitist strategy, as well as extinction and immigration strategy to improve the disadvantage of basic genetic algorithm. When searching the controller’s parameters, the improved genetic algorithm can decrease the probability of going into the local optimality, and elevate the probability of searching the approximately best solution substantially as well. Besides, one of disadvantages of genetic algorithm is the time spent to compute. The thesis utilizes parallelization program to solve this main disadvantage of genetic algorithm. In the meantime, based on the consideration of load balance, we divides the program by using a method of partition to decrease the genetic algorithm computing time and to analyze its parallelization efficiency. The author uses conventional inverted pendulum system and motor speed control system to verify the experiment. The simulation results show the superiority of the proposed method.