批次自我學習法於移動模糊控制器之設計

Motion Fuzzy Controller Design by a Batch Self-Learning Method

陳冠華(Kuan-Hua Chen)

指導教授 ： 翁慶昌

淡江大學/工學院/電機工程學系碩士班/碩士(2009年)

https://doi.org/10.6846/TKU.2009.00351

摘要

本論文提出一個批次自我學習法來自動產生一個可以有效控制輪型機器人移動之移動模糊控制器。此方法在每一代均有控制階段、受控體模糊鑑別階段與控制器學習階段等三個階段。此方法主要有兩個模糊系統，其中一個為用來鑑別輪型機器人之移動模糊鑑別器，另一個為用來控制輪型機器人移動之移動模糊控制器。在第一階段之控制階段，移動模糊控制器將控制實際的輪型機器人，並且蒐集輪型機器人的輸出入對，其將包含機器人在場地上實際移動的特性。在第二階段之受控體模糊鑑別階段，利用第一階段所得到之輸出入對來鑑別輪型機器人，並且用基因遺傳演算法來得到一個可以逼近這些輪型機器人的輸出入對之模糊鑑別器。在第三階段之控制器學習階段，將第二階段所得到之移動模糊鑑別器來取代實際的輪型機器人，並且用基因遺傳演算法來得到一個讓此模糊鑑別器具有不錯控制性能的移動模糊控制器。在到達最大代數之前，此三階段之學習方法將一直重複的進行。模糊系統的一些可調參數被視為一個參數集，本論文用基因遺傳演算法來分別找出具有最佳逼近性能之模糊鑑別器與具有最佳控制性能之模糊控制器。移動模糊控制器之參數將可以由所提之批次自我學習法來自動產生，所選取的移動模糊控制器將可以使得輪型機器人之控制具有最佳的移動性能。此外，不同的輪型機器人在不同的環境下，所提之方法仍然讓控制器具有不錯的控制性能，所以所提之方法具有不錯的適應性。從FIRA 3維機器人足球模擬器之模擬結果可以驗證所控制的輪型機器人可以有效的從起始位置順利到達目的地位置。

關鍵字

移動控制 ； 基因遺傳演算法 ； 模糊鑑別器 ； 模糊控制器

並列摘要

In this thesis, a batch self-learning method is proposed to automatically determine a motion fuzzy controller so that the motion of the controlled wheeled robot has a good performance. This method in each generation can be separated into three states: a control state, a controlled plant identification state, and a controller learning state. There are two fuzzy systems in this method. One is a fuzzy identifier to identify the real wheeled robot and the other is a fuzzy controller to control the wheeled robot. In the first state of control state, the motion fuzzy controller will control the real wheeled robot and the input and output data of the robot is collected, where the motion characters of the robot on the plane are included. In the second state of the controlled plant identification state, based on the obtained input and output data in the first state, a fuzzy identifier with a good approximation performance is obtained by Genetic Algorithm. In the third state of controller learning state, the fuzzy identifier obtained in the second state is viewed as the controlled plant and a fuzzy controller with a good control performance is obtained by Genetic Algorithm. These three states of the learning method will be repeated until the maximum generation is reached. The antecedent and consequent parameters of the fuzzy system are viewed as a parameter set and Genetic Algorithm is proposed to choose appropriate parameter sets so that the selected fuzzy identifier has a good approximation and the selected fuzzy controller has a good control performance. The parameters of fuzzy controller can be automatically selected by the batch self-learning method so that the obtained motion fuzzy controller has a good control performance. Moreover, different robots in different environment using the proposed method also can let the obtained controller have a good control performance. Therefore, the proposed self-learning method has a good adaptability. Some simulation results in FIRA 3D robot soccer simulator are presented to illustrate that the controller can control the robot to the destination smoothly.

並列關鍵字

Motion Control ； Genetic Algorithm ； Fuzzy Identifier ； Fuzzy Controller

參考文獻

[4] K.S. Narendra and K. Parthasarathy, “Identification and control of dynamical system using neural network,” IEEE Transactions on Neural Networks, vol.1, pp.4-27, Jan. 1990.

[5] C.F. Juang, “A TSK-Type recurrent fuzzy network for dynamic systems processing by neural network and genetic algorithms,” IEEE Transactions on Fuzzy Systems, vol.10, no.2, pp.155-170, Apr. 2002.

[6] C.T. Lin and C.S.G. Lee, “Neural-network-based fuzzy logic control and decision system,” IEEE Transactions on Computers, vol.40, pp. 1320-1336, Dec. 1991.

[7] J.S. Jang, “ANFIS: Adaptive-network-based fuzzy inference systems,” IEEE Transactions on Systems, Man and Cybernetics, vol.23, no.3, pp. 665-685, 1993

[8] C.C. Wong and C.C. Chen, “A GA-based method for constructing fuzzy systems directly from numerical data,” IEEE Transactions on Systems, Man and Cybernetics Part B: Cybernetics, vol.30, no.6, pp.904-911, 2000.