林巍聳臺灣大學：電機工程學研究所田姸君Tien, GloriousGloriousTien2007-11-262018-07-062007-11-262018-07-062005http://ntur.lib.ntu.edu.tw//handle/246246/52983本研究的目的為利用雙重試錯規劃法為一新型雙輪驅動運輸車設計模糊控制器的參數。雙重試錯規劃法為最佳化控制中可調式評斷法的一種種類；藉由閉迴路控制系統中訊號的傳遞，雙重試錯規劃法可自行學習如何評斷控制器效能同時自行調整控制器參數。在此研究中，選定雙重試錯規劃法的學習模組為可調式模糊推論系統；其模糊規則採用線性組合形式，其中模糊規則之線性組合參數即為需要調整之參數，並採用線上學習方式使每一組新輸入資料都可觸發學習動作。由於雙重試錯規劃法乃衍生自動態規劃法，因此設計的控制器可以控制受控物狀態在易於控制範圍，使整體時序上控制器表現為最佳化。根據拉格朗日法推導出此新型雙輪驅動運輸車之數學模型。此模型依照有無負載分為重心在輪軸之下與重心在輪軸之上兩類。利用雙重試錯規劃法分別為其設計控制器，並利用重心在輪軸之上時的天生不穩定特性來驗證雙重試錯規劃法自行尋找系統平衡點的能力。電腦模擬和驗證顯示利用雙重試錯規劃法為此雙輪驅動運輸車設計模糊控制器為可行的。The goal of this research is to develop a Dual Heuristic Programming (DHP) design of self-learning fuzzy controller for maneuvering a two-wheeled transporter. DHP, which is a category of the adaptive critic method, is an optimal control algorithm. The fuzzy inference system containing fuzzy if-then rules of Takagi and Sugeno’s type is chosen as the learning model. A pattern-learning algorithm is developed to implement the learning process. With the DHP learning scheme, the adaptive parameters in the learning models are updated at every time step to satisfy the Bellman equation. Based upon the state feedback signals, the controller trained by the DHP learning scheme has an ability to find the possible optimal solution into the future. The mathematical model of a two-wheeled transporter is derived by Lagrange formalism. The two-wheeled transporter depending on payload could have its intermediate body with center of gravity resting below or above the wheel axle. Therefore, the resulting configuration could be open loop unstable. DHP design of the transporter controller has been derived and implemented successfully. Simulation results show the resulting controller can learn from crash and comply with changes in the configuration of the transporter.Chapter 1 Introduction...................................1 1.1 Background..........................................1 1.2 Motivation and Contributions........................4 1.3 Organizations of This Thesis........................6 Chapter2 Self-Learning Adaptive Fuzzy Controller Based on Dual Heuristic Programming Method.......................7 2.1 Introduction........................................7 2.2 Basic concepts and notation.........................8 2.2.1 Representation of plants..........................8 2.2.2 Dynamic programming: Bellman’s principal of optimality .............................................9 2.2.3 Adaptive networks: pattern learning algorithm.....12 2.2.4 Dual heuristic programming: an optimal controller design algorithm..................14 2.3 Adaptive fuzzy controller design by DHP.............19 Chapter 3 Dynamic Model of a Two-Wheeled Transporter.....27 3.1 Introduction........................................27 3.2 Mathematical modeling by the Lagrange formalism.....30 3.2.1 Transporter without payload.......................31 3.2.2 Transporter with payload..........................34 3.3 Configuration of the control system.................35 3.4 Plant model for the DHP method......................37 Chapter 4 DHP-based Self-Learning Control of a Two-wheeled Transporter...............................43 4.1 Introduction........................................43 4.2 DHP-based adaptive fuzzy controller for steering and posture control.........................................44 4.2.1 Structure of the adaptive fuzzy controller........44 4.2.2 Structure of the critic model ....................47 4.2.3 Learning Strategy.................................47 4.2.4 Straight line tracking control....................49 4.3 Transporter without payload.........................52 4.3.1 Parameter of the transporter......................52 4.3.2 The primary utility function......................53 4.3.3 Simulation of the learning process and tracking the reference velocity profile..............................53 4.3.4 Simulation result of the straight line tracking...60 4.4 Transporter with payload............................62 4.4.1 Parameter of the transporter..................... 62 4.4.2 The primary utility function......................62 4.4.3 Simulation of the learning process and tracking the reference velocity profile..............................65 4.4.4 Simulation result of the straight line tracking...72 Chapter 5 Conclusion.....................................75 References..............................................77 Publications............................................79en-US雙重試錯規劃法可調式模糊推論系統雙輪驅動運輸車Dual Heuristic ProgrammingAdaptive fuzzy inference systemTwo-wheeled transporterthesis