无人机控制器的建模与比较研究

摘要

本文采用分层控制方法设计一种用于无人机控制的PID和神经网络控制器。利用Matlab控制系统工具箱指令构建系统平衡点和状态空间模型的线性化模型。通过特征向量分析识别出系统各状态间的相关性强弱，以发现降阶系统模型。使用时常数分离系统模型为控制器开发进行进一步降阶，并利用简化模型设计了纵向和横向控制器。控制器参数可由二阶和三阶的系统方程和系统的阻尼比与带宽计算得到。使用MATLAB SIMULINK进行完整的线性化纵向和横向状态空间模型的仿真实验以测试本文设计控制器的性能。为设计模型参考神经控制器进行了系统识别和神经训练。利用近似的俯仰姿态系统和可变频的输入需求对PID和神经网络控制器之间性能的相似性和差异进行了研究。基于PID控制器的设计方法表明，比例参数位于反馈路径而不在前向路径上可优化输入，并在输入需求阶跃变化时产生最小的过冲。本文研究了神经控制器对PID控制器缺点的抑制作用，并对分层控制系统设计中采用神经控制器的重要性进行了分析。

目录

声明

摘要

Abstract

Table of Contents

1 Introduction

1．1 Controller Design Methodology

1．1．1 PID Controller

1．1．2 Proportional and Integral Parameters

1．1．3 Derivative Parameter

1．2 Model Reference Neural Controller

1．2．1 Plant Neural Net Model

1．2．2 Reference Model

1．2．3 Neural Net Controller

1．3 Project Objective

1．4 Notations and Definitions

2 UAV Modeling

2．1 Mathematical Model

2．2 Modeling Environment and Implementation Algorithm

3 Linear Model Analysis

3．1 Linearization

3．2 System Modes and Stability

3．2．1 Longitudinal System Modes

3．2．2 Latitudinal System Modes

3．3 System Step Response

3．3．1 Longitudinal Transfer Functions and Step Response

3．3．2 Latitudinal Transfer Functions and Step Response

3．4 Linear Model Approximation

3．4．1 Longitudinal Linear Model Approximation

3．4．2 Transfcr Function for Pitch Attitude and Pitch Rate

3．4．3 Transfer Function for Speed

3．4．4 Transfer Function for Elevation

3．4．5 Latitudinal Linear Model Approximation

3．4．6 Transfer Function for Roll Attitude and Roll Rate

3．4．7 Transfer Function for Sideslip Angle

3．4．8 Transfer Function for Heading

4 Linear Controller Design

4．1 Longitudinal Controllers

4．1．1 Pitch Attitude Controller Design

4．1．2 Speed Controller Design using Pitch Attitude

4．1．3 Speed Controller Design Using Throttle

4．1．4 Elevation Controller Design

4．2 Latitudinal Controller Design

4．2．1 Roll Attitude Controller Design

4．2．2 Heading Controller Design

5 Neural Controller Design

5．1 Model Reference Pitch Attitude Neural Controller

5．2 Plant and Controller Neural Net Model

5．3 Reference Model

5．4 Simulation and Training

5．5 PID and Neural Net Controller Comparison

Conclusion

References

Appendix

Acknowledgement