基于BP神经网络的倒立摆智能控制研究  

Research on intelligent control of inverted pendulum based on BP neural network

作　　者：韩治国[1] 陈能祥 许锦 李伟[1] HAN Zhiguo;CHEN Nengxiang;XU Jin;LI Wei(School of Aeronautics,Northwestern Polytechnical University,Xi’an 710072,China;Shenzhen Qianhai Gezhi Technology Co.,Ltd.,Shenzhen 518052,China)

机构地区：[1]西北工业大学航天学院,陕西西安710072 [2]深圳前海格致科技有限公司,广东深圳518052

出　　处：《实验技术与管理》2021年第6期101-106,共6页Experimental Technology and Management

基　　金：陕西省教育科学“十三五”规划课题(SGH20Y1024);2020年产学合作协同育人项目(202002039009、202002252005);西北工业大学2021年度校级教育教学改革研究项目(2021JGY07);陕西省重点研发计划项目(2020GY-047)。

摘　　要：本文首先介绍了一种典型3层BP神经网络模型及其基本原理,然后在惯性坐标系内应用经典力学理论建立了直线一级倒立摆的动力学模型,得到以小车加速度作为输入的系统状态空间方程,在MATLAB中利用lqr函数求得系统最优反馈增益矩阵K并实现了稳定控制。在实时控制过程中不断采集数据样本并保存,用.m文件编写BP神经网络训练程序,对数据样本进行训练并自动生成BP神经网络控制器模块,最后在Simulink中搭建仿真模型进行仿真验证,通过分析仿真结果并进行实物验证,证明了所设计的BP神经网络控制器的合理性与有效性。This paper first introduces a typical three-layer BP neural network model and its basic principle, then establishes the dynamic model of linear inverted pendulum in inertial coordinate system by applying classical mechanics theory and obtains the system state space equation with the acceleration of the car as the input. In MATLAB, the lqr function is used to obtain the optimal feedback gain matrix K and realize the stability control. In the process of real-time control, data samples are collected and saved continuously, and BP neural network training program is written with. M file to train data samples and generate BP neural network controller module automatically. Finally, a simulation model is built in Simulink for simulation verification. Through the analysis of simulation results and physical verification, the rationality and effectiveness of the designed BP neural network controller are proved.

关 键 词：BP神经网络 直线一级倒立摆 LQR控制 MATLAB仿真 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]