基于深度强化学习的固高直线一级倒立摆控制实验设计  

Experimental Design of Googol's Linear Single Inverted Pendulum Control Based on Deep Reinforcement Learning

作　　者：冯肖雪 谢天 温岳 李位星[1] FENG Xiaoxue;XIE Tian;WEN Yue;LI Weixing(School of Automation,Beijing Institute of Technology,Beijing,100086 China)

机构地区：[1]北京理工大学自动化学院,北京100086

出　　处：《科技资讯》2023年第23期4-10,共7页Science & Technology Information

摘　　要：为适应各高校人工智能专业学生对于机器学习领域的学习需求,同时兼顾固高科技直线一级倒立摆控制系统可操作性、实时性和安全性,设计了一套基于深度强化学习的固高直线一级倒立摆控制实验方案。首先采用深度强化学习算法的无模型控制结构搭建控制器并进行虚拟仿真实验。考虑倒立摆电机驱动刷新频率的限制以及提高样本处理速度,进一步设计了基于离线Q学习算法的平衡控制器实现倒立摆实物稳定控制。该实验方案既加深了学生对人工智能领域知识的理解,也适应了固高科技直线一级倒立摆的应用场景。In order to meet the learning needs of students majoring in artificial intelligence in colleges and universities in the field of machine learning,and take into account the operability,instantaneity and safety of the linear single inverted pendulum control system of Googol Tech,this paper designs an experimental plan for Googol's linear single inverted pendulum control based on deep reinforcement learning.Firstly,this paper uses a model-free control structure of the deep reinforcement learning algorithm to construct a controller and conduct virtual simulation experiments.Considering the limitation of the refresh frequency driven by the inverted pendulum motor and the improvement of sample processing speed,it further designs a balance controller based on the offline Q-Learning algorithm to achieve the physical stability control of the inverted pendulum.This experimental plan not only enhances studnets'understanding of the knowledge in the field of artificial intelligence,but also adapts to the application scenario of the linear single inverted pendulum of Googol Tech.

关 键 词：直线一级倒立摆 深度强化学习 Deep Q Network算法 Q学习算法 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]