连续型强化学习与PID控制的应用对比分析:以一阶倒立摆系统为例  

Comparison and Analysis of Respective Applications of Continuous Reinforcement Learning and PID control:a Case Study of First-Order Inverted Pendulum System

作　　者：程丽梅 贾文川[1] 

机构地区：[1]上海大学机电工程与自动化学院,上海200444

出　　处：《工业控制计算机》2021年第10期20-22,共3页Industrial Control Computer

摘　　要：作为典型的强人工智能方法,强化学习已经开始被尝试应用于真实控制任务中。为了探究连续型强化学习在物理系统控制中的效果,以一阶倒立摆系统的平衡控制为实例,使用强化学习方法和经典PID控制方法对倒立摆系统分别进行建模与控制实现,并着重对控制表现进行对比分析,结果表明在该控制任务中连续强化学习训练所得的控制策略对扰动的反应更迅速,并具有更强的稳定性。As a typical method of strong artificial intelligence,reinforcement learning has been tried to be applied to real control tasks.In order to explore the effect of continuous reinforcement learning in the control of physical system,this paper takes the balance control of first-order inverted pendulum system as an example,uses reinforcement learning method and classical PID control method to model and control the inverted pendulum system respectively,and focuses on the comparative analysis of the control performance.The results show that the control strategy obtained by continuous reinforcement learning has a faster response to disturbance and a stronger stability.

关 键 词：强化学习 倒立摆系统 连续性 PID控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] TP18[自动化与计算机技术—控制科学与工程]