基于强化学习的倒立摆控制系统设计2.docVIP

PAGE PAGE III PAGE I 摘 要 强化学习可以算作是这几年来在人工智能和机器学习研究领域内非常热门的项目，而在强化学习与监督学习中有着许多不同之处。强化学习不需要教师信号，但是它也不等同于只能够完成有限功能的无监督学习，而这种无监督学习在学习过程中无法得到提示。强化学习强调能够获得评价性反馈信号，并且使用评价信息来达到行为决策上的优化。显而易见，强化学习具有广阔的前景。 复杂系统控制中有许多经典的问题，倒立摆就是其中之一。倒立摆系统是一种绝对不稳定的，具有非线性、多变量等多种性质的系统。而在控制过程中，倒立摆系统也是一个验证各种控制理论的很理想的模型之一。它可以反映例如可镇定性、鲁棒性和随动性等等问题。近代以来，倒立摆系统广泛应用于我们的生活当中。卫星的运行、机器人的行走等都是运用了类似倒立摆系统稳定控制的例子。显而易见倒立摆的研究具有其深刻的理论意义和重要的工程意义。 在本次的课题中，我们会以大学四年相关的学习内容作为基础，把强化学习作为研究对象，将倒立摆系统作为实验模型，进行系统科学的实验研究。我们会通过对一级单臂倒立摆的平衡控制进行研究，使系统具备学习能力，并且可以在运行的过程中获得新的内容和信息，具有类似生物的运动控制技能。本课题在强化学习和Python语言的基础上，提出了基于Q学习的的强化学习系统。我们用Pycharm进行实验仿真，证明提出的强化学习系统具有认知倒立摆系统的平衡控制技能的能力。 关键词 强化学习，Q学习算法，倒立摆系统 ABSTRACT There are many classic problems in complex system control, the inverted pendulum is one of them. The inverted pendulum system is an absolutely unstable system with nonlinear and multivariable properties. In the control process, the inverted pendulum system is also one of the ideal models to verify various control theories. It can reflect such problems as stability, robustness and follow - up. Since modern times, the inverted pendulum system has been widely used in our life. Satellite operation, robot walking and so on are the use of the inverted pendulum system stability control examples. Obviously, the research of inverted pendulum has profound theoretical significance and important engineering significance. In this project, we will take the relevant learning content in the four years of college as the basis, take reinforcement learning as the research object, and take the inverted pendulum system as the experimental model to conduct systematic scientific experimental research. We will study the balance control of a one-arm inverted pendulum to make the system capable of learning and acquire new contents and information in the process of operation, with the movement control skills similar to those of living creatures. Based on reinforcement learning and Python language, this paper proposes a reinforcement learning s