基于Q学习算法和BP神经网络的倒立摆控制
Learning to Control an Inverted Pendulum Using Q-Learning and Neural Networks
-
摘要: Q学习是Watkins[1]提出的求解信息不完全马尔可夫决策问题的一种强化学习方 法.将Q学习算法和BP神经网络有效结合,实现了状态未离散化的倒立摆的无模型学习控 制.仿真表明:该方法不仅能成功解决确定和随机倒立摆模型的平衡控制,而且和Anderson[2] 的AHC(Adaptive Heuristic Critic)等方法相比,具有更好的学习效果.Abstract: Q-learning is a reinforcement learning method to solve Markovian decision problems with incomplete information. This paper presents a novel method to control an inverted pendulum with unquantized states by using Q-learning and neural networks. Simulation results are included to show that the new method can not only balance the determined or stochastic inverted pendulums successfully but also lead to a better effect of learning when compared with Anderson's AHC method.
-
Key words:
- Q-Learning /
- BP neural network /
- learning control /
- inverted pendulum /
- Gaussian noise
计量
- 文章访问数: 3510
- HTML全文浏览量: 190
- PDF下载量: 1890
- 被引次数: 0