基于神经网络及强化学习的智能控制系统  

An Intelligent Control System Based on Neural Networks and Reinforcement Learning

作　　者：钟宇平[1] 王丽丹[1] 段书凯[1] 张凤运 李波[1] 

机构地区：[1]西南大学电子信息工程学院,重庆400715

出　　处：《西南大学学报（自然科学版）》2013年第11期172-179,共8页Journal of Southwest University(Natural Science Edition)

基　　金：国家大学生创新创业计划(201210635123);国家自然科学基金资助项目(60972155;61101233);中央高校基本科研业务费专项资金资助项目(XDJK2012A007;XDJK2013B011);重庆市高等学校青年骨干教师资助计划和优秀人才支持计划(渝教人〔2011〕65号);教育部"春晖计划"(z2011148);留学人员科技活动项目择优资助经费(渝人社办〔2012〕186号);重庆市高等教育教学改革研究重点项目(09-2-011)‘西南大学教育教学改革研究项目(2012JY201)的资助

摘　　要：将神经网络与强化学习结合,提出了一种新型算法模型.该模型应用于智能控制系统中将使智能体具有自主性、高效率、大容量等综合优势.最后,利用计算机软件仿真实验,验证了方案的有效性.本次实验对机器人使用基于试错改进机制的强化学习方式,与传统机器人研究领域大多使用的基于导师机制的监督学习相比,提高了机器人自主适应环境的能力,使机器人更加智能化.同时,将神经网络引入到强化学习中,使该智能系统较其他强化学习系统具有更快的处理信息的速率.With the rapid development of science and technology today, humans require robots to be more intelligent. Reinforcement learning is a kind of automatic learning algorithm, by which the intelligent agent can accumulate experience and improve strategies through constant trial and error, and ultimately get the optimal action strategies. Artificial neural network can process input data in parallel and has the advantage of high computation speed. Combining the neural network with reinforcement learning, this paper propo- ses a new algorithm model. Applied in the intelligent control system, this model may enable robots to have the advantage of automatic decision-making capacity, high efficiency of information processing and large capacity for input data. Finally, a simulation experiment based on computer software verifies the effective- ness of the scheme. In this study, the robot employs the reinforcement learning algorithm based on the mechanism of trial-and-error improvement. Compared with the traditional algorithm in robotics research areas such as supervision learning which is based on guiding, reinforcement learning improves the robots＇ ability to automatically adapt to the environment and makes the robot more intelligent. Meanwhile, the neural network is introduced into reinforcement learning, so that the intelligent system will have a higher rate of data processing than other reinforcement learning systems.

关 键 词：强化学习 神经网络 数值仿真 智能控制 

分 类 号：TP273.5[自动化与计算机技术—检测技术与自动化装置]