基于Q-学习的模糊神经网络控制器

神经模糊系统在机器人的智能控制中具有巨大的应用潜力,但已有的系统构造方法几乎都面临着样本资源匮乏这一巨大困难。为克服传统系统构造方法可能因样本获取困难而引起的“维数灾难”等问题,该文在模糊神经网络中引入了Q-学习机制,提出了一种基于Q-学习的模糊神经网络模型,从而赋予神经模糊系统自学习能力。文章最后给出了其在菅野模糊小车控制中的仿真结果。实验表明,在神经模糊系统中融入智能学习机制Q-学习是行之有效的;它可以被用来实现机器人智能行为的自学习。值得一提的是,该文的仿真实验在真实系统上同样是容易实现的,只要系统能提供作为评价信号的传感信息即可。

A Q-learning Based Neural Fuzzy Controller

This paper proposes a novel three-layered parallel fuzzy inference model QL_NNFI ,which is structured by Q-learning method.Under Q-learning mechanism,a neuro-fuzzy system can be structured without train patterns,which has been preventing us from constructing efficiently neuro-fuzzy systems all the while.This paper also offers a digital experiment that emulating the control of the SUGENO fuzzy car with this model.According to the experiment result,the amalgamation of the proposed model and the Q-learning mechanism,which makes the fuzzy inference model has self-learning ability,is successful.And it can also be simply applied to the intelligent control of a real robot system.