论文部分内容阅读
提出一种基于强化学习的 ART2神经网络(RL-ART2),使其利用强化学习的特性通过与环境交互而无需训练样本即可进行在线学习,同时给出该神经网络的学习算法.当 ART2神经网络运行时,通过内部竞争学习得到输出的分类模式,随后通过与环境交互得到神经网络分类模式的运行效果并对其进行评价.通过这种不断与环境的交互学习,当经过在线学习足够的时间和次数后,ART2神经网络即具有相当的识别率.移动机器人路径规划仿真实验表明,使用 RL-ART2后与未使用前相比大大减少了机器人与障碍物的碰撞次数,实践证明该方法的合理性和有效性.
A ART2 neural network based on reinforcement learning (RL-ART2) is proposed, which makes it use the characteristics of reinforcement learning to interact with the environment without online training samples and gives the learning algorithm of the neural network.When ART2 nerve When the network is running, through the internal competitive learning, the output classification model is obtained, and then through the interaction with the environment to get the operation effect of the neural network classification mode and evaluate it.Under this continuous interaction with the environment, when online learning enough time ART2 neural network has considerable recognition rate.Mobile robot path planning simulation experiments show that the use of RL-ART2 compared with before the use of greatly reduced the collision between the robot and obstacles, the practice proved that the method is reasonable Sexuality and effectiveness.