论文部分内容阅读
将模糊控制和神经网络相结合,形成模糊神经网络(fuzzy neural network,FNN)控制器,利用联想记忆进行离线训练,用来记忆预先利用强化Q学习(Q-Learning,QL)在线训练获得的移动机器人自适应行为的模糊控制规则。FNN经过离线训练后,把规则隐含地分布在整个网络之中,在控制应用时,不必进行复杂的规则搜索和推理,无需查表,只需通过高速并行的分布计算就可产生最佳输出的自适应行为。仿真结果表明,由于输入模糊子集接近于网络所用的训练模糊子集,所以输出几乎和该条训练规则的结果相同。
Combining fuzzy control and neural network, a fuzzy neural network (FNN) controller is formed, and offline training is performed by using associative memory, which is used to memorize the movement acquired in advance through Q-Learning (QL) online training Fuzzy control rules of robot adaptive behavior. FNN offline training, the rules implicitly distributed throughout the network, in the control of the application, without the need for complex rules search and reasoning, without looking up the table, only through the distribution of high-speed parallel computing can produce the best output Adaptive behavior. The simulation results show that the output is almost the same as the training rule because the input fuzzy subset is close to the training fuzzy subset used by the network.