论文部分内容阅读
针对传统模型参考自适应控制存在的鲁棒性问题和神经网络结构庞大因而计算量膨胀的问题,提出了一种变结构神经网络L1自适应控制方法,其中变结构神经网络用于在线辨识系统存在的未知非线性函数,该网络通过对节点进行唤醒与催眠以动态调节结构,以最少的节点数进行有效的逼近,降低计算复杂度;L1自适应控制用于网络权值学习与系统非线性补偿,反馈回路中设有一个低通滤波器,只要满足L1增益条件,就能确保系统的输入输出信号的瞬态响应和稳态跟踪性能与一个期望的线性时不变系统的响应保持一致。通过对四旋翼飞行器进行仿真,验证了该方法的有效性。
Aiming at the problem of robustness of traditional model reference adaptive control and the huge structure of neural network, an adaptive control method based on variable structure neural network (L1) is proposed, in which variable structure neural network is used to identify the existence of the system online , The network dynamically adjusts the structure by awakening and hypnotizing the nodes and effectively approximates with a minimum number of nodes to reduce the computational complexity. L1 adaptive control is used for network weight learning and system nonlinear compensation The feedback loop includes a low-pass filter that ensures that the transient response and steady-state tracking of the system’s input and output signals are consistent with the response of a desired linear time-invariant system as long as L1 gain conditions are met. Through the simulation of quadrotor, the effectiveness of this method is verified.