In view of the strong non-linear and time-varying characteristics of linear ultrasonic motors, a fuzzy recurrent neural network controller is put forward. In order to improve the control of the convergence speed and robustness of the proposed controller, the differential evolution algorithm is used for the rapid preliminary off-line optimization of the weights and parameters of network,and then combines the adaptive law and BP algorithm and gradient descent optimization mechanism for learning and accurate optimization. The results show that the fuzzy recurrent neural network controller is better than PID and self-structuring neural network controller, and has a good anti-interference capability.
Read full abstract