Autolanding Control Using Recurrent Wavelet Elman Neural Network

This paper develops a control system with the recurrent wavelet Elman neural network (RWENN) that improves the capabilities of a commercial aircraft to land automatically (autoland) when it is subjected to severe wind disturbances and faults. The proposed RWENN controller is used for the autolanding control, as its real-time learning ability is better than a conventional neural network. The parameters of the RWENN are: 1) translations and dilations of the hidden layer's wavelet functions and 2) the weights between the hidden and output layers. These parameters are learned online using the gradient descent method. The adaptive laws of learning rates are derived from the Lyapunov theorem; hence, system stability can be guaranteed. Moreover, optimal learning rates provide the fastest convergence of parameters. Simulation results show that the RWENN-based control scheme can achieve better performance than other control schemes for the autolanding system in the presence of severe disturbances and faults.