Neural-network-based nonlinear inverse flight control

The use of neural networks (NNs) as an integral part of nonlinear dynamic inverse flight control is examined. The approach is to train the NNs to carry out the inversions from desirable responses to required controls. A backpropagation algorithm was used and several NNs of different complexity were tested. A single-hidden-layer 25-neuron network was selected and trained until the NNs converged. A control is illustrated with simulations using a model of an advanced research aircraft. The results were very good despite indication that the NN did not provide a good inversion mapping