Adaptive critic neural network for feedforward compensation

The paper is concerned with the application of adaptive critic techniques to feedback control of nonlinear systems using neural networks (NN). No initial model for the nonlinear system is necessary. The work shows how to cope with nonlinearities through adaptive critics with no preliminary off-line learning required. The proposed scheme consists of a feedforward action generating NN that compensates for the unknown system nonlinearities. The learning of this NN is performed online based on a signal from a second neural network acting as a critic. The critic NN tunes itself online using the performance measure of the system. Both NN tuning algorithms are based on backpropagation, which must be modified to guarantee closed-loop performance. The two algorithms are derived using nonlinear stability analysis, so that both system tracking stability and error convergence can be guaranteed in the closed-loop system.