A neural network regulator

The paper presents an architecture for neural network regulators. The back-propagation algorithm has been used in an integral hierarchical structure to perform I-O mapping and adaptation of controller parameters. This avoids the use of the sign of the plant errors during the back-propagation procedure, and prevents the generation of excessive control signals. It does not require a reference model or inverse system model, or the application of any probing signals. The neural network regulator has a compact structure, which can easily be extended to cater for more complex dynamic systems or additional control loops. The regulator has been evaluated by simulation on a detailed nonlinear model of a turbogenerator system. The results illustrate the good performance which may be achieved with the neural adaptive controller