The application of direction-basis-function neural networks for a stable recursive nonlinear identification technique

A recursive identification technique for nonlinear discrete dynamical systems is developed in this paper. The technique utilizes the direction-basis-function (DBF) neural nets as a generic discrete nonlinear model structure. DBF nets have enabled the use of some conventional adaptive control ideas to devise a prediction based weight updating rule that guarantees the convergence of both the prediction and weight errors. The key issues, such as the choice of the number of DBF neurons and their parameters, are addressed along with the detailed convergence analysis and a few illustrative examples.