Identification of dynamic nonlinear systems using artificial neural networks incorporating a priori knowledge

A prescriptive technique is developed that sets or constrains the weights of a particular layer, designated the gray layer, according to a known or partially known function that is related in some manner to the system of interest. An algorithm is derived for backpropagating the error and updating adjustable parameters within a general gray layer that is consistent with existing learning techniques. As a practical application, gray layer technology is applied to the identification of switched reluctance motor characteristics on the basis of experimental measurements. The results demonstrate that artificial neural networks using gray layers converge faster, require fewer adjustable weights, and more accurately predict the unknown system. As a further illustration, a gray layer is applied to the identification of two differential partially known dynamic systems. Results exhibit a significant improvement in the quality of the identification model and an increase in the rate of convergence