Developing the Theory of a Model-Based Dynamic Recurrent Neural Network
Marc Karam, Mohamed Ali Zohdy · Proceedings · 2007
The theory lying behind a model-based dynamic recurrent neural network (MBDRNN) previously used to improve the linearized models of nonlinear systems is developed in this paper. The MBDRNN is initially based on the linearized system model, and then is trained to represent the system's nonlinearities by adapting the weights of its nodes' activation functions using back-propagation . The details of the various computations necessary for a successful operation of the MBDRNN are presented.