A second order recursive prediction error algorithm for diagonal recurrent neural networks
Yongji Wang, Gregor Fernholz, Sebastian Engell · 2002
A recursive prediction error (RPE) learning algorithm with second order of convergence for diagonal recurrent neural networks (DRNN) is presented. A guideline for the choice of optimal learning rate is derived from convergence analysis based on Lyapunov theory. With application of this method to model a batch distillation column, the results show that the RPE based DRNN has higher modeling precision and requires a shorter computation time compared to backpropagation (BP) based training of multilayer perceptron nets (MLP).