An improved approach for nonlinear system identification using neural networks

The ability of a neural network to realize some complex nonlinear function makes them attractive for system identification. In the recent past, neural networks trained with back-propagation learning algorithm have gained attention for the identification of nonlinear dynamic systems. However, the conventional back-propagation algorithm suffers from a slow rate of convergence. In this paper, we present an improvement to the back-propagation algorithm based on the use of an independent, adaptive learning rate parameter for each weight with adaptable nonlinear function. Simulation results show that the learning speed is increased significantly by making the slope of nonlinearity adaptive since it amplifies those directions in weight space that are successfully chosen by gradient descent. The results demonstrate that the suggested method gives better error minimization and faster convergence.