Application of neural networks to dynamic system parameter estimation

This paper shows that parameters of a dynamic system can be estimated as an output of a neural network excited by the system response to a predetermined input signal. Performance of the heteroassociative analog memory thus defined is investigated using computer simulated second-order system responses contaminated by Gaussian noise. With a single-hidden layer feedforward network the estimation errors are comparable to those obtained with a standard LSE method, without a need for iterative calculations.