Neural networks in local state estimation

The problem of state estimation with partially unknown system dynamics is proposed and discussed. The unknown part of nonlinear functions in system dynamics are approximated by neural networks (NN´s). System state and the NN unknown parameters are estimated simultaneously in real-time without necessity of any off-line training process of NN. Three local estimators were applied to the problem of NN based state estimation. The estimation qualities of the considered estimators are compared within three numerical examples covering a different kind of uncertainty in the model description.