Modeling method for nonlinear stochastic dynamic system based on neural network and extended Kalman filter

This paper presents a modelling method for nonlinear stochastic dynamic system (NSDS) modelling based on a neural network and an extended Kalman filter (EKP). Using this method, the data contaminated by noise can be filtered by the EKP. A dynamic neural network (DNN) which is a good approximation to the deterministic part of the NSDS can be obtained. Meanwhile the DNN can be used as a state estimator for the NSDS.<
>