Optimal and adaptive estimation using on-line training neural networks

This paper is concerned with optimal and adaptive estimation by using on-line training neural networks. The conventional least-squares estimation algorithms for estimation of random vectors and the algorithms based on the neural networks are compared. The result obtained allows the linear optimal algorithm to be treated as on-line trained linear neural network. The neural estimation algorithms give the common decision of the problem for nonlinear, non-Gaussian case. Adaptive neural state estimator with on-line adaptation scheme is shown. The efficiency of applying the neural networks to the nonlinear estimation problems is investigated by two examples.