Parameter estimation of the sparse data systems using a smoothed-likelihood estimator

A new approach for the parameter estimation of linear stochastic dynamic models from limited data is described in this paper. The method formally incorporates dynamic process noise as well as output error in defining the estimator, and is motivated by previous work on dynamic model maximum likelihood estimation for sparse data systems. The proposed estimator (smoothed-likelihood estimator) uses a smoothing algorithm to estimate the state of the system and its covariance. Simulation results are presented, evaluating the performance of the smoothed-likelihood estimator, the maximum likelihood estimator, and a regression model estimator.