Comparison of multiple Kalman filter and moving horizon estimator for the anesthesia process

In this paper, a new method to estimate the states and the parameters of the anesthesia process is proposed and compared to a Moving Horizon Estimator (MHE) approach. The proposed method makes use of multiple extended Kalman filters (MEKF) where each EKF uses a different set of system parameters whose selection is based on a predictive performance criterion. In view of usage in a closed loop, the comparison between the two methods is based on a metric quantifying the capability of the estimators to predict the future behavior of the system. The metric is also used as a performance measure for tuning the hyperparameters of the estimators. While the results on simulated data are similar, the MEKF method outperforms MHE on clinical data. Tests show that the MEKF method can better predict the future trajectory of the system during the whole induction, on average for all the patients but also for the worst scenario.