Sensitivity of Ljung´s algorithm for estimating blood flow

Use of Ljung´s modification [1] of the extended Kalman filter in the estimation of pulmonary blood flow has been described elsewhere [2]. Many applications of the extended Kalman filter have been made to systems with parameters that are varying with time. We have adjusted Ljung´s modification to this application, and also added small non-linear correction terms [3]. The evaluation of such identification algorithms for time-varying parameters must eventually reduce to an analysis of their performance in each practical application. The preliminary evaluation in [2] was based on experiments in human beings undergoing surgery, in which the pulmonary blood flow was relatively constant. Here we evaluate the performance of the algorithm on experiments in dogs, where drugs and other means were used to change cardiac output. Also, sources of error are discussed and their effect quantified with regard to the overall error. Sensitivities with respect to parameter variations are computed and graphed. The effect of model order is evaluated. Finally, the effect of starting the algorithm with different initial parameter values is investigated.