Learning the Dynamic Bayesian Networks Structure of Ventilation Process in a Linear System Identification Perspective

Ventilation process model is important for automatically appropriate ventilation for patients residing in the intensive care unit (ICU). Based on other researchers’ work, we try to build a data driven ventilation model under the framework of dynamic Bayesian networks (DBNs). All the variables in our model are suggested by the doctor in ICU. They are all noninvasive measurements obtained directly from bedside devices and cover nearly all frequently used parameters. Compared with the system dynamic model, our model is comprehensible, compact and simple, it provides doctor the instant needed information; compared with the single input-output model; our model covers more dynamics of the ventilation process; compared with knowledge based model, our model can predict the trend and be used as a simulator as well. The preliminary results show that our model is satisfactory.