Approximate Model-based State Estimation in Simplified Receding Horizon Control

Abstract

Multiple variable systems often have to be so controlled that in the lack of appropriate sensors no satisfactory information is available for the complete estimation of their state variables. Normally only their certain components are kept under observation and control, while the other ones evolve according to the consequences of the exerted control signal. In control-based treatment of patients suffering from "Type 1 Diabetes Mellitus (T1DM)", the only directly measured quantity is the subcutaneous glucose concentration in the blood controlled by a single control signal, the insulin ingress rate. The applied model may use several components in the state variable. The traditional "Receding Horizon Controller (RHC)" requires the estimation of the complete state variable for the calculation of the control signal. In this paper preliminary simulations are persented in which the operation of the RHC is studied in the control of two vertically connected, oscillating masspoints so coupled by springs that only the state of the upper one is observed and directly controlled. Instead sensor-based observations, the lower point’s coordinate is calculated by the use of an available "rough" model. Preliminary calculations were made for a particular human glucose-insulin model, too. In the implementation of the RHC special simplifications were introduced. In our further work we wish to apply this method for investigating various T1DM treatment models.