An adaptive plasma glucose controller based on a nonlinear insulin/glucose model

Publisher Website
Google Scholar
Abstract
The design of plasma glucose controllers traditionally relies on linear approaches. The implementation of an appropriate nonlinear model of the insulin/glucose regulatory system into an adaptive controller should predict the insulin-dependent glucose removal more reliably and hence provide better control over a wide spectrum of insulin signals. A discretized form of the model leads to a two-step procedure. First, the measured plasma glucose levels associated with the exogenous glucose infusion rates are used in the estimation of the past removal rates which, in turn, can be expressed as a weighted sum of past insulin inputs and previous values of the removal rate. Parameters of the sum are adjusted on-line by a recursive method of estimation which features a prefiltering of data to account for a corrupting coloured process noise. The same equation is in turn used to predict the time course of the insulin-dependent fractional rate of glucose removal. The performance of the controller, tested in vivo in three pigs, is presented for various intravenous or subcutaneous rapid injections and staircase infusions of insulin. Plasma glucose is maintained at an average level of 99.9 +/- 8.7% of the target value (% set point +/- coefficient of variation). The controller reacts promptly to large and rapid variations in insulin action. Although control improves with the number of glucose measurements, the prediction of glucose removal allows for some flexibility in the monitoring of the plasma glucose. Sampling frequency varied from a 2 min interval during transient periods to 7 min as steady states were reached.

This publication has 12 references indexed in Scilit: