A. Bhattacharjee^1,, A. Sutradhar¹

Modern close loop control for blood glucose level in a diabetic patient necessarily uses an explicit model of the process. A fixed parameter full order or reduced order model does not characterize the inter-patient and intra-patient parameter variability. This paper deals with an online frequency domain kernel estimation method for modeling a nonlinear dynamic system of multivariable glucose-insulin process in a type-1 diabetic patient that captures the process dynamics in presence of uncertainties and parameter variations. The present work proposes a frequency domain kernel estimation of a Volterra model using the harmonic excitation input by taking FFT on the input data sequence from the glucose-insulin process of the patient. Volterra equations up to second order kernels with extended input vector for Volterra model are solved online by adaptive recursive least square (ARLS) algorithm. Twice the length of the extended input vector for the glucose-insulin process is considered for finding the frequency domain kernels that can be directly used as the Volterra transfer function and are useful for closed loop internal model control. The input-output data taken from the 19th order first principle model of the patient in intravenous route, have been used to identify the system with a short filter memory length of M=2 and the validation results have shown good fit both in frequency and time domain responses with nominal patient as well as with intrapatient parameter variations.

System identification, nonparametric model, glucose-insulin interaction, Volterra model, frequency domain kernels, Volterra transfer function