Page 87 - Control Theory in Biomedical Engineering

74 Control theory in biomedical engineering

physiological signals in response to physical activities in the PMM-AP sys-
tem. The controller set-point is set at 110 mg/dL except during exercise
when it becomes 160 mg/dL.
The quantitative evaluation of the closed-loop operation based on the
proposed algorithms is presented in Table 3. The purpose of these simula-
tions is to show that the PMM-AP is robust and reliable in handling signif-
icant disturbances to the BGC. The average percentage of time spent in the
target ranges of [70, 140] mg/dL and [70, 180] mg/dL are 50.4% and 75.4%
for all subjects. There is no hypoglycemic event as the BGC is never less than
70 mg/dL. The minimum and maximum observed BGC values are 73 and
279 mg/dL. The average minimum and maximum observed BGC values
across all experiments during the whole simulation are 88 and 252 mg/
dL, respectively. Overall, the results demonstrate that the proposed
PMM-AP is able to regulate BGC effectively in presence of significant
unknown disturbances caused by the diverse timing and amounts of meals
and exercise specifications while mitigating severe hypoglycemic and hyper-
glycemic excursions. The closed-loop results for all subjects for the last day
of simulations are shown in Fig. 4. The AL-mAP can proactively keep the
CGM values in a safe range during exercise periods due to the learning fea-
ture of the PMM-AP controller (Hajizadeh et al., 2019a, d). This is done by
using a safe (higher) controller set-point in advance, which is defined based
on historical data that predicts the presence of exercise. For one select sub-
ject, the closed-loop simulation results for all 30 days are shown in Fig. 5.
This figure also shows that the PMM-AP guarantees the safety and reliability
of the insulin delivery system especially during exercise periods. The pre-
dicted hypoglycemic episodes warn the user to consume rescue carbohy-
drates about 20 min before the potential hypoglycemic episode. Overall,
the PMM-AP system can also regulate the BGC with a minimum need
for hypoglycemia treatments (Table 4, average 28 hypoglycemic episodes
that necessitate rescue carbohydrates are predicted, about one rescue carb
per day).
In this work, the insulin compartment of Hovorka’s model is integrated
with the recursive subspace identification technique. The adaptive and indi-
vidualized PIC estimates provide accurate information on the amount of
active insulin present in the body and appropriate information for the model
identification. The proposed modeling technique also utilizes other
additional variables such as biosignals to consider the effects of PA on
BGC. Furthermore, this PIC information is used to define limits on the
computed insulin doses. The PIC bounds, along with the risk indexes used