Event-triggered model predictive control for embedded artificial pancreas systems
The development of artificial pancreas (AP) technology for deployment in low-energy, embedded devices is contingent upon selecting an efficient control algorithm for regulating glucose in people with type 1 diabetes mellitus (T1DM). The energy consumption of the AP can be lowered by reducing updates of the control model: the number of times the decisionmaking algorithm is invoked to compute an appropriate insulin dose.
1. An artificial pancreas, comprising:
a glucose sensor configured to output glucose data related to a concentration of glucose in the bloodstream of a subject;
an insulin pump;
a memory containing machine readable medium comprising machine executable code having stored thereon instructions for performing a method of delivering insulin; and
a control system coupled to the memory comprising one or more processors, the control system configured to execute the machine executable code to cause the processor to:
store, in the memory, a control action for determining an amount of insulin to inject;
receive, from the glucose sensor, a set of glucose data;
process the set of glucose data to determine whether an update event is triggered;
update the control action to determine a first insulin dose value if the update event has been triggered;
determine a second insulin dose value if the update event has not been triggered, wherein the determination of the second value uses less power than the updating of the control action; and
send a command to the insulin pump to deliver an amount of insulin based on either the first insulin dose value determined by the updated control action or the second insulin dose value.
2. The artificial pancreas of claim 1 , wherein process the glucose data further comprises:
determine whether a suspension event is warranted; and
suspend the insulin pump if the suspension event is warranted.
3. The artificial pancreas of claim 2 , wherein if suspension is warranted, the control system does not trigger the update event.
4. The artificial pancreas of claim 1 , wherein the control action is associated with a Model Predictive Control (MPC) algorithm, and wherein the update of the control action updates the MPC algorithm with state feedback.
5. The artificial pancreas of claim 1 , wherein the update event is an insulin-on-board (IOB) upper bound, and wherein if the IOB upper bound is reached, the pump is suspended and the update event is not triggered.
6. The artificial pancreas of claim 1 , wherein the update event is triggered when an output estimation error norm |Δy k | exceeds a pre-specified threshold, wherein y k is a scalar instantaneous measure at time k of blood glucose level y.
7. The artificial pancreas of claim 1 , wherein the update event is triggered after a first lmax control actions in U 1:N u |k opt are implemented; wherein l denotes a counter initialized at one; wherein lmax is greater than or equal to 1, and less than or equal to N u , or less than or equal to 3; wherein U 1:N u |k opt denotes an optimal sequence of control actions from time step k+1 to k+N u ; and wherein N u is a control horizon of the control system.
8. The artificial pancreas of claim 1 , wherein the artificial pancreas is wearable/implantable in the human body.
9. The artificial pancreas of claim 1 , wherein the control system is configured to be idle at least 55% of its operation time.
10. The artificial pancreas of claim 1 , wherein the second insulin dose value is determined via a stored previous control action.
11. A method of implementing an artificial pancreas using a control system with one or more processors, the method comprising:
storing, in a memory, a sequence of control actions for an insulin delivery pump;
receiving from a glucose sensor, at the control system, a set of glucose data;
processing, by the control system, the set of glucose data to determine whether an event is triggered for updating the sequence of control actions for the insulin pump to output an updated sequence of control actions by determining a first insulin dose value from the updated sequence of control actions;
wherein the event is not triggered,
determining a second insulin dose value via the control system, wherein the determination of the second value uses less power than the updated sequence of control actions; and
sending either the first insulin dose value or the second insulin dose value to the insulin pump.
12. The method of claim 11 , wherein the second insulin dose value is determined via a previous stored sequence of control actions.
13. The method of claim 12 , wherein updating the control actions comprises processing the glucose data for state feedback of updating a model predictive control (MPC) algorithm.
14. The method of claim 12 , wherein processing, by the control system, the set of glucose data to determine whether an event is triggered further comprises using a linear model to predict the blood glucose based at least on the set of glucose data.
15. The method of claim 12 , wherein the memory is flash memory.
16. The method of claim 12 , wherein the control system is on an embedded platform or mobile device.
17. The method of claim 12 , wherein the control system is on an insulin pump.
18. The method of claim 12 , wherein the control system is configured to be idle at least 55% of its operation time.
19. The artificial pancreas of claim 1 , further comprising storing the updated control action in the memory.