A Predictive Neural Network Control Approach in Diabetes Management by Insulin Administration

In this paper, a neural network based predictive control structure is proposed and applied to treat the problem of diabetes management. The control approach is to predict future plant behavior, hence it specifies accurate control actions necessary to stabilize slow process systems such as physiological systems. The controller adapts to a specific plant and to any changes in its behavior during the control process. When the system was trained using experimental data its overall performance improved continuously during its work. The system proved to be applicable to this particular problem, managing efficiently all selected test cases. It was noticed during the controller test that the diabetic patient is affected by a far history of his blood glucose concentration and insulin treatment profiles, and the controller was able to deal successfully with this phenomena in the control process