Closed-control Based On Model Migration For Artificial Pancreas

Diabetes has gradually became a global disease which harms human health and society.The artificial pancreas system is one of the most promising methods which can be used for diabetes treatment.With the development of continuous glucose monitoring system and insulin pump,the hardware part of the artificial pancreas system has satisfied the clinical requiroements.However,the closed-loop control algorithm of the artificial pancreas system design faces huge challenges since human blood glucose regulation process has nonlinear and unstead dynamic with strong disturbances as well as the risk of blood glucose asymmetry.Traditional modeling methods of blood glucose control need sufficient data of patients for model identification.It will impose heavy burden to patients and doctors and the overall modeling cost will be higher for repetitively collecting sufficient data of blood glucose.While traditional closed loop control algorithm is just considering the innovation of method itself without considering the risk of blood glucose value asymmetry enough.Therefore,the focus of this article is how to make use of prior knowledge to reduce the cost of individual modeling and how to reduce the hypoglycemia event of patients.Based on the above motivation,this paper mainly focouses on the following aspects:(1)Arapid and economic modeling method of T1DM patients for glucose control is first proposed in cosidertation of constrains for the glucose control model based on the predecessors' work.The proposed method can develop an appropriate model with a small amount of incentive data for specific subject using the idea of model migration with PSO,which can solve the problem of traditional modeling method for a lot of modeling efforts of using a sufficient incentive data.(2)The method proposed a zone-MPC inblood glucose control algorithm based on model migration with PSO.Experimental results show that the control performance of the proposed method using only a small amount of incentives(two hours)modeling is approximately same as the method using a lot of incentive data(five days)of the ARX modeling method.The proposed modeling method based on the base model algorithm has high efficiency and economy,especially in the absence of blood glucose data of patients.The proposed PSO model migration algorithm can effectively replace the traditional subject dependent modeling method,and can realize the automatic closed-loop blood glucose control combining the zone-MPC control algorithm.(3)The paper proposes a zone-MPC algorithm of closed loop control based on asymmetric risk function and model migration with PSO considering the risk of blood glucose asymmetry.The proposed method can adjust weight coefficient of the hyper/hypoglycemia and control variables in objective function of MPC by designing asymmetric risk function.The proposed method were evaluated using in silico subjects,and the results showed that the algorithm can effectively reduced the frequency of the patients with hypoglycemia,especially in the big diet scenario.