Delay optimal blood glucose control based on reinforcement learning

It is significantly important to develop effective measures for surveillance, prevention and control of diabetes and its complications. In this paper, considering the delay influence of external sugar source input on blood glucose level, we propose a novel blood glucose regulation model with time-delay. Based on this, we present a delay optimal control problem, where the control variables include the rates of glucagon release and insulin release, and the cost function aims to achieve stable blood glucose levels. To solve this problem, we then develop an improved reinforce learning algorithm combining deep Q-network with soft update mechanism. Finally, numerical results show that the proposed algorithm is highly effective in solving blood glucose level control problem, which provides a new paradigm to manage the blood glucose for diabetes patients.