Huang, H.-P.H.-P.HuangLiu, S.-W.S.-W.LiuChien, I.-L.I.-L.ChienLin, C.-H.C.-H.LinI-LUNG CHIEN2018-09-102018-09-102010http://www.scopus.com/inward/record.url?eid=2-s2.0-80051708713&partnerID=MN8TOARShttp://scholars.lib.ntu.edu.tw/handle/123456789/356541An artificial neural network (ANN) model for the prediction of glucose concentration in a glucose-insulin regulation system for type 1 diabetes mellitus is developed and validated by using the Continuous Glucose Monitoring System (CGMS) data. This network consists of structured framework according to the compartmental structure of the Hovorka-Wilinska model (HWM), and an additional update scheme is also included, which can improve the prediction accuracy whenever new measurements are available. The model is tested on a real case, as well as long term prediction has been carried over an extended time horizon from 30 minutes to 4 hours, and the quality of prediction is assessed by examining the values of the four indexes. For instant, the overall Clarke error grid (CEG) Zone A value is up to 100% for the 30-min-ahead prediction horizon with update. Therefore, for practical purpose, our results indicate that the promising prediction performance can be achieved by our proposed structured recurrent neural network model (SRNNM). ? 2009 IFAC.Neural network; Type 1 diabetes[SDGs]SDG3Artificial neural network models; Continuous glucose monitoring; Glucose concentration; Long-term prediction; Prediction accuracy; Prediction horizon; Prediction performance; Quality of predictions; Structured recurrent neural networks; Time horizons; Type 1 diabetes; Type 1 diabetes mellitus; Forecasting; Insulin; Process control; Recurrent neural networks; Glucoseconference paper10.3182/20100705-3-BE-2011.0084