In artificial pancreas (AP) systems, continuous glucose monitoring (CGM) data are used to compute the required insulin amount to be infused with an insulin pump to regulate blood glucose concentration of people with type 1 diabetes (T1D). Real-time plasma insulin concentration (PIC) estimations will facilitate calculation of more realistic insulin infusion rates and prevent hypoglycemia caused by overdosing of insulin. Our objective is to develop a method to estimate PIC in real time from CGM and infused insulin data in real-time by using a mathematical model. Thirteen datasets from nine different subjects with type 1 diabetes (T1D) which are based on two euglycemic clamps with (seven datasets) and without (six datasets) an insulin infusion site warming device (IISWD) are used. Hovorka's model that describes glucose-insulin dynamics in different parts of the human body has been incorporated into a continuous-discrete extended Kalman filter (CDEKF) to provide a PIC estimate. Furthermore, because of variability in system dynamics over time, some uncertain model parameters that have significant effect on PIC estimates are considered as new states in Hovorka's model to be estimated by CDEKF. Partial least squares models are developed for the initial guess of the time-varying unknown model parameters used in the nonlinear CDEKF estimator. The performance of proposed method is tested with clinical data by computing the Pearson product-moment correlation coefficient, root mean square error and mean absolute relative error. The method will be beneficial for an AP system with realtime PIC estimates for preventing excess insulin infusions.