The introduction of continuous glucose monitoring (CGM) devices for glucose level measurement accelerated the application of artificial intelligence methods in predicting advanced time blood glucose levels by providing lots of continuous structured data needed to train the methods. Advanced time blood glucose level prediction enables diabetic patients to better manage their blood glucose levels and receive early warnings about the wrong treatments and adverse conditions such as hypoglycemia or hyperglycemia. In this study, an artificial neural network is trained for 30- and 60-min prediction horizon by using physiological models for insulin injection, carbohydrate intake, and physical activity in addition to past CGM data for each of six real T1D patients. The mean of the prediction error for six patients is obtained as 18.81 mg/dL and 30.89 mg/dL for 30- and 60-min prediction horizons, respectively. These results are better than the other studies in the literature that use real patient data, and the model is computationally simpler compared to the deep learning-based methods. Therefore, in this study, a model that can be implemented on the mobile or embedded device, learn the patient's physiologic dynamics, and make accurate predictions during the measurements is developed and presented.