This study is dedicated to developing an optimized insulin dosage management system for patients with Type 1 Diabetes, integrating a novel composite neural network glucose prediction model with a fuzzy PID control algorithm. To address the critical need for precise insulin dosage adjustments amid fluctuating blood glucose levels, we introduce a comprehensive solution that combines advanced predictive algorithms with autonomous control strategies. Our approach employs a hybrid neural network model that merges Convolutional Neural Networks (CNN) and Transformers, enabling accurate predictions of future blood glucose trends The efficacy of our proposed solution was rigorously tested on a group of simulated individuals, including 10 adults with Type 1 Diabetes, using the UVA/Padova simulator. The experimental results indicate a significant enhancement in blood glucose management, as evidenced by a marked reduction in the risk of hyperglycemia—the frequency of hyperglycemic episodes and the duration of high blood glucose states were decreased, respectively 51.1% and 46.9%.