A Deep Q-Network-Assisted Artificial Hummingbird Optimized PID Controller with Derivative Filter for Blood Glucose Regulation

Abstract

Blood glucose regulation is one of the most critical challenges in artificial pancreas systems due to the nonlinear characteristics of glucose-insulin dynamics and the difficulty of maintaining appropriate controller performance under varying operating conditions. This study proposes a hybrid control strategy combining the Artificial Hummingbird Algorithm (AHA) and a Deep Q-Network (DQN)-assisted PID controller with derivative filter for nonlinear blood glucose regulation. Initially, the controller parameters were optimally tuned using the AHA according to the Integral Square Error (ISE) performance criterion. Subsequently, instead of redesigning the entire controller, the DQN was employed to adaptively determine a weighting factor for the derivative gain during online operation, while the remaining controller parameters remained fixed. The proposed controller was evaluated using the nonlinear APMonitor blood glucose regulation model under identical simulation conditions. The simulation results demonstrated that the proposed approach improved blood glucose tracking performance by reducing the tracking error and oscillatory behavior compared with the conventionally optimized PID controller. Furthermore, the DQN training process exhibited stable convergence within a limited number of training episodes, indicating the effectiveness of the proposed adaptive learning framework. By combining offline global optimization with online adaptive learning, the proposed method preserves the simplicity and robustness of the classical PID controller while enhancing its regulation capability for nonlinear blood glucose systems. The obtained results indicate that the proposed hybrid strategy provides an effective and computationally efficient solution for intelligent blood glucose regulation.