Development of a cognitive blood glucose–insulin control strategy design for a nonlinear diabetic patient model

Abstract The novel cognitive blood glucose–insulin control technique presented in this work uses five layers in the controller’s structure to monitor and control the blood glucose levels of various diabetic patients’ types. The first layer is the cognitive dataset that represents the attributes of the control system. The second layer is the identifier neural network model that represents the different types of nonlinear Bergman diabetic patient models. The third layer is the feedforward neural network controller based on an identifier neural network model to find the maximum insulin-infusion level for each meal in each sample. The fourth layer is the feedback PID-RBF-NN controller based on the radial basis function neural network model to find the optimal insulin-infusion value and to keep the blood glucose level in the normal state. The fifth layer is the nonlinear patient Bergman minimal model. To train this controller, the grey wolf optimization meta-heuristic technique is employed. The results of the MATLAB simulations for three distinct patients showed the effectiveness and robustness of the suggested control algorithm in monitoring the dynamic behaviour of the diabetic patients’ blood glucose levels. Additionally, the comparison results demonstrated that the suggested cognitive glucose–insulin control algorithm improved the time to reach a normal physiological blood glucose level by 10% compared to the fuzzy logic and the fractional-order proportional integral derivative (PID) control algorithms, by 25% compared to the type-2 fuzzy control algorithm, and by 6% compared to the PID-particle swarm optimization control algorithm.