Non-Reduced Order Method to Parameterized Sampled-Data Stabilization of Inertial Neural Networks With Actuator Saturation
Runan Guo, Yanzheng Zhu, Choon Ki Ahn · IEEE Transactions on Circuits and Systems I Regular Papers · 2024
This article focuses on the stabilization issue of inertial neural networks (INNs) with actuator saturation. Two novel sampled-data stabilization controller design methods are proposed without resorting to traditional variable transformation. The nonlinearity of the activation function significantly influences system performance. Different from the simple bounding techniques used to handle the activation function in the existing sampled-data results for INNs, in this paper, the activation function is represented as a weighted form with weight functions based on the parameterized approach. The designed controller gains depend on affinely transformed weight parameters, leveraging the sector nonlinearity information of the activation function. Constraints on these weight parameters are also considered in the form of linear matrix inequalities (LMIs). By constructing new looped functionals and inequality estimation techniques, two sufficient local stability criteria guaranteeing$H_{\infty }$performance are proposed in the form of LMIs using parameterized techniques. Additionally, corresponding optimization algorithms for enlarging the basin of attraction are presented. Finally, the feasibility of the obtained results is validated through a numerical example with two different classes of common activation functions.