Pattern-Based Neural Control for Sampled-data Nonlinear Systems

In real-world industrial control processes, the inherent dynamics and parameters of the system frequently experience fluctuations. As a result, conventional control strategies tailored for a static environment often face challenges in adapting to intricate and dynamic operating conditions. This study focuses on challenges related to pattern-based learning and control in a class of sampled-data systems featuring multiple control scenarios. The control approach consists of two distinct phases, encompassing identification, and another phase dedicated to recognition and pattern-based control. In the initial identification stage, we develop sampled neural network (NN) controllers tailored to various control scenarios. Leveraging the deterministic learning (DL) theory for sampling, NNs accurately approximate the system’s internal unknown dynamics. Following this, a library of knowledge-based candidate controllers is established. In the subsequent identification phase, a set of estimators accurately identify the closed-loop system dynamics under the normal controller. In the phase of recognition and control, swift and precise detection of changes in control scenarios is accomplished through a comparison of the ongoing system dynamics with constructed recognizers, employing the minimum residue principle. Subsequently, the appropriate learning controller is chosen to guarantee system stability and achieve high-performance control. Simulation studies are presented to demonstrate the effectiveness of the proposed scheme.