Pattern-based learning and control for a class of sampled-data nonlinear systems

In industrial control processes, system dynamics and parameters often change over time. As a result, traditional control schemes designed for fixed environments struggle to handle complex, dynamic operating conditions. This paper focuses on pattern-based learning and control challenges in sampled-data systems under multiple operating scenarios. The proposed control strategy involves two identification phases followed by a recognition and control phase. In the first identification phase, sampled-data neural network (NN) controllers are designed for different control scenarios. Leveraging deterministic learning (DL) theory, the unknown internal dynamics of the system are accurately approximated by neural networks (NNs). Subsequently, a knowledge-based candidate controller library is constructed. In the second identification phase, the closed-loop system dynamics under the normal controller are precisely identified using a set of estimators. During the recognition and control phase, changes in control scenarios are rapidly and accurately detected by comparing current system dynamics with pre-constructed recognizers, based on the minimum residual principle. Subsequently, a suitable learning controller is selected to ensure system stability and high-performance control. Simulation studies demonstrate the validity of the proposed approach.