Hybrid Data-Event Mechanism for Optimal Control of Constrained Overhead Crane Systems

Abstract

This article presents a nearly optimal control scheme in a hybrid data-event mechanism for constrained nonlinear overhead crane systems. Initially, an identifer neural network (NN) is designed to acquire the knowledge of original constrained systems through input-output data. Then, a modified non-quadratic value function is constructed to tackle the input constraints having asymmetric bounds. Meanwhile, a dynamic event-sampling condition is formulated, which facilitates the derivation of an event-sampled Hamilton-Jacobi-Bellman equation (ES-HJBE) tailored for such nonlinear optimal control problems. To solve the ES-HJBE, a unique critic NN is utilized in the reinforcement learning framework, where its weight parameters are updated via a combination of the gradient descent approach and concurrent learning. After that, Lyapunov approach is employed to prove uniform ultimate boundedness stability of the closed-loop system and the critic NN’s weight estimation error. Finally, simulations of constrained overhead crane systems are provided to validate the proposed event-sampled control strategy. Note to Practitioners—Overhead crane systems have been widely utilized in modern industrial and logistics applications, such as automotive manufacturing, mining operations, and warehousing systems. These systems are sensitive to variations in the load weight, the hoisting speed, and the operational coordination. This characteristic poses significant challenges in building accurate models. In addition, designing adaptive controllers for such systems often needs to take mechanical constraints, computational costs, and economic expenses into account. To address these issues, this article develops a reinforcement learning-based optimal control strategy for constrained nonlinear overhead crane systems in a hybrid data-event mechanism. The present optimal control scheme offers two crucial advantages: First, it ensures the stability of constrained overhead crane systems only through input-output data. Second, it reduces the computational burden compared to the conventional static event-sampled control approaches.