Performance Optimization of Maglev Train’s Electromagnetic Levitation System: Control Structure and Algorithm

Abstract

This article investigates the performance optimization structure and algorithm of the electromagnetic levitation (EML) system. Firstly, a functionalized control structure is proposed by the incremental attachment of a controller gain system. Under this control structure, the predesigned controller is designed for nominal performance while the controller gain system handles performance optimization. This is of great help because the system trajectory of the highly dynamic and open-loop unstable EML system can be contained in the admissible region during the optimization of the controller gain system. Secondly, it is demonstrated that optimizing the incrementally attached controller gain system (under the proposed control structure) for performance optimization is equivalent to optimizing the predesigned controller. The advantage of the equivalence is that the predesigned control system needs not to be modified and the incremental attachment can be performed even when the EML system is running. Furthermore, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Q</i> -learning algorithm is presented for the real-time implementation of the proposed structure and the incrementally attached controller gain system. Finally, the effectiveness of the structure and algorithm is validated on the EML system. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —The EML system is key to maglev transportation. It is open-loop unstable, highly dynamic, and its accurate model is generally unavailable. On the other hand, the levitation performance may not be optimal given the predesigned controller. Moreover, during the long-term operation, it is desired in practice that the controller of EML system can be upgraded in a friendly fashion. This article presents a control structure that optimizes the levitation performance in an incremental style, such that the predesigned controller does not need to be modified. The incremental attachment can be realized via the onboard CAN network or UDP channel. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Q</i> -learning algorithm optimizes the incrementally attached controller gain system without distabilizing the levitation stability of the EML system under the proposed control structure. In such a way, the open-loop and highly dynamic characteristics can be handled and the friendly implementation style can be achieved.