"Dataset for Sequence Surrogate Model-Assisted Multiobjective Optimization of HTS Maglev Systems"

"High-temperature superconducting (HTS) maglevsystems are highly promising for next-generation transportation;however, the design optimization of the permanent magnetguideway (PMG) is severely bottlenecked by the computationallytime-consuming finite element method (FEM) required to resolvethe strongly flux-pinning hysteresis. To bridge this gap, this paperproposes a data-driven surrogate modeling framework basedon Long Short-Term Memory (LSTM) networks. First, builtupon an experimentally validated H-formulation FEM model, adataset is generated using a Maximin Latin Hypercube Samplingstrategy. To address the hysteretic multi-valuedness inherent insuperconducting systems, the dynamic data is decoupled bymotion phases and augmented with directional features. Bycapturing the path-dependent sequential patterns, the proposedLSTM surrogate achieves high predictive accuracy (R2 > 0.999)for levitation force under diverse geometric and operationalconditions, significantly outperforming the XGBoost baseline(R2 > 0.96). Furthermore, by coupling this surrogate with theNSGA-II algorithm, a constrained design optimization of thePMG is executed. The results demonstrate that the proposedframework can reduce the PMG cross-sectional area by 13.2%while satisfying both the levitation and guidance force requirements.This work provides a practical data-driven framework forefficiently optimizing complex engineering systems with stronghysteresis effects."