MLP-based modeling and prediction of space tether-net capture probability

This study addresses the challenge that capture-probability assessment in space tether-net capture missions still heavily relies on high-fidelity dynamics simulations, leading to prohibitive computational costs for large-scale parameter studies and real-time in-orbit decision-making. An efficient prediction framework based on a Multilayer Perceptron (MLP) ensemble surrogate model is proposed. First, two continuous geometric indices, the capture enclosure index and the deployment area index, are introduced to construct an interpretable capture-criterion model and to establish a high-quality probabilistic labeling scheme via logistic regression. Second, an MLP-based ensemble surrogate model is constructed using 13-dimensional input features that combine initial launch uncertainties and launch distance. To enhance fitting in critical decision regions, a weighted binary cross-entropy loss is adopted together with an ensemble strategy in which the top-ranked base models on the validation set are aggregated by simple averaging. Experimental results show that the surrogate model achieves a coefficient of determination of R 2 = 0.9466 on an independent test set. Compared with tether-net dynamics simulations, the surrogate achieves a computational speed-up on the order of 10 5 , reducing the per-query cost from minutes to milliseconds. The proposed approach provides an effective tool for system tolerance design, mission planning, and real-time in-orbit decision-making in space tether-net capture applications.