Surrogate-based rapid accessibility assessment of Kuiper Belt object transfers

Rapid accessibility assessment is needed for Kuiper Belt Objects (KBOs), because direct multiple-gravity-assist trajectory optimisation is too costly for screening large target populations. This study developed a surrogate-based method for rapid assessment of prescribed KBO transfers. High-fidelity multiple-gravity-assist trajectory optimisation with deep-space manoeuvres provided labels for a Gaussian process regression (GPR) model that predicted the deep-space-manoeuvre-only transfer objective and time of flight from target-state representations. Seven representations were compared, including orbital elements and Cartesian, cylindrical, and spherical states sampled at fixed epochs. A multi-epoch Cartesian representation gave the most balanced performance across prediction and ranking tasks. Logarithmic target scaling improved prediction accuracy in the low-cost region, whereas uncertainty-guided adaptive sampling improved screening precision and recall at the tested 10% and 20% screening fractions. On an independent test set, GPR achieved a global mean absolute transfer-objective error of 0.673 km/s and a mean absolute time-of-flight error of 537.5 days. The mean absolute error was 0.133 km/s for cases whose validated transfer objective was below 1 km/s, and 0.245 km/s for cases selected by the surrogate as having a predicted transfer objective below the same threshold. The larger global mean reflected the sparse high-cost tail of the objective distribution. Applied to 2223 representative catalogue targets, the surrogate identified ten targets with the lowest predicted transfer objectives for subsequent multiple-gravity-assist and deep-space-manoeuvre validation. These results provide a rapid screening approach for the prescribed Earth–Jupiter–KBO transfer scenario.