Task-aware conformal power envelopes for multi-horizon spacecraft monitoring
- 1School of Computer Science, Northwestern Polytechnical University, Xi’an, China
- 2Ningbo Institute, Northwestern Polytechnical University, Ningbo, China
Abstract
Reliable spacecraft power monitoring requires uncertainty estimates across forecast horizons and consistency between primitive electrical states and derived power. We present a task-aware conformal framework that selects joint residual geometry according to downstream power-envelope efficiency. A residual iTransformer produced point centres for five primitive electrical states at 1, 5, and 15 min. Candidate geometries were fitted, selected, and calibrated on chronologically separated complete-orbit blocks, and the final evaluation comprised 25,278 forecast origins from the chronologically subsequent July 2023 telemetry. Across three random-seed runs, the task-aware selector chose a joint box, whereas a residual-volume control selected the sample-covariance ellipsoid. Orbit-balanced primitive-state joint coverage was 0.892 (95% complete-orbit bootstrap confidence interval [CI], 0.883–0.899) at the 90% nominal level and 0.947 (95% CI, 0.943–0.952) at the 95% nominal level. At 90% and 95%, simultaneous coverage of both power channels across all horizons was 0.913 and 0.957. Relative to residual-volume selection, the task-aware method reduced normalized power interval score by 34.3% and 30.3%; the paired absolute reductions were 1.522 (95% CI, 1.505–1.540) and 1.552 (95% CI, 1.534–1.570). Power envelopes were computed from synchronized primitive current and bus-voltage sets. For the selected non-negative box, zero algebraic gap and full interval-image containment follow by construction; the empirical analyses quantify coverage and sharpness. The results demonstrate task-aware geometry selection for BUPT-1 multi-horizon power monitoring.
Related
- SPACR: Single-Pass Adaptive Training of Uncertainty-Aware Conformal Regressors
- Monitoring for Shifts at Unknown Times: Parameter-Free Conditional Conformal Martingales
- Regime-Aware Conformal Uncertainty Quantification for Irregular ADS-B Trajectories in Terminal Airspace
- From fleet trajectories to single-cell promises: hierarchically weighted conformal prediction for battery SOH monitoring
- Missingness-Aware Conformal Prediction Under Cross-Hospital Distribution Shift
BibTeX
@article{lu2026taskaware,
title = {{Task-aware conformal power envelopes for multi-horizon spacecraft monitoring}},
author = {Nan Lu and Lehan Cui and Xinian Zhi},
journal = {PLOS One},
year = {2026},
month = oct,
doi = {10.1371/journal.pone.0360910},
url = {https://doi.org/10.1371/journal.pone.0360910},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Task-aware conformal power envelopes for multi-horizon spacecraft monitoring
AU - Nan Lu
AU - Lehan Cui
AU - Xinian Zhi
PY - 2026
DA - 2026/10/13
JO - PLOS One
DO - 10.1371/journal.pone.0360910
UR - https://doi.org/10.1371/journal.pone.0360910
AB - Reliable spacecraft power monitoring requires uncertainty estimates across forecast horizons and consistency between primitive electrical states and derived power. We present a task-aware conformal framework that selects joint residual geometry according to downstream power-envelope efficiency. A residual iTransformer produced point centres for five primitive electrical states at 1, 5, and 15 min. Candidate geometries were fitted, selected, and calibrated on chronologically separated complete-orbit blocks, and the final evaluation comprised 25,278 forecast origins from the chronologically subsequent July 2023 telemetry. Across three random-seed runs, the task-aware selector chose a joint box, whereas a residual-volume control selected the sample-covariance ellipsoid. Orbit-balanced primitive-state joint coverage was 0.892 (95% complete-orbit bootstrap confidence interval [CI], 0.883–0.899) at the 90% nominal level and 0.947 (95% CI, 0.943–0.952) at the 95% nominal level. At 90% and 95%, simultaneous coverage of both power channels across all horizons was 0.913 and 0.957. Relative to residual-volume selection, the task-aware method reduced normalized power interval score by 34.3% and 30.3%; the paired absolute reductions were 1.522 (95% CI, 1.505–1.540) and 1.552 (95% CI, 1.534–1.570). Power envelopes were computed from synchronized primitive current and bus-voltage sets. For the selected non-negative box, zero algebraic gap and full interval-image containment follow by construction; the empirical analyses quantify coverage and sharpness. The results demonstrate task-aware geometry selection for BUPT-1 multi-horizon power monitoring.
SN - 1932-6203
ER -