Answers are grounded in this paper and cite their sources.
Uncrewed aerial vehicles operating in cold regions face a fundamental scheduling challenge: battery efficiency drops with ambient temperature, causing the effective range to vary dynamically over the day. Existing routing methods, however, treat range as a fixed or static parameter, producing plans that may pass nominal feasibility checks but fail under actual low-temperature execution. This article addresses the multiplatform UAV scheduling problem with time windows and payload constraints, and proposes a temperature-aware adaptive large neighborhood search (TA-ALNS) that converts the temperature-to-capacity relationship into a time-varying effective range and embeds this signal directly into the search-decision layer. Endurance is verified segment by segment at each leg’s departure time using a battery efficiency function, so that battery state participates in route feasibility evaluation rather than serving merely as a post-hoc correction. Three new operators are designed—an energy-risk guided destroy operator, a temperature-aware insertion repair operator, and a temperature-aware platform migration repair operator—that embed range margins, temperature risk, and time slack into search decisions to actively avoid high-risk legs and optimize platform assignments. On Solomon benchmark instances under real winter temperature traces, TA-ALNS achieves lower objective values in 43 out of 48 paired comparisons against the static baseline (STB), with average composite-cost improvements of 19.37% in single-platform and 10.15% in multiplatform scenarios (strongest under deep low-temperature exposure; see Section VI-G). STB solutions exhibit service-failure risks under real low-temperature execution, whereas TA-ALNS effectively circumvents high-risk long-distance legs. Ablation experiments confirm that embedding battery-aware information into search decisions, rather than using it for post-hoc evaluation only, is the key driver of performance gains, with the in-search feasibility/evaluation layer contributing most of this benefit and the temperature-aware operators a further increment. Robustness checks across two additional winter days and seven alternative battery efficiency function families indicate that the direction of the benefit is not specific to a single weather sample or a single battery model—the V-shaped model used in the headline comparison sits at the conservative end of the cross-model spectrum. We further introduce a Thompson sampling-based adaptive operator selection mechanism with principled Bayesian exploration, positioned as a run-to-run reliability enhancement: it reduces the across-run standard deviation by 46%, while remaining at least as good on the best-of-five objective used throughout the article. The method is most effective in single-platform, large-scale scenarios.
No takes yet. Share an insight, caveat, or question.
Fang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: