1 School of Intelligent Engineering, Shenzhen Campus of Sun Yat-sen University, Shenzhen 518107, China
2 Shenzhen Information Security Management Center, Shenzhen Municipal Government Service and Data Administration, Shenzhen 518000, China
Show less
文章历史+
Received
Published
Revised
2026-01-03
2026-11-08
2026-05-09
Online Date
2026-11-19
PDF (7188K)
摘要
电动垂直起降航空器(electric vertical take-off and landing,eVTOL)的高效调度是低空交通高质量发展的关键,为解决当前低空调度研究中存在的约束模型构建不全面、传统算法易陷入局部最优问题,开展了管制低空场景下以能耗和延误为优化目标的多目标调度方法研究。首先,基于任务时空、资源分配、序列关系和冲突避免等多维度约束构建了eVTOL调度模型。其次,提出了一种融合强化学习与禁忌搜索的改进多目标遗传算法求解该模型,通过任务串预处理策略降低问题复杂度,引入基于Q-learning的参数自适应策略动态调整交叉与变异率,同时设计基于关键任务串的局部禁忌搜索策略以增强算法跳出局部最优的能力。最后,构建仿真场景,通过实验验证了关键策略的有效性与算法整体的有效性,相较于传统NSGA-Ⅱ算法在反世代距离与超体积指标上分别取得了36.2%与25.6%的提升。研究成果可为低空交通调度系统的优化设计提供参考。
Abstract
Efficient scheduling of electric vertical take-off and landing (eVTOL) aircraft is regarded as the key to high-quality low-altitude traffic. However, incomplete constraint models are found in current scheduling research. Additionally, local optima issues are easily encountered by traditional algorithms. Therefore, a multi-objective scheduling method regarding energy consumption and delay was investigated for controlled low-altitude scenarios. First, an eVTOL scheduling model was constructed. Multi-dimensional constraints were included, such as task spatio-temporal factors, resource allocation, sequence relationships, and conflict avoidance. Second, an improved multi-objective genetic algorithm was proposed. Reinforcement learning and Tabu search were integrated into the algorithm. The problem complexity was reduced by a task string preprocessing strategy. A parameter adaptive strategy based on Q-learning was introduced to adjust crossover and mutation rates dynamically. Meanwhile, a local Tabu search strategy based on critical task strings was designed. The ability to escape local optima was enhanced by this strategy. Finally, simulation scenarios were constructed. The effectiveness of the key strategies and the overall algorithm is verified through experiments. Compared with the traditional NSGA-Ⅱ algorithm, an improvement of 36.2% is achieved in the inverted generational distance (IGD) metric. Additionally, an improvement of 25.6% is obtained in the hypervolume (HV) metric. A reference is provided by these research results for the optimization of low-altitude traffic scheduling systems.
CUI Shi-ya, LIU Qiang, LIU Hui, LIAO Jun-chen, XUE Ke-yi.
Low-altitude Manned Scheduling Method Based on Reinforcement Learning Co-evolutionary Algorithm[J].
Science Technology and Engineering, 2026, 26(31): 13689-13699 DOI:10.12404/j.issn.1671-1815.2600018
设计了融合Q-learning与变邻域搜索的改进NSGA-Ⅱ算法(Q-learning and variable neighborhood search based NSGA-Ⅱ,QVNSGA-Ⅱ)。该算法通过设计问题特化的高效整数编码、引入增强局部搜索能力的变邻域搜索算子,构建基于强化学习的参数自适应调节机制,实现对复杂解空间的有效探索,算法框架如图2所示。
将QVNSGA-Ⅱ与现有算法进行比较,分别为NSGA-Ⅱ、基于分解的多目标进化算法(multiobjective evolutionary algorithm based on decomposition,MOEA/D)和增强型强度帕累托进化算法(strength Pareto evolutionary algorithm Ⅱ,SPEA-Ⅱ),图9展示了四种算法的帕累托解集对比情况。
BauranovA, RakasJ. Designing airspace for urban air mobility: a review of concepts and approaches[J]. Progress in Aerospace Sciences, 2021, 125: 100726.
[2]
ZhangM, ZengY L, WangK, et al. Learned unmanned vehicle scheduling for large-scale urban logistics[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(7): 7933-7944.
[3]
LiW, ChengR G, HuangH H, et al. Energy consumption modeling and optimization of an eVTOL aircraft: integrating weight, motor, and battery dynamics[J]. Energy, 2025, 325: 136229.
[4]
PangY T, ZhaoP, HuJ M, et al. Machine learning-enhanced aircraft landing scheduling under uncertainties[J]. Transportation Research Part C: Emerging Technologies, 2024, 158: 104444.
[5]
PaulS, ChowdhuryS. A graph-based reinforcement learning framework for urban air mobility fleet scheduling[C]//Proceedings of AIAA AVIATION 2022 Forum. Chicago: AIAA, 2022: 3911.
[6]
MurthyS, GaoZ Y, ClarkeJ, et al. A reinforcement learning approach to quiet and safe UAM traffic management[C]//Proceedings of AIAA SCITECH 2025 Forum. Orlando: AIAA, 2025: 2118.
[7]
ShihabS A M, WeiP, JuradoD, et al. By schedule or on demand? a hybrid operation concept for urban air mobility[C]//Proceedings of AIAA Aviation 2019 Forum. Dallas: AIAA, 2019: 3522.
[8]
FatuleE C, ArifM A, WangF Y, et al. A mathematical model for scheduling electrical vertical takeoff and landing vehicle flights[C]//Proceedings of the 56th North American Power Symposium. El Paso: IEEE, 2024: 1-6.
ZhongGang, HuaJunming, DuSen, et al. Research on optimization scheduling of urban low-altitude flight plan based on complex network[J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(11): 531479.
[11]
PradeepP, WeiP. Heuristic approach for arrival sequencing and scheduling for eVTOL aircraft in on-demand urban air mobility[C]//Proceedings of 2018 IEEE/AIAA 37th Digital Avionics Systems Conference. London: IEEE, 2018: 1-7.
[12]
GeG Y. Research and design of eVTOL garage scheduling and management systems for urban air mobility[C]//Proceedings of 2024 4th International Signal Processing, Communications and Engineering Management Conference. Montreal: IEEE, 2024: 270-275.
WangFei, XuHaofan, WangJingshuo. Vehicle-drone collaborative distribution route optimization based on the NSGA-Ⅱ algorithm[J]. Journal of South China University of Technology (Natural Science Edition), 2026, 54(6): 42-53.
Civil Aviation Administration of China. Basic classification method for national airspace[R/OL]. (2023-12-21)[2025-08-23]. http://www.aopa.org.cn/usr/1/upload/file/20231227/17036672816906355.pdf.
LiYaohua, TanNa. Optimization method of aircraft integrated planning based on genetic algorithm[J]. Control Engineering of China, 2017, 24(2): 435-440.
ZhangGuohui, LiZhixiao, ZhangLiping, et al. Reinforcement learning based co-evolutionary algorithm for solving flexible job shop energy efficient scheduling problem[J]. Application Research of Computers, 2025, 42(7): 2039-2047.
[25]
YangY J, LiK X, ZhanH X, et al. A Q-learning-based improved multi-objective genetic algorithm for solving distributed heterogeneous flexible job shop scheduling problem with dual-resource constraints[J]. Expert Systems with Applications, 2025, 261: 125516.