Real-world Vehicle Routing Problems (VRPs) are characterized by complex road networks and frequently varying travel speeds. Considering the spatial heterogeneity and frequent temporal variations of travel speeds on real-world road networks significantly increases the computational complexity of solving such problems. To efficiently address such Time-Dependent VRPs (TDVRPs) in real-world networks, this paper proposes a novel attention-based model with an encoder-decoder structure named the Dimension-Reduction Residual Attention Model (DRRAM). The model accepts node coordinates, demands, and travel time information as inputs. Specifically, a dimension-reduction attention mechanism is employed to extract key features from the inputs, thereby enhancing the representational capacity of node feature vectors while reducing computational costs. Furthermore, a residual attention mechanism is introduced to facilitate better gradient backpropagation and accelerate model convergence. The proposed model is trained and tested using extensive TDVRP instances derived from a time-dependent speed network in Chengdu, which comprises 408 intersections, 1250 directed road links, and 240 consecutive time slots. Experimental results show that the proposed model improves performance by up to approximately 7.8% compared to recent advanced DRL models. Moreover, compared to the state-of-the-art Tabu Search algorithm, it reduces computation time to at most 1/1500 and improves solution quality by up to 9.6%.
求解TDVRP问题的常见方法主要包括精确算法、启发式与元启发式算法。精确式算法可以得到理论上的最优解。Liu等[1]在考虑时间窗的绿色车辆的TDVRP问题(Time-dependent Green Vehicle Routing Problems with Time Window, TDGVRPTW)时,引入了TD(Time-dependent)弧这一概念描述服务客户的过程,并可以将无限多的TD弧简化为一组有限的非支配 TD 弧。他们采用分支定价算法求解这一问题,并能在客户数为100的情况下具有有效性。Monemi 等[2]引入逻辑型Benders分解算法(Logic-Based Benders Decomposition, LBBD)进行求解:主问题负责生成满足容量约束的可行路径(忽略时间依赖性),子问题则针对每条路径验证其在时间依赖交通条件下的可行性,并通过生成最优性割或不可行性割反馈至主问题迭代优化。他们以加拿大魁北克市的实际路网和历史交通数据为基础构建实例,验证了算法能够在50的客户规模上能取得最优解。文献[3-5]将路径规划期分为7个时段并采用分支定价算法求解考虑不同问题特征的TDVRP问题,Vidal等[3]解决了非完全连接无向路网上的TDVRP问题,客户数不超过75;Wölck等[4]和Zhang等[5]考虑了最多50个客户数,分别解决了考虑货物拣选和装载用时与考虑货物二维形状限制的TDVRP问题。精确算法虽然可以保证解的最优性,但由于VRP问题的NP-hard属性[6],上述研究都没有求解客户数超过75的问题,而实际中可能面对更大客户数的情况。
LiuY, YuY, ZhangY, et al. Branch-cut-and-price for the time-dependent green vehicle routing problem with time windows[J]. Inf J Comput, 2023, 35(1): 14-30.
[2]
CastellucciP B, DarvishM, CoelhoL C. A Benders decomposition algorithm for the time-dependent vehicle routing problem[M]. Montreal:Bureau de Montreal, Université de Montreal, 2021.
[3]
VidalT, MartinelliR, PhamT A, et al. Arc routing with time-dependent travel times and paths[J].Transp Sci2021, 55(3): 706-724.
[4]
WölckM, MeiselS. Branch-and-price approaches for real-time vehicle routing with picking, loading, and soft time windows[J]. Informs J Comput,2022, 34(4): 2192-2211.
[5]
ZhangX, ChenL, GendreauM, et al. Learning-based branch-and-price algorithms for the vehicle routing problem with time windows and two-dimensional loading constraints[J]. Informs J Comput, 2022, 34(3): 1419-1436.
[6]
LenstraJ K, KanA H G R. Complexity of vehicle routing and scheduling problems[J]. Networks, 1981, 11(2): 221-227.
[7]
ZhangM T, ZhangJ, HeC F. A review of metaheuristic algorithms research[J]. Computer Engineering and Applications,2026,62(2): 40-53.
GeX L, ZhangH. Study on the optimization of vehicle routing problem in urban real time traffic network [J]. Industrial Engineering and Management, 2018,23(3):140-149+156.
WuY, MaZ J. Time-dependent production-delivery problem with time windows for perishable foods[J]. Systems Engineering-Theory & Practice, 2017, 37(1): 172-181.
ZhangJ T, DingY. Simulated annealing with variable neighborhood for time-dependent vehicle routing problem with time window[J]. Operations Research and Management Science, 2019, 28: 77-84.
GmiraM, GendreauM, LodiA, et al. Tabu search for the time-dependent vehicle routing problem with time windows on a road network[J]. Eur J Oper Res, 2021, 288(1): 129-140.
[16]
ZhangX Y, JinP, HuX X, et al. Research on the time-dependent multi-depot open vehicle routing problem with time windows[J]. Chinese Journal of Management Science, 2024, 32(1): 146-157.
FanL. A two-stage hybrid ant colony algorithm for multi-depot half-open time-dependent electric vehicle routing problem[J]. Complex Intell Syst, 2024, 10(2): 2107-2128.
[21]
VinyalsO, FortunatoM, JaitlyN. Pointer Net works[C]//29th Conference on Neural Information Processing Systems, 2015:2692-2700.
[22]
BelloI, PhamH, LeQ V.Neural combinatorial optimization with reinforcement learning[C]//5th International Conference on Learning Representations, 2017: 1-13
[23]
WilliamsR J. Simple statistical gradient-following algorithms for connectionist reinforcement learning[J]. Mach Learn, 1992, 8(3): 229-256.
[24]
NazariM, OroojlooyA, SnyderL, et al. Reinforcement learning for solving the vehicle routing problem [C]//32nd Conference on Neural Information Processing Systems, 2018: 9861-9871.
[25]
KoolW, van HoofH, WellingM. Attention, learn to solve routing problems![C]//7th International Conference on Learning Representations, 2019: 1-12.
[26]
VaswaniA, ShazeerN, ParmarN, et al. Attention is all you need[C]//31st Advances in Neural Information Processing Systems, 2017: 5998-6008.
[27]
HelsgaunK. An extension of the lin-Kernighan-helsgaun TSP solver for constrained traveling salesman and vehicle routing problems[R]. Roskilde: Roskilde Universitet, 2017.
[28]
LiJ, WuG, FanM, et al. Heterogeneous attention-based graph convolutional network for solving asymmetric pickup and delivery problem[J]. IEEE Trans Autom Sci Engin, 2025.
[29]
LiJ, XinL, CaoZ, et al. Heterogeneous attentions for solving pickup and delivery problem via deep reinforcement learning[J]. IEEE Trans Intell Transport Syst, 2021, 23(3): 2306-2315.
[30]
LiX, LuoW, YuanM, et al. Learning to optimize industry-scale dynamic pickup and delivery problems [C]//2021 IEEE 37th International Conference on Data Engineering (ICDE), 2021: 2511-2522.
[31]
HeM L, YangM, WuX H. Vehicle routing problem with simultaneous pickup-delivery and soft time windows from low-carbon perspective[J]. Journal of Jiangsu University(Natural Science Edition), 2026, 47: 283-291.
LeiK, GuoP, WangQ X, et al. End-to-end deep reinforcement learning framework for multi-depot vehicle routing problem[J]. Application Research of Computers,2022,39(10):3013-3019.
XuY, FangM, ChenL, et al. Deep reinforcement learning for solving the heterogeneous capacitated vehicle routing problem[J]. IEEE Trans Cybern, 2022, 52(12): 13572-13585.
[38]
LiJ, DaiB, NiuY, et al. Multi-type attention for solving multi-depot vehicle routing problems[J]. IEEE Trans Intell Transport Syst, 2024, 25(11): 17831-17840.
[39]
LinB, GhaddarB, NathwaniJ. Deep reinforcement learning for the electric vehicle routing problem with time windows[J]. IEEE Trans Intell Transport Syst, 2022, 23(8): 11528-11538.
[40]
HeR, RavulaA, KanagalB, et al. Realformer: Transformer likes residual attention[C]//Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, 2021: 929-943.
[41]
KingaD, AdamJ B. A method for stochastic optimization[C]//Int Conf Learn Represent (ICLR). 2015.
[42]
ZhangD, WallaceSW, GuoZ, et al. On scenario construction for stochastic shortest path problems in real road networks[J]. Transp Res Part E Logist Transp Rev, 2021, 152: 102410.
[43]
GuoF, ZhangD, DongY, et al. Urban link travel speed dataset from a megacity road network[J]. Sci Data, 2019, 6: 61.
[44]
HuangY, ZhaoL, Van WoenselT, et al. Time-dependent vehicle routing problem with path flexibility[J]. Transp Res Part B Methodol, 2017, 95: 169-195.
[45]
WangC, CaoZ, WuY, et al. Deep reinforcement learning for solving vehicle routing problems with backhauls[J]. IEEE Neur Net Lear Syst, 2025, 36(3): 4779-4793.
[46]
Ben TichaH, AbsiN, FeilletD, et al. Empirical analysis for the VRPTW with a multigraph representation for the road network[J]. Comput Oper Res, 2017, 88: 103-116.