基于时空预测的多策略网约车调度算法  

Multi-strategy ride-hailing dispatching algorithm based on spatio-temporal prediction

作　　者：郭羽含 钱一炀 钱亚冠 Guo Yuhan;Qian Yiyang;Qian Yaguan(School of Science,Zhejiang University of Science&Technology,Hangzhou 310012,China)

机构地区：[1]浙江科技大学理学院,杭州310012

出　　处：《计算机应用研究》2025年第4期1034-1043,共10页Application Research of Computers

基　　金：国家自然科学基金资助项目(12271484);浙江省自然科学基金重点项目(LZ22F020007);浙江省教育厅科研项目(Y202454572);浙江科技大学教学研究与改革重点项目(2024-JG13)。

摘　　要：准确预测网约车订单需求与实施高效车辆调度策略,对于提升运营效率、降低成本和保证服务质量至关重要,是优化资源配置、增强乘客满意度的关键途径。然而,现存研究在模型构建上往往侧重单一维度分析,调度算法的求解效率及解空间探索能力有待提升,限制了对复杂出行场景的适应性和解决方案的全面性。针对上述问题,构建了时空融合图卷积预测模型,通过集成注意力机制,深度挖掘并利用时空维度综合信息,准确捕捉影响订单需求的隐含特征;同时设计了多策略解搜索算法,基于A*算法生成的代价矩阵选用多样化解搜索策略进行求解,增强了算法在复杂情境下的收敛性与求解质量。基于大量真实网约车数据的实验结果表明,相较于对比模型,所提预测模型MAE平均提升为1.87%,RMSE平均提升为1.92%。多策略解搜索算法较之对比算法,在最优解数量、解超体积以及解间距上平均优化提升率分别为13.88%、32.48%、17.61%,求解效率平均提升21.13%。Accurately predicting ride-hailing order demand and implementing efficient vehicle scheduling strategies are critical for improving operational efficiency,reducing costs,and ensuring service quality.Existing studies often focus on single-dimensional analysis in model construction,and scheduling algorithms exhibit limited efficiency and solution space exploration.To address these issues,this paper developed a spatiotemporal graph convolutional prediction model to integrate attention mechanisms,enabling deep exploration and utilization of spatiotemporal information to capture latent features affecting order demand.Additionally,it designed a multi-strategy solution search algorithm to employ diverse search strategies based on a cost matrix generated by the A*algorithm.This approach enhanced the algorithm’s convergence and solution quality in complex scenarios.Experimental results on large-scale real-world ride-hailing datasets demonstrate that the proposed prediction model achieves an average improvement of 1.87% in MAE and 1.92% in RMSE compared to baseline models.The multi-strategy solution search algorithm outperforms comparison algorithms,achieving average improvements of 13.88% in the number of feasible solutions,32.48% in hypervolume,and 17.61% in spacing.Furthermore,the algorithm’s computational efficiency is enhanced by an average of 21.13%.

关 键 词：需求预测 多目标优化 车辆调度 交通运输 可行解搜索 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP399[自动化与计算机技术—控制科学与工程]