基于多模深度森林和迭代Kuhn-Munkres的动态上车点推荐算法  

Dynamic pick-up point recommendation based on multi-modal deep forest and iterative Kuhn-Munkres algorithm

作　　者：郭羽含 朱茹施 Guo Yuhan;Zhu Rushi(School of Science/School of Big Data Science,Zhejiang University of Science&Technology,Hangzhou 310023,China)

机构地区：[1]浙江科技大学理学院/大数据学院,杭州310023

出　　处：《计算机应用研究》2024年第12期3634-3644,共11页Application Research of Computers

基　　金：国家自然科学基金面上项目(12271484);浙江省自然科学基金重点资助项目(LZ22F020007);浙江科技大学青年科学基金资助项目(2023QN022)。

摘　　要：针对现存动态上车点配置模型在大规模算例的全局最优和求解效率方面存在瓶颈的问题,基于乘客步行距离、乘客步行时间、上车点路况指标以及至乘客目的地所需成本四个关键影响因子进行建模,并提出了基于多模深度森林的动态上车点预测算法和一种迭代Kuhn-Munkres上车点配置算法。预测算法融合了多模态决策树结构和深度学习技术以提升模型预测准确性;配置算法通过多场景自适应机制自动调整边权重并选择最优边进行增广,以得到所有乘客和上车点的最优配置。实验结果表明,相较于其他主流预测模型,该预测算法平均绝对误差降低2.705,均方误差降低5.915,可决系数提升0.214,解释方差提升0.195;配置算法在乘客数量占优条件下的平均调度效果相较于实验中其他方案提高了2.04%。这表明预测算法和配置算法具有较高的实用性,且配置算法在处理大规模实例上具有明显优势。To address the bottleneck issues of global optimality and computational efficiency in existing dynamic pick-up point allocation models for large-scale scenarios,this paper developed a model based on four key influencing factors:passenger walking distance,passenger walking time,pick-up point road conditions,and the cost to the passenger’s destination.This paper proposed a multi-modal deep forest-based dynamic pick-up point prediction algorithm and an iterative Kuhn-Munkres pick-up point allocation algorithm.The prediction algorithm integrated a multi-modal decision tree structure with deep learning techniques to enhance prediction accuracy.The allocation algorithm utilized a multi-scenario adaptive mechanism to automatically adjust edge weights and selected the optimal edges for augmentation to achieve the optimal allocation for all passengers and pick-up points.Experimental results demonstrate that the proposed prediction algorithm reduces the mean absolute error by 2.705,the mean squared error by 5.915,increases the coefficient of determination by 0.214,and improves the explained variance by 0.195 compared to other mainstream prediction models.Under conditions where passenger quantity advantage,the allocation algorithm improves average scheduling effectiveness by 2.04%compared to other schemes tested in the experiments.These results indicate that the proposed algorithms are highly practical,with the allocation algorithm shows significant advantages in handling large-scale instances.

关 键 词：上车点推荐 多模深度森林 迭代Kuhn-Munkres算法 网约车 城市交通 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]