融合时空流差的网约车双模式混合调度算法

Dual-Mode Hybrid Scheduling Algorithm for Online Car-Hailing Fusion Spatiotemporal Flow Difference

服务车辆时空分布与出行需求的不一致严重影响网约车平台的服务效率,降低平台和服务车辆的收益以及乘客的服务体验。针对该问题,提出融合空间、时间和天气3种维度影响因素的时空流差计算方法,并构造双层深度森林模型对时空流差进行准确预测。双层深度森林模型通过集成两种不同输入数据的深度森林模型来提升模型预测准确性。基于时空流差预测,设计一种在线局部调度与离线全局调度相结合的双模式混合调度算法。在线局部调度采用集成并行和n阶段求解模式对正在等待订单的车辆进行实时调度,离线全局调度则通过遗传匹配算法对可提前预测的车辆进行离线全局调度。依据遗传算法获取最优路径以及车辆对应子空间的最优匹配值,设计一种迭代Kuhn-Munkres算法和更新机制得到所有车辆和子空间的最优匹配。实验结果表明,该预测模型较其他预测模型解释方差平均提升0.13,确定系数平均提升0.16,平均绝对误差平均减少2.39,均方误差平均减少100.44;调度算法可将全局供需差异降低57.16%,司机接单率提升88.4%。

The inconsistency between the spatial and temporal distribution of service vehicles and travel demands has a substantial detrimental effect on the operational efficiency of ride-hailing platforms.This leads to a reduction in platform and service vehicle revenue as well as an inferior passenger experience.To address this issue,this study proposes an innovative approach that incorporates spatial,temporal,and weather-influencing factors to accurately calculate spatiotemporal flow differences.To improve prediction accuracy,a double-layer deep forest model is developed.The double-layer deep forest model improves the accuracy of model prediction by integrating two deep forest models with different input data.Building on the predicted spatiotemporal flow differences,a dual-mode hybrid dispatching algorithm that combines online local dispatching with offline global dispatching is devised.Online local dispatching utilizes integrated parallelism principles and n-stage solution modes to achieve real-time vehicle dispatching,whereas offline global dispatching uses a genetic-matching algorithm.A genetic algorithm is used to determine the best route and matching values for the corresponding vehicle subspaces.Subsequently,an iterative Kuhn-Munkres algorithm is developed,and the mechanism to determine the best matching for all vehicles and subspaces is updated.The experimental results demonstrate the superiority of the proposed prediction model over existing approaches.The model achieves an increase in the average variance of 0.13,an average determinability coefficient enhancement of 0.16,an average absolute error reduction of 2.39,and an average meansquare error improvement of 100.44.Furthermore,the proposed dispatching algorithm reduces the global supplydemand difference by 57.16%and significantly enhances the driver's order acceptance rate by 88.4%.