ABSTRACT

Forecasting transportation demands can aid online car-hailing platforms to dispatch their service vehicles in advance to areas with more potential orders. This results in a reduction in passengers’ waiting time and better utilisation of transportation resources. However, the complexity and dynamics of multi-dimensional influential factors make the forecasting and dispatching procedures challenging. This paper addresses these issues by using machine learning techniques and an effective probabilistic dispatching strategy. Multiple influential factors were identified in spatial, temporal, and meteorological dimensions, and effective machine learning algorithms were applied to predict the number of passenger orders. The fusion of the multi-dimensional features enables the proposed algorithms to better reveal the spatiotemporal characteristics and their correlations. A sensing-area-based strategy was introduced to dispatch available service vehicles to high demand-intensity regions efficiently with respect to the global demand-supply-balance and the individual probability of receiving orders. Finally, extensive experiments with large-scale real-world datasets were conducted to evaluate the performance of the machine learning algorithms and the effectiveness of the dispatching strategy. Overall, this paper extensively studies the forecasting of the spatiotemporal demand in multiple cities using point-of-interest data and the dispatching of available service vehicles based on such information for online car-hailing platforms.

摘要

预测运输需求可以帮助在线汽车租赁平台提前将其服务车辆调度到具有更多潜在订单的区域。这样可以减少乘客的等待时间并更好地利用运输资源。但是，多维影响因素的复杂性和动态性使预测和调度程序具有挑战性。本文通过使用机器学习技术和有效的概率调度策略来解决这些问题。在空间，时间和气象方面确定了多个影响因素，并应用了有效的机器学习算法来预测乘客的订单数量。多维特征的融合使所提出的算法能够更好地揭示时空特征及其相关性。引入了基于感知区域的策略，以针对全球需求-供应平衡和接收订单的个别可能性，将可用服务车辆高效地分配到高需求密集区域。最后，针对大规模的真实世界数据集进行了广泛的实验，以评估机器学习算法的性能和调度策略的有效性。总体而言，本文使用兴趣点数据以及基于此类信息的可用服务车辆对在线汽车​​叫车平台的调度，广泛研究了多个城市的时空需求预测。引入了基于感知区域的策略，以针对全球需求-供应平衡和接收订单的个别可能性，将可用服务车辆高效地分配到高需求密集区域。最后，针对大规模的真实世界数据集进行了广泛的实验，以评估机器学习算法的性能和调度策略的有效性。总体而言，本文使用兴趣点数据以及基于此类信息的可用服务车辆对在线汽车​​叫车平台的调度，广泛研究了多个城市的时空需求预测。引入了基于感知区域的策略，以针对全球需求-供应平衡和接收订单的个别可能性，将可用服务车辆高效地分配到高需求密集区域。最后，针对大规模的真实世界数据集进行了广泛的实验，以评估机器学习算法的性能和调度策略的有效性。总体而言，本文使用兴趣点数据以及基于此类信息的可用服务车辆对在线汽车​​叫车平台的调度，广泛研究了多个城市的时空需求预测。