When studying the vehicle routing problem, especially for on-time arrivals, the determination of travel time plays a decisive role in the optimization of logistics companies. Traffic Internet of Things (IoT) connects ubiquitous devices and collects data from various channels like traffic cameras, vehicle detectors, GPS, sensors, etc. that can be used to analyze real-time traffic status and eventually increase the efficiency of logistics management for Logistics 4.0. However, big IoT data contain joint features that interact non-linearly and complicatedly, thus increasing the stochastic nature and difficulty of determining travel time on real-time basis. This research proposes a novel method (named the gradient boosting partitioned regression tree model) to forecast travel time based on big data collected from the industrial IoT infrastructure. The proposed method separates the global regression tree model based on the gradient boosting decision tree into several partitions to capture the time-varying features simultaneously – that is, to subdivide the non-linearity into fragments and to characterize the feature interactions in a manageable way with recursive partitions. We illustrate several analytical properties with manageable advantages in terms of big data analytics of the proposed method and apply it to real traffic IoT data. Findings of this research show that the proposed method performs successfully at enhancing the predictive accuracy of travel time after empirically comparing it with other computational methods.

在研究车辆路线问题时，特别是对于准时到达的路线，确定旅行时间对物流公司的优化起着决定性的作用。交通物联网（IoT）连接无处不在的设备并从交通摄像头，车辆检测器，GPS，传感器等各种渠道收集数据，可用于分析实时交通状况并最终提高物流的物流管理效率4.0。但是，大的物联网数据包含非线性复杂地相互作用的联合特征，从而增加了随机性，并增加了实时确定旅行时间的难度。这项研究提出了一种基于从工业物联网基础设施收集的大数据来预测出行时间的新颖方法（称为梯度提升分区回归树模型）。所提出的方法将基于梯度增强决策树的全局回归树模型分为几个分区，以同时捕获时变特征，即将非线性细分为片段，并以可管理的方式描述特征交互。递归分区。我们根据提出的方法的大数据分析来说明几种具有可控优势的分析属性，并将其应用于实际流量IoT数据。这项研究的结果表明，与其他计算方法进行实证比较后，所提出的方法在提高旅行时间的预测准确性方面取得了成功。