Pragmatic real-time logistics management with traffic IoT infrastructure: Big data predictive analytics of freight travel time for Logistics 4.0
Introduction
It is difficult (or expensive) to consistently and accurately predict travel time (Bartolini et al., 2020), because it varies either exogenously due to traffic congestion, weather conditions, moving targets, or mobile obstacles, or endogenously whenever the decision maker can set the vehicle speeds (e.g., in order to trade-off between fuel consumption and travel time); see, Gendreau et al. (2015), for example. This research is particularly interested in the stochastic nature of travel time (see Kenyon and Morton, 2003, for example) in regards to travel and service times. Beyond the classical formulation of the vehicle routing problem, several studies propose solutions to the time-constrained routing problem with a pre-defined total travel time limit following the least expected time (Gao and Huang, 2012), the legal constraint of traffic laws (Gendreau et al., 2015), and the company's own travel time limits from internal constraints (Karsten et al., 2015). Lefever et al. (2019) treat travel times as uncertain data and model them as random variables that take values in a symmetric and bounded interval around their mean value. Manseur et al. (2020) point out that the formulation of least expected time is risk-neutral without taking into account travel time variations. Travel time on many roads may vary due to changes in traffic conditions and differ with the dynamics of traffic congestion (Carrion and Levinson, 2012), especially during peak hours (i.e., time-dependent effect), which increase the difficulty of obtaining a reliable estimate of travel time. Vidal et al. (2020) note that inadequate management of travel times in routing remains a major barrier to applications, and that producing accurate predictions and performing rapid travel-time queries on large-scale networks raise significant methodological challenges.
Real-time information can enable adaption to change traffic conditions and make better routing decisions in uncertain networks (Gao and Huang, 2012). Several recent studies show that real-time routing approaches can significantly improve decision quality in logistics (Bock, 2019; Ferrucci and Bock, 2016). Real-time logistics refers to the concept of contemporary logistics, e.g., Logistics 4.0 (Winkelhaus and Grosse, 2020), of using ICT (information and communication technology) and modern logistics technology to actively eliminate the delay in management and execution of logistics business processes, thereby improving the response speed, the logistics capabilities, and the competitiveness of enterprises. Real-time logistics is not only concerned with the cost of the logistics system, but also with the speed and value of the overall business system; see Bogataj and Grubbström (2013). It has become the core competitiveness for generating additional profits aside from attracting customer purchases (Niu et al., 2019), particularly in this era of Logistics 4.0.
With information and communication technologies increasingly being applied over a wide array of sectors, coordinating travel route and time is now possible for a given number of traveling vehicles with given origins and destinations (see Speranza, 2018, for example). A common in-vehicle device that supports drivers to select path is based on a digitalized road network map and a Global Positioning System (GPS) aerial (see Lersteau et al., 2016, 2018). Winkelhaus and Grosse (2020) highlight big data-based systems that comprise data-driven decision making (DDDM) in Logistics 4.0, and many studies address the significance of big data in supply chain applications (see Papadopoulos et al., 2017; Dubey et al., 2020, for example). This is also known as data-driven decision management or data-directed decision that involves collecting, organizing, engineering, extracting, and analyzing key datasets (Sun et al., 2015) to (1) inform a decision-making process with a decision extrapolated from insights and projected efficacy of data and to (2) validate a course of action before committing to it (Chen et al., 2020). However, we cannot deny the various challenges brought about by data-driven decision making, particularly with big data (see Lin et al., 2020, for example). Pragmatic real-time logistics was first proposed by Bosch in20191 in order to overcome the shortfalls of document-flow logistics between ERP systems with many media disruptions and different interfaces. Pragmatic real-time logistics is based on easy-to-use and affordable IoT technology that delivers the data from the supply chain processes, enables the material flow to be visible to logistics staff, and empowers employees to actively steer it with real-time tools and services.
To mitigate the impact of unpredictable variations of stochastic travel time and to produce an accurate prediction for maximizing the reliability of freight travel time, we aim to propose a novel data-driven method that combines real-time road traffic information (big traffic data) to accurately predict near-future travel time (at a specific point of time). We focus on the following three research questions in this study:
- 1.
For any conjoint locations, if there are several different IoT devices between them, then how can one accurately predict the travel time through big data collected by the IoT devices?
- 2.
What is the best data-driven method that allows real-time processing without having priori knowledge of traffic flow theory?
- 3.
How robust is the proposed method with respect to different traffic situations to support the pragmatic real-time logistics management of Logistics 4.0?
We propose a new method for the predictive analytics of stochastic travel time – that is, the gradient boosting partitioned regression tree (GBPRT), which separates the global regression model based on the gradient boosting decision tree into several partitions to capture the time-varying features simultaneously. This is done through a recursive partition and a cell of the partition, subdividing the non-linearity into fragments, and characterizing the feature interactions in a manageable way. We conduct an empirical study with real big traffic data to evaluate the proposed method and show its superior performance in comparison with alternative data-driven and model-driven methods.
We organize the rest of the paper as follows. Section 2 first introduces the industry background and relevant literature focusing on three strands of state-of-the-art studies. Section 3 describes the newly proposed gradient boosting partitioned regression tree (GBPRT) for big data analytics. Section 4 presents an empirical study that implements big traffic IoT data to conduct short-term prediction of freight travel time. We show that our proposed method indeed meets the goal of big data analytics in search of manageable tools for logistics professionals. Section 5 discusses our major contributions, implications, limitations, and future works. Section 6 concludes.
Section snippets
Bachground and related literature
This section first introduces the relevant background of current smart traffic systems of interest to the industry and the big IoT data generated therein. Next, we discuss the value of information (and IoT) technology in the literature and its decisive role in the construction of contemporary smart logistics. Finally, we present the methods used in the literature to model and predict random travel events.
The new methodology
Based on the conventional method of gradient boosting regression trees, we introduce a procedure of categorical partitioning to make the traditional regression tree model more flexible in order to cope with complex situations that can be classified on the feature space. Section 1 briefly introduces the conventional regression tree method. We discuss a new procedure of categorical partitioning in Section 3.2 based on the feature space. Integrating the categorical partitioning procedure, we
Empirical study
In this study, we apply the proposed method with traffic IoT data and show its performance. We introduce the data in Section 4.1. We show the computational methods in Sections 4.2. We report the results in Section 4.3 and discuss their robustness in Section 4.4.
Major contributions
Several studies show that proper considerations of time-varying dependence can efficiently eliminate systematic errors and increase predictive precision of dynamic travel time forecasting in comparison with methods that do not take it into account (see Nair and Dekusar, 2020, for example). There are two ways to describe such a dependent structure. One way defines the covariance between observations as a cyclic function of their deviation over time, and another assumes that auto-covariance
Conclusion
This research proposes a data-driven framework for predicting near future travel time with the gradient boosting partitioned regression tree (GBPRT) method, which modifies the existing regression tree method in machine learning, in order to improve real-time logistics management for the requirement of Logistics 4.0. The data-driven framework relies on the data collected from the growing V2X system. We show how data can be leveraged with GBPRT for solving problems on three different levels:
Acknowledgement
The authors express their gratitude to Matthias Hülsmann, Vice President of Connected Logistics of BOSCH Connected Industry, for providing the related information. The authors sincerely thank the three anonymous reviewers for their insightful comments on the manuscript. This work was financially supported by the Center for Open Intelligent Connectivity from the Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education (MOE)
References (97)
- et al.
Impact of transportation lead-time variability on the economic and environmental performance of inventory systems
Int. J. Prod. Econ.
(2014) - et al.
Rule-based autoregressive moving average models for forecasting load on special days: a case study for France
Eur. J. Oper. Res.
(2018) - et al.
Coordination in humanitarian relief chains: practices, challenges and opportunities
Int. J. Prod. Econ.
(2010) - et al.
Revealing personal activities schedules from synthesizing multi-period origin-destination matrices
Transp. Res. Part B Methodol.
(2020) - et al.
Mitigating risks of perishable products in the cyber-physical systems based on the extended mrp model
Int. J. Prod. Econ.
(2017) - et al.
Transportation delays in reverse logistics
Int. J. Prod. Econ.
(2013) - et al.
Value of travel time reliability: a review of current evidence
Transport. Res. Pol. Pract.
(2012) - et al.
Aggregation and travel time calculation over large scale traffic networks: an empiric study on the grenoble city
Transport. Res. C Emerg. Technol.
(2018) - et al.
Adopting a platform approach in servitization: leveraging the value of digitalization
Int. J. Prod. Econ.
(2017) - et al.
Merging anomalous data usage in wireless mobile telecommunications: business analytics with a strategy-focused data-driven approach for sustainability
Eur. J. Oper. Res.
(2020)