Container terminal daily gate in and gate out forecasting using machine learning methods
Introduction
In the past decades, the Chinese government has responded to the shifting global economic environment by introducing various economic policies, such as “One Belt One Road”, “Go West.” etc., that have created a new economic landscape. These new policies have also gradually resulted in changes in port governance in China’s seaport system, resulting in a more commercialized and competitive environment (Notteboom and Yang, 2017). Container transportation is progressively becoming one of the essential modes of freight transportation as cargo transport containerization advances (Vasiliauskas and Barysiene, 2008). The container terminal serves as the hub of the maritime transportation network. Various policies and strategies have been established over the past few decades to improve the efficiency of container port operations. Container throughput is wildly recognized as a critical indicator for determining a port’s competitiveness and efficiency (Intihar et al., 2017, Gao et al., 2016, Paflioti et al., 2017). It is utilized to analyze the port’s production and operation status and the consequences of planned execution. It is an ideal foundation for port planning and layout in the future as an indicator for comprehensive port development assessment (Lam et al., 2004). The terminal’s extremely dynamic external environment necessitates a greater emphasis on resilience, adaptability, and flexibility. Even a minor adjustment in port management can have a significant economic impact (Shankar et al., 2019). As a result, terminal managers gradually pay more attention to examining future market trends and forecasting strategies (Paixao and Marlow, 2003, Vonck and Notteboom, 2016). Decision-makers can readily estimate traffic flow in the container terminal over time by using reliable forecasting models (Levine et al., 2009, Petering, 2009).
In a container terminal, a storage yard is an area to store containers temporarily. It consists of marshaling yards and container yards, where the marshaling yard stores containers which are waiting to be transported and loaded to the specified vessel in an orderly manner. Container yards are used for container handover, security checking, classification, and other activities. Container throughput at a terminal includes handling imports, exports, and transshipments. An export container experiences series movement from the shipper to the vessel. It is first loaded by the shipper and transported to the container terminal via a truck. The movement of the trucker delivering the laden container to the container terminal is the container “gate in”. Fig. 1 shows the container flow direction of the gate in (red arrow) and gate out (blue arrow) procedure. Gate is the boundary of a container terminal from the landside (Carlo et al., 2014). The gated in containers are stored in the container yard and eventually loaded to the pre-assigned vessel via quay crane after the vessel arrives. According to the vessel’s estimated time of arrival (ETA), the terminal sets up a gate in deadline for the particular shipment of the vessel. The deadline is generally a few days before the actual time of arrival (ATA), mainly for security checking or document work purposes. It is essential to meet this deadline. Otherwise, the container may get rolled over to the next sailing date, typically one week later. For each vessel, the total gate in containers should be almost equal to the exports at vessel’s arrival. Gate out containers follow the opposite moving flow. After the arrival time of the vessel, the containers are unloaded from the vessel via quay crane, stored in the container yard, and eventually picked up by the consignee. Gate out represents the containers transferred from the container yard to the landside. For each vessel, the gate out containers should be approximately equal to the imports over a time period after vessel’s arrival.
Note that the total number of imports and exports containers over a certain period is the same as the number of gate out and gate in containers. We first access the literature on container throughput forecasting. Many studies have done extensive research on predicting the container throughput. These predictions are generally made on a long-term basis, for instance, one or two years or even longer, for strategic decision making. While this paper focuses on a very short-term forecasting, namely, a one-day based gate in and gate out containers. Specifically, this paper studies the one-day-ahead forecasting of gate in and gate out containers for a container terminal. In the container terminal, gate in and gate out are the main container flows in a port. With this prediction, there will be a good estimation of the yard stock (the sum of gated in and non-gated out containers). These containers are stacked in the container yard, and handling these containers accounts for a major workload at the terminal. Therefore, the short-term forecasting results can directly be used in daily terminal operations, such as the terminal’s allocation and arrangement of workers and machines, and the assignment of equipment, yard spaces, and materials. Additionally, the traditional inventory scheduling and planning strategies such as safety stock, overtime working, and outsourcing are not eligible for a terminal since the capacity and limitation of a terminal are fixed and unchangeable over a short time period. Therefore, forecasts on a short-term basis are essential for the scheduling a container terminal system, and for the operator in decision making and planning. The accuracy of predicting container flow volume contributes to higher utilization of assets, increased reliability, and flexibility, lower lead time and reduced costs of transport chain for containers.
Although there are tons of research on container throughput prediction, there has not been much research focused on the next day gate in and gate out forecasting in the literature to the best of our knowledge. The aim of this paper is to emphasize the importance of gate in/out container forecasting and provide a comprehensive tool for that purpose. We also provide managerial insights based on the proposed forecasting methods and experiment results. Inspired by the literature for container throughput forecasting, we choose a machine learning method, XGBoost, which is well suited for the target problem of this paper. We collect historical data from the database system of Ningbo-Zhoushan port, one of the largest seaports in China. The quantity of gate in and gate out containers in a day is affected by natural factors such as the weather and climatic conditions as well as terminal operational factors such as the estimated arrival time of vessels or gate in/out containers in the past days. In this paper, we put these features into machine learning-based methods to predict the next day gate in and gate out container quantities. The prediction results are compared to a benchmark time series-based method, ARIMA.
We organized this paper as follows: Section 2 provides a review of related literature. We summarize the literature on terminal throughput forecasting methodology and the insights on terminal management. In Section 3, we state the mathematical structure of our selected forecasting methods. Section 4 describes the empirical study at the Ningbo-Zhoushan terminal, the experiment design, computational results, and results analysis. Section 5 concludes the paper and gives recommendations about facilitating new transport policy-making.
Section snippets
Literature review
Predictions of port container throughput have risen in popularity in recent decades. Previous researchers have attempted to use various quantitative methodologies for this topic. We list the literature on forecasting models between 2001 and 2021 in Table 1. In the early stage, error correction models (ECM) (Fung, 2002, Hui et al., 2004) and vector error correction model(VECM) (Fung, 2001) was adopted for container throughput forecasting.
The classical time series models, Autoregressive
Methodology formulation
In this section, we first propose an overall framework of our methodology, in which we decompose the problem into forecasting on each day and each vessel. We then describe the mathematical procedure of the two machine learning methods in the framework.
Data description and experiment design
In this section, we perform an empirical study on Ningbo Zhoushan Port Beilun Second Container Terminal. We collected operational data of the terminal between Jan 10th, 2017. and Dec 25th, 2017 (340 days in total). We divide the 340 days of operational data into a training set of 300 days (40281 records) and a test set of 40 days (5362 records). We note that the data of 2017 is a bit out of date. However, the main contribution of this paper is providing a methodology to predict the daily
Conclusion and future work
This paper proposes a decomposition-ensemble framework to predict the daily gate in and gate out containers at a container terminal. We divide the daily problem into a sub-problem of vessel-based forecasting each day. The intention is to use some related temporal features of vessels in the selected machine learning method XGBoost. Accordingly, we aggregate the result of every individual in the vessel set as the forecasting gate in/out amount for the following day, which is the expected
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
We thank two anonymous reviewers for their insightful suggestions, which have greatly improved the results of this paper. The first author acknowledges the financial support of this research by the Ningbo Science and Technology Bureau Ningbo Natural Science Programme 2021J183, National Natural Science Foundation of General Program 72071116, and Ningbo Science and Technology Bureau 2025 major projects Fund 2019B10026 and 2021Z089.
References (55)
- et al.
Machine learning for international freight transportation management: a comprehensive review
Res. Transp. Bus. Manag.
(2020) - et al.
Storage yard operations in container terminals: Literature overview, trends, and research directions
European J. Oper. Res.
(2014) - et al.
Forecasting container throughputs at ports using genetic programming
Expert Syst. Appl.
(2010) - et al.
A modified regression model for forecasting the volumes of Taiwan’s import containers
Math. Comput. Modelling
(2008) - et al.
Container throughput forecasting using a novel hybrid learning method with error correction strategy
Knowl.-Based Syst.
(2019) - et al.
Forecasting with model selection or model averaging: a case study for monthly container port throughput
Transp. A: Transp. Sci.
(2016) - et al.
A decision support system using soft computing for modern international container transportation services
Appl. Soft Comput.
(2010) - et al.
GMDH-based hybrid model for container throughput forecasting: Selective combination forecasting in nonlinear subseries
Appl. Soft Comput.
(2018) - et al.
A novel hybrid decomposition-ensemble model based on VMD and HGWO for container throughput forecasting
Appl. Math. Model.
(2018) - et al.
Port governance in China since 2004: Institutional layering and the growing impact of broader policies
Res. Transp. Bus. Manag.
(2017)
Box dynamics: A sectoral approach to analyse containerized port throughput interdependencies
Transp. Res. A
A comparison of univariate methods for forecasting container throughput volumes
Math. Comput. Modelling
Effect of block width and storage yard layout on marine container terminal performance
Transp. Res. E
From predictive to prescriptive analytics: A data-driven multi-item newsvendor model
Decis. Support Syst.
A combined approach to forecast container throughput demand: Scenarios for the Hamburg-Le Havre range of ports
Transp. Res. A
A novel hybrid ensemble learning paradigm for nuclear energy consumption forecasting
Appl. Energy
Modeling of container throughput in Northern Adriatic ports over the period 1990–2013
J. Transp. Geogr.
Panarchy within a port setting
J. Transp. Geogr.
Data characteristic analysis and model selection for container throughput forecasting within a decomposition-ensemble methodology
Transp. Res. E
A novel decomposition-ensemble model for forecasting short-term load-time series with multiple seasonal patterns
Appl. Soft Comput.
A game theoretic model and a coevolutionary solution procedure to determine the terminal handling charges for container terminals
Comput. Ind. Eng.
Comparison of ARIMA model and xgboost model for prediction of human brucellosis in mainland China: a time-series study
BMJ Open
Stock price prediction using the ARIMA model
M., time series analysis
Xgboost: A scalable tree boosting system
Competition between the ports of Hong Kong and Singapore: a structural vector error correction model to forecast the demand for container handling services
Marit. Policy Manag.
Forecasting Hong Kong’s container throughput: An error-correction model
J. Forecast.
Cited by (0)
- 1
These authors have equal contributions to this work.