The study of congestion factors for optimal entrance gate allocation in a seaport: a micro-level scenario model analysis

Abstract

Purpose

This paper seeks to evaluate the factors that contribute to congestion at port entrances, propose a comprehensive approach to managing port gates that addresses the factors causing traffic jams and assess the outcomes of resolving the issue through an optimal model for incoming container truck traffic.

Design/methodology/approach

The study employed a one-way ANOVA and a one-way MANOVA to examine the impact of congestion-causing factors on the waiting time of trucks in each lane at the entrance gate. The purpose of this was to comprehend the intricate issue and demonstrate the outcomes of the resolution. We used the identified factors that were causing congestion to develop a management strategy for the port gate. As part of this strategy, we implemented a policy where traffic flows in the opposite direction in certain lanes. The Simulation of Urban Mobility program introduced the microscopic traffic simulation model as a discrete event simulation.

Findings

The examination of variables influencing the congestion at the port entrance revealed that there were four factors contributing to the congestion: (1) the quantity of lanes; (2) the level of bookings; (3) the factors related to the traffic signal cycle and (4) the assignment of lane types. The one-way MANOVA analysis of the three factors yielded significant evidence for a single pair of interactions. (1) The factors to consider are the quantity of lanes, the level of booking and the assignment of lane types. If the entrance to the rear alley consists of two lanes with a width of 1.85 at the 50% capacity level, and if the critical value of the uneven queue coefficient is reached, it can result in a maximum reduction of the average waiting time by 15.02%.

Originality/value

This study is unique because it examines the surrounding environment and operational behavior of the port to identify how individual and group congestion factors interact. It uses various statistical tools to determine the allocation of the number of port entrances with a reversible lane policy and appointment level. Additionally, it analyzes the detailed results using microscopic traffic simulation modeling. The established foundational model can assist operators in simulating the queue length and mean waiting time of trucks for this specific waiting line in other ports with comparable entrance characteristics.

Citation

Chuchottaworn, N. and Raothanachonkun, P. (2024), "The study of congestion factors for optimal entrance gate allocation in a seaport: a micro-level scenario model analysis", Journal of International Logistics and Trade, Vol. 22 No. 3, pp. 134-155. https://doi.org/10.1108/JILT-12-2023-0079

1. Introduction

Laem Chabang Port, located in Thailand, holds the distinction of being the largest deep-sea port in the country. It boasts an impressive annual throughput of more than 7,000,000 TEUs. Gate No. 1 at the port is currently facing significant congestion, particularly in the afternoon. This is caused by an imbalance between the varying number of trucks arriving and the limited capacity of the gate (Phan and Kim, 2016). The port stakeholders' efforts to resolve the issue by implementing an advanced booking system have been impeded by the port’s restricted access to technology and resources, preventing them from effectively addressing the problem (Zhao et al., 2015). Hence, it is imperative to investigate techniques for mitigating the ambiguity associated with incoming container trucks (Xu et al., 2021a, b). In addition, each port authority has prioritized the management of traffic lanes and reducing the duration of time that officers spend at the entrances of the ports (Chamchang and Niyomdecha, 2022). The study conducted by Huynh and Chu (2017) proposed a method for alleviating congestion at the entrance of the port. They developed a model to ascertain the optimal number of traffic lanes required. Prior research has focused on analyzing factors in the surrounding environment or policies implemented by ports in different situations, utilizing congestion factors to identify the optimal solution. The objective of this study is to analyze the factors contributing to congestion at the entrance of Laem Chabang Port in Thailand, a major port that shares routes with nearby communities and industrial estates. The findings will inform the development of effective policies to alleviate traffic congestion in this area. We employ a microscopic traffic simulation model to assess the effectiveness of the queue at the port entrance, considering the unpredictable and constantly changing truck arrival patterns.

An analysis of studies revealed that it is still possible to utilize statistical methods in conjunction with traffic simulation models to examine the interplay of factors that influence congestion. Hence, the author postulates that (1) each congestion component exerts distinct effects on the port entry gates, and (2) tailoring the functioning of the port entrance gates according to congestion factors can mitigate the congestion at the pier entrance. These assumptions encompass tactics that can effectively assist ports in managing both pre-booked and non-pre-booked container trucks. We evaluate the mean waiting time and the quantity of trucks in each scenario using the conventional queuing model for comparison. We employ a discrete traffic simulation model to quantify the outcomes, given the intricate nature of the process. The research is segmented into six sections:

• Section 1 presents the necessary context and important concepts for addressing congestion at port gates through the use of microscopic traffic simulation.

• Section 2 provides a comprehensive analysis of previous research on the causes and remedies of port congestion.

• Section 3 provides a comprehensive explanation of the assumptions. Empirical data sets and potential scenario modeling Validation of the model.

• Section 4 provides a comprehensive analysis of how to examine the interplay between individual and group congestion causes. To decide the port entry allocation policy, the results of factors impacting congestion must be utilized.

• Section 5 presents a comparison between the conventional queuing model optimization method and the novel congestion factor gate allocation method.

• Section 6 provides an overview of potential areas for future research.

2. Literature review

This section provides a summary of the pertinent literature on strategies to address congestion at port gates and the creation of different scenarios using microscopic traffic simulation. In addition, we analyze the existing literature on conventional approaches, with a specific focus on investigating the issue of truck queuing congestion at port entrances. Our analysis is centered around container truck management and gate operation.

Several studies have emphasized the problem of congestion in ports globally, specifically regarding the arrival of container trucks (Saharidis and Konstantzos, 2018). The efficiency of container port operations is constrained by port gates, which act as bottlenecks. The influx of container trucks entering the port has led to significant queueing issues at the gates (Essi et al., 2021). While expanding infrastructure is the most efficient way to decrease port congestion, it is also the most expensive and time-consuming. Therefore, analyzing the factors that contribute to congestion in the current operational setting can yield long-term advantages (Xu et al., 2021a, b). In general, the congestion problem at container terminal gates can be alleviated by increasing the number of service lanes. Nevertheless, the limited availability of space and the exorbitant expenses associated with construction hinder the expansion of container terminal gates in numerous ports (Wang et al., 2018). Consequently, identifying the method for transforming flexible door operation has emerged as a practical and cost-effective solution. The study topics can be categorized into two main areas: (1) container truck management and (2) port gate operation.

This applies to the management of container trucks as well as the operation of port gates (Phan and Kim, 2016). The purpose of this guideline is to facilitate the negotiation process for distributing container truck arrival times among multiple trucking companies, with the objective of minimizing the influx of trucks during peak hours (Azab and Eltawil, 2016). A simulation model was developed to analyze the effects of truck arrival patterns and provide operators with strategies to decrease turn times while maintaining port gate performance Chen and Jiang (2016). We introduce a method for managing truck trips that takes into account the timing of truck and ship services. This is achieved by adjusting the time window and conducting numerical experiments to evaluate different strategies. Our study highlights the significance of coordinating planning efforts and the effectiveness of various strategies (Liu et al., 2023). Preemptively scheduling truck appointments can efficiently mitigate congestion. Panida and Hussen (2021) propose a method to break out the activities of port operators in order to analyze and modify container truck inspections.

Regarding the operations and management of terminal gates, Fleming et al. (2013) we introduce an agent-based simulation that represents the integration of road infrastructure enhancements and physical planning strategies as a solution to the problem (Kwateng et al., 2017). In order to mitigate the issue of excessive unidirectional traffic, it is necessary to implement road enhancements and increase the capacity of container parking (Kiunsi, 2013). Expanding the number of lanes at ports during busy periods when the length of truck queues may exceed the capacity of the roads (Grubišić et al., 2020). Research indicates that one potential solution to congestion is the establishment of alliances between ports, allowing for the sharing of port space. This approach has the potential to enhance the capacity for trucks entering the port (Lin et al., 2022). Alternatively, endeavor to regulate the movement of trucks and facilitate the transition of traffic from daytime to nighttime (Nze and Onyemechi, 2018). Port congestion can be attributed to inconsistent planning and regulation, as well as the implementation of technologies that are not compatible with the behavior of truckers.

Nevertheless, the implementation of the reversing lane system has been chosen due to its straightforwardness and efficiency in reducing traffic congestion, without the requirement for significant new infrastructure. As an illustration, Zhao et al. (2015) examined the technique of allocating reversing lanes along a route and optimizing signal control methods as a conventional approach to traffic management at an intersection that is both acceptable and capable of accommodating the future increase in traffic volume en route to the port. Lu et al. (2018) devised a lane diversion technique at traffic lights to mitigate traffic congestion by selectively opening or closing lanes with high traffic volume.

The literature review concludes that existing research predominantly focuses on the management of incoming trucks and the improvement of port gate operations' efficiency. This encompasses tasks such as coordinating the timing of container truck arrivals, optimizing port gate operations, implementing advanced technologies to enhance productivity, and exploring alternative approaches like reversible lanes. Reversible lane timing has the most minimal adverse effect on port operations and stakeholders when compared to other methods. Implementing a suitable approach to regulate the activation of reversible lanes can significantly mitigate congestion at port gates.

This research is distinctive because it examines the correlation between congestion factors using a detailed traffic model at a small scale and the unique characteristics of Laem Chabang Port operations in Thailand. The following disparities are presented: (1) An analysis of the variables that influence congestion at the entrance gate of the port; (2) a suggestion for the distribution of port entrance gates based on the factors that cause congestion. The findings of this study will aid in identifying the variables that contribute to congestion and propose a solution by determining the most efficient number of inbound, outbound, and reversible lanes within the given limitations.

3. Problem statement and assumptions

The operating hours of Laem Chabang Port are continuous, spanning 24 h a day. The port operates in three distinct shifts: 08.00–16.00, 16.00–24.00, and 24.00–08.00. At present, there are a total of 8 service lanes available for entering the inspection gates, which can be accessed through shared access roads (as shown in Figure 1). Only three specific types of container trucks are permitted to enter. 1. Universal cargo trucks 2. Refrigerated container trucks 3. Vacant container trucks. There are both planned and unplanned trucks arriving. At present, none of the inspection gates classify traffic lanes based on the type of vehicle. Consequently, container trucks arrive ahead of schedule in order to vie for entry into the inspection gates and remain in the port vicinity. This behavior leads to significant traffic congestion in the port (Figure 1).

Figure 2 displays the ratio of each category of incoming vehicle. The peak number of vehicles was observed between 15:00 and 16:00 h. The majority, exceeding 80%, of the incoming vehicles consisted of general cargo containers and empty trucks.

Traffic congestion at port gates is a result of random port gate selection behavior and the unpredictable arrival of container trucks. Because most container vans do not make appointments in advance. Complex vehicle movements result in mathematical constraints, whereas discrete event modeling methods can more appropriately handle uncertain and dynamic processes (Trunfio, 2011). Furthermore, the port’s limited understanding of the variables that contribute to congestion will result in erroneous policy choices aimed at resolving the issue.

3.1 Assumptions

As the number of vehicles on the road increases, both the average waiting durations and the number of vehicles waiting will grow more intricate. Consequently, the assumptions will be derived from the most up-to-date operational data. These assumptions will be applied in developing a range of service gate types and levels of advance reservations (Table 1).

3.2 Simulation model

In this study, the Simulation of Urban Mobility (SUMO) software was selected as the detailed microscopic traffic simulation program, customized to match the specific characteristics of Laem Chabang Port in Thailand. The author has collected and analyzed data in collaboration with relevant authorities of the Laem Chabang Port in Thailand to identify factors that might influence congestion at the entrance gates. This approach aligns with the study by Gracia et al. (2017), which investigated factors contributing to delays in port operations. allocation of entrance gates took into account the results of the congestion-related factors. The internal operational structure of the port is depicted in Figure 3.

Table 2 displays the percentage of each category of truck that arrived at the port between 08:00 and 20:00. This data was gathered by the gate officers of Laem Chabang Port during the study period. The analysis revealed that dry trucks accounted for the largest proportion, with a total of 2,912 trucks, representing 48.01% of the total. Empty container trucks followed closely behind with 2,669 trucks, making up 44.06% of the total. Temperature-controlled container trucks had the smallest proportion, with only 485 trucks, accounting for 7.93% of the total. Additionally, the period between 13:01 and 18:00 had the highest volume of trucks.

The researcher examines the distribution of truck arrivals. The test cycle was partitioned into 12 intervals, with each interval having a duration of one hour. The p-value was determined using the Kolmogorov-Smirnov test, with the following assumptions:

H0.

The sample follows a distribution Fitting

Ha.

The sample does not follow a distribution Fitting

Table 3 displays the analysis of the vehicle arrival distance for each time period using the Kolmogorov-Smirnov method. By segmenting the time period into one-hour intervals, ranging from 08:00 to 20:00, we discovered that the p-value for each interval surpassed the significance threshold of 0.05. Thus, we fail to reject the null hypothesis (H0), indicating that the distribution conforms to the required type. The skewness and kurtosis test conducted on the complete data set produced results around ±1.96.

Upon analyzing the bar chart (histogram) of the data set for each time period, it was observed that the landscape exhibits a normal distribution (normality) as the majority of the data is concentrated in the central region of the chart. The box plot for each time period displays a symmetrical distribution of data for vehicle arrival times. The majority of the data is concentrated at the middle of the chart. The two components of data, namely the upper quantities and lower quantities, exhibit comparable proportions. The data used for testing exhibits a symmetrical distribution. The entire data set adheres to the normal distribution. Hence, the traffic model that has been created can utilize the data distribution throughout all 12 time periods as its input.

Table 4 presents the data regarding the proportion of door selections made by trucks when they enter the port. Upon examination, we observe that the eight gates display different degrees of selection. Gate B has the highest selection proportion at 18.70%, while gate E has the lowest selection proportion at 9.86%. Typically, there are distinct traffic lanes for scheduled and unscheduled vehicles at each entrance gate area. Out of all the container trucks that enter the area, 89.25% do not have scheduled appointments, while the remaining 10.75% do.

3.3 Model validation

Table 5 presents the validation of the model by applying the calibration criteria from the Design Manual for Roads and Bridges (DMRB) to calculate the Geoffrey E. Havers (GEH) value. The GEH value is a traffic engineering statistic used for calibrating traffic volume. The analysis indicates that the simulation results of the developed model exhibit a satisfactory level of error, specifically with a GEH value of less than 5. Consequently, we can utilize it to strategize alternative scenarios that improve the functioning of the entrance gate (Chanchaichujit et al., 2016).

4. Experimental design

Building upon the assumptions and principles for mitigating congestion issues at the port’s entrance gates as discussed earlier, this section presents an experimental design approach used to analyze congestion factors for gate allocation at Laem Chabang Port in Thailand. The results of the congestion analysis will be utilized in the design of gate allocation. Microsimulation models of traffic flow will be employed to process the results under varying scenarios.

• Step 1: Problem Definition and Identifying Bottlenecks

The first step involves defining the issues and delays occurring simultaneously at the port’s entrance gates while considering relevant constraints in the gate allocation context. To do this, statistical data and field surveys conducted at Thailand’s Laem Chabang Port will be utilized.

• Step 2: Review of Literature and Data Collection

In the second step, a comprehensive review of past literature is conducted, and data is collected in collaboration with port authorities to identify factors influencing congestion at the port’s entrance gates. The analysis of the relationship between congestion factors is carried out using analysis of variance (ANOVA) and Multivariate Analysis of Variance (MANOVA).

• Step 3: Building Microscopic traffic simulation Model

The third step involves constructing a microscopic traffic simulation model using the Simulation of Urban Mobility software. This is accomplished by integrating geographical data from the Geographic Information System (GIS) program, QGIS, which contains road network data leading to Laem Chabang Port and its surroundings. In order to create the base model, the current inspection gate locations are defined, along with the traffic patterns of all four vehicle types. Subsequently, the model is subjected to testing for accuracy and precision.

• Step 4: Employing Results to Define Alternative Scenarios

In the fourth step, the outcomes of the congestion factors are utilized to establish alternative scenarios. This process is repeated for all possible situations and includes the determination of advance booking levels and parking locations both inside and outside the port area. Drawing from prior research, it was found that gate-sharing strategies can be highly effective in alleviating congestion (Wang et al., 2022). Hence, these methods serve as valuable indicators for application and performance comparison in this study.

• Step 5: Comparative Analysis of Experiment Results

In the fifth step, the experimental results are compared through microscopic traffic simulation models between the base model and various scenario models. This comparison considers average waiting times and average queue lengths at the port entry gates for different time periods.

4.1 Examination of variables influencing port congestion

4.1.1 Number of traffic lanes (factor A)

Currently, Laem Chabang Port is situated on a level terrain, featuring eight traffic lanes for incoming vehicles and three lanes for outgoing traffic. The inbound side of the inspection gates is divided into two staggered levels, each containing four gates (booths). The outer checkpoint consists of inspection gates E-H, while the inner checkpoint consists of inspection gates A-D. Thus, the researcher will modify the ratio of the increase in traffic lanes on the inbound side by 10–30% based on the sensitivity analysis results of the model, in order to examine the interaction of various factors (Chamchang and Niyomdecha, 2022).

4.1.2 Vehicle reservation levels (factor B)

The trucks are scheduled to arrive at the inspection gate according to their appointment time, which ranges from 10% to 30% based on the sensitivity analysis results of the model, in order to test the interaction of factors. Modifying the booking level will have a distinct impact on the distribution of both categories of trucks in each situation. Trucks that have scheduled appointments will have a greater ratio and the option to utilize the inspection gate on any route. Trucks that do not have appointments will have a reduced proportion and may opt to exclusively utilize inspection gate E-H. This measure prevents the trucks with scheduled appointments from forming a line together with the trucks without appointments, and it does not affect the accuracy of the model being analyzed (Torkjazi et al., 2018).

4.1.3 Traffic light cycles (factor C)

At present, container trucks entering any of the gates at Laem Chabang Port in Thailand must go through intersections that are regulated by stationary traffic lights. Field data indicates that traffic volumes at intersections on both the inbound and outbound routes of the port exhibit significant variability, with traffic light cycles lasting between 37 and 52 s per cycle. In order to examine the factor interaction, the researchers decided to augment the traffic light cycles at the intersections surrounding Laem Chabang Port by 10–30% (Zhao et al., 2015).

4.1.4 Gate operation ability (factor D)

Our investigation revealed that the eight doors' operational capability could handle different types of trucks and turnaround times. The scheduled trucks took 55–65 s, while the non-scheduled trucks took 90–150 s. Consequently, the researchers augmented the door operating capacity by 10–30% in order to examine the correlation between these variables (Chamchang and Niyomdecha, 2022).

4.1.5 Parking space size (factor E)

Currently, the parking capacity within Laem Chabang Port is distributed as follows: 71.04% inside the port, 22.34% on the roadside within the port, and 6.62% outside the port before entering. Consequently, the researcher has modified the dimensions of the parking area within the port by a range of 10–30% in order to examine the interplay of the variables (Xu et al., 2021a, b).

4.1.6 Traffic lane type assignment (factor F)

At present, the Laem Chabang Port has established specific criteria for the selection of the initial gates. Trucks with appointments have the option to utilize the inspection gates on all routes. Nevertheless, trucks lacking appointments are restricted to utilizing inspection gates E-H exclusively. Consequently, the researcher has augmented the ratio of traffic lanes designated for trucks without appointments by 10–30% by introducing gates A and B (Gracia et al., 2017).

4.1.7 Types of vehicles sharing the route (factor G)

At present, both local residents' private vehicles and container trucks use the same route to access Laem Chabang Port. Thus, in order to assess the potential effects of these factors, the researchers augmented the percentage of regular vehicles on the route by 10–30% (Habib, 2013).

4.2 Analysis of variance of congestion factors with ANOVA technique

This study proposes a hypothesis to examine the elements contributing to congestion in port entry allocation. A microscopic traffic simulation model is utilized to observe the effects of modifying these factors by dividing the test group into many groups. The number 7 is determined based on the count of factors and assumptions.

• H0: The average waiting time and average queue length were not different among all groups tested.

• Ha: The average waiting time and average queue length were different for all groups tested.

In accordance with the results of the model sensitivity analysis in the previous section, we will test each group of models by adjusting the proportion of the factor increase from 10% to 30%. The study posits that the mean waiting time and queue length are equivalent across all groups under examination. In all tested groups, the 7 congestion factors exhibit a variance in both the average waiting time and the average queue length that deviates from the base model. The data collection is divided into 15-min intervals from 13.00 to 18.00, with each interval containing 20 data sets. We employ the ANOVA technique to examine the variation among simulations from different groups in order to assess the relationship between the groups. This statistical tool is well-suited for analyzing group variances and is commonly used in research operations with a confidence level of 95%, which is a standard criterion (Gracia et al., 2017).

We anticipate that specific factor groups will have a unidirectional effect on the congestion at the entrance of the container truck inspection gate. The ANOVA technique will be employed to compare the critical F value from the degree of freedom table of the F-Table with the results obtained from the traffic model, which tested the proportion of the increase in the factor of 10–30%. The analysis indicates that the computation of F0.5,6,160 produces a value of 13.06 for the quantity of vehicles in the waiting area and 6.54 for the mean length of the queue. The critical F value of 2.16 is equal to the value found in the degree of freedom table for F-Table with values of 0.5, 6,160. Thus, we reject the null hypothesis (H0) and accept the alternative hypothesis (Ha) when the calculated F value surpasses the critical F value. This demonstrates the existence of at least one distinct pair of data points, indicating the presence of at least one factor that yields different results for the average waiting time and average queue length compared to the other groups that were tested. Consequently, the researcher performed experiments by segregating each set of data in order to analyze the fresh outcomes. Ultimately, it was determined that the computed F value was lower than the critical F value (computed F < critical F). Consequently, it was imperative to exclude the data from groups 4, 5, and 7 (factors D, E, and G) as they exhibited distinct outcomes compared to the other factors examined (Table 6).

Upon reprocessing all four factors, we determined that the values F_{0.5, 3, 80} obtained from the calculation were equivalent to 1.69 for the number of automobiles waiting and 1.13 for the average waiting length. The critical F_{0.5, 3, 80} are equivalent to 2.72 when comparing them to the degree of freedom table values of F-Table. Thus, if the calculated F value is lower than the critical F value (calculated F < critical F), we accept the null hypothesis (H0), which suggests that the period is The mean waiting time and queue length exhibited uniformity among all examined groups. The findings are displayed in Table 7.

Table 8 displays the outcomes of the analysis of variance conducted on the variables of average waiting time and average queue length, which are anticipated to impact the congestion at Gate No. 1. All factors were found to have a p-value less than 0.05, indicating that the average waiting time and average queue length were not statistically different among all tested groups. The congestion value was influenced by the following factors, listed in ascending order of impact: The factors to consider are: (1) Number of lanes (Factor A), (2) Reservation level (Factor B), (3) Traffic signal cycle factor (Factor C), and (4) Lane type assignment (Factor F). Thus, it can be inferred that the reversible lane management of average waiting time has a significant impact on the considerable variation observed in the four aforementioned factors.

Table 9 demonstrates how the 1-way MANOVA technique combines the four factors for analysis, assessing their variance, error, and correlation to establish whether the response variable is associated with the number of queues and waiting times of vehicles entering the inspection gate. The researcher opted to employ the Wilks test to ascertain the impact of each pair of factors on the various responses. Table 9 displays the p-values for the combinations of factors that were examined. At the 0.05 significance level, there was strong evidence of interaction between two pairs: (1) the number of lanes (Factor A) and booking level (B), and (2) the booking level (Factor B) and lane type assignment (Factor F). Nevertheless, the number of lanes (Factor A) and traffic signal frequency (Factor C) had the least impact on all response variables.

Significant evidence was discovered for one pair of interactions in the three-factor analysis: the number of lanes (Factor A), reservation level (C), and lane type assignment (Factor F) at the 0.05 significance level. The number of lanes (Factor A), reservation level (B), and traffic light cycle (Factor C) were found to be the least significant factors across all response variables.

The mean responses were evaluated for differences across various factor levels using eigen analysis. The Eigen analysis results revealed a significant disparity in the sum of the Eigen values of queue length. The pairwise factors exhibited an interaction of 1.025, while the groupwise factors showed an interaction of 1.215. The average waiting time exhibited minimal variation across the majority of factors, with a negative difference of −8.868 for the pairwise factors and a difference of 0.0096 for the groupwise factors.

4.3 Revers lane policy

The reversible lane arrangement is a strategy used to optimize road capacity by increasing the number of lanes in a direction with high traffic volume and non-intersecting traffic flows. A reversible lane configuration is appropriate for situations where the traffic volume in both directions is distributed at ratios of 70/30, 65/35, or 60/30. Hence, the implementation of the reversible lane concept will bring about a novel strategy for managing port gate operations, considering the disparity between the number of vehicles entering and exiting the port (Liu et al., 2014).

This study employs the unbalanced queuing coefficient to assess the queuing situation at the port entrance and ascertain the appropriate time for RL to alter the traffic direction. The variable CR, or congestion ratio, is calculated by dividing the average queue length in the inbound direction by the average queue length in the outbound direction. Once the unbalanced queuing coefficient, represented by the variable CR*, reaches a specific value known as the critical value, the traffic lane will switch direction, as long as there are no vehicles in the lane.

Nevertheless, in order to mitigate the issue of frequent changes in direction during actual tasks, there exist techniques to regulate the traffic direction of reinforcement learning (RL) as outlined below: (1) Modify the indicated lane direction to be inbound when the current ratio (CR) is greater than the reference current ratio (CR*); (2) Modify the indicated lane direction to be outbound when the CR is less than the reciprocal of the reference current ratio (1/CR*); and 3) Keep the lane direction unchanged when the CR is between the reciprocal of the reference current ratio and the reference current ratio (1/CR* ⩽ CR ⩽ CR*). The researcher conducted a study on the exit route of Laem Chabang Port in Thailand and found that there are currently two lanes available for vehicles to change their route against the current. At present, the port lacks a policy regarding the utilization of traffic lanes in the opposite direction of the current. The recommendations from the port authorities indicate that the feasibility of operating traffic lanes against the current can be assessed by testing two lanes simultaneously, exclusively for trucks with scheduled appointments (Figure 4).

4.4 Model principles

The procedure verifies the state of the reversible lane activation upon the arrival of a container truck at the port. If the ratio of the average length of the line in the incoming direction to the average length of the line going outward (CR) exceeds the critical value (CR*) of the unbalanced queuing coefficient, the reversible lane is opened for incoming traffic at the port. If Figure 5 is not present, the reversible lane will not be activated.

4.5 Initial decision variable data definition

The heuristic solution process initialization necessitates the identification of four decision variables (nin,nout,nRL, and CR*). The initialization of the decision variables has an impact on two aspects of the port gate-related processes: the number of vehicles entering and exiting the port, which determines the activation or deactivation of the initial number of reversible lanes, and establishes the initial critical value of the unbalanced queuing coefficient (CR*) at three specific values: 2.33 (70:30), 1.85 (65:35), and 1.50 (60:40). The four decision variables are the data set used to solve the problem. The initial values of the four decision variables are defined as ⁽⁰⁾ = (nin(0), nout(0), 0, CR*⁽⁰⁾), with the parameter definitions provided in Table 10. The optimization model follows.

(1)Min WT={∑i=1N(Si)(Win+Wout+WRL)} T

(2)Win=Wg−in+Wf−in

(3)Wout=Wg−out+Wf−out

(4)WRL=Wg−RL+Wf−RL

(5)S.t. nin+nout+nRL=N

(6)nin≥nout

(7)Si=(nin,nout,nRL,CR*),i=1,2,. . .,N

(8)CR* > 1

The primary goal of the objective function (1) is to minimize the total waiting time in both the inbound (first segment) and outbound (second segment) directions from Laem Chabang Port, Thailand. Additionally, it aims to minimize the total waiting time in the activated reversible lane (third segment).

Equation (2): The average gate operation waiting time per truck in the inbound direction is the sum of the average waiting time for vehicles entering the gate and the average waiting time for vehicles entering the gate in the inbound direction.

Equation (3): The mean waiting time for gate operation per truck in the reversing lane is equivalent to the mean waiting time of vehicles entering the gate plus the mean waiting time of vehicles approaching the gate in the reversing lane.

Equation (4): The mean waiting time for gate operation per truck in the outbound direction is equivalent to the mean waiting time of the vehicles entering the gate plus the mean waiting time of the vehicles approaching the gate in the outbound direction.

Equation (5) restricts the total number of open outbound, inbound, and reversible lanes to a fixed value.

Equation (6) imposes a constraint that the channel capacity in the inbound direction of the port must not be smaller than the channel capacity in the outbound direction.

Equation (7) assigns values to the four decision variables (nin,nout,nRL, and CR*) within the range of 1 to N.

Equation (8) demonstrates the necessary condition for the unbalanced Q coefficient to be positive, referred to as the critical value.

Based on the given information, it can be concluded that the initial number of inbound lanes (nin(0)) is 5, the number of outbound lanes (nout(0)) is 3, the initial number of reversible lanes (nRL(0)) is 0, and the initial critical value of the unbalanced queuing coefficient CR*⁽⁰⁾ is set at 2.33 (in a 70:30 ratio). Hence, the initial decision variables ⁽⁰⁾ are (5, 3, 0 และ 2.33).

The model’s solution comprises four decision variables (nin,nout,nRL and CR*). The model process will repeatedly simulate the outcomes by modifying the values of the four decision variables using the following steps:

(1) Specify the starting value of the computation interval T=T0. (2) Establish the initial settings for the decision variables nin,nout,nRL and CR*, represented by the variable ⁽⁰⁾. (3) Simulate the outcome of the situation using variable ⁽⁰⁾ to determine the value of Si. (4) To determine the value of Si, iterate through the steps starting from n = 1 and continuing until n reaches R., (5) Cease the processing and provide a concise summary of the outcomes.

5. Experimental results

The authors utilized the findings from the 1-way MANOVA technique to ascertain the solution approach through the implementation of the reversible lane policy method. We developed specific scenarios for each simulation model to assess potential variations in waiting time and average queue size at the port entrance. These results were then compared to the current operation, which serves as the baseline model. Table 11 presents a concise overview of the experimental findings.

According to the data presented in Table 11, among the six suggested solutions, the solution ⁽⁵⁾ = (5,1,2,1.85) at a 50% booking level is the most appropriate. It has an average waiting time of 41.02 min/vehicle for the inbound lane and 7.04 min/vehicle for the outbound lane. The cumulative waiting time for both inbound and outbound directions is 48.06 min per vehicle. Thus, it can be inferred that the optimal average waiting time of the port gate system (CT) occurs when the decision variable u is set to (5,1,2,1.85), the critical value of the unbalanced queue coefficient is 1.85 (with a ratio of 65:35), and there are two reversible lanes. When comparing the initial solution ⁽⁰⁾ =(5,3,0,2.33) with the most suitable solution, it is found that the average waiting time can be reduced by 15.02%.

6. Summary and conclusions

This study examines the factors that contribute to congestion at the port entrance gate by analyzing the factors that impact port congestion. Analyzed using a microscopic traffic simulation model, the results of the study are based on data collected from the Laem Chabang Port unit in Thailand. The microscopic traffic simulation model utilizes data gathered by the port personnel, and statistical tests are employed to validate the model’s accuracy. The model’s results are assessed, and the influence of each method is determined in various scenarios based on the average waiting time and the number of trucks in the queue. A summary of these findings is provided.

Based on the research findings, we can infer that each congestion factor has an impact on the congestion at the port entrance, specifically in terms of the average waiting time of all the groups tested. This observation aligns with the initial hypothesis. Implementing a reversible lane policy and determining an appropriate appointment level can greatly decrease the average waiting time (Chen et al., 2013).

This study employed one-way ANOVA and one-way MANOVA statistical analyses as the primary methods to investigate the factors influencing congestion at the port entrance. There was substantial evidence supporting two pairs of interactions: 1. Number of traffic lanes (Factor A) and Vehicle reservation levels (Factor B), and 2. Vehicle reservation levels (Factor B) and Traffic Lane Type Assignment (Factor F). In the three-factor group analysis, one pair of interactions showed significant evidence. 1. Number of traffic lanes (Factor A), Vehicle reservation levels (Factor B) and Traffic Lane Type Assignment (Factor F) at a significance level of 0.05. Hence, it can be inferred that there exists a clear correlation among the aforementioned three factors. Hence, in order to alleviate the congestion, it is imperative to consider all three factors when formulating the problem-solving policy, which may vary in each port.

The acquired findings are utilized to establish the design principles for the solution, with the objective of minimizing the overall waiting duration. We construct an optimal model to ascertain the most advantageous quantity of reversible lanes and the most advantageous scheduling rules at the port gate. Next, we suggest employing a heuristic approach to address the optimal model. We utilize the microscopic traffic simulation model to ascertain parameters that cannot be determined through conventional analytical methods. To accurately determine the number of reversible lanes at the reservation level of 10–40%, it is recommended to have 5 inbound lanes, 2 outbound lanes, and 1 reversible lane when the unbalanced queue coefficient reaches the critical value of 1.85. To achieve a reservation level of 50–60%, the system should have 5 inbound lanes, 1 outbound lane, and 2 reversible lanes when the unbalanced queue coefficient reaches the critical value of 1.85. For a reservation level between 70 and 90%, the configuration includes 5 inbound lanes, 1 outbound lane, and 2 reversible lanes when the unbalanced queue coefficient reaches a critical value of 1.5. Given that the decision variable ⁽⁵⁾ is (5,1,2,1.85), the unbalanced queue coefficient has a critical value of 1.85 (65:35), the appointment level is 50%, and there are 2 reversible lanes. It has the capability to decrease the average waiting time by a maximum of 15.02%.

This study offers valuable insights to port authorities regarding the benefits of managing reversible lanes in relation to vehicle volumes. By doing so, ports can effectively reduce congestion at container terminal gates and mitigate the impact of congestion factors. The findings can be utilized by individuals in positions of authority to improve the quality of service at port entrances. It is observed that reversible lanes work best when the truck appointment level is at 50%. This suggests that additional analysis should be conducted at different appointment levels in the future. Based on the findings of this study, port authorities can discern the variables that contribute to congestion and devise strategies to enhance the efficiency of port gates. Moreover, the implementation of reversible lane management can provide a theoretical basis for operational planning and future advancements, resulting in innovative ideas for investing in technology to enhance port capacity.

Figures

Figure 1

Overview of Laem Chabang port (July 2022 to January 2023)

Figure 2

Percentage of each arrival vehicle type

Figure 3

The structure of the simulation model

Figure 4

Queuing system with revers lane policy

Figure 5

Traffic direction change logic of RLs

Table 1

Main assumptions of our simulation model

Bases model assumptions
1.1	The operating hours of Laem Chabang Port span 24 h
1.2	All types of container trucks can be serviced via the eight inspection doors
1.3	At the beginning of the simulation, the traffic lane is consistently regarded as unoccupied, and its capacity is equivalent to the length of the port access road
1.4	The Port Administration Division of Laem Chabang Port obtained historical data on truck arrival patterns and service capacity by conducting data surveys. During the period of July 2022 to January 2023
1.5	The distribution of incoming container truck types was as follows: 48.16% for ordinary containers, 8.71% for another type, and 43.13% for a different type. Refrigerated containers and unoccupied trucks
1.6	the trucks with planned appointments ensure that they pre-register the necessary paperwork and licenses in the system
1.7	The operating hours of the crew at the inspection gate vary based on the nature of the appointment
1.8	Trucks without a prior appointment will exclusively utilize E-H doors

Source(s): Authors’ own work

Table 2

Truck arrival patterns

Time period	Vehicle types			Total	Proportion (%)
	Dry	Reefer	Empty
08:01–09:00	209	35	191	435	7.17
09:01–10:00	224	37	205	466	7.68
10:01–11:00	204	34	187	425	6.99
11:01–12:00	221	37	203	461	7.60
12:01–13:00	228	38	209	475	7.85
13:01–14:00	274	46	251	571	9.41
14:01–15:00	282	47	258	587	9.67
15:01–16:00	299	50	274	623	10.26
16:01–17:00	280	47	257	584	9.63
17:01–18:00	251	42	230	523	8.62
18:01–19:00	185	31	170	386	6.35
19:01–20:00	135	23	124	282	4.65
Total	2,912	485	2,669	6,066	100.00
%	48.01%	7.93%	44.06%

Source(s): Authors’ own work

Table 3

Truck arrival distribution fitting

Time period	Distribution fitting	Kolmogorov-Smirnov test (p-value)	Skewness (Pearson)	Kurtosis (Pearson)
08:01–09:00	BETA(0.975, 0.91)	0.9392	0.0884	−1.1942
09:01–10:00	BETA(0.973, 0.924)	0.8948	−0.0468	−1.2702
10:01–11:00	BETA(0.925, 0.799)	0.8948	−0.1418	−1.2454
11:01–12:00	BETA(0.922, 0.864)	0.9861	0.0864	−1.2226
12:01–13:00	Uniform (0.001,15)	0.1541	−0.1505	−1.1492
13:01–14:00	BETA(0.883, 0.821)	0.8716	−0.0301	−1.2913
14:01–15:00	Uniform (0.04, 12.48)	0.4145	−0.0504	−0.9961
15:01–16:00	BETA(0.899, 0.823)	0.5158	−0.0954	−1.2598
16:01–17:00	Uniform(0.09, 14.16)	0.2978	0.0033	−1.1114
17:01–18:00	BETA(0.88, 0.814)	0.9555	−0.0578	−1.2227
18:01–19:00	BETA(0.961, 0.884)	0.9124	−0.0785	−1.1854
19:01–20:00	BETA(1.21, 1.16)	0.9709	−0.0380	−1.1451

Source(s): Authors’ own work

Table 4

Proportion of destination terminal

Non-booking						Booking
Terminal gate	Vehicles	Proportion (%)	Terminal gate	Vehicles	Proportion (%)	Terminal gate	Vehicles	Proportion (%)	Terminal gate	Vehicles	Proportion (%)
A0	426	7.75	B1	453	8.24	A0	60	1.09	B1	52	0.95
A1	448	8.15	B2	462	8.41	A1	64	1.16	B2	58	1.06
A2	458	8.33	B3	459	8.35	A2	54	0.98	B3	60	1.09
A3	460	8.37	B4	459	8.35	A3	53	0.96	B4	48	0.87
A4	430	7.82	B5	411	7.48	A4	54	0.98	B5	36	0.66
A5	439	7.99				A5	52	0.95
Total				4,905	89.25	Total				591	10.75

Source(s): Authors’ own work

Table 5

Comparisons of observed and simulation

Time period	Traffic data (observed)	Traffic data (Simulated)	GEH	Average queue (observed)	Average queue (simulated)	GEH	Calibration results
08:01–09:00	435	377	0.142	8	4.5	0.567	Pass
09:01–10:00	466	371	0.227	14	18.7	0.286	Pass
10:01–11:00	424	343	0.211	25	19.8	0.230	Pass
11:01–12:00	461	397	0.149	27	19.4	0.326	Pass
12:01–13:00	476	410	0.148	25	34.8	0.326	Pass
13:01–14:00	571	512	0.109	38	31.2	0.197	Pass
14:01–15:00	586	558	0.049	40	30.9	0.257	Pass
15:01–16:00	622	648	0.041	39	31.1	0.227	Pass
16:01–17:00	584	489	0.177	34	25.1	0.301	Pass
17:01–18:00	523	423	0.211	31	33.9	0.089	Pass
18:01–19:00	385	238	0.472	29	22.6	0.247	Pass
19:01–20:00	281	323	0.139	27	15.5	0.542	Pass

Source(s): Authors’ own work

Table 6

The findings indicate the mean waiting time and mean length of the 7 groups

Source	Average queue number					Average waiting time
	SS	df	MS	F Cal	F crit	SS	df	MS	F Cal	F crit
Between groups	845953.41	5	169190.7	13.060	2.167423	254.9951	6	42.49919	6.549107	2.167423
Within groups	1787730.2	138	12954.57			863.0783	133	6.489311
Total	2633683.6	143				1118.073	139

Source(s): Authors’ own work

Table 7

The findings indicate the mean waiting time and mean length of the 4 groups

Source	Average queue number					Average waiting time
	SS	df	MS	F Cal	F crit	SS	df	MS	F Cal	F crit
Between groups	23.132386	3	7.710795	1.692	2.724944	23.19232	3	7.730773	1.131838	2.724944
Within groups	346.29435	76	4.556505			519.1014	76	6.830282
Total	369.42674	79				542.2938	79

Source(s): Authors’ own work

Table 8

Factor analysis of variance shows average queue length and wait time

Average queue number						Average waiting time
Factor	df	SS	p-value	%	Cum. (%)	Factor	df	SS	p-value	%	Cum. (%)
A	1	588.15	0.0000014	29.96	29.96	A	1	546.1	0.0000053	27.82	27.82
B	1	524.94	0.0000173	26.74	56.70	B	1	473.04	0.0000223	24.10	51.91
C	1	463.64	0.0000510	23.62	80.31	C	1	420.01	0.0000614	21.39	73.31
F	1	386.49	0.0006144	19.69	100.00	F	1	350.47	0.0009471	17.85	91.16
Total	4	1963.22				Total	4	1789.62

Source(s): Authors’ own work

Table 9

MANOVA results

Source	Wilks’ test (p-value)	Eigenvectors		Source	Wilks’ test (p-value)	Eigenvectors
		Queue	Waiting			Queue	Waiting
Factor AB	0.002	0.416	1.11	Factor ABC	0.686	0.038	0.003
Factor AC	0.641	0.024	0.184	Factor ABF	0.009	0.317	0.002
Factor AF	0.344	0.059	1.865	Factor ACF	0.140	0.245	0.002
Factor BC	0.637	0.026	0.151	Factor BCF	0.248	0.338	0.001
Factor BF	0.030	0.209	−13.878	Factor ABCF	0.056	0.277	0.002
Factor CF	0.087	0.293	1.665

Source(s): Authors’ own work

Table 10

Model parameters

Symbol	Parameters	Symbol	Parameters	Symbol	Parameters
nin, nout	Truck lanes entering and leaving the port	WT	Container terminal entrance gate system waiting time (min)	WRL	Average container terminal gate operation waiting time per truck in the Reverse Lane (min)
nRL	Number of Reverse Lane	Win, Wout	Average truck inbound and outbound container terminal door operation waiting time (min)	Wg-RL	The average wait time for vehicles entering service at the container terminal gate’s in the Reverse Lane
CR	The ratio of inbound to outbound average queue length	Wg-in, Wg-out	Average Container Terminal gate entrance and exit wait time for vehicles entering service (min)	Wf-RL	Average container terminal gate vehicle entry time in the Reverse Lane (min)
CR*	Critical unbalanced queuing coefficient value	Wf-in, Wf-out	Average container terminal gate waiting time for inbound and outbound vehicles (min)	N	Total gate lanes
i	Channel number at container terminal entrance	T	Period of calculation	Si	Traffic lanes inbound, outbound, reverse lane, and the critical unbalanced queuing coefficient

Source(s): Authors’ own work

Table 11

Simulation results

Decision variable	Average waiting time of all lane (minutes/truck)
	Booking 10%			Booking 20%			Booking 30%
	Enter	Departure	Total	Enter	Departure	Total	Enter	Departure	Total
u(0) =(5,3,0,2.33)	51.49	6.49	57.98	51.26	6.26	57.52	51.60	6.60	58.20
u(1) =(5,2,1,2.33)	46.34	7.01	53.35	45.81	6.98	52.79	44.81	7.61	52.41
u(2) =(5,2,1,1.85)	45.31	7.53	52.84	44.64	7.50	52.14	43.96	8.16	52.12
u(3) = (5,2,1,1.5)	45.47	8.05	53.51	44.79	7.93	52.73	44.21	8.54	52.75
u(4) =(5,1,2,2.33)	43.72	9.50	53.22	42.62	9.88	52.50	44.40	9.08	53.48
u(5) =(5,1,2,1.85)	46.86	7.33	54.19	45.18	7.25	52.43	45.14	7.82	52.97
u(6) = (5,1,2,1.5)	46.39	8.11	54.51	45.84	7.99	53.84	45.74	8.62	54.36

Decision variable	Booking 40%			Booking 50%			Booking 60%
	Enter	Departure	Total	Enter	Departure	Total	Enter	Departure	Total
u(0) =(5,3,0,2.33)	50.16	5.16	55.32	50.78	5.78	56.56	51.11	6.11	57.22
u(1) =(5,2,1,2.33)	42.08	6.14	48.22	42.29	7.60	49.89	46.3	7.0	53.4
u(2) =(5,2,1,1.85)	43.78	6.57	50.35	42.40	7.57	49.96	45.3	7.5	52.8
u(3) = (5,2,1,1.5)	42.02	6.88	48.91	42.21	7.93	50.14	45.5	8.0	53.5
u(4) =(5,1,2,2.33)	43.31	9.29	52.60	42.41	7.33	49.74	43.7	8.5	52.2
u(5) =(5,1,2,1.85)	43.00	8.32	51.32	41.02	7.04	48.06	43.9	7.3	51.2
u(6) = (5,1,2,1.5)	43.58	8.88	52.46	43.63	7.93	51.56	46.4	8.1	54.5

Decision variable	Booking 70%			Booking 80%			Booking 90%
	Enter	Departure	Total	Enter	Departure	Total	Enter	Departure	Total
u(0) =(5,3,0,2.33)	51.43	6.43	57.86	51.45	6.45	57.89	51.59	6.59	58.18
u(1) =(5,2,1,2.33)	45.4	7.0	52.3	44.8	7.6	52.4	48.1	6.1	54.2
u(2) =(5,2,1,1.85)	43.8	7.4	51.2	43.6	8.1	51.7	46.6	6.4	53.1
u(3) = (5,2,1,1.5)	44.5	7.9	52.4	43.3	8.6	51.9	45.7	6.9	52.5
u(4) =(5,1,2,2.33)	45.1	8.4	53.5	44.4	9.1	53.5	45.0	7.3	52.3
u(5) =(5,1,2,1.85)	45.8	8.3	54.1	45.2	7.9	53.1	43.1	6.3	49.4
u(6) = (5,1,2,1.5)	41.2	8.3	49.5	42.1	7.6	49.7	42.3	6.9	49.3

Source(s): Authors’ own work