Dynamic flow analysis and crowd management for transfer stations: a case study of Suzhou Metro

Abstract

Transfer stations are important nodes in the metro network, and it is of great significance to study the coordinated organization scheme between passenger demand and facility configuration, in order to improve the transportation efficiency. Taking the Dongfangzhimen station of the Suzhou Metro as the research object, this paper starts by analyzing the configuration of service-oriented facilities, and then dissects the spatiotemporal distribution characteristics and individual behavior characteristics of passengers via data processing, where a meticulous analysis has been carried out based on the automatic fare collection data and the field investigation data. On this basis, we have discussed the performance and bottlenecks under the current facility configuration and flow organization scheme, constructed the simulation scene of a representative peak hour in holidays using AnyLogic, and put forward a volume-based path organization scheme considering the coordination with flow demand, as well as a plan of multi-stage crowd management. As evidenced by the results, the improved scheme exhibits a higher adaptability to passenger flow and a greater balance among facility utilization, where the maximum queue length, the average time consumption and the average standing density have been reduced by 31%, 14% and 6%, respectively. The proposed methods of data analysis, flow organization and strategy decision are reliable and applicable to the management of metro transfer stations.

1 Introduction

Metro transit has become the heavyweight of urban public transportation due to its advantages of mass capacity, low energy consumption and high operational punctuality, providing an effective approach to alleviate urban traffic congestion. Under the develo** trend of network construction and operation, transfer stations are crucial nodes for the interconnection among different transit lines. However, faced with the increasing network mileage and passenger volume, the incompatibility between flow demand and infrastructure supply at transfer stations is emerging accordingly, which will affect the network transportation efficiency. Therefore, it is of great significance to study the crowd management methods at transfer stations for the sake of enhancing the proactivity of daily metro management.

Current literatures on the problem of crowd management at metro transfer stations can be divided into three aspects: passenger characteristics analysis, passenger flow management, and scheme optimization. Passenger characteristics analysis is an important work before optimizing organization schemes, which can be classified into individual behavior characteristics and volume distribution characteristics (Sakellariou et al. 2012). The individual behavior characteristics under crowded scenarios include waiting time duration, walking speed and area density (Al-Ahmadi 2018), where parameters like queue length, path selection and service frequency should be given further discussion for subway stations. In addition, Toto et al. (2016) proposed a real-time prediction framework based on the time-invariant circumstance parameters and historical station passenger data. To describe the temporal distribution of inbound and outbound passengers, Yang et al. (2021) calibrated pedestrian source generation considering the dynamic variation of boarding and alighting passengers during peak hours. Krstanoski (2014) found that the multinomial probability distribution model can better describe the boarding and alighting procedure on the train platform. Similarly, Zhang et al. (2008) and Li et al. (2020) further studied the resistance effect and the non-compliance behaviors during the boarding and alighting procedure based on micro simulation models, in order to adjust the organization rules under different volume conditions. Surveys and experiments have also revealed that physical station layout, passenger demographics, psychometric propensities, and operational plans have significant impacts on passenger behavior and activities, especially during the peak hour (Kim et al. 2014; Loukaitou-Sideris et al. 2016; Seriani et al. 2017).

As to the metro flow management, Heekyukim et al. (2016) pointed out that the station structure and facilities layout are the basis of path organization; namely the flow organization and management should be coordinated with the static infrastructure. Chen et al. (2017) further studied the collaborative organization between train stop plan and station passenger control considering transportation profits. Besides, in order to improve the level of safety management, quite a few scholars (Kallianiotis et al. 2018; Shi et al. 2012) studied the evacuation strategies under emergent scenarios like tunnel fires and rain floods.

In the perspective of scheme optimization, methods of mathematical modeling and dynamic simulation are widely applied, where the former concentrates on establishing programming models to decide parameters like path control nodes, facility numbers and operation plans, and the latter focuses on the integrated scenario simulation to optimize systematic organization schemes in a more intuitive way. In order to minimize the system cost of organization schemes, Zhu et al. (2014) formulated a streamline optimization model with considerations on route choice analysis and volume assignment. Asano et al. (2010) took the view that the multi-directional collision avoidance between passengers should be given additional consideration, and established the optimal path choice model to facilitate flow organization in crowded situations. In terms of the simulation-based optimization, Gao and Xu (2010) developed an event-driven model to describe the complicated passenger behaviors and service processes, in which the events include entering the station, buying tickets, checking tickets, selecting pathways to the platform, waiting for trains, etc. Similar methods were also adopted by Liu and Peng (2018) via the Anylogic software to recognize and solve the problems of ticket gate configuration based on capacity analysis. Based on a Petri net simulator, Tessitore et al. (2022) put forward a real-time rescheduling framework considering passenger demands and timetable perturbations under stochastic disturbance scenarios.

From the forgoing literature review, it is obvious that the key to managing crowds at a metro station is to achieve the coordination between supply and demand. However, due to the constant change of passenger characteristics and the diversity of station layouts, a current research gap lies in the discrepancy between general decision methods and practical management feasibility, particularly in terms of coordinating infrastructure layout, facility utilization, flow organization and passenger demand. Additionally to the necessary analysis of passenger characteristics and facility configurations, this paper aims to put forward an applicable crowd management scheme for metro transfer stations through a specific case study, and establish a referable methodology of performance evaluation and threshold judgement, validated by a micro pedestrian simulation.

The remainder of this paper is structured as follows. By taking a key transfer station of Suzhou Metro as the research object, Sect. 2 provides a detailed introduction and analysis of station layout and facility configuration, followed by the passenger characteristics analysis from the perspective of volume distribution and individual behavior in Sect. 3. Section 4 points out the current organization bottlenecks under crowded scenarios and puts forward basic rules and approaches for reorganizing different paths. A micro simulation model is presented in Sect. 5, where feasible enhancement strategies are systematically devised, along with a volume-based strategy implementation mechanism considering the dynamics of passenger volumes. Finally, Sect. 6 concludes the paper by discussing the main contributions and possible future work.

2 Station facilities configuration

The metro station is a distribution place consisting of different kinds of equipment and facilities, the layout of civil architecture and facilities determines the set of available paths for passengers entering, exiting and transferring. Each possible path is composed of a series of connected nodes and links. Therefore, the structural layout of station facilities plays a crucial role in analyzing the passenger characteristics and adjusting the passenger streamlines.

2.1 Station entrances distribution and surrounding land use

The Dongfangzhimen (DFZM) station is an important station within the network of the Suzhou rail transit system. It serves as a transfer station for Line 1 and Line 3, as shown in Fig. 1a. The DFZM station is a three-level underground station with 10 entrances. The station halls of Line 1 and Line 3 are located on the basement floor, the platform of Line 1 is located on the second floor, and the platform of Line 3 is located on the third floor. As indicated in Fig. 1b, the main body of the DFZM station lies beneath the Suzhou shop** center, where the north and south are residential lands, the east is the recreation space belonging to the city square, and the west is commercial land. Therefore, this station primarily serves the trip demand of commuters and recreational passengers.

Fig. 1

Basic illustration and location of DFZM metro station

2.2 Facility classification and layouts

Generally, a metro station is composed of facilities and infrastructures from different systems (Chopra et al. 2016), such as the automatic fare collection (AFC) system, the integrated monitoring system, the signal control system, the environmental control system, the building automation system, etc. Targeting the coordination by organizing passenger volumes, this paper primarily focuses on the service-oriented facilities which have interactions with passengers during the transportation service. To ensure a meticulous crowd management, the service-oriented facilities are classified into the following three types based on their behavior characteristics:

• 1. Distributing facility. The distributing facilities consist of the station hall and the train platform, which are used for passenger volume exchange, such as passengers entering and exiting, as well as passengers alighting and boarding.

• 2. Queuing facility. The queuing facilities refer to the facilities or equipment where passengers are prone to form queues to receive service, usually including the ticket vending machines (TVMs), the security inspection devices and the automatic gate machines (AGMs).

• 3. Passing facility. The passing facilities usually function as the connection parts between different service areas, including the passages, escalators and stairs, where the passengers are in a dynamic movement by human power or external forces.

Among the service–oriented facilities listed in Table 1, the inbound and outbound AGMs are key facilities, whose configuration and layout will affect the passenger distribution, path selection and the flow organization (Wan et al. 2013). Meanwhile, the AGMs are interconnected with the transit ticketing system, where the big data of passengers checking time, trip origination and trip destination are recorded in detail, namely the automatic fare collection (AFC) data. The AGM data is an important base for analyzing passenger characteristics. The Line 1 and Line 3 halls are both equipped with 6 AGM groups according to the station layout. Figure 2 shows the AGMs distribution of the Line 1 hall, where G1 and G5 are inbound AGM groups, G2, G3 and G4 are outbound AGM groups under normal conditions. In addition, AGMs in G2 and G4 are bidirectional gates, which can be converted to entering mode for massive inbound volumes.

Table 1 Normal service-oriented facility configuration inside the DFZM station

Fig. 2

AGMs distribution and direction of Line 1 hall

Representative infrastructures and facilities are shown in Fig. 3, including the corridors, AGMs and platforms. The site photographs were taken during a non-peak period at the night of August 5th, 2021.

Fig. 3

Representative infrastructures and facilities at DFZM station

2.3 Performance indicators for facility configuration

According to the classification of facilities, indicators reflecting the service level and efficiency of a metro station are proposed along with corresponding calculation methods. Capacity usage, passenger density, queue length, and time consumption are the main indicators used to evaluate the performance of the facility configuration scheme.

• 1. Capacity usage

  The capacity usage is a fundamental indicator to evaluate the facility performance, especially for the passing facilities and queuing facilities. The capacity usage coefficient is defined as the ratio of actual number of served passengers to the theoretical number of served passengers, as shown in Eq. (1).

  $$ \eta_i = \frac{Q_i }{{C_i^s }} $$

  (1)

  where Q_i denotes the served passenger volume of the ith facility per hour, pax/h; $C_i^s$ refers to the standard capacity of the related facility which is acquired from the Chinese Code for Metro Design (GB 50157-2013), pax/h. For example, the standard capacity of a 1-m-wide escalator running at 0.5 m/s is 6720 pax/h, and the capacity of every AGM is 1500 pax/h.

• 2. Passenger density

  The passenger density is a dynamic indicator considering the stochastic volume distribution. For distributing areas like the hall and the platform, the passenger density is calculated by:

  $$\rho_i = \frac{{m_i (0) + \int_0^T {\left[ {r_i^a (t) - r_i^d (t)} \right]{\text{d}}t} }}{w_i \cdot l_i }$$

  (2)

  where $m_i (0)$ represents the initial number of passengers of the ith distributing area, pax; T is the observed time period, 5 min preferred in this study; $r_i^a (t)$ and $r_i^d (t)$ denote the time-varying passenger arrival rate and departure rate, pax/min; w_i and l_i correspond to the width and length of the current area, m.

• 3. Queue length

  The maximum queue length is equal to the number of passengers waiting in front of the service facilities, which can indicate the crowd status of corresponding facilities. The queue length of a service facility is usually determined by the number of configurations, the time it takes for personal service, and the rate of passenger arrival. Considering the stochastic passenger arrivals, the service procedure under large volume scenarios can be regarded as an M/M/N queuing system with multiple paths and multiple servers (Gamarnik and Goldberg 2013). When N is 1, it becomes the M/M/1 system, such as the security check service. In the M/M/N system, the average queue length can be theoretically calculated according to Eq. (3). As to the maximum queue length, it is obtained by adding a counter in the simulation system.

  $$\overline{q_i } = \frac{{\sigma_i^{n_i + 1} }}{n_i !n_i } \cdot \frac{P_i (0)}{{(1 - {{\sigma_i } / {n_i }})^2 }} \cdot \sigma_i = \frac{a_i }{{\lambda_i }}$$

  (3)

  where n_i is the configuration number of the ith service facility; $\sigma_i$ denotes the service intensity, defined as the ratio of the average arrival rate $a_i$ to the average service rate $\lambda_i$; $P_i (0)$ is the probability when the facility queuing area is empty.

• 4. Time consumption

  Both the flow path and passenger activity should be taken into account when calculating the time consumption (Zheng et al. 2014). In this paper, the inbound time consumption is composed of the passengers walking time, queuing time and service time, as expressed in Eq. (4).

  $$t_i = \frac{{d_i^{{\text{ep}}} }}{v_i } + \frac{{h_i^{{\text{ep}}} }}{v_i \cdot \sin \alpha } + \sum_{j = 1}^k {t_{i,j}^q } + \sum_{j = 1}^k {t_{i,j}^s }$$

  (4)

  where $d_i^{{\text{ep}}}$ and $h_i^{{\text{ep}}}$ denote the average horizontal distance and the vertical height from entrance to platform on the ith passenger path, m; α is the angle between the stairs and the ground; $v_i$ denotes the average walking speed on the current path, m/s; $t_{i,j}^q$ refers to the average queuing time at the jth service facility on the current path, and $t_{i,j}^s$ refers to the average service time, s; k is the number of service facilities, including the ticket vending machine, security inspection device, check-in gates, etc.

• 5. Platform crowdedness

  The platform crowdedness depends on factors such as the size of the platform, the train headway, passenger arrival rates, and alighting rates. More passengers will get detained when the entering arrival rate and the transfer-in arrival rate increase, and the increased alighting passenger rate will cause more walking conflicts and increase the standing density in the waiting area. In China, passengers who want to board the train should let the train passengers alight first. Therefore, the most crowded moment occurs at the end of passengers alighting when both upstream trains and downstream trains arrive at the platform simultaneously. This instantaneous platform density can be calculated by:

  $$d_{i + 1} = \frac{{q_i + I_{i,i + 1} \cdot (\alpha_{in} + \alpha_{tr} ) + 2\beta_{al} }}{w_p \cdot l_p - n_e \cdot w_e \cdot l_e - n_c - 4n_s } \cdot \gamma$$

  (5)

  where ${d}_{i+1}$ denotes the maximum platform density when the i + 1th train dwells, pax/m²; ${q}_{i}$ is the number of detained passengers when the ith train leaves, pax; ${I}_{i,i+1}$ is the headway between the ith train and the i + 1th train, min; ${\alpha }_{in}$ and ${\alpha }_{tr}$ represent the entering arrival rate and the transfer-in arrival rate, pax/min; ${\beta }_{al}$ is the average alighting passengers per train, pax; ${w}_{p}$ and ${l}_{p}$ are the width and the length of the platform, m; ${n}_{e}$ is the number of escalator and stair groups; ${w}_{e}$ and ${l}_{e}$ are the width and the length of one escalator and stair group, m; ${n}_{c}$ is the total size of columns inside the platform area, where the occupied area of one column is about 1 m²; ${n}_{s}$ is the number of seating areas placed in the platform, where one seating area is about 4 m²; $\gamma$ denote the spatial disequilibrium factor, e.g., the passenger density near the escalators is higher than the other platform areas, according to the simulation results, the disequilibrium factor ranges from 1.24 to 1.47 during the peak period.

In Eq. (5), the parameter q_i is a dynamic value varying with the volume rate and the train frequency. In order to estimate q_i, the equations below are established by considering the difference between the accumulative arriving passengers and the aggregate boarding passengers.

$$q_i = \sum_{k = 1}^{i - 1} {I_{k,k + 1} \cdot (\alpha_{in} + \alpha_{tr} )} - \sum_{k = 1}^i {P_k }$$

(6)

$$P_k = \min \left\{ {\frac{t_d }{2} \cdot m \cdot n_d \cdot \alpha_b ,(1 - \eta_k ) \cdot m \cdot C_c } \right\}$$

(7)

where ${P}_{k}$ is the tolerable number of boarding passengers for the kth train, pax; m is the train marshaling number; ${t}_{d}$ denotes the average train dwelling time during the peak period, s; ${n}_{d}$ is the number of doors per train; ${\alpha }_{b}$ is the average passengers boarding rate, pax/s, usually takes 1.5 to 2 pax/s; ${\eta }_{k}$ is the average carriage occupancy exclusive of the alighting passengers during the peak hour; ${C}_{c}$ represents the standard capacity of one train carriage, pax/car, depending on the train type, e.g., the capacity of a Chinese B-type metro carriage is 210 pax/car.

3 Passenger characteristics deconstruction

The spatiotemporal distribution and individual behavior characteristics of passengers can be analyzed in detail using AFC data and field investigation data (including the questionnaire data and video monitoring data). Gras** the demand of station passengers is the basis for achieving the equilibrium between static facility configuration and dynamic flow organization.

3.1 Spatiotemporal distribution characteristics

3.1.1 Daily temporal distribution

Due to the COVID-19 pandemic, urban public transportation in 2021 and 2022 has been greatly affected, especially the mass rail transit systems. Consequently, the data during 2019 is utilized in this paper. The daily temporal distributions of inbound passengers and outbound passengers during the holiday (October 3rd, 2019) and weekday (October 18th, 2019) are indicated in Fig. 4. The inbound arrival rates for holidays are extremely higher than for weekdays, and the corresponding temporal distribution takes on an obvious bimodal pattern (afternoon and evening). As to the outbound volume, both weekday and holiday show the bimodal pattern (morning and evening), and a weekday’s morning peak is approximately 2 h ahead of the holiday, which conforms to the surrounding land use.

Fig. 4

Daily volumes temporal distribution of inbound passengers and outbound passengers

To further study the temporal variation and support the dynamic management, the Mahalanobis distance (McLachlan 1999) is introduced to describe the differences among time series data, which can effectively calculate the similarity between samples. The Mahalanobis distance for a continuous arrival rate vector a(t) can be calculated as follows.

$$D_M (t) = \sqrt {{[a(t) - \mu (t)]^T S_{(t)}^{ - 1} [a(t) - \mu (t)]}}$$

(8)

$$\mu (t) = \frac{1}{{\left| {R_t } \right|}}\sum_{a \in R_t } a$$

(9)

$$S_{(t)} = \frac{{\left| {R_t } \right|}}{{\left| {R_t } \right| - 1}}\sum_{a \in R_t } {\left[ {\frac{aa^T }{{\left| {R_t } \right|}} - \mu (t)\mu (t)^T } \right]}$$

(10)

where $D_M (t)$ is the Mahalanobis distance depending on the time of the day, $a(t)$ and $\mu (t)$ denote the observed and the average volume arrival rate, respectively, pax/min; $S_{(t)}$ is the covariance matrix, and $R_t$ is the reference data set.

The Mahalanobis distance outperforms the Euclidean distance in detecting outliers, especially when the data is not normally distributed (Kumar and Khani 2021). The probability distribution of a centroid deviation usually follows the $\chi^2$ distribution. Using the distance-based outlier detection method, the confidence level α and the degree of freedom p are set as 0.05 and 15, respectively, and the corresponding threshold value $\chi_p^2 (\alpha )$ is 7.26.

Taking the inbound volume as an example, the Mahalanobis distance distribution between holiday samples is indicated in Fig. 5. Compared with the reference dataset, the major outliers time ranges at October 3rd 2019 are distributed in periods of 6:00–8:00, 12:30–14:30 and 22:00–23:00, which reflects a different demand of inbound passengers. As for the instantaneous status variation, the Mahalanobis distance distribution between adjacent short-term arrivals is shown in Fig. 6. It is indicated that the inbound arrival rates fluctuate violently in the periods of 6:00–9:00, 12:30–14:00, 18:15–18:45 and 20:30–22:30, which corresponds to the festival trip demand considering the surrounding land use. It should be noted that the outliers time range does not equate to the volume peak period; since the function of outlier detection is to monitor the dynamic variation of the arrival rate.

Fig. 5

Outlier detection of the daily inbound volume data between holiday samples

Fig. 6

Outlier detection of daily inbound arrival rates (every 15 min)

3.1.2 Arrival rate distribution

The statistical distribution of inbound and outbound volumes is the basis of subsequent parameter calibration, especially for the generation sources of passengers. Taking the representative peak hours of Line 1 and Line 3 as examples, the characteristics of inbound volume peak and outbound volume peak are analyzed, respectively.

• 1. Inbound volume (Line 1)

  During the inbound peak hour from 20:00 to 21:00, the cumulative number of inbound passengers of Line 1 is 6607, where the total arrival rate is randomly ranging from 68 to 153 pax/min, and the western arrival rate is 22% higher than the eastern one, as indicated in Fig. 7a. Due to the stochasticity of the number of inbound passengers, the corresponding statistical distribution takes on an obvious normal distribution, shown in Fig. 7b, where the mean value μ is 110.12 pax/min, and the standard error δ is about 17.71 pax/min.

• 2. Outbound volume (Line 3)

  During the outbound peak hour from 18:45 to 19:45, the number of cumulative outbound passengers of Line 3 is 2090, with an obvious rule of impulsive fluctuation under the statistical interval of 1 min, as indicated in Fig. 8a. More specifically, the interval between the peaks of arrival rates ranges from 4 to 6 min, which tallies with the planned train headway. Meanwhile, it is obvious that during the peak hour, 96.3% of the passengers choose to exit from the southern gates of Line 3. As indicated in Fig. 8b, the discrete arrival rate of outbound volumes obeys the exponential distribution, with the mean value μ around 34.78 pax/min.

Fig. 7

Arrival rates distribution of Line 1 inbound passengers

Fig. 8

Arrival rates distribution of Line 3 outbound passengers

Based on Figs. 7a and 8a, the inbound arrival rates fluctuate randomly, while the outbound arrival rates take on an impulsive distribution in accordance with the train arrivals. It can be referred that the latter is more regular because the outbound passengers alighting from the arriving trains are more predictable than the inbound passengers entering from different corridors and entrances. Figures 7b and 8b give the basic distribution form of inbound arrivals and outbound arrivals, respectively, which can provide support for the subsequent parameter calibration of different passenger source modules in the simulation platform.

3.1.3 Gate volume distribution

Due to the large amount of original multi-source data, this section shows the daily distribution of AGMs by taking the AFC data of inbound passengers of Line 1 during China’s National Day (September 29th to October 9th, 2019). The daily distribution and variation of passenger volumes are indicated in Fig. 9a, b, respectively. During the festival, the volumes of every AGM group take on an obvious trend of increasing first and then decreasing, with the peak appearing on October 3rd. As to the variation of passengers served by AGMs, the volume proportion of western G5 first decreased and then increased, and the volume proportion of western G1 shows an opposite variation trend, indicating that the inbound volumes on both sides are more evenly distributed on holidays when compared to weekdays. Meanwhile, G2 and G4 undertook a small part of inbound volumes during October 2nd to October 4th, due to their conversions from exiting to entering.

Fig. 9

The volume distribution and proportion variation of entering AGM groups (Line 1)

3.2 Individual behavior characteristics

Because most metro stations are underground, the movement and activity in a limited space will be affected by a variety of factors (Gulhare and Vedagiri 2018; Moussaid et al. 2011), such as passenger preference, personal familiarity, congestion status, etc. These uncertain factors lead to the heterogeneity in individual behaviors such as walking velocity, route choice and facility service utilization. In a crowd scenario, the walking speeds of different passengers are not significantly impacted by their trip purposes, but rather by the volume and location. E.g., passengers in corridors walk faster than those near stairs, and the speed of a walking individual is dependent on the speed of the group flow. Table 2 indicates the individual walking characteristics within distributing facilities. The 251 passengers investigated during the peak hour were randomly selected from the recorded CCTV monitoring videos. Note that the velocity is used as the initial velocity (without crowdedness) of passenger source in the simulation model, including the entering flow generation sources and the alighting flow generation sources. In order to guarantee the heterogeneity of different passenger groups, the triangular distribution is used to describe the walking speed, consisting of the values of lower bound, majority value and upper bound.

Table 2 Walking velocity of different passenger groups

As we know, not every passenger needs to buy tickets on a vending machine due to the existence of public transportation IC card or QR code. However, for a metro station, it is inevitable to provide a ticket selling service for passengers who are not familiar or accustomed with the current metro system. Table 3 lists some key service characteristics of TVM usage during the investigated peak hour, providing a solid base for subsequent path modeling and adjustment.

Table 3 Service characteristics of representative TVM groups during the investigated peak hour

Similarly, the service characteristics at other facilities like the security device and the AGM are also indispensable, as shown in Table 4. The usage of security inspection devices should also consider the ratio of passengers with knapsacks, which determines the extra passengers assigned to pass through the X-ray device. For Line 3, the check-in interval of the south AGM group is nearly the half of the interval of the north AGM group, which means that the corresponding passenger volume is higher. Since the scenario parameters are dynamic, a basic regression function between AGM service interval and passenger arrival rate has been established in order to calibrate facility attributes dynamically, as shown in Fig. 10.

Table 4 Observed service characteristics of security inspection devices and entering AGMs of Line 3

Fig. 10

Regression function between arrival rate and AGM service interval

4 Flow organization analysis

In order to put forward a practical and effective organization scheme, the analysis of the current flow organization scheme and corresponding bottlenecks becomes necessary, as well as the basic rules and approaches.

4.1 Current flow organization during peak hours

4.1.1 Flow organization schemes

Upon field investigation, the outburst mass volumes at the DFZM station include four scenarios, namely the crowded inbound volume, the crowded outbound volume, the crowded transfer volume and the simultaneous inbound and transfer volume. The crowded inbound volume and the crowded transfer volume are frequently encountered during daily station management, with the following organization schemes.

• 1. The crowded inbound volume can be divided into two categories: one-sided large volume and bilateral large volume. Key control points E3, E4 and E8 are responsible for limiting the number of entering passengers according to the distribution of inbound volumes. General measures include entering only, batched entering and temporary closure. Meanwhile, two backup security devices will be activated to lessen the pressure on the original security areas.

• 2. Due to the limited hall and platform of Line 1, the crowded transfer volume at the DFZM station primarily concentrates on the transfer volume from Line 3 to Line 1. The relevant measures lie in the path regulation, where a clip-shaped isolation belt is additionally set in the downstream of the transfer path, and the southern entering AGMs of Line 3 near E3/4 will close to reduce the interference between inbound volumes and transfer volumes.

4.1.2 Potential organization bottlenecks

Based on the foregoing analysis of station facilities and passenger characteristics, the potential problems of the current flow organization are listed as follows.

• 1. The disequilibrium of AGM utilization at Line 1 hall

  According to Fig. 9, the daily volumes distribution of AGM groups takes on an obvious disequilibrium at the station hall of Line 1, where the proportion of entering passengers from G4 and G5 accounts for 55–70%. This disequilibrium will substantially increase the queue length and delay at western security inspection devices and check-in gates, resulting in an imbalance between passenger demand and AGM capacity during peak hours.

• 2. The frontal conflict at the south corridor of Line 1 hall

  According to the current flow organization scheme, if the number of inbound passengers continues to increase during peak hours, two backup security inspection devices will be implemented along the south corridor, and the AGM groups G2 and G4 will convert to check-in status. Because the corridor width is about 3 m, the frontal conflicts between outbound passenger flow from G3 and inbound flows to G2 and G4 are inevitable at both ends of the southern corridor.

• 3. The limitation of Line 1 station space

  Due to the historical problems of antecedent demand forecast and station design, the interior space size of Line 1 is relatively small when compared with the space size of Line 3. The current platform length of Line 1 is designed for trains marshalled with 4 carriages (about 80 m), while the platform length of Line 3 targets at trains marshalled with 6 carriages (about 120 m). The limitation of the platform size and carrying capacity is prone to cause passenger congestion and detention, especially under large outbound and transfer-in volumes.

• 4. The passivity of flow organization during inbound passenger peak times

  Commercial and square lands dominate the area surrounding the current metro station, which leads to more recreational and shop** trip demands during weekends and holidays, with an hourly inbound passenger volume over 9000 pax/h, which is 2–3 times higher than the volume during weekdays. Faced with the outpouring of passengers, current flow organization measures are not active enough to efficiently counter the dynamics and complexity.

4.2 Basic flow organization approaches

The key to passenger flow organization inside a station or a public building lies in avoiding or resolving the path interferences among different flows (Li and Zhou 2013). As shown in Fig. 11, the path interferences between different passenger flows can be classified into three types: conflicted interference, frictional interference, and blocked interference.

• 1. The conflicted interference is defined as the vertical or lateral conflict between different passenger paths, where passengers on the non-dominated flow path will experience a longer delay, due to the continuity of passengers on the dominated flow path.

• 2. The frictional interference usually takes place between two frontal passenger flows in a narrow corridor, which will affect the flow velocity and evacuation capacity to some extent.

• 3. The blocked interference refers to the phenomenon where passengers on one path occupy the walking space of the opposite path, forming a flabellate crowd. This kind of interference typically occurs near those entrances or escalator groups with no specific barricades.

Fig. 11

Three interference types between pedestrian traffic flows

When confronted with the three interference types, effective approaches are taken into account from the following three aspects.

• 1. Adjust facility allocation

  Under the influence of peak-time passenger volumes, the phenomenon of queuing before facilities like escalators, TVMs or security devices will prolong the passenger waiting time when the number or location of corresponding facilities are not reasonably allocated. In order to improve the coordination between passengers and facilities, the passing bottlenecks on different paths should be immediately identified via a capacity usage analysis, as the basis of adjusting the facility allocation.

• 2. Enhance passenger guidance

  The passengers of a metro station are usually composed of inbound passengers, outbound passengers and cross-street passengers, while the transfer passengers between different lines should be regarded closely at transfer stations. Faced with the individual behavior of passengers on different paths, active internal guidance is indispensable to avoid delays caused by passenger hesitation and aimless behavior. Passenger guidance should be considered together with path adjustments by duly setting up clear signs and displaying specific service messages.

• 3. Design unobstructed paths

  Passenger flow paths are traditionally designed according to the principle of the shortest path to minimize the walking distance and reduce the spatiotemporal occupation inside the station (Li et al. 2014; Stubenschrott et al. 2014). However, the shortest path does not equate to the unobstructed path due to the above-mentioned path interferences. Therefore, the adjustment of every pedestrian path should take into account both the traffic efficiency and the relationship with other paths from a global perspective.

5 Simulation-based scheme improvement

5.1 Station physical modeling and parameter calibration

The whole simulation model is built in Anylogic 5.8, a micro simulation tool for discrete and systematic systems based on agents, which have been widely used in the field of transportation management (Karaaslan et al. 2018; Lupin et al. 2014). By importing the base map of the DFZM station, the layouts of different layers can be portrayed in detail according to a certain scale, as shown in Fig. 12a. The facilities and equipment of station halls include station entrances, passageways, security areas, TVMs, AGMs, escalators, stairs, columns, isolation belts, etc. The platforms are mainly composed of waiting areas, escalators, stairs, columns, seats and rails. Through systematic analysis on passenger paths and facility layouts, the basic framework is architected in Fig. 12b.

Fig. 12

The 3D model and agent framework of DFZM station in Anylogic 5.8

Table 5 shows the basic input peak-hour volumes for the simulated scenario, which is the most representative scenario on January 1st, 2019 (the New Year’s Day), where the inbound and outbound volumes are both at a high level, the internal transfer volumes are relatively significant, and the total station volume is approximately 3000 pax/h higher than the peak-hour volume of the National Day shown in Fig. 4. Since the parameters of different agent modules vary depending on the scheme of path organization and facility configuration, it is necessary to calibrate them considering relevant passenger characteristics and demands. The major modules that need calibration are pedestrian generation, queue service, route selection, facility attributes, and train operation, and some of the parameters in these modules are indicated in Fig. 13.

Table 5 Input passenger volumes of DFZM during the peak hour (15:00–16:00, January 1st, 2020)

Fig. 13

Parameter calibration of representative agent modules

5.2 Flow organization improvement methodology

5.2.1 Method framework

The method framework of our research is shown in Fig. 14, where the adaptability analysis is performed simultaneously with the simulation procedure, and the possible improvement measures mainly include the exit control, facility configuration, flow reorganization and operation enhancement. Note that the proposed strategy framework is only available for anticipated crowd scenarios without any unexpected accidents.

Fig. 14

The framework of alleviating anticipated station crowdedness

5.2.2 Available station management strategies

According to the simulation results of current bottleneck recognition, the hall and platform of Line 1 are the most problematic areas. Therefore, the improvement measures target path rerouting and facility adjustments to alleviate congestion. The major strategies are presented as follows, and the opportune moment for different strategies will be discussed later along with the graded volumes.

Strategy 1: Faced with the facility pressure at the western hall, E1 and E4 changes to the state of entering only, while E6 keeps exiting only, in order to regulate passenger entries and exits. For the inbound volumes entering the south corridor, entrances E1 and E2 are both available; for the outbound volumes leaving the southern corridor, exits from E1 are forbidden to avoid frictional or blocked interference between inbound volumes and outbound volumes. Similarly, outbound passengers from the northern hall are limited to choose E6 for exiting.

Strategy 2: In the perspective of passenger guidance, the underground commercial corridors surrounding the station hall have been taken into account to coordinate the cross-street pedestrians and station passengers. Figure 15 shows the cross-street paths from east to west, where cross-street volumes are divided into the northern flow and southern flow under necessary guidance and entrance control. More specifically, the northern cross-street pedestrians have to pass through E5 since E6 is exit-only under Strategy 1, and the southern cross-street pedestrians can choose E1 and E2 randomly.

Fig. 15

Possible path distribution for cross-street pedestrians considering surrounding corridors

Strategy 3: Considering the conflicts between queuing passengers and walking passengers, it is necessary to adjust the location and number of some TVMs. The location of four AGMs in the western hall is moved to the southwest corner to avoid conflicts between the queues and entering volumes from E1. Besides, in order to reduce the average queue length and improve the space utilization, the number of AGMs in the northwest corner is increased from 2 to 3, where the extra TVM can be draft from other idle TVM groups placed in Line 3 hall. The improved configuration of TVMs is illustrated in Fig. 16.

Fig. 16

TVMs configuration adjustment in Line 1 hall

Strategy 4: The backup security inspection areas need further adjustments. Under the original scheme, two backup security devices at the south corridor are put into use, and the nearby AGM groups G2 and G4 are converted to the entering state, see Fig. 9a. However, the simulation results indicate that this scheme will cause severe conflicts between entering passengers and exiting passengers in the middle of the station hall. Figure 17 shows the modified scheme, with G3 serving as the check-in gates with isolations, where the outbound passengers can choose E6 and exits of line 3 to leave. Compared with the original scheme, the modified scheme can reduce the path conflicts and increase the queue lengths, while also limiting the arrival rate of entering passengers by shutting down G2 and G4, which will accordingly mitigate the volume impact on the platform of Line 1.

Fig. 17

Integrated flow organization and facility adjustment for crowd management

Strategy 5: As introduced before, the station hall and platform of Line 1 are designed too small to meet the passenger demands during peak hours. To further minimize the instantaneous volume impact on Line 1 facilities, passengers transferring from Line 3 to Line 1 are allowed to pass through the transfer corridor in batches, where the duration and interval of batches are 50 s and 10 s, respectively. The control point is located upstream at the hall of Line 3.

Strategy 6: Under the foregoing measures, the southern AGMs serving as check-in gates will cause inconveniences for outbound passengers towards E3 and E4, and the corresponding selection rate of outbound passengers accounts for 24.46%. This problem can be solved by guiding these outbound passengers towards E3 or E4 through the transfer corridor and the hall of Line 3. However, the volume passing through the transfer corridor will exceed 4000 pax/h, taking into account the bi-directional transfer passengers, which will cause partial jams both near and inside the corridor. Therefore, an anti-clockwise organization for transfer volumes has been devised through concerted efforts of the authors and station managers, as indicated in Fig. 17. Due to the size limitation of the hall and platform of Line 1, passengers from Line 3 to Line 1 are guided through the east corridor with a longer transfer distance, which will reduce the arrival rate to Line 1; while passengers from Line 1 to Line 3 are able to utilize the transfer corridor for the convenience of a quick evacuation.

According to the foregoing analysis, it is inevitable to increase the walking time of partial passengers under the exit control and flow reorganization strategies. Inbound and outbound passengers will get affected to a different degree, and the detour time of different flows has been estimated, as shown in Table 6. Taking Strategy 6 as an example, original passengers leaving from G3 to E3 have to walk through the transfer corridor and choose the check-out gates of Line 3 to exit, and passengers alighting at the platform of Line 3 and leaving from the exits of Line 1 are also affected.

Table 6 Increased walking time cost under exits control and path reorganization

5.2.3 Feasibility of train frequency adjustment

Besides the station-level strategies mentioned above, the train operation scheme can be optimized to enhance the distributing efficiency of platform passengers by compressing the train headway. However, the feasibility of this strategy needs to be validated under different scenarios. For the convenience of train dispatching decisions, a model to calculate the maximum compressible time of train headway has been established as follows, where (11) and (12) show that the compressible headway should be limited by both the number of backup carriages and the signal control requirement.

$$\frac{T \cdot m}{{I_0 - \Delta I}} - \frac{T \cdot m}{{I_0 }} \le n_r \Rightarrow \Delta I \le \frac{n_r \cdot I_0^2 }{{n_r I_0 + T \cdot m}}$$

(11)

$$\Delta I \le I_0 - I_{\min }$$

(12)

$$\Delta I_{\max } = \min \left\{ {\frac{n_r \cdot I_0^2 }{{n_r I_0 + T \cdot m}}, \, I_0 - I_{\min } } \right\}$$

(13)

where T is the turnaround time of a single routing metro line in min; m is the marshaling number of trains in operation; I₀ denotes the average train headway under the original dispatching scheme during peak hours in min; $\Delta I$ is the compressible time of the current train headway in min; n_r is the number of remaining available train carriages for the current line; $I_{\min }$ is the minimum train headway under the technology of signal control and communication in min; $\Delta I_{\max }$ is the maximum compressible time of train headway in min.

According to the Code for Metro Design (GB 50157-2013) released by the Chinese government, the number of maintenance carriages should account for 15% of the number of total carriages, which is equivalent to 85% of the carriages being able to be marshalled for operation. Suzhou Metro Line 1 is equipped with 47 trains, equivalent to 188 carriages. Therefore, the number of available carriages is about 160. Under the original scheme, the train headway is 3 min 25 s (205 s), the marshaling number m is 4, and the turnaround time is 96 min (bi-directional journey time & turn-back operation time), hence the number of carriages in operation is 112. The number of the remaining available carriages is 48. The minimum headway required for signal control is 2 min. Therefore, it can be inferred that the maximum compressible headway is 1.017 min (about 61 s).

The decrease of train headway will decrease the platform queue length, because a shorter interval between arriving trains corresponds to a higher capacity remaining of train carriages, which means that more passengers can board the train without secondary waiting, thus the platform crowdedness can be alleviated. However, the smaller train headway calls for more operating carriages, which will consequently increase the operation investment cost. Aiming at objectives of the average platform queue length and the number of operating carriages, the optimal value for train headway can be figured out, as indicated in Fig. 18, where the suggested headway for trains running on Line 1 is 2 min 50 s. Similarly, the train headway of Line 3 is considered to be 5 min, compared to the original 5 min 50 s.

Fig. 18

The variations of platform queue length and turnover carriages with train headway (Line 1)

5.3 Simulation results discussion

5.3.1 Global evaluation

The simulation results under the improved scheme are listed in Tables 7 and 8, where the indicators of capacity usage, volume density and time consumption are calculated by the corresponding equations in Sect. 2.3, and the maximum queues are extracted from the data counters. The major enhancements are discussed as follows.

Table 7 Capacity usage analysis of key equipment at Line 1 before and after improvement

Table 8 Crowd indicators at different locations under original and improved schemes

The capacity usage of two western security checks has decreased, and the usage of two eastern checks increases, where the current western check gains an obvious decline of over 22%. The improvement has resulted in a more balanced spatial distribution, largely due to the strategies of entrance control and path guidance. Meanwhile, by adding an extra TVM in the east hall, the capacity usage and maximum queue length of the corresponding TVM group have decreased 28% and 36%, respectively, which has effectively reduced the interference between ticket-buying queues and inbounding volumes.

Under the improved scheme, the maximum queue length of the western security check of Line 1 is reduced to 250 pax from 360 pax, with a decreasing amplitude of 31%, which has directly saved the time spent on the station entering path. As indicated in Table 8, the average time consumption from entrance to platform for Line 1 inbound passengers has decreased to 444 s from 516 s. However, the higher inbounding efficiency means a higher passenger arrival rate for the platform, and this is exactly why we have compressed the train headway from 3 min 25 s to 2 min 50 s, in order to alleviate the platform congestion by compressing the train headway.

Since the major flow organization bottlenecks are related to the layouts and facilities of Line 1, most indicators of Line 3 remain consistent with the original scheme, while some parameters appear to have a slight decrease, like the passenger density on the platform and the transfer corridor. According to Strategy 6, the east corridor is used to detour the transfer volumes from Line 3 to Line 1, and the partial outbound passengers of Line 1 are converged into the transfer volumes from Line 1 to Line 3. Therefore, the related density inside the transfer corridor has decreased by 0.2 pax/m² on average.

5.3.2 Direct effects of applied strategies

Strictly speaking, each strategy is a contribution to the global improvement, and it is hard to quantify the contributions due to the coupling effects among strategies since some applied strategies are complementary to each other. However, it is possible to analyze the direct influence of each strategy considering their local effects, as listed in Table 9. It should be noted that not every strategy has a positive direct effect, with some having positive effects, some having negative effects, and some having both positive and negative effects. While from a global perspective, the negative effects of a strategy may also help alleviate the station crowdedness, e.g., under the clockwise flow organization, the increase in transfer passengers’ walking time will lead to a decrease in the volume rate from the Line 3 hall into the Line 1 hall, which in turn leads to a decrease in passenger density, accordingly.

Table 9 Direct effects analysis for major applied strategies

5.4 Volume-based crowd management scheme

The proposed measures can effectively alleviate some local congestions caused by the disequilibrium between passenger paths and facility layouts under original organization schemes. However, considering the limitation of station size and facility capacity, as well as the dynamics of passenger demands shown in Figs. 5 and 6, the applicable volume ranges and coordinated measures should be identified to achieve volume-based management and control. The applicable strategy implementation is illustrated in Fig. 19. Note that Strategy 3 is a static measure of TVM configuration adjustment, which is suggested as a long-term adjustment to reduce the conflicts between the queues of ticket buyers and the flows of inbound or outbound volumes near the corridors.

Fig. 19

The volume-based strategy implementation mechanism

Since the DFZM station is particularly prone to peak-time inbound passenger volumes, the organization schemes are classified into the following five stages according to the 15 min arrival rate from Line 1 west.

• 1. Q_{15 min} ≤ 650 pax. This stage does not involve adopting any additional strategies. The scenario simulation shows that the queue length in the western security area is about 150 m, which is within the acceptable capacity limit for the prescribed queueing area. The simulation results of the inbound volume scenarios from 500 pax/15 min to 650 pax/15 min are indicated in Fig. 20.

• 2. 650 < Q_{15 min} ≤ 1000 pax. Strategy 1 to 3 should be adopted in this stage to reduce the path interferences in the western hall, because the western inbound volume usually accounts for over 65% during peak hours. When Q_{15 min} reaches 1000 pax, the western security inspecting queue length is about 250 pax, and the eastern ticket vending queue length increases to 10 per TVM, which has approached the capacity limit of the current guiding strategies.

• 3. 1000 < Q_{15 min} ≤ 1125 pax. Stage 2 allows Strategy 4 and 6 to redirect some inbound passengers into the south corridor from their original check-in paths. Because the queue areas of the backup security check channels are constrained by the narrow southern corridors, the corresponding shifted volumes are limited accordingly, where the aggregate effective served volume is around 500 pax/h, namely 125 pax/15 min.

• 4. 1125 < Q_{15 min} ≤ 1600 pax. To achieve an efficient demand management, all proposed strategies, except Strategy 5, should be enabled organically. When Q_{15 min} reaches 1600 pax, the average platform passenger density is about 2.73 pax/m², which is over 90% of the standard unit capacity for island platforms (3 pax/m²). Representative simulation results are shown in Fig. 21.

• 5. When the inbound volume exceeds 1600 pax/15 min, the suggested measure is to temporarily shut down partial entrances or close the station to avoid passenger injuries.

Fig. 20

Spatial distribution of the passenger density on Line 1 hall (Q_{15 min} ≤ 650 pax)

Fig. 21

Simulation results of platform passenger distribution (1125 < Q_{15 min} ≤ 1600 pax)

6 Conclusion

Oriented on the passenger demands at a transfer station, this paper presents a coordinated improvement of path organization and facility configuration by taking a typical transfer station as an example. The major contributions lie in the following two aspects. On one hand, the aggregate spatiotemporal characteristics and disaggregate individual characteristics are analyzed in detail upon the multisource data of AFC data, monitoring data and field investigation data, as the basis of parameter calibration. On the other hand, based on the capacity usage analysis of different hall and platform facilities, an applicable mechanism of strategy implementation is presented considering the coordination between volumes and facilities under crowd scenarios. The simulation results during holiday peak hours have validated that the improved scheme outperforms the original in adapting passenger volumes and achieving facility usage equilibrium. The proposed methods of data analysis, facility adjustment, path reorganization, and train headway compression are related. As to the practical implications, this paper presents a systematically designed method from passenger demand analysis, performance quantification, volume control and flow organization for the crowd management at metro stations. When applying the proposed method to other stations, basic rules, principles and strategies can be followed based on the scenario characteristics.

Future means in the planning and operation of metro transfer stations will largely depend on the integration of BIM (Building information modeling) and SMS (Smart metro station). By combining BIM and SMS, a digital twin platform of a metro station can be established to depict the autonomic monitoring, forecasting, analyses and decision-making processes, where different planning schemes and operation strategies can get accurate tests and effective assessments. Our forthcoming research will emphasize improving global strategies and creating emergent evacuation plans for abnormal situations, including screen door malfunctions, station fires, and train accidents. Meanwhile, it would be beneficial to study the psychological changes of different passengers under emergent scenarios, which will be helpful for meticulous parameter calibration and passenger evacuation. Anyway, the presented technology of spatiotemporal data analysis and simulation-based optimization can be used as the decision basis of metro station management for the policy makers.