A cloud model-based method for passenger flow control at subway stations: A real-world case study

Abstract

In the condition of large passenger flow, subway station managers take measures of passenger flow control organization for reducing high safety operation risks at subway stations. The volume of passenger flow in urban railway network operation continues to increase and the Congestion of passenger flow is very high. Passenger flow control measures can greatly give birth to the pressure of transportation and ensure an urban rail transit system’s safe operation. In this paper, we develop a cloud model-based method for passenger flow control, which extends the four-level risk-control grade of a large passenger flow at facilities by considering its fuzzy and stochastic characteristics. Then, an efficient passenger flow control strategy for subway stations is made, where the control time and locations are simultaneously determined. Finally, a station in the Beijing subway is studied to test the validity of the proposed approach. The results show that the time of maximum queuing length is much shorter and the density of passenger flow is lower than existing methods in practice. With the in-depth study of complex network controllability, many studies have applied to control judgment and real network optimization. This paper analyzes the cloud-model-based method for passenger flow control at subway stations and therefore a new method can be incorporated for developing and optimizing control strategies. A few researchers have attempted to find the solution to the problem of crowding risk classification and the passenger flow control strategy. The focus of some studies simultaneously solves the passenger flow control with multiple stations.

Keywords

Subway stations passenger flow control risk-control grade identification cloud model crowding control problem normal cloud

1 Introduction

Crowding occurs at subway stations, particularly the Beijing subway [1, 2] during peak hours, because the passenger volume is so large that not all passengers can be timely served with limited resources. Thus, passenger flow control should be considered by subway managers. The existing crowding risk-control grades for facilities and qualitative methods to handle large passenger flows do not work well without considering the stochastic and fuzzy characteristics of crowding risk and management experience [3 –6]. Therefore, a new method that incorporates these factors must be developed to optimize control strategies.

There are many methods to identify the risk-control grade for a single facility [5 , 7–10]. For example, the crowding on platforms in subway stations was studied during the morning peak hours by simulation [7]. The level of service is used to study the congestion degrees on platforms and examine the passenger responses to discomfort in crowded vehicles and on congested platforms based on stated preference surveys [5]. However, these studies do not consider the uncertain and fuzzy characteristics of crowding risk at facilities despite their advantage of transformation from imprecision or approximate information to quantitative data analysis.

Passenger flow control addresses the crowding problem of stations or trains because of the excess demand in subways [2]. Some studies focused on simultaneously solving the passenger flow control with multiple stations. Furthermore, operations control strategies were considered to reduce passenger crowding pressure [4]. Nevertheless, these determinacy methods usually neglect some uncertain factors or qualitative rules; thus, their usability or robustness is not sufficient. Passenger demand is the major uncertainty factor, in the passenger flow control process. The complexity of the system means that transportation activities frequently deal with a lot of dynamics and uncertainties. Congestion is directly related to the following four parameters for an urban rail transit route with uncertain passenger demand: outside arrival passengers, transfer passengers from neighbouring lines, the schedules of participating trains, and the train capacity.

The stochastic and fuzzy-based risk classification has been successfully applied to transportation systems [11 –15]. To our knowledge, few researchers have attempted to solve the problem of crowding risk classification and strategies of passenger flow control considering the fuzzy and stochastic characteristics of the subway system despite its significance in passenger flow control. The head orientation and the pedestrian for the situation of daily life will manage the data field to make the spatial models to resolve the viral transmission. The risk value of Covid-19 will be ranked for the observed situation of the busy streets for the fewer risk in the street cafes on the markets This redesign will be explored by the venue and the street. For the fairly street to raise in the street cafes for the markets [20].

The cloud model, which was first designed by [16, 17], realizes the bidirectional transformation between qualitative and quantitative objects based on probability statistics and fuzzy set theory, which was used in information systems, human emotion models, and data mining. This paper aims to present a new cloud-based method to distinguish the crowding risks and provide optimal strategies for passenger flow control.

The main contributions of our research can be summarized as follows:

A cloud-based method is developed for passenger flow control to extend the four-level risk-control grade of a large passenger flow at facilities by considering its fuzzy and stochastic properties. An effective passenger flow control strategy for subway stations is then developed, and the control time and locations are simultaneously determined. With the help of this paper’s analysis of the cloud-model-based method for controlling passenger flow at subway stations, a new approach can be used to build and develop control strategies. According to the results, passenger flow density is lower using this method than it is with the current methods in use, and the maximum queuing time is much shorter.

A new cloud-based method is proposed to address the uncertain safety risks problem for operation at subway stations because of the large passenger flow during rush hours. For the first time, different risk indices of facilities are explicitly modelled in a cloud-based model to make control strategies for large passenger flows.

The cloud-based method can solve the passenger flow control problem with the randomness and fuzziness of concepts in traditional analytical methods. The normal cloud model can effectively integrate the randomness and fuzziness of the concepts via its numerical characteristics, which ensures the applicability and practicality of the developed model.

The following parts are organized as follows: Section II describes the problem of the risk-control grade classification of large passenger flows at facilities. In Section III, the cloud-based algorithm is introduced, which combines the risk-control grade identification with the passenger flow control. In Section IV, a real-world case of large passenger flow control at Huixinxijie Beikou Station of the Beijing subway is discussed. Section V provides the conclusions. The sharing of the bike system will maintain the large volume in the trajectory data for the bike returns in the scheduling of the convolution neural networks for the intelligent fashion. This will do by the optimization algorithm in the SDF significant outcomes for the prediction accuracy for the machine learning techniques [21 –24].

2 Problem description

2.1 Crowding risk of facilities

According to the safety standard [18], there are four levels of passenger flow at facilities: free, unobstructed, congested, and heavily congested. Thus, the risk-control grade of the passenger flow is from low to high (I, II, III and IV), as shown in Tables 1 –3. The most crucial technology for increasing metro management standards and performance levels is passenger flow forecasting. It is a vital technological tool for maintaining the constant and sustainable growth of urban transportation. This [25] paper employs mathematical and neural network modelling methods to predict metro passenger flow based on land uses around metro stations, as well as considering the spatial correlation of metro stations within the metro line and the temporal correlation of time series in passenger flow prediction. The passenger density (D), occupied area per person (OA), walking speed (WS), passenger flow volume per meter (PFV), and average queuing length (AQL) is considered indices to assess the safety risk of the facilities. Unfortunately, these values of measure indices are difficult to accurately estimate their stochasticity. Moreover, there may be different criteria for the risk-control grade in individual stations because of the typical features of passengers and the operation requirement of each station [10]. Thus, the fuzzy characteristic of the suggested thresholds should be considered during the identification of a crowding risk.

Table 1
Risk-control grade classification of large passenger flows at platforms and corridors

Risk -control grade Flow state Platform Corridor

OA(m2/p) AQL(d) D (p/m2) WS (m/s)

I Free ≥0.7 ≤1/4d ≤0.7 ≥1.27

II Unobstructed 0.3-0.7 1/4d-1/2d 0.7-1.1 1.15-1.27

III Congested 0.2-0.3 1/2d-3/4d 1.1-1.6 0.78-1.15

IV Heavy congested ≤0.2 ≥3/4d ≥1.6 ≤0.78

Risk -control grade	Flow state	Platform	Corridor
I	Free	≥0.7	≤1/4d	≤0.7	≥1.27
II	Unobstructed	0.3-0.7	1/4d-1/2d	0.7-1.1	1.15-1.27
III	Congested	0.2-0.3	1/2d-3/4d	1.1-1.6	0.78-1.15
IV	Heavy congested	≤0.2	≥3/4d	≥1.6	≤0.78

Note: AQL of a platform is the ratio of its width d.

Table 2

Risk-control grade classification of large passenger flows at ticket equipment areas

Risk -control grade	Flow state	OA(m2/p)	AQL(p)	WS (m/s)
I	Free	≥0.7	≤1	≥1.33
II	Unobstructed	0.3-0.7	1-13	0.5-1.33
III	Congested	0.2-0.3	13-20	0.3-0.5
IV	Heavy congested	≤0.2	≥20	< =0.3

Note: AQL (p) of ticket equipment is the average number of queuing passengers, and p means person.

Table 3

Risk-control grade classification of large passenger flows at stairs

Risk -control grade	Flow state	Up stair			Downstair
		OA	PFV	WS	OA	PFV	WS
		(m2/p)	(p/m/min)	(m/s)	(m2/p)	(p/m/min)	(m/s)
I	Free	≥1	≤32	≥0.57	≥1	≤36	≥0.63
II	Unobstructed	0.6-1	32-43	0.42-0.57	0.7-1	36-48	0.48-0.63
III	Congested	0.3-0.6	43-54	0.22-0.42	0.3-0.7	48-60	0.28-0.48
IV	Heavy congested	≤0.4	Uncertain	Uncertain	≤0.4	Uncertain	Uncertain

Figure 1 shows, the classification of large passenger flows at platforms and corridors, in this table’s representation based on platform and flow state Table 2.

Fig. 1

Classification of large passenger flows at platforms and corridors.

Figures 3 shows, the classification of large passenger flows in the ticket equipment area, in this graphical representation based on flow state and risk control Table 3.

Fig. 2

Classification of large passenger flows at ticket equipment areas.

Fig. 3

Classification of large passenger flows at stairs.

2.2 Crowding risk control strategies for stations

For the entire station, there are rules to manage the risk grade of large passenger flows based on the facilities. In the Beijing subway, the three-grade control of large passenger flows at stations is used as shown in Table 4. According to this method, some control strategies are predefined at the assigned facilities when the station is confronted with different combinations of risk-control grades of facilities.

Table 4
Risk-control grade and corresponding facilities where a passenger flow must be controlled

Risk-control grade Platforms Corridors/stairs Ticket equipment areas

Frist-grade √ × ×

Second-grade √ √ ×

Third-grade √ √ √

Risk-control grade	Platforms	Corridors/stairs	Ticket equipment areas
Frist-grade	√	×	×
Second-grade	√	√	×
Third-grade	√	√	√

The process of the three-grade control method is as follows. The first-grade risk control strategy is used at the entry of platforms if the platform has a high risk of crowded passenger flow. Meanwhile, if corridors or stairs are also at a high safety risk, second-grade control must be applied at the entry of gates. If the worst scenario occurs, where all platforms, corridors, stairs, and ticket equipment areas are under high risk, the third-grade control must be immediately executed at the entrances of the station. The flow chart of the method is shown in Fig. 4. Grades III and IV of facilities are considered high risk in practice. Three control points are involved in passenger flow control.

Fig. 4

Overview of the three-level risk control strategies at stations.

2.3 Discussion

First, the existing methods cannot handle the fuzzy and stochastic crowding risk-control grade of facilities because of the passenger behaviour, and different measured indicators simultaneously change, as mentioned in Section II-A.

Second, the three-level risk control method consists of intractable rules based on three typical scenarios, whose threshold values are fuzzy and not easy to distinguish. Moreover, the impact analysis of the risk control strategies at a given control point remains unclear according to related studies [2]. The unknown effect of risk control strategies on different control points should also be considered.

Third, the three-level risk control method neglects the critical element that the crowding risk-control grades of different facilities react with one another through passenger flow propagation. The unknown reaction that inevitably affects the effect of risk control strategies should be incorporated.

3 Cloud-based method of passenger flow control

The figure illustrates the cloud-based model for controlling excess passengers in the subway railway station. Its fuzzy and stochastic characteristic gives a unique way for passenger control safety as shown in Fig. 5.

Fig. 5

Cloud-based model for passenger control.

A cloud-based method to control large passenger flows is proposed in this section to identify the crowding risk-control grade with fuzziness and stochasticity.

Model-based assessment:

Pedestrians for the orientation of the head will collect in the daily life for the spatially viral transmission for the situation which is ranked by the 19 infections in the fairly busy will make very much few risks compared to other markets this will manage the venue and the street design for the assessment exploration. Here the FFM peg software will be managing d in the area of the orientation for the pedestrians in the head orientation for lower than 20 per cent of the linear interpolation in the python.

3.1 Cloud model

A cloud can be defined as follows. Set U as the domain of a quantitative set that is expressed by precise values. C is a qualitative concept of U. For any element x of U, there is a tendency random number μ (x) ∈ [0, 1], which shows the certain degree of x to U; the distribution of certainty degree on U is called the cloud [12, 17].

The on-demand availability of computer system resources, in particular processing power and data storage (cloud storage), without the user’s direct active oversight, is known as cloud computing. Many times, large clouds divide their operations among several locations, each of which is a data centre. The transmission of hosted services through the internet is referred to as “cloud computing.” The three primary types or categories of cloud computing are services Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS). The cloud model integrates fuzziness with randomness based on three digital characteristics: expected value Ex, entropy En, and hyper entropy He, which is denoted by C (Ex, En, He). Ex is the centre of all cloud drops x, En is the granularity scale of the cloud, and He is the uncertainty of the cloud granularity. From the viewpoint of fuzzy set theory, Ex is the expected sample of a cloud with a membership degree, En is the uncertainty of samples for the cloud, and He is the uncertainty of the membership degree. For example, if one rule “3En is accepted, then the domain (Ex - 3En, Ex + 3En) of all elements is in total contribution to the concept C.

The normal cloud model with the normal distribution and the Gaussian membership function is one of the commonly used cloud models [19]. For example, if x, En, and He² obey the normal distributions, which are probability density functions of En, the following formulas can be obtained:

f_{E n} (x) = \frac{1}{\sqrt{2 π} H e} e^{{(x - E n)}^{2} / 2 H e^{2}}

(1)

f_{x} (x | E n) = \frac{1}{\sqrt{2 π} E n^{'}} e^{{(x - E x)}^{2}} / 2 E {n^{'}}^{2}

(2)

where the variable En′ is first generated from the normal distribution of N (En, He²), and a cloud drop x is produced from the normal distribution of N (Ex, En′).

The degree of real-world value’s membership μ (x) of cloud drop x to concept C is calculated by: $μ (x) = e^{\frac{- {(x - Ex)}^{2}}{2 {({En}^{'})}^{2}}}$ (3) In addition, if there are similar clouds in the domain denoted by C_i (Ex_i, En_i, He_i), the composition (parent) cloud is as followed:

$\begin{matrix} C (Ex, En, He) = C_{1} (E x_{1}, E n_{1}, H e_{1}) \circ \\ C_{2} (E x_{2}, E n_{2}, H e_{2}) \circ \dots \circ C_{N} (E x_{N}, E n_{N}, H e_{N}) \end{matrix}$ (4) where ∘ is the merge operation of clouds, and N is the total number of child clouds.

If the weight parameter of each child cloud is known and defined θ_j, j = 1, 2, . . , N, then the parameters of the parent cloud can be formulated as follows: $Ex = \frac{\sum_{j = 1}^{N} E x_{j} \times E n_{j} \times θ_{j}}{\sum_{j = 1}^{N} E n_{j} \times θ_{j}}$ (5) $En = \sum_{j = 1}^{N} E n_{j} \times θ_{j}$ (6) $He = \frac{\sum_{j = 1}^{N} H e_{j} \times E n_{j} \times θ_{j}}{\sum_{j = 1}^{N} E n_{j} \times θ_{j}}$ (7)

Algorithm for Passenger Flow Control

Input; Set U –quantitative set, C is a qualitative concept of U. x of U, μ (x) ∈ [0, 1] Ex –centre, En –granularity scale, He –uncertainty

If (“3En” ⟶ accepted) then

Domain (Ex - 3En, Ex + 3En) € C

End if

If (x, En, and He² obey⟶normal distributions)

Obtain probability density functions of En as Equations (2)

Compute real-world value’s membership μ (x) as equation (3)

Else If (similar clouds in the domain)

Compute composition cloud as

$\begin{matrix} C (Ex, En, He) = C_{1} (E x_{1}, E n_{1}, H e_{1}) \circ C_{2} (E x_{2}, \\ E n_{2}, H e_{2}) \circ \dots \circ C_{N} (E x_{N}, E n_{N}, H e_{N}) \end{matrix}$

End if

If (θ_j, j = 1, 2, . . , N)

Parameters of the parent cloud formulated as Equations (5)–(7)

End if

3.2 Cloud-based crowding risk identification algorithm

Let I be the total number of risk-control grades for a large passenger flow and J be the number of risk indicators of the large passenger flow at a facility. Let δ = (δ₁, δ₂, . . . , δ_I-1) be the threshold vector of these grades, where δ _i is the vector of risk indicators at rank i of the large passenger flow at a facility and denoted by (δ_i1, δ_i2, . . . , δ_iJ) ^T. Let Ex_ij, En_ij, and He_ij be the expectation, entropy, and hybrid entropy, respectively, of the cloud of the jth risk indicator at the ith grade for the passenger flow. Let x_j be the real-world value of the jth risk index for the large passenger flow at a facility and $δ_{ij}^{s}$ be the standard value of the threshold δ_ij.

A crowding risk identification algorithm is developed based on the cloud model. First, the characteristics of the cloud model for the jth risk indicator at the ith grade level of large passenger flow, which is denoted by TC_ij (Ex_ij, En_ij, He_ij), are estimated. A normal distribution cloud is built to model the risk-control grades from the second to the penult, where the first and last grades are described by a half-upgrade and a half-degrade normal distribution cloud, respectively. Heij is assumed to be constant with a predefined value. Then, the quantitative characteristics of the risk classification cloud for a passenger flow are as follows:

$\begin{matrix} E x_{ij} = \\ {\begin{matrix} \begin{matrix} (δ_{ij}^{s} + δ_{i - 1 j}^{s}) / 2 & \forall i = 2, 3, . . ., I - 1 \end{matrix} \\ \begin{matrix} δ_{ij}^{s} & \forall i = 1 & or & i = I \end{matrix} \end{matrix}, \forall j = 1, 2, . ., J \end{matrix}$ (8)

$\begin{matrix} E n_{ij} = \\ {\begin{matrix} \begin{matrix} (δ_{i - 1 j}^{s} - δ_{ij}^{s}) / I & \forall i = 2, 3, . . ., I - 1 \end{matrix} \\ \begin{matrix} δ_{ij}^{s} & \forall i = 1 & or & i = I \end{matrix} \end{matrix}, \forall j = 1, 2, . ., J \end{matrix}$ (9)

where $δ_{ij}^{s}$ is provided as follows: $\begin{matrix} δ_{ij}^{s} = \\ \frac{δ_{ij} - min {δ_{ij} | i = 1, 2, . ., I - 1}}{max {δ_{ij} | i = 1, 2, . ., I - 1} - min {δ_{ij} | i = 1, 2, . ., I - 1}}, \\ \forall j = 1, 2, . ., J \end{matrix}$ (10)

As mentioned in Section II, different indices are applied to measure the risk of a large passenger flow at facilities, so the related clouds should be merged with a parent (composition) cloud to analyze its performance. Thus, we develop a template (parent) cloud model of risk indicators for a large passenger flow, which is denoted by TC_i (Ex_i, En_i, He_i) and determines whether the current risk grade is up to the minimum threshold for the passenger flow control. Based on Equations (4)–(7), the characteristics of the cloud are as follows: $E x_{i} = \frac{\sum_{j = 1}^{J} E x_{ij} \times E n_{ij} \times θ_{j}}{\sum_{j = 1}^{J} E n_{ij} \times θ_{j}}, \forall i = 1, 2, . ., I$ (11) $E n_{i} = \sum_{j = 1}^{J} E n_{ij} \times θ_{j}, \forall i = 1, 2, . ., I$ (12) $H e_{i} = H e_{ij}, \forall i = 1, 2, . ., I$ (13)

Furthermore, we build the real-world indicator cloud (Composition) of passenger flow TC (Ex, En, He) based on real data as the following sub-algorithm 1:

CROWDING RISK IDENTIFICATION –algorithm 1:

Step 1: risk index x_j, normal value

$x_{j}^{s}$ Is calculated by Equation (10),

Compute membership degree ( $μ_{i} (x_{j}^{s})$ ) is determined by Equations (1)–(3)

If (x_j > max {δ_ij|i = 1, 2, . . , I - 1})

$x_{j}^{s} = 1$

if x_j < min {δ_ij|i = 1, 2, . . , I - 1},

$x_{j}^{s} = 0$

Step 2: Considering the fuzziness and randomness of the cloud model $w_{i} = \frac{μ_{i} (x_{j}^{s})}{\sum_{i} μ_{i} (x_{j}^{s})}$ (14) Obtain passenger flow ${TC}_{j}^{R} ({Ex}_{j}^{R}, {En}_{j}^{R}, {He}_{j}^{R})$ by (15)–(17). ${Ex}_{j}^{R} = \frac{\sum_{i = 1}^{I} E x_{ij} \times E n_{ij} \times w_{i}}{\sum_{i = 1}^{I} E n_{ij} \times w_{i}}$ (15) ${En}_{j}^{R} = \sum_{i = 1}^{I} E n_{ij} \times w_{i}$ (16) ${He}_{j}^{R} = H e_{ij}, \forall i = 1, 2, . ., I$ (17)

Step 3: For all risk indicators based on real data do

Repeat Step 1 and Step 2 obtained Weight parameters generated ⟶ real-world indicator cloud $Ex = \frac{\sum_{j = 1}^{J} {Ex}_{j}^{R} \times {En}_{j}^{R} \times θ_{j}}{\sum_{j = 1}^{J} {En}_{j}^{R} \times θ_{j}}$ (18) $En = \sum_{j = 1}^{J} {En}_{j}^{R} \times θ_{j}$ (19) $He = H e_{i}, \forall i = 1, 2, . ., I$ (20)

Return, a real-world indicator

Finally, the crowding risk by the similar possibilities between the real-world indicator and the template cloud based on sub-algorithm 2 is identified as shown in Table 5. Larger similar possibilities between the real-world indicator and a particular grade level of template cloud correspond to more possibilities to which the risk level belongs

Table 5

Procedures of sub-algorithm 2

Step 1: Generate a cloud drop x_l from the real-world indicator cloud model TC (Ex, En, He);

Step 2: Draw a random number En_i′ based on a normal distribution N (En_i, (He_i) ²) using the template cloud TC_i (Ex_i, En_i, He_i) at the ith risk-control grade of passenger flow at a facility;

Step 3: Calculate the degree membership to the ith risk-control grade of passenger flow for cloud drop x_l (y_l∈i) according to Equation (3);

Step 4: Repeat the above steps until the total number of cloud drops reaches M;

Step 5: Calculate the similar possibilities between the ith template and real-world indicator cloud y_i by: y_i = ∑y_l∈i/M;

Step 6: The normal similar possibilities r_i can be given by:

r_{i} = y_{i} / \sum_{i = 1}^{I} y_{i}

3.3 Large passenger flow control based on could model

Let r_i⩾k be the possibility that the real-world risk-control grade is no lower than the kth grade of the large passenger flow at a facility, whereas r_i<k is the opposite possibility. We can obtain these values at a given k according to the algorithm in Section III B.

The crowding risk of large passenger flow at a facility is at the kth risk-control grade when the rule r_i⩾k ⩾ r_i<k is satisfied based on the cloud model. Thus, we can identify the crowding risk of a facility from its threshold of k in Section II-A.

After the crowding risk of large passenger flow at a facility is identified, control strategies can be obtained by implementing the three-grade control at stations as mentioned in Section II B.

4 Case study

In this section, Huixinxijie Beikou Station in the Beijing subway is used to illustrate the application of the cloud-based method. The proposed method outperforms existing passenger flow control methods in practice in terms of the lasting time of the maximum queuing length and the density of passenger flow when the crowding risk is handled during peak hours.

4.1 Station description and parameter analysis

Huixinxijie Beikou Station has a two-layer structure with side platforms, three exits and entrances, and two types of passenger flow including inbound and outbound flows, as shown in Table 6 and Fig. 6. There are eight typical streamlines of passenger flow in two directions (up and down) in this station, whose proportions are provided in Table 7, according to the related data collected in 5-minute intervals from 7 : 00 am to 9 : 00 am. Using Matlab and Analogic software, the passenger distribution propagation in the station is obtained, and the risk-control grades at the facilities are evaluated based on Section II-A. Then, the critical facilities with a higher density of passenger flow are identified, such as the A/B entrance and platform. The down direction with a higher density is selected to demonstrate the risk-control grade identification and passenger flow control in the following sections. Moreover, one peak moment with the highest density, which is 7 : 35 am, is selected to simplify the analysis.

Table 6
Parameters of the facilities in Huixinxijie Beikou Station

No Type of facility Size (m) Density (p/m2) Capacity (p/m/min)

A, B, C Exit &Entrance – – 60

23–24 Walkway 4×50 1.67 60

21–22 Security Machine 1×5 2.56 32

19–20 Ticket gate 1×6 2.56 35

13, 15, 16, 18 Stair 1.5×2 2.33 40

14, 17 Escalator 1 2.33 60

1–12 Gate at the platform – 2 –

No	Type of facility	Size (m)	Density (p/m2)	Capacity (p/m/min)
A, B, C	Exit &Entrance	–	–	60
23–24	Walkway	4×50	1.67	60
21–22	Security Machine	1×5	2.56	32
19–20	Ticket gate	1×6	2.56	35
13, 15, 16, 18	Stair	1.5×2	2.33	40
14, 17	Escalator	1	2.33	60
1–12	Gate at the platform	–	2	–

Fig. 6

Directed topology graph of Huixinxijie Beikou Station.

Table 7

Proportions of different passenger flow streamlines at Huixinxijie Beikou Station

No	Streamline description	Streamline	Proportion
1	1# Inbound in the up direction	A/B-24-22-20-18-11/12	15%
2	2# Inbound in the up direction	C-23-21-19-16-7/8/9/10	5%
3	1# Inbound in the down direction	A/B-24-22-20-15-5/6	65%
4	2# Inbound in the down direction	C-23-21-19-13-1/2/3/4	15%
5	1# Outbound in the up direction	11/12-18-20-24-A/B	35.5%
6	2# Outbound in the up direction	7/8/9/10-16/17-19-23-C	19.1%
7	1# Outbound in the down direction	5/6-15-20-24-A/B	29.5%
8	2# Outbound in the down direction	1/2/3/4-13/14-19-23-C	15.9%

For the platforms, the weight parameters of the cloud models are $θ_{OA}^{P} = 1 / 3$ $θ_{AQL}^{P} = 1 / 3$ and. According to the collected data, we set $x_{OA}^{P} = 0.32$ m2/p $x_{AQL}^{P} = 2 / 3 d$ , where d is the length of the platform. For the downstairs, the weight parameters are $θ_{OA}^{S} = θ_{PFV}^{S} = θ_{WS}^{S} = 1 / 3$ . The actual values of the risk indicators are $x_{OA}^{S} = 0.58$ m2/p, $x_{PFV}^{S} = 45$ p/m/min, and $x_{WS}^{S} = 0.52$ m/s.

4.2 Passenger flow control based on the cloud-based model

First, we determine whether the first-grade passenger flow control should be used. According to the risk-control grades of the platforms in Table 1, the clouds for the risk indicators of the large passenger flow at the platform (TC_ij (Ex_ij, En_ij, He_ij)) are generated, whose quantitative characteristics are calculated using Equations (8)–(10). Then, a template cloud model of risk indicators (TC_i (Ex_i, En_i, He_i)) is made, as shown in Table 8. Next, the real-world indicator cloud TC (Ex, En, He) is obtained, whose quantitative characteristics are provided in Table 9. The similarity possibility between the real-world indicator and the template cloud is obtained by generating 5000 cloud drops based on sub-algorithm 2 as follows: $[r_{1}, r_{2}, r_{3}, r_{4}] = [0.0000, 0.0002, 0.8898, 0.1100]$ (21) where the rule r_i⩾III ⩾ r_i<III is satisfied as shown in Fig. 3a. Thus, the crowding risk of a large passenger flow at the platform is at grade III, which satisfies the criteria of the first-grade passenger flow control.

Table 8

Quantitative characteristics of template clouds for the risk indicators at the facilities

Risk -control grade	Flow state	Platform			Downstair
		Ex	En	He	Ex	En	He
I	Free	1.000	0.078	0.01	1.000	0.079	0.01
II	Unobstructed	0.764	0.078	0.01	0.761	0.079	0.01
III	Congested	0.269	0.089	0.01	0.263	0.087	0.01
IV	Heavy congested	0	0.089	0.01	0	0.087	0.01

Table 9

Quantitative characteristics of real-world risk indicator clouds at the facilities

Index	Platform			Downstair
	OA	ALL	Composition	OA	PFV	WS	Composition
Ex	0.271	0.249	0.260	0.249	0.750	0.674	0.552
En	0.1	0.083	0.089	0.083	0.083	0.071	0.079
He	0.01	0.01	0.01	0.01	0.01	0.01	0.01

Second, we check whether the second-grade passenger flow control is applied. The quantitative characteristics of risk indicators for the downstairs are shown in Tables 8 9. The similarity possibility between the template and the real-world indicator cloud is as follows: $[r_{1}^{S}, r_{2}^{S}, r_{3}^{S}, r_{4}^{S}] = [0.0028, 0.7645, 0.2325, 0.0002]$ (22)

Thus, the rule $r_{i ⩾ III}^{S} < r_{i < III}^{S}$ is satisfied at the downlink stairs as shown in Fig. 3b. The crowding risk of a large passenger flow at the downlink stairs is at grade III, and thus, the condition for the second-grade control is not satisfied.

In summary, the first-grade strategy of the passenger flow control should be used, whereas the second-grade control is not necessary.

Semantic features of the geographic features:

The graph structure will manage the knowledge embedding for the social sensing and the geographic extension will make the semantic information for the convolutional neural network this manages the social media data as shown in Tables 8 9. Image semantic segmentation is a technique for dividing images into numerous distinct regions with their features and extracting the objects of interest. Image semantic segmentation technology can segment and label certain targets in remote sensing images to collect specific data in remote sensing image research. A neural network contains at least two physical components: processing elements and connections between them. Neurons are the components used for processing, while links are the connections that connect them. There is a weight parameter assigned to each link. The three main components of the neural network are the input layer, the processing layer, and the output layer. This needs KE-CNN for determining a large number of people in the traditional machine learning process for the seamless and social sensing of the social information as shown in Fig. 7.

Fig. 7

Quantitative characteristics of template clouds for the risk indicators at the facilities.

Figure 7 shows, the quantitative characteristics of template clouds for the risk indicators at the facilities, in this graphical representation is based on flow state and risk control as shown in Fig. 8.

Fig. 8

Similarity possibilities between the template and real-world clouds at the platform and downstairs.

4.3 Analysis of passenger flow control strategies

In this section, we discuss the facilities where passenger flow control strategies should be used. Two commonly used types of control facilities are illustrated: ticket equipment areas and entrances [2].

First, we use the first-grade passenger flow control at ticket equipment areas and evaluate the performance of the platform in terms of the AQL, as shown in Fig. 8. Compared with the scenario without passenger flow control, the AQL with flow control is much shorter, and the density of the large passenger flow is lower, which implies that this control strategy helps to alleviate the crowding risk at the platform as shown in Fig. 9.

Fig. 9

Queuing length distribution at the platform.

After the control strategy at the ticket equipment area is used, more passengers will gather here, and a new crowding risk will generate. Then, other strategies should be considered. Thus, we introduce a passenger flow control strategy at the A/B entrances and evaluate the following performance of the ticket equipment area. The time that the maximum queuing length lasts significantly decreases, and the density of the passenger flow decreases after the additional strategy execution at the entrances, as shown in the Figs. 10 11. For this case, the combination control at the ticket equipment areas and entrances is a better choice.

Fig. 10

Queuing length at the ticket equipment area.

Fig. 11

Density distribution in the north hall.

5 Conclusions

In this paper, the problem of passenger flow control at stations was discussed. First, a cloud-based method was developed to identify the crowding risk at facilities based on the existing four-level risk-control grade classification. Then, a strategy of passenger flow control based on this cloud model was made. Finally, a real-world case illustrates the process of risk-control grade identification and large passenger flow control, which demonstrates the effectiveness of our proposed method. The cloud model is first used to grasp the fuzziness and probability of the crowding risk for large passenger flows, which will help to improve the accuracy. Accordingly, the strategies of large passenger flow control are made to decrease the crowding risks at stations.

In future, it is worthwhile to formulate a systematic analysis framework of crowding risk control with large passenger flows. Furthermore, passenger flow propagation models can be integrated into the dynamic process of crowding risk identification.

Footnotes

Acknowledgments

This work is partially supported by the National Key R&D Program of China (No. 2020YFB160070X). The authors also gratefully acknowledge the helpful comments and suggestions of the reviewers, which have improved the presentation.

References

, Li

, Liu

, Ran

and Qin

, Passenger flow control with multi-station coordination in subway networks: algorithm development and real-world case study, Transport Metrica B 7(1) (2019), 446–472.

, Liu

, Li

and Jiang

, Capacity-oriented passenger flow control under uncertain demand: Algorithm development and real-world case study, Transportation Research Part E: Logistics and Transportation Review 87 (2016), 130–148.

Yin

, Tang

, Yang

, Gao

and Ran

, Energy-efficient metro train rescheduling with uncertain time-variant passenger demands: An approximate dynamic programming approach, Transportation Research Part B: Methodological 91 (2016), 178–210.

, Lam

, Cheung

C.Y.

and Lam

C. F.

, A study of crowding effects at the Hong Kong light rail transit stations, Transportation Research Part A: Policy and Practice 33(5) (1990), 401–415.

, Gao

, Xiao

and Li

, Modeling the pedestrian’s movement and simulating evacuation dynamics on stairs, Safety Science 70 (2014), 189–201.

, Li

, Xu

, Liu

and Ran

, Modeling departure time choice of metro passengers with a smart corrected mixed logit model –A case study in Beijing, Transport Policy 69 (2018), 106–121.

Jiang

C.S.

, Deng

Y.F.

, Hu

, Ding

and Chow

W.K.

, Crowding in platform staircases of a subway station in China during rush hours, Safety Science 47(7) (2009), 931–938.

H”anseler

F.S.

, Bierlaire

and Scarinci

, Assessing theusage and level-of-service of pedestrian facilities in trainstations: A Swiss case study, Transportation Research Part A: Policy and Practice 89 (2016), 106–123.

Wang

, Huang

and Guo

, Optimization of emergency materialdispatch from multiple depot locations to multiple disaster sites, Sustainability (Switzerland) 9(11) (2017), 1–8.

10.

Zhang

, Jia

and Qin

, Optimal number and location planning of evacuation signage in public space, Safety Science 91 (2016), 132–147.

11.

, Yang

, Gao

and Li

, Robust train regulation for metro lines with stochastic passenger arrival flow, Information Sciences 373 (2016), 1339–1351.

12.

Karyotis

, Doctor

, Iqbal

, James

and Chang

, A fuzzy computational model of emotion for cloud-based sentiment analysis, Information Sciences 433–434 (2018), 448–463.

13.

, Xie

, Li

and Qin

, Learning the route choice behaviour of subway passengers from AFC data, Expert Systems with Applications 95 (2018), 324–332.

14.

Wang

, Qin

, Xu

and Jia

, A fuzzy optimization model for high-speed railway timetable rescheduling, Discrete Dynamics in Nature and Society (2012).

15.

Zhang

, Jia

, Wang

and Xu

, Energy consumption optimizationof train operation for railway systems: algorithm development andreal-world case study, Journal of Cleaner Production 214 (2019), 1024–1037.

16.

Yang

, Yan

, Peng

and Gao

, Encoding words into Cloud models from interval-valued data via fuzzy statistics and membership function fitting, Knowledge-Based Systems 55 (2014), 114–124.

17.

, Meng

and Shi

, Membership clouds and membership cloud denerators, Journal of Computer Research and Development 32(6) (1995), 15–20.

18.

China National Standard, GB 50157-2003. Code for Design of Metro 2013.

19.

Wang

, Xu

and Li

, Generic normal cloud model, Information Sciences 280 (2014), 1–15.

20.

Garcia

, Mendez

, Fray

and Nicolas

, Model-based assessment of the risks of viral transmission in non-confined crowds, , Safety Science 144 (2021), 105453.

21.

Qiao

, Han

, Huang

, Yue

, Mao

, Shu

and Wu

, A dynamic convolutional neural network based shared-bike demand forecasting model, ACM Transactions on Intelligent Systems and Technology (TIST) 12(6) (2021), 1–24.

22.

Liu

, Qin

, He

, Mi

and Yu

, A new multi-data-driven spatiotemporal PM2. 5 forecasting model based on an ensemble graph reinforcement learning convolutional network, Atmospheric Pollution Research 12(10) (2021), 101197.

23.

, Wang

and Fang

, GE-STDGN: a novel Spatio-temporal weather prediction model based on graph evolution, Applied Intelligence (2021), 1–15.

24.

Zhu

, Pivaro

, Gupta

and Canova

, Safe Model-based Off-policy Reinforcement Learning for Eco-Driving in Connected and Automated Hybrid Electric Vehicles. arXiv preprint arXiv:2105.11640 (2021).

25.

Lin

, Wang

, Wu

and Gong

, Passenger flow prediction based on land use around metro stations: a case study, Sustainability 12(17) (2020), 6844.

A cloud model-based method for passenger flow control at subway stations: A real-world case study

Abstract

Keywords

1 Introduction

2 Problem description

2.1 Crowding risk of facilities

Table 4 Risk-control grade and corresponding facilities where a passenger flow must be controlled Risk-control grade Platforms Corridors/stairs Ticket equipment areas Frist-grade √ × × Second-grade √ √ × Third-grade √ √ √

3 Cloud-based method of passenger flow control

Model-based assessment:

4 Case study

4.1 Station description and parameter analysis

Semantic features of the geographic features:

Footnotes

Acknowledgments

References

Table 4
Risk-control grade and corresponding facilities where a passenger flow must be controlled

Risk-control grade Platforms Corridors/stairs Ticket equipment areas

Frist-grade √ × ×

Second-grade √ √ ×

Third-grade √ √ √