A Statistical Study of the Pedestrian Distribution in a Commercial Wholesale Centre Based on the Traffic Spatial Structure

Abstract

Designing reasonable commercial transportation space is of great significance to enhancing the overall value of commerce. This study takes the classic cases of three typical plans of commercial wholesale centres (CWCs) as the research object, uses space syntax to analyse the connectivity of the architecture of the traffic space, simulates the current situation of the pedestrian flow distribution through ArcGIS, and constructs a multiple regression model for the association between connectivity and pedestrian flow distribution. The results of this study show that for CWCs with a single business type and a focus on traffic efficiency, the distribution of the pedestrian flow is most affected by the main entrance and the vertical traffic distribution. For different types of planes, when the commercial traffic space has strong symmetry, its group relationship is more concise. While a uniform form is more conducive to achieving a balanced distribution of commercial traffic, the asymmetry of the plan layout helps to establish a psycho-spatial map for the visitors. In addition, the commercial value of the first floor is greatly influenced by the layout of the building entrance, and the commercial value of the second floor is greatly influenced by the location of the vertical transportation. In conclusion, when commercial building development planning is in the stage of traffic flow design, the use of space connectivity traffic data can assist in the selection of construction plans and predict the distribution of the economic value in CWCs.

1. Introduction

A professional wholesale market is a commercial area that mainly sells one kind of commodity and its substitutes and complements, adopting the retailing and wholesaling of commodities as the dominant mode of operation, with few other modes of business [1]. A commercial wholesale centre (CWC) can be described as a fifth-generation integrated commercial platform based on professional markets, including commodity procurement, exhibition, and promotion, office services, and modern logistics service functions. There are differences between a CWC and traditional shopping centres in terms of the commercial operation: (1) CWCs do not have traditional commercial anchor shops, there are fewer differences in the use of space for shops of different sales units, and there is no significant difference between a single type of business and the provision of services in attracting commercial flows. (2) CWCs pay more attention to traffic efficiency and have low requirements for spatial interest and experience; so, they mostly adopt a grid-type traffic structure mode, where the traffic flow line has a balanced layout [2]. (3) Brands with similar commercial value are mostly placed on the same floor for sales, and there is no obvious difference in the customer demand for goods in different areas on the same floor. Therefore, compared with traditional shopping centres, the commercial viability of different areas of a CWC is mainly affected by the spatial layout and traffic structure, and the differences in the types of businesses, shop sizes, and services provided on the same floor have a limited impact on the commercial viability. The vitality of a commercial shop mainly comes from the flow of people, and the number of visits converted from the flow of people constitutes a positive relationship with the commercial value of the shop [3]. It has been shown that shops in all areas within a mature commercial centre receive better pedestrian flow, balancing commercial vitality and contributing to the maximisation of comprehensive commercial value [4,5]. In the early stage of building design, commercial managers can take the initiative to optimise the balance in the distribution of pedestrian flow after completion through the optimisation of the traffic structure in order to balance the commercial value and make the building as a whole have a uniform commercial vitality [6,7].

Existing research on pedestrian flow statistics includes short-term and long-term monitoring, with a difference in the application scenarios; short-term monitoring of the pedestrian flow is mainly applied in real-time traffic flow control, including traffic diversion, congestion prevention, tourist diversion, and disaster prevention and control [8,9]; long-term monitoring of the pedestrian flow is mainly applied as periodic traffic flow prediction for urban management and planning measures, and it is mostly used in urban facility planning, commercial market analysis, and functional space planning [10,11]. The pedestrian flow in a CWC is affected by a combination of factors, and the total pedestrian flow changes all the time. However, the distribution of the pedestrian flow in each area within the CWC is mainly affected by the functional layout, the traffic flow, the commercial activities, and the management diversion measures [4]. In the operation of a mature CWC, the above factors change less, and the distribution of the pedestrian flow shows a strong regularity and periodicity [12,13,14]. Therefore, the research object is in line with the selection of the characteristics of the long-term monitoring of pedestrian flow.

There are different methods for the short-term and long-term monitoring of flows. Short-term monitoring includes the following: (1) Monitoring network and mobile device pedestrian positioning research, with the use of surveillance, infrared, physical excitation- and electric field sensing-based techniques, mobile phone Bluetooth, WIFI, etc., by recording the spatial localisation of customer foot traffic in the building during a certain time period [15,16,17]. Xiaobing Ding and Zhigang Liu predicted the changes in the metro passenger flow during peak hours and proposed optimisation suggestions by building a data collection device integrating high-definition image acquisition and WIFI detection technologies [18]. (2) Providing public transport management services for intelligent transport system research, such as the use of the LSTM-BILSTM method to collect urban road traffic flow information and provide intelligent guidance to alleviate traffic congestion and reduce environmental pollution [19]. Methods for the long-term monitoring of pedestrian flow include the following: (1) Artificial intelligence for pedestrian flow distribution learning, where the collection method combines different algorithms with a large number of graphs to obtain the spatial distribution characteristics of pedestrian flow [20]. Notable research results in recent years, such as the YOLO-V3 target detection method and the multi-target tracking (MTT) algorithm, have introduced deep learning to footfall statistics [21,22]. Andres Musalem used an efficient and scalable video analytics approach to determine the relationship between customer visits, consumption behaviour, and service support [23]; Zhiyuan Wei used fine-grained demographic mobility data and proposed a phased and disaggregated data-driven framework for analysing people–mobility patterns at the community level to aid in the prediction of service demand at key facilities [24]. (2) Building or urban transport network accessibility research through quantitative indicators of the spatial structure of transport to generate models of human flow distribution and carry out classification research [25]. Guibo Sun and Chris Webster took Hong Kong as an example, with the establishment of a three-dimensional pedestrian network, to study the role of three-dimensional transport in reshaping network connectivity and the integration of a high-density urban walking environment [26].

The above studies on pedestrian flow statistics have certain limitations: (1) The integrity of the original data on pedestrian flow was high; however, areas or specific buildings with a low degree of data openness often suffered from prediction errors due to the lack of data on pedestrian flow. (2) Image acquisition could not be carried out widely because it involves personal data security and privacy issues. Therefore, in this study, we adopted the combination of an easily accessible building–traffic network structure and long-time pedestrian flow statistics, which has the advantages of moderate data demand, easy accessibility, and no need for post-completion statistics, i.e., it was carried out at the design stage of commercial building programmes.

The transportation spatial network system serves as an important component of cities and buildings, and a large number of studies have confirmed the strong correlation between the transportation network and the development of urban areas and functional layout, which directly affects social phenomena and residents’ behaviours [27,28]. Current research methods using transport networks can be divided into three types:

• First, multivariate data-based urban street network analysis. Li Qian took the street network of Chengdu, China, as the unit of analysis, combined taxis, shared bicycle trajectories, user comments, cultural facilities, and points of interest with multi-source data for spatial identification, and mined the centres of high social, economic, and cultural vitality through the establishment of a multivariate regression model [29]. Andres Sevtsuk synthesised existing and proposed environments to predict pedestrian volumes using an intermediate exponential model of network analysis, which was effectively validated in community planning and large-scale development projects and provided a guiding direction for pedestrian amenity improvements and public investment [30].
• Second, the comprehensive study of urban three-dimensional spatial networks. Spatial design network analysis (sDNA) was used to analyse urban vertical multi-storey road networks to predict social behaviour and traffic flow characteristics [31]. Srilalitha Gopalakrishnan used a spatial network spatial analysis framework to carry out a connectivity analysis of vertical open spaces and high-rise infrastructure in Singaporean cities to study the promotion of the efficient integration of vertical public spaces on enhancing the social and public spatial benefits [32]. Ian Harvey and Scott Orford combined multiple hybrid spatial design network analysis (MH-sDNA) to carry out the modelling of urban population change, pioneering the prediction and validation of the migration of an urban functional layout over time [33].
• Third, the synthesis of spatial syntax and related methods. Mahbub Rashid sampled the street networks of 25 independent UK cities and found that urban compactness based on spatial network descriptions correlated with social equity indicators, with large differences in the correlations between different types of cities [34]. In addition, SeoJin Ha and SuJin Jang synthesised the space syntax and eigenvector ratio of the adjacency matrix to study the status of the association between the indoor commercial space characteristics of large-scale commercial centres and the efficiency of revenue generation [35]. Andi synthesised space syntax and VGA maps to consider the factors of pedestrian flow, horizontal traffic complexity, vertical traffic complexity, tenant type distribution, visual quality, retail layout, and the anchor layout of shopping centres and proposed spatial layout adjustments to optimise the shopping experience in shopping centres [36].

It can be seen that multivariate data network model analysis is applicable to public spaces with social, economic, and land use functional attributes for urban-scale and area-wide traffic research; spatial design network analysis has better research potential for multi-layer three-dimensional spaces on an urban scale; and the integrated research of spatial syntax can be applied in behavioural research, triggered by people’s perception of the environment on an architectural spatial scale, e.g., in the research on the environmental behaviours of commercial centres, museums, and other public service facilities. Although a large number of scholars have used road networks to predict urban traffic flow, there is a relative lack of research on the distribution of pedestrian flows inside buildings and their commercial economic value. Therefore, this study makes use of the easily accessible data of the internal traffic network to study the characteristics of pedestrian flow distribution and provide commercial managers and designers with a fast and efficient method to assess commercial value, which can be carried out at the stage of design proposals. The ultimate goal is to optimise the traffic network structure to achieve a balanced flow of commercial traffic and enhance the comprehensive commercial value. The main research content of this paper includes the following: (1) taking the CWC as the research object, based on the spatial syntax method, we study the connectivity characteristics of the building traffic structure and screen the proxy variable indicators; (2) combined with the actual operation factors of a commercial operation, we construct a multiple regression model to predict the pedestrian flow.

2. Materials and Methods

We took the yearbook sample of “China’s top 100 commodity markets” as the research object and first carried out a typological study to determine the typical representatives of each type as the research unit. Second, we analysed the spatial grouping relationship of the research units and obtained the internal traffic connectivity variables of the businesses through the spatial syntax method. Finally, a prediction model for the distribution of commercial foot traffic was constructed, with the actual observed and counted foot traffic as the dependent variable, while the proxy variables of traffic connectivity, building entrances, and vertical traffic were used as the independent variables to construct a multivariate regression model and to test the validity of the prediction (Figure 1).

2.1. Research Unit

For this study, we collected the top 100 CWCs in terms of the annual transaction value, according to the China Commodity Exchange Market Statistical Yearbook 2021, as the research object. Through the traffic space network structure and topological graphic characteristics inside commercial buildings, the typology of the top 100 CWCs in China was studied (See Appendix A,

Table A1. China’s 2021 top 100 commercial wholesale centres list.

Number	Commercial Wholesale Centres	Forms
1	Zhejiang Yiwu Xiaoshangpin Market	Axisymmetric grid
2	Zhejiang Shaoxin Qingfang Centre	Axisymmetric grid
3	Wuhan Hankou Guoji Shangpin Centre	Axisymmetric grid
4	Changsha Gaoqiao Market	Asymmetric grid
5	Baiding Baigouzhen Market	Axisymmetric grid
6	Suzhou Dongfang Sichou Market	Asymmetric grid
7	Haining Pige Centre	Axisymmetric grid
8	Changshu Fuzhuang Centre	Cross-symmetric grid
9	Tongxiang Puyuan Yaomaoshan Market	Asymmetric grid
10	Xinfadi Nongfu Chanpin Market	Cross-symmetric grid
11	Jintian Yangguang liansuo Market	Asymmetric grid
12	Yuyao Suliao Centre	Asymmetric grid
13	Nanjing Nongfuchangping Centre	Axisymmetric grid
14	Shenzhen Huaqiang Dianzi Centre	Asymmetric grid
15	Nantong Dieshiqiao Guoji Jiafang Centre	Cross-symmetric grid
16	Chengdu Guoji Shangmao Centre	Axisymmetric grid
17	Zhoushan Guoji Shuichang Centre	Asymmetric grid
18	Changzhou Dongshou Shangmao Centre	Cross-symmetric grid
19	Xinfadi Nongfuchangpin Wuliu Market	Axisymmetric grid
20	Shandong Kaisheng Nongchangpin Centre	Asymmetric grid
21	Shenzhen Chayue Centre	Axisymmetric grid
22	Zhenhai Dazhong Jiaoyi Centre	Asymmetric grid
23	Changsha Hongxin Market	Axisymmetric grid
24	Yongkang Keji WujinCentre	Asymmetric grid
25	Shangqu Zhongxin Market	Cross-symmetric grid
26	Wantian Jiancha Shanmao Centre	Asymmetric grid
27	Weierkang Roulei Shuichang Market	Asymmetric grid
28	Liaoning Xiliu Fuzhuang Centre	Cross-symmetric grid
29	Wuxi Buxiugang Wuliu Centre	Asymmetric grid
30	Xuzhou Xuanwu Market	Cross-symmetric grid
31	Hangzhou Yifa Fushi Centre	Asymmetric grid
32	Anqing Guangcai Market	Axisymmetric grid
33	Hunan Gangcai Market	Asymmetric grid
34	Shaoxin Yuezhou Qingfang Market	Axisymmetric grid
35	Jiangsu Lingjiatang Market	Cross-symmetric grid
36	Tangshan Yahongqiang Market	Axisymmetric grid
37	Ningbo Huadong Wuzi Centre	Asymmetric grid
38	Wuxi Wuzhou Guoji Zhuangshi Centre	Cross-symmetric grid
39	Lianyungang Donghai Shuijing Centre	Asymmetric grid
40	Haiwaihai Hangzhou Qiche Centre	Axisymmetric grid
41	Zhuji Huadong Guoji Zhubao Centre	Cross-symmetric grid
42	Dongyang Mudiao Centre	Axisymmetric grid
43	Weifang Haode Maoyi Market	Axisymmetric grid
44	Jinan Likou Fuzhuang Centre	Asymmetric grid
45	Qianqing Qingfang Yuanliao Centre	Axisymmetric grid
46	Langfang Huancheng Guoji Qiche Centre	Axisymmetric grid
47	Haozhou Zhongyaocai Market	Cross-symmetric grid
48	Guiyang Xinan Shanmao Centre	Asymmetric grid
49	Dongyang Huamu Jiaju Centre	Asymmetric grid
50	Changzhou Zhouqu Dengju Market	Asymmetric grid
51	Liaocheng Guangcai Market	Axisymmetric grid
52	Zhejiang Datang Qingfang Centre	Cross-symmetric grid
53	Wenzhou Yide Xiebo Centre	Cross-symmetric grid
54	Guangzhou Jiangnan Guocai Market	Asymmetric grid
55	LuQiao Riyongpin Market	Cross-symmetric grid
56	Wuxi Jinshu Cailiao Market	Axisymmetric grid
57	Shenzhen Shuibei Zhubao Centre	Asymmetric grid
58	Zhejiang Linhang Jinshu Centre	Asymmetric grid
59	Weifang Qingzhou Huahui Centre	Axisymmetric grid
60	Wuxi Chaoyang Market	Axisymmetric grid
61	Ningbo Shihua Changping Market	Asymmetric grid
62	Shenzhen Haijixing Wuliu Centre	Axisymmetric grid
63	Zibo Shafa Jiaju Market	Axisymmetric grid
64	Ningbo Shihua Changpin Market	Asymmetric grid
65	Linzhe Ruixin Qimo Centre	Axisymmetric grid
66	Guangzhou Baima Fuzhuang Market	Asymmetric grid
67	Ningbo Yetihuagong Jiaoyi Market	Axisymmetric grid
68	Qingdao Jimo Fuzhuang Market	Axisymmetric grid
69	Handan Yongnian Biaozhunjian Market	Axisymmetric grid
70	Hangzhou Liangyou Wuliu Market	Cross-symmetric grid
71	Suzhou Nanhuanqiao Pifa Market	Cross-symmetric grid
72	Changsha Huangxin Nongchangpin Market	Axisymmetric grid
73	Zhejiang Jili Tongzhuang Market	Cross-symmetric grid
74	Zhexin Huamu Centre	Asymmetric grid
75	Shandong Taishan Gangcai Market	Asymmetric grid
76	Jiaxin Changsichou Market	Cross-symmetric grid
77	Jiangsu Huagongpin Jiaoyi Market	Asymmetric grid
78	Shenzhen Huanan Cheng	Cross-symmetric grid
79	Tengzhou Nongfu Changpin Centre	Asymmetric grid
80	Jiangsu Huadong Wujin Centre	Axisymmetric grid
81	Dezhou Haitao Shangcheng Company	Axisymmetric grid
82	Zhejiang Ruian Centre	Cross-symmetric grid
83	Dongyue Guoji Huamu Centre	Asymmetric grid
84	Dongguang Xinli Guoji Maoyi Centre	Asymmetric grid
85	Jinan Zhongheng Centre	Axisymmetric grid
86	Pinghu Fuzhuang Centre	Axisymmetric grid
87	Jinxiang Dasuan Guoji Jiaoyi Market	Cross-symmetric grid
88	Guangzhou Guoji Qingfang Centre	Asymmetric grid
89	Xintai Qinghe Yangrong Centre	Asymmetric grid
90	Hangzhou Xinshidai Jiaju Centre	Asymmetric grid
91	Jiangyin Guangzhuan Jingsu Market	Asymmetric grid
92	ZhangJiagang Fangzi Yuanliao Market	Asymmetric grid
93	Linzhe Xiaoshangpin Centre	Axisymmetric grid
94	Linpin Jiangnan Guoji Sichou Centre	Axisymmetric grid
95	Changzhou Xiaxi Huamu Market	Asymmetric grid
96	Tengzhou Jiayu Shangmao Centre	Asymmetric grid
97	Shijiazhuang Xinji Pige Centre	Axisymmetric grid
98	Chongqin Yuzhong Daronghui	Asymmetric grid
99	Haining Jiafang Centre	Cross-symmetric grid
100	Zhejiang Nanxun Jiancai Market	Asymmetric grid

) and categorised into three types: an asymmetric grid structure, an axisymmetric grid structure, and a cross-symmetric grid structure.

Type I: asymmetric grid structure, where the building plan traffic shows a high degree of freedom, according to the urban flow of the layout of the building entrances, the flexible organisation of the grid-like commercial space, the traffic space, and the auxiliary space. Type II: axisymmetric grid structure, where the building traffic entrance caters to the main flow of people, the plan space pattern is mirrored by the core traffic axis, showing obvious left–right symmetry, commercial and auxiliary spaces are distributed in a grid, and the regional division is obvious. Type III: the cross-symmetric grid structure belongs to a special type of axial symmetric grid structure, and the functional spaces inside the building strictly follow the symmetry law. There is a lack of consideration for the actual distribution of urban pedestrian flow, and the entrance space, functional layout, and commercial shops focus on the logical relationship of regular grid graphics.

From the above three types of CWC, samples with a similar construction scale, better business conditions, and more comprehensive information data were selected for this study. Due to the limitations of the research and to ensure the credibility and validity of this study, the first and second floors of each sample were selected as the research unit (

Table 1. General plan classification of commercial wholesale centres.

Forms	Asymmetric Grid	Axisymmetric Grid	Cross-Symmetric Grid
Topology graphics
Sample
	CWC1	CWC2	CWC3
	Guangzhou Guojiqingfang	Yiwu Guoji Shangmao	Shenzhen Huanan Centre

).

Through onsite interviews and research and based on the technical floor plans provided by the staff of the management centres of the three study subjects, the researchers drew a sketch of the floor layout of the first and second floors of each sample (

Table 2. Spatial type division.

Forms	Asymmetric Grid	Axisymmetric Grid	Cross-Symmetric Grid
First floor	CWC1-1	CWC2-1	CWC3-1
Second floor
	CWC1-2	CWC2-2	CWC3-2

). This study drew on the introduction to architectural planning and the related literature to analyse the space typology and classified the internal use space of the CWCs into three categories: A space (activity space); B space (block space) and C space (circulation space), referred to as BC Space; and D space (auxiliary space). Among them, the A space was the main use space of the CWC, i.e., the fixed commercial shops. The BC space was a horizontal and vertical transportation space that connected the commercial functions, transported the flow of people to the various areas of the CWC, and balanced the interests of each merchant. The D space was the auxiliary service space, including offices, logistics, public evacuation, parking spaces, etc., which ensured the normal operation of the business. The BC space was the main carrier of the organisation of the traffic line, which directly affected the distribution of the commercial flow and the overall commercial value. Therefore, the BC space was studied as the focal space of this research.

Space syntax research suggests that the least number of axes can be used instead of actual traffic corridors to measure the connectivity of traffic structures [37]. This study established axial models for three types of sample BC spaces as a base model for quantitative analysis (

Table 3. Axis map of traffic space.

Item	CWC1	CWC2	CWC3
First floor
Second floor

).

2.2. Data Acquisition

The general building plan data were obtained from Google Maps on 1 January 2023, while the plan of the first and second floors of the buildings was drawn through field research, onsite measurements, records, and comprehensive comparisons. The study sample footfalls were obtained through measured statistics.

A total of 62 investigators were dispatched for the study, in 3 groups, and statistical research was conducted on each sample. Matching was conducted for 20 observation points on the first floor and 20 on the second floor of CWC1, 20 observation points on the first floor and 20 on the second floor of CWC2, and 22 observation points on the first floor and 22 on the second floor of CWC3. Footfall counts were conducted on a daily basis during commercial business hours (9 a.m.–6 p.m.) at the selected traffic locations within the building using mobile phones to record videos of traffic passing through the lateral sections of the traffic corridors and conduct footfall counts. The onsite statistics found that rapid changes in foot traffic occurred half an hour after the start of business and half an hour before the end of business each day at the CWC, which interfered with the statistical results. In addition, holidays had an impact on the overall distribution of commercial foot traffic. Therefore, the date of counting was chosen to be a non-holiday, with all the shops in normal operation status; so, the distribution of pedestrian flow in each area was more stable [12,38]. The final recording dates were scheduled for 9 October 2023 to 13 October 2023 (Monday to Friday, working days), and the time slots for the morning recording were 10:00 a.m.–10:15 a.m. and 11:00 a.m.–11:15 a.m., with 3:00 p.m.–3:15 p.m. and 4:00 p.m.–4:15 p.m. for the afternoon, for a total of four times. The duration of a single recording was 15 min, and a total of 60 min was recorded at each observation point (see Appendix B, Figure A1, Figure A2 and Figure A3).

3. Results

3.1. Descriptive Statistics Analysis

For this study, the YOLO-V3 (Python loader package) target detection method was used to count the pedestrian flow information on the video, and the actual measured pedestrian flow counts of the three samples totalled 119,097. Based on the cross-section foot traffic statistics of each sample, the foot traffic values were entered using kriging spatial interpolation [37], and an Arc GIS geographic information map of the foot traffic was created to simulate the distribution of the foot traffic on the first and second floors of each sample as the basis of the data statistics (

Table 4. Interpolation analysis of the pedestrian flowrate.

Item	CWC1	CWC2	CWC3
First floor
Second floor

).

The first floor of sample CWC1 showed a trend of a gathering distribution of pedestrian flow, mainly concentrated in the core atrium and the central channel of the east–west and north–south directions, and the central commercial entrances in the west and south sides played a more obvious role in the gravitational force of the pedestrian flow. The distribution of the second floor was more balanced than that of the first floor, the atrium was still the place of convergence, the distribution of the west side of the pedestrian flow decreased gradually, the east side of the distribution of the pedestrian flow contour was dense, and the phenomenon of a sudden decrease in the pedestrian flow in this area occurred. The first floor of sample CWC2 concentrated the core of the pedestrian flow in the atrium traffic intersection between the commercial sub-districts and formed a continuous commercial dynamic line with the transverse traffic channels of the sub-districts, through which the pedestrian flow penetrated into the interior of the sub-districts. The second floor of the pedestrian flow was concentrated in the core of the interior of the sub-districts, and the layout of the atriums between the sub-districts impeded the formation of a transverse traffic dynamic line, which constituted relatively independent commercial sub-districts. For sample CWC3, the first floor of the east side was adjacent to the main road and public transportation hub, and the flow of people was from the east side of the building, showing the phenomenon of density in the east and sparseness in the west; at the same time, due to a lack of commercial penetration, the first floor away from the centre had a rapid decline in the flow of people. The second-floor distribution of pedestrian flow was more balanced, along the core of the core of the central traffic flow line of the horizontal distribution, and the north–south sides had a limited penetration of the second floor at the edge of the plane area. There was insufficient pedestrian flow in the edge area of the second floor.

Through the onsite data statistics, we determined that the distribution of pedestrian flow had a close correlation with the layout of the traffic dynamic elements of the CWC, and the space area adjacent to the main commercial entrance, general entrance, escalator, lift, and staircase showed a significant increase in the commercial pedestrian flow. Therefore, in order to enhance the credibility of pedestrian flow prediction, it is necessary to introduce commercial traffic entrances and traffic elements as variable factors.

According to the theory of spatial syntax, the axial map can be interpreted using the four basic variables of integration, connectivity value, depth value, and choice value as the traffic connectivity indicators [39]. Integration is expressed using Rn, which responds to the degree of closeness of the connectivity with other spaces centred on a certain space, reflecting its attraction to the surrounding traffic flow, and it determines whether the traffic accessibility is high or low. The connectivity value is expressed using Cn, which is the number of direct connections between a certain area of traffic and adjacent traffic within a building, and its value reflects the potential chance of being selected as a traffic path, i.e., the superiority or inferiority of the connectivity. The average depth value is expressed by Mn, which is used to evaluate the topological distance from any space to any other space, with the value reflecting the ease of reaching that space. The choice value is expressed using Ch, which is the total frequency of a transportation space as the shortest topological distance between two nodes in a transportation system, and its value reflects the frequency of that transportation space being selected for the shortest path planning.

Related studies have shown that integration degree (Rn) axis maps can be used for core traffic and regional accessibility analysis [38,40,41]. This study calculated the axial model of each sample through Depthmap and obtained the integration metric model of each layer of the sample, whose central axis was closer to red with higher Rn values and closer to blue with lower Rn values, as shown in

Table 5. Rn value axis map.

Item	CWC1	CWC2	CWC3
First floor
Second floor
Rn value

. For the asymmetric grid structure sample CWC1, the first layer of the integration of high-value axes was located in the central region of the core traffic, with the formation of a small area of traffic circulation, surrounding secondary traffic away from the core, and the integration gradually reduced. The second layer of the integration of high-value axes was located in the plane of the left side of the north–south tandem traffic, the right side of the traffic was blocked by the atrium, and the integration value was intermediate. The axisymmetric grid sample CWC2 had a similar distribution of high-integration axes on the first and second floors, which were located in the central core traffic area of each sub-district and were efficiently connected through the east–west tandem traffic lines of each sub-district; however, the integration of the traffic axes on the upper side of the plane was lower. The axis with high values of integration between the first and second floors of the cross-symmetric grid sample CWC3 was located in the central core traffic in the east–west direction. Because the first floor was affected by the atrium layout in many places, the integration values of the traffic axes on the upper and lower sides were lower than the values of the traffic axes in similar spatial areas on the first floor, and the high-value integration axes on the first floor formed a loop connecting all the zones.

A comparative statistical analysis of the values of Rn, Cn, Mn, and Ch in the axes of the layers of the study sample was carried out (

Table 6. Central tendency and statistical dispersion for metrics.

Item	Floor	Metrics	Min	Max	Average	Std Dev	Median
CWC1 (N = 118)	1F (N = 43)	Rn	0.742	2.491	1.428	0.461	1.267
		Cn	1	9	2.967	2.389	2
		Mn	2.033	4.467	2.987	0.601	3.006
		Ch	0.003	0.466	0.094	0.115	0.046
	2F (N = 75)	Rn	1.048	2.987	1.602	0.386	1.532
		Cn	1	15	3.718	3.003	3
		Mn	2.229	4.501	3.414	0.534	3.412
		Ch	0.002	0.661	0.066	0.108	0.038
CWC2 (N = 263)	1F (N = 125)	Rn	0.767	2.431	1.627	0.367	1.599
		Cn	1	13	5.673	3.029	5
		Mn	2.691	6.361	3.665	0.644	3.561
		Ch	0.003	0.296	0.063	0.071	0.0317
	2F (N = 138)	Rn	0.784	2.547	1.615	0.332	1.521
		Cn	1	13	5.261	3.034	5
		Mn	2.681	6.464	3.772	0.611	3.816
		Ch	0.003	0.359	0.058	0.072	0.036
CWC3 (N = 217)	1F (N = 100)	Rn	1.794	3.848	2.383	0.631	2.137
		Cn	1	28	6.012	5.282	5
		Mn	1.987	3.117	2.689	0.361	2.784
		Ch	0.002	0.391	0.052	0.072	0.032
	2F (N = 117)	Rn	1.521	3.918	2.336	0.632	2.241
		Cn	1	30	4.871	5.783	5
		Mn	2.062	3.734	2.896	0.455	2.695
		Ch	0.002	0.306	0.035	0.058	0.026

). In the asymmetric grid structure sample CWC1, the first floor was connected by a single corridor, and the second floor was connected by an atrium combined with a double sub-corridor. The overall spatial accessibility of the commercial stores on the second floor (Rn average = 1.602) was better than that of the first floor (Rn average = 1.428), and the spatial connectivity of the second floor (Cn average = 3.718) was better than that of the first floor (Cn average = 2.967). However, the spatial selectivity of the first floor (Ch average = 0.094) was better than that of the second floor (Ch average = 0.066). The pedestrian flow on the second floor penetrated into the neighbouring areas on the same floor through the double sub-corridors, and the traffic structure of the first floor easily led to the concentration of the pedestrian flow in the middle of the core; however, the average depth of the second floor was larger (Mn average = 3.414), and the commercial value of the stores on the edge of the plane showed a rapid decrease. The first floor of the axisymmetric grid structure sample CWC2 and the traffic of the second floor were connected by a single channel. Due to the obstruction of lateral traffic movement caused by the atriums of each partition on the second floor, the Rn, Cn, Mn, and Ch values of the first floor were slightly better than those of the second floor, and the topological distances of the end areas of the second floor were lower than those of the first floor. The cross-symmetric grid structure sample CWC3 first and second floors had the same traffic mode, with the core traffic using the atrium combined with a double corridor and the rest of the area using a single-channel hybrid layout mode. The second floor of the commercial store space division scale was more flexible and diverse than the first floor, the density of traffic was greater, and the Rn, Cn, and Mn values were slightly better than those of the first floor; however, the Ch value of the first floor was better than that of the second floor.

When comparing the first-floor traffic connection characteristics of the three types horizontally, the CWC3 cross-symmetric grid structure had the best spatial accessibility (Rn average = 2.383), which showed that the stores in each commercial area had a reasonable chance for contact with the flow of people, and the CWC2 axisymmetric grid structure had the best traffic connection (Cn average = 5.673), which better promoted the diffusion of commercial flows to each area. The average depth value (Mn average = 2.689) was the lowest for CWC3, which meant there was easier access to the areas with lower commercial value at the edge of the plane. The CWC1 asymmetric grid structure had the best spatial selectivity (Ch average = 0.094), reflecting a more balanced and reasonable opportunity for each traffic space to be selected for penetration.

Professor Hillier, who proposed the theory of spatial syntax, pointed out that whether for urban or architectural spaces, residents need to gradually build up a mental picture of the whole spatial system by walking through it [42]. The theory defines the degree of spatial comprehensibility as the strength of the ability to reflect the overall spatial structure through the local spatial structure [43]. If the global connectivity is highly positively correlated with the integration degree, it indicates that the spatial system is better understood; in contrast, when the information obtained through the local space easily misleads pedestrians, this reflects the poor comprehensibility of the spatial system. In terms of urban spatial planning, the comprehensibility can effectively identify urban historical core areas and new areas [44] and can be used as an evaluation index of urban pedestrian space quality, which can help to improve the spatial walking experience in pedestrian-friendly cities [45]. In commercial building space research, some scholars have used comprehensibility to analyse the spatial recognition of shopping centres, optimise visitors’ walking experience, and effectively improve the defects of a poor commercial building atmosphere [36]. A large number of studies have verified the wide applicability of comprehensibility in urban and architectural spaces [46]. CWC pedestrians can establish a spatial mental map of the overall business through localised area walking experiences. The creation of a mental map assists walkers in quickly determining their spatial orientation and exploring the surrounding area, which aids in the radial penetration of commercial activities.

According to the theory of spatial syntax, we quantified the comprehensibility of the BC space in the CWCs by calculating the numerical ratio of the global integration and the connectivity indexes of the spatial axes, establishing the Rn–Cn two-dimensional scatter coordinates map (

Table 7. Rn–Cn scatter coordinates.

Item	CWC1	CWC2	CWC3
First floor
Second floor
Rn value

), and obtaining its corresponding fitted straight line with the regression coefficient R² [42,47]. When R² < 0.5, the local space fails to reflect the global space; when 0.5 < R² < 0.7, the local space reasonably reflects the global space; when 0.7 < R² < 1.0, the local space better reflects the global space. The regression coefficients of the first and second layers of the sample of the asymmetric grid structure of CWC1 were both higher than 0.7, and the spatial structure of the traffic space was helpful for the commercial flow to quickly build up spatial mental maps and carry out commercial activities. The CWC2 axisymmetric grid structure had a higher regression coefficient than 0.7. The R² values of the regression coefficients of the first and second floors of the CWC2 axisymmetric grid structure and the CWC3 cross-symmetric grid were all located in the range of 0.5–0.7, reflecting that commercial flows could build up a spatial mental map after a local tour, which could effectively assist the infiltration of commercial behaviours. The spatial comprehensibility of the CWC1, CWC2, and CWC3 samples decreased sequentially, reflecting that spatial diversity and asymmetry help walkers to establish local spatial characteristics that differentiate them from neighbouring areas.

3.2. Correlation Analysis

The correlation study on the connectivity indicators of different floors of the samples showed that the Rn, Cn, Mn, and Ch values had significant correlations, so it was necessary to identify the redundant variables and obtain the proxy scalar with the highest correlation with the flow rate to help optimise the flow rate prediction model (

Table 8. Pearson’s correlation analysis of each metric.

Forms	Metrics	Rn	Cn	Mn	Ch	Pedestrian Flowrate
CWC1-1	Rn	1	0.829 **	−0.981 **	0.858 **	0.829 **
	Cn	0.829 **	1	−0.803 **	0.806 **	0.645 *
	Mn	−0.981 **	−0.803 **	1	−0.889 **	−0.732 *
	Ch	0.858 **	806 **	−0.889 **	1	0.619
	Pedestrian Flowrate	0.829 **	0.645 *	−0.732 *	0.619	1
CWC1-2	Rn	1	0.869 **	−0.977 **	0.893 **	0.863 **
	Cn	0.869 **	1	−0.909 **	0.689 **	0.621
	Mn	−0.977 **	−0.909 **	1	−0.821 **	−0.804 **
	Ch	893 **	0.689 **	−0.821 **	1	0.806 **
	Pedestrian Flowrate	0.863 **	0.621	−0.804 **	0.806 **	1
CWC2-1	Rn	1	0.780 **	−0.993 **	0.740 *	0.691 *
	Cn	0.780 **	1	−0.776 **	0.745 *	0.510
	Mn	−0.993 **	−0.776 **	1	−0.695 *	−0.683 *
	Ch	0.740 *	0.745 *	−0.695 *	1	0.621 *
	Pedestrian Flowrate	0.691 *	0.510	−0.683 *	0.621 *	1
CWC2-2	Rn	1	0.289	−0.997 **	0.801 **	0.657 *
	Cn	0.289	1	−0.272	0.576	0.592
	Mn	−0.997 **	−0.272	1	−0.793 **	−0.655 *
	Ch	0.801 **	0.176	−0.793 **	1	0.641 *
	Pedestrian Flowrate	0.657 *	0.592	−0.655 *	0.641 *	1
CWC3-1	Rn	1	0.503	−0.994 **	0.859 *	0.623
	Cn	0.503	1	−0.490	0.588	0.509
	Mn	−0.994 **	−0.490	1	−0.843 **	−0.619
	Ch	0.859 *	0.588	−0.843 **	1	0.819 **
	Pedestrian Flowrate	0.623	0.509	−0.619	0.819 **	1
CWC3-2	Rn	1	0.529	−0.735 *	0.890 **	0.656 *
	Cn	0.529	1	−0.630	0.649 *	0.542
	Mn	−0.735 *	−0.630	1	−0.926 **	−0.637 *
	Ch	0.890 **	0.649 *	−0.926 **	1	0.629 *
	Pedestrian Flowrate	0.656 *	0.542	−0.637 *	0.629 *	1

* p < 0.05, ** p < 0.01.

). In CWC1, the correlation of Rn, Cn, Mn, and Ch was significant, the correlation between Rn, Cn, and Ch and Mn was negative, the correlation of the first floor Ch with the flow rate was weak, and the correlation of Rn with the flow rate was the most significant. Ch had a weak correlation with the flow of people, and Rn had the most significant correlation with the flow of people. In CWC2, there was a correlation between Rn, Cn, Mn, and Ch in the first layer, Cn had no correlation with Rn, Mn, and Ch in the second layer, and Rn had a reasonable correlation with the flow of people in the first and second layer. In CWC3, there was no correlation between Cn and Rn and Mn in the first and second layer, Rn, Ch, and Mn had a negative correlation, and Rn was more significantly correlated with the flow of people. The correlation was more significant. Summarising the above data, it can be seen that Rn has more potential as a proxy variable for connectivity than Mn, Cn, and Ch. For the asymmetric grid structure samples, Rn had a significant correlation with the pedestrian flowrate, and the correlation level between the axisymmetric and cross-symmetric grid structures was reasonable but not significant. Therefore, in order to improve the validity of this study’s predictions, after analysing the onsite pedestrian flow statistics, we introduced the two variables of commercial entrances and vertical traffic to improve the accuracy of the predictions.

3.3. Multiple Regression

In order to improve the validity of this study’s predictions, we introduced two variables, commercial entrances and vertical traffic, to improve the accuracy of the predictions. According to the floor plan layout of the CWC1, CWC2 and CWC3 samples, the introduction of main entrances and ordinary entrances corrected the values of the spatial axes that were closely related to them, which were assigned values 1–2 in turn; the introduction of staircases, vertical lifts, and escalators corrected the values of the spatial axes that were closely related to them, which were assigned values 1–3 in turn.

Based on the analysis of the connectivity indicators as proxy variables, we selected commercial entrances (X1), vertical traffic (X2), and integration (X3) as the independent variables and foot traffic (Y) as the dependent variable to establish a multiple linear regression model. According to the regression linear model data of the first and second floors of the three types of research samples (

Table 9. Topology forms and study sample.

Sample	Variable	B	t	p	VIF	Sig.	R2	Durbin–Watson
CWC1-1	X1	6.838	1.100	0.297	2.201	0.000	0.844	2.121
	X2	19.849	4.513	0.001	1.102
	X3	29.881	2.292	0.045	2.071
CWC1-2	X1	2.097	0.541	0.601	1.027	0.001	0.806	1.748
	X2	8.027	2.259	0.047	1.724
	X3	10.592	3.113	0.011	1.729
CWC2-1	X1	5.074	1.873	0.075	1.304	0.000	0.726	2.321
	X2	9.899	3.456	0.002	1.270
	X3	12.444	3.007	0.007	1.348
CWC2-2	X1	6.636	3.310	0.000	1.482	0.000	0.781	2.297
	X2	4.199	2.463	0.003	1.461
	X3	4.688	3.344	0.109	1.964
CWC3-1	X1	0.331	1.631	0.017	2.535	0.001	0.604	1.941
	X2	5.744	1.318	0.020	3.001
	X3	5.174	0.599	0.055	1.307
CWC3-2	X1	10.133	2.239	0.035	1.379	0.000	0.615	2.135
	X2	6.192	2.298	0.031	1.486
	X3	6.807	2.529	0.019	1.651

), the Durbin–Watson test statistic was between 1.5 and 2.5, which indicated that there was no autocorrelation among the independent variables X1, X2, and X3, and it passed the test of independence; the covariance test statistic of VIF was between 1 and 5, which was in a reasonable range, indicating there was no multicollinearity problem between the independent variables. The significant coefficients S of the regression linear models were all lower than 0.05, indicating statistical significance. The coefficients of determination of the first- and second-level linear regression models of CWC1 (R² = 0.844, 0.806) were greater than 0.8, with a high degree of fit, and the independent variables explained more than 80% of the changes in the dependent variables. The coefficients of determination of the first- and second-level linear regression models of CWC2 (R² = 0.726, 0.753) and CWC3 (R² = 0.604, 0.615) were greater than 0.6, which were in a reasonable range of fit, with the independent variables explaining more than 60% of the variation in the dependent variable. When comparing the multiple linear regression models, the asymmetric grid structure was better than the axisymmetric grid structure, which was better than the cross-symmetric grid structure.

The results of the multiple linear regression analysis showed that when conducting the assessment of the commercial vitality of a CWC, the statistics of the pedestrian flow in different areas on the same floor of a CWC can be obtained with a linear regression model with an explanatory level of no less than 60% by adopting the degree of integration in the spatial syntax as a proxy variable and taking into account the commercial entrances and the vertical traffic as the dependent variables.

4. Discussion

This paper took CWCs as the research object and established an in-building pedestrian flow prediction model by introducing three variable factors, namely, commercial entrances, vertical traffic, and integration, which is a technical method with the advantages of efficiency, easy implementation, and the ability to be carried out at the stage of scheme design. The research method used here was quite different from the widely used research techniques of a monitoring network, mobile device pedestrian positioning technology, and artificial intelligence crowd distribution learning. The application advantages are mainly reflected in the following aspects: (1) In the application of statistical practice, it is applicable to the indoor traffic flow prediction of commercial buildings. In the design stage of the building scheme, it provides the basis for predicting the pedestrian flow after the completion of the building, assisting commercial operators and builders in reducing subjective judgement to optimise the traffic space of the scheme [36]. (2) In terms of the difficulty of realising research, for the areas or places where the degree of openness of the public data is relatively low, by obtaining the spatial structural plan layout and the main cross-section of the human flow, it can assist in the speculation of the remaining layout of the human flow space. This method can be used as an effective alternative when it is difficult to obtain multivariate information data [48]. (3) In terms of the protection of personal information security, the prediction model is built without the need to collect pedestrian flow information through surveillance or mobile devices, which respects the personal information and privacy of the counted walkers [49].

Compared with existing studies, the technical methods used in this study have the following disadvantages: (1) In terms of statistical efficiency, network monitoring, WIFI detection, mobile phone Bluetooth, and the infrared detection of pedestrian flow statistics technology can provide real-time feedback on the latest changes in pedestrian flow [15,16,17], whereas the present research method is less efficient in predicting efficiency, and it needs to prioritise the acquisition of building traffic structures to carry out the subsequent prediction. (2) In terms of statistical accuracy, this study does not have the real-time location of pedestrian distribution with high confidence, and it is suitable for fuzzy judgement prediction with moderate accuracy requirements, which is in line with the cyclical prediction over a longer term. In addition, in the research object of this paper, a CWC, the sales content is of the same type, providing a single service, with a similar shop space area and no traditional commercial main shop setup. The commercial value of similar brands is mostly placed on the same floor for sales, with customers on the same floor, and the demand for different areas is not significantly different. Therefore, the attraction ability of the A space to the flow of people is similar. However, for common commercial types, the impact of the A space on the distribution of foot traffic is an important consideration.

This research had certain limitations. The proxy variables of the pedestrian distribution prediction model established in this study were based on the spatial syntactic topological relationship, while the distribution of the pedestrian flow in reality should take into account commercial buildings’ operation cycles and management mode factors; hence, this research method does not fully reflect the actual traffic situation. Second, the research object was a CWC, and the applicability to an ordinary commercial plaza needs to be verified. In addition, the factors that determine the pedestrian flow in a commercial space are not only integration, entrances and exits, and vertical traffic distribution, but they also need to incorporate the external commercial atmosphere of the building, the urban traffic, and so on, into the system, which is a comprehensive and complex problem. Furthermore, this study was limited by conducting comprehensive statistical analyses and the validation of forecasting models for the top 100 research samples of China’s merchandise market in 2021. Finally, although the focus of this paper was on the B and CC spaces, the impact of the difference between the A space and D space on the distribution of pedestrian flow remains to be verified.

To address the above shortcomings, the research accuracy can be improved by supplementing and validating the research cases, and the influence of A space differences on the distribution of pedestrian flow needs to be reassessed. In order to extend the study to general commercial plazas, its applicability can be improved by introducing more and generally applicable parameters.

5. Conclusions

As a special type of business, a CWC differs from a traditional shopping centre in terms of its business operations, spatial layout, and industry settings. In this study, we selected the top 100 commercial wholesale centres in China, in terms of annual turnover, as the research object. This study was carried out on the basis of the traffic space network structure and topological graphic characteristics inside commercial buildings, which were divided into three types: an asymmetric grid structure, an axisymmetric grid structure, and a cross-symmetric grid structure. The vitality of commercial shops mainly comes from the flow of people, and the shops in each area inside a mature CWC obtain a better flow of people, achieving the goal of balancing the commercial vitality and maximising the comprehensive commercial value. Compared with traditional shopping centres, the pedestrian flow in different areas of a CWC is mainly affected by the spatial layout and traffic structure, and the differences in the types of businesses, shop sizes, and services provided on the same floor have a limited impact on the commercial vitality. Therefore, the BC space was determined as the focus of this study, with the A space and D space as subsequent research branches. The traffic space structure analysis technique adopted in this study has the advantages of easy implementation, moderate data demand, and the protection of public privacy compared with the current mainstream traffic counting techniques (including extensive network monitoring, mobile device positioning, and artificial intelligence distribution learning); however, it has lower statistical timeliness and statistical accuracy.

Based on the spatial syntax theory, the BC space of the three types of CWC samples was interpreted in terms of connectivity metrics, in which the quantitative metrics of Rn, Cn, Mn, and Ch were all correlated with the flow of the respective samples, and the Rn correlation was the most significant. The cross-sectional comparative analysis revealed that the cross-symmetric structure sample performed better in integration and average depth, and the asymmetric structure sample performed best in comprehensibility. In addition, the onsite statistics of the pedestrian flow of each sample found that, in addition to the traffic structure network, commercial entrances and exits and different types of vertical traffic modes were the key factors affecting the distribution of the pedestrian flow. Therefore, commercial pedestrian flow was selected as the dependent variable, Rn as the representative variable of traffic connectivity, and building entrances and vertical traffic as the independent variables; together, they formed a multiple regression model to assess visitor statistics. The results of this study show that the introduction of the dependent variable effectively improved the statistical prediction ability of pedestrian distribution in a CWC (CWC1: R² = 0.844/0.806, CWC2: R² = 0.726/0.753, CWC3: R² = 0.604/0.615), which effectively guaranteed the scientificity of the prediction of pedestrian flow under a certain spatial organisational relationship. In addition, the main conclusions of this study are as follows:

• The influence of the spatial grouping relationship on the spatial accessibility and comprehensibility is more significant. When the grouping relationship of a CWC is more concise and the morphology is more unified, it effectively guides consumers to specific commercial behaviours in different spatial environments.
• The spatial asymmetry of the commercial interior, through local space perception, helps to make clear the pedestrian’s location and to establish a psycho-spatial map that matches the actual situation; hence, it is easier for pedestrians to obtain spatial information.
• In areas that are more difficult for people to reach, the installation of vertical transport facilities, such as escalators, lifts, and stairs, is conducive to enhancing their spatial accessibility, supplementing the flow of people, and balancing the distribution of commercial value.

The results of this study provide a novel approach to the evaluation of the commercial vitality and pedestrian flow statistics of a CWC. It has a certain guiding value in the planning of traffic movement and space layout inside the building. On the one hand, when carrying out the traffic space design of a CWC, the simplicity and clarity of the traffic space relationship should be emphasised; symmetry and unity of the form are more conducive to a balanced distribution of commercial vitality. The design should give priority to the commercial space layout with a high degree of global integration, while auxiliary service functions can be located in an area with high structural attributes and low commercial vitality. In addition, for CWCs that need to be upgraded in terms of traffic, this research method can be used to compare and analyse the characteristics of the pedestrian distribution before and after the adjustment of the traffic scheme so that a comparison of multiple schemes can be made to assist in the screening of schemes.