Statistical Modeling of Traffic Flow in Commercial Clusters Based on a Street Network

Abstract

Traffic flow characterizes vitality in commercial clusters, and the accurate prediction of traffic flow based on the street network has significant implications for street planning and vitality regulation in commercial clusters. However, existing studies are limited by certain problems, such as difficulty in obtaining traffic flow data and carrying out technical methods. The purpose of this study is to use urban physical data to study traffic flow so as to quickly and effectively estimate the traffic flow in commercial clusters. This study takes the street networks of 100 commercial clusters in China as the research objects and classifies them into three forms according to the theory of “A city is not a tree”. Taking typical commercial clusters in these three forms as the research unit, space syntax was used to study five metrics of street network connectivity, and integration (Dn) was selected as a proxy variable for street network connectivity. The results show that the traffic flow in the three forms of commercial clusters can be predicted using the multiple regression models established based on the three metrics of integration, the traffic level, and the operation cycle. This study establishes the connection between the street network form and the traffic flow, which enables the possibility of obtaining the traffic flow of commercial clusters quickly and effectively. For areas with poorly structured urban data, the results can help urban planning administrators to predict the potential economic attributes using easily accessible street network data in commercial clusters.

1. Introduction

A commercial cluster is an urban functional area that integrates commodity purchase, marketing display, entertainment experience, commercial offices, and regional logistics, rendering it the fifth generation of merchandising markets in China. Studies have shown that improving the number of visits to commercial stores, increasing the assortment of businesses, and balancing the homogeneity of commercial vitality can contribute to the effective operation of a commercial cluster [1,2,3]. The commercial vitality of a mature commercial cluster needs to be well-balanced, and store visits need to be evenly distributed among the commercial clusters [4,5]. Visits to stores are mainly dependent on traffic flow, and the higher the traffic flow is, the higher the number of shoppers and the more expeditious the turnover of merchandise on-site will be [6,7]. Therefore, the study of traffic flow in commercial clusters can help to achieve the stable and rapid development of commercial clusters.

Existing research on traffic flow mainly focuses on three aspects: the detection of traffic flow, analysis of traffic flow characteristics, and research on urban problems based on traffic flow. (1) Two main methods are applied to capture traffic flow patterns in order to establish predictive analysis models. The first method, based on ultrasonic and infrared sensing, GPS, etc., records the number of vehicles passing through a certain segment in unit time as the traffic flow of that segment by transmitting the driving signal to the terminal through the vehicle [8,9,10]. The other is the prediction of traffic flow statistics based on images. Since 2012, CNNs (convolutional neural networks) have been successful in the field of target detection, enabling traffic flow detection to enter the era of convolutional neural network methods. In recent years, the representative algorithms have become YOLO, SSD, the R-CNN target detection method and SORT, and DeepSort multi-target tracking algorithms [11,12]. (2) A large number of studies have been conducted on traffic flow based on four dimensions: stochasticity, periodicity, spatiotemporal characteristics, and environmental characteristics. The aim of this research was to explain the spatiotemporal variations in, and pattern of, traffic flow, to control the error rate, and to improve the accuracy of prediction results through empirical studies [13,14,15]. (3) Using the detection results of the traffic flow, including information on the traffic flow, vehicle travel speed, and urban spatial distribution, researchers aimed to study the problem of urban carbon emissions, urban traffic, and urban infrastructure services and facilities [16,17,18]. The abovementioned studies on traffic flow are limited in the following ways: (1) They require the completeness of the original data, making it difficult to obtain traffic information for a wide range of areas in locations with low data openness, resulting in prediction errors. (2) Although the study of the characteristics of traffic flow itself improves the accuracy of traffic flow prediction, the number of variables to be considered renders the calculation complicated and unsuitable for projects requiring fast prediction with low accuracy. (3) Due to the poor openness of urban big data and other factors, most studies only take into account the actual traffic flow factors to study urban problems, ignoring the study of the urban physical environment and socio-economic and time-space factors as variables.

The street network is the skeleton of the city and is one source of easily accessible data on the physical form of the city. Studies have been conducted to confirm the relevance of street networks to the development of cities and that street networks directly affect the behavior of people and vehicles, including the three following studies [19,20]. Firstly, street networks were analyzed using social network analysis and betweenness centrality [21,22]. The degree, closeness, and betweenness centrality were found to have a strong correlation with the population density and street density, and this was used to predict the co-ridership rate of online vehicles and evaluate the resilience of street networks [23,24]. Secondly, the social network analysis method was introduced to the investigation of buildings and physical spaces using space syntax, targeting building interior spaces in the early stage. It improved significantly after 2006, with the focus shifting to street networks. For example, some scholars used spatial syntax to predict the spread of the COVID virus phenomenon in Hong Kong and the touring paths of tourists in Chinese historical districts [25,26,27,28]. Thirdly, spatial design network analysis (sDNA) was used to adapt the calculation method and data structure of SNA and (space syntax) to develop two indicators of closeness and betweenness [29]. Notably, sDNA can analyze the accessibility of street networks and calculate the potential of streets for pedestrian, bicycle, motor vehicle, and public transport use. This approach has been applied to different spatial networks, such as transit rail, urban street, and 3D pedestrian spaces [30,31]. This suggests that the use of SNA street networks is suitable for the study of socioeconomic attributes and transportation on the urban scale, and space syntax is suitable for the study of behavior triggered by human perceptions of the city on the urban mesoscopic scale, while sDNA has better potential for the study of urban micro-scale, three-dimensional spaces.

Although a series of studies on street network connectivity and spatial perception have been conducted by many scholars using various methods, few studies have been conducted on the traffic flow and economic attributes of cities using their connectivity characteristic metrics. Thus, the purpose of this study is to explore the traffic flow of commercial clusters using easily accessible street network data in an attempt to provide urban planners with a simple and effective method that can be used to predict the traffic flow in the region at the early stage of planning and during the operation period and to adjust the street network according to the results in order to achieve the stable and rapid development of commercial clusters. Therefore, the main contents of this paper include the following: (1) taking commercial clusters as the research object, we studied street network connectivity characteristics based on the space syntax method to simplify its connectivity characteristic indicators and (2) using the simplified indicators and street network attributes, we constructed a multiple regression model to predict the traffic flow.

2. Research Methods

In this study, 100 business clusters in China were analyzed. Firstly, they were typed, and three kinds of representatives were identified as the study units. Secondly, the traffic flow in the region was measured, and data on the street network in the region were obtained. Furthermore, the street network connectivity was analyzed using the space syntax method, and the connectivity index was simplified. Finally, a prediction model was developed, and a multiple regression model (Figure 1) was constructed with the traffic flow as the dependent variable and the output of the indicator from the space syntax and the street network properties as the independent variables.

2.1. Research Unit

In his article “A City Is Not a Tree”, Professor Christopher Alexander applied mathematical aggregation models to analyze the structural characteristics of “natural cities” and “man-made cities” and then summarized two important types of cities, the “semi-grid” and “tree”, and highlighted that a city with vitality should take the shape of a semi-grid. Based on the abovementioned classical theories, the top 100 commercial clusters were studied in terms of their annual turnover using the China Commodity Market Statistical Yearbook 2021. Based on Christopher Alexander’s A New Theory of Urban Design, the street network development time sequence and topological graphic characteristics of the 100 commercial clusters were typified (see Appendix A,

Table A1. List of 100 commercial clusters in China.

	Commercial Cluster	Forms
1	Shanghai Jinsu Shichang	Semilattice
2	Zhangjiagang Jonglong yuan	Lattice
3	Changshu Fuzhuang Cheng	Semilattice
4	Suzhou Sichou Cheng	Lattice
5	Binzhou Heibai Tei cheng	Semilattice
6	Yiwu Guoji Shangmao	Tree
7	Suzhou Huagong Jiaoyi Zhongxin	Lattice
8	Dalian Shiyou Jiaoyi Suo	Tree
9	Shaoxin Qingfang Cheng	Tree
10	Baoding Baigou Cheng	Lattice
11	Tianjin Xiyou Jinshu Shichang	Tree
12	Nantong Zhihao Shichang	Lattice
13	Chengduo Guoji Shangmao	Tree
14	Jinhua Keji Wujin Jituan	Lattice
15	Taian Gangcai Shichang	Semilattice
16	Ningbo Suliao Cheng	Semilattice
17	Beijing Xingfa Zhongxin	Semilattice
18	Shanghai Shihua Jiaoyi Zhongxin	Semilattice
19	Wuxi Dongfang Gangcai Cheng	Lattice
20	Shijiazhuang Xinhua Shangmao	Semilattice
21	Nantong Guoji Jiafang Cheng	Tree
22	Shanxin Qingfang Cheng	Lattice
23	Wuxi Gangkou Wuliu Yuan	Semilattice
24	Anshan Xiliu Fuzhuang Cheng	Lattice
25	Zhengzhou Guoji Nongye Zhongxin	Tree
26	Beijing Ershou Qiche Shichang	Lattice
27	Wuxi Jinsu Cailiao Shichang	Semilattice
28	Tianjin Zhongchu Fazhan Shichang	Tree
29	Shenyan Wuai Shangpin Shichang	Semilattice
30	Wuxi Nanfang Buxiugang Shichang	Lattice
31	Shijiazhuang Santiao Shichang	Semilattice
32	Hefei Huishang Gangcai Shichang	Lattice
33	Hangzhou Jinshu Cailiao Shichang	Semilattice
34	Yuncheng Yudu Shichang	Semilattice
35	Shanghai Shiyou Jiaoyi Gongsi	Semilattice
36	Linyi Wangbao Gangcai Shichang	Semilattice
37	Huanan Kuangwu Wuliu Shichang	Lattice
38	Wuxi Huadong Shihua Shichang	Tree
39	Zhenjiang Guoji Gangtie Zhongxin	Lattice
40	Suzhou Fangzhi Yuanliao Shichang	Semilattice
41	Chongqin Chaotianmen Shichang	Tree
42	Dalian Shangpin Cheng	Semilattice
43	Tianjin Konggang Qiche Shichang	Lattice
44	Shanxin Qingfang Gongmao Yuan	Lattice
45	Nanchang Hongcheng Shichang	Lattice
46	Changsha Gaoqiao Shichang	Lattice
47	Nanjing Zhongcai Nongye Shichang	Lattice
48	Anqing Guangcai Shichang	Lattice
49	Langfang Xianghe Jiaju Cheng	Tree
50	Haozhou Zhongyao Jiaoyi Zhongxin	Lattice
51	Changsha Hongxin Nongye Shichang	Lattice
52	Beijing Nongfu Changpin Shichang	Lattice
53	Xuchang Xinqu Gangcai Shichang	Semilattice
54	Chanzhou Lingjiatang Shichang	Lattice
55	Guangzhou Zhongda Shichang	Semilattice
56	Hangzhou Xiaoshan Shangye Cheng	Semilattice
57	Wuxi Zhaoshang Chengshi Shichang	Lattice
58	Changsha Gangtie Shichang	Lattice
59	Fuzhou Nanfang Gangcai Zhongxin	Tree
60	Changzhou Fangzhi Touzhi Zhongxin	Lattice
61	Guangzhou Guocai Pifa Shichang	Lattice
62	Chongqin Guangyinqiao Nongmao	Semilattice
63	Hangzhou Chengbei Jinshu Shichang	Semilattice
64	Xingjiang Hengyuan Wuliu Yuan	Tree
65	Foshan Xiqiao Qingfang Cheng	Semilattice
66	Changsha Sanxiang Nanhu Shichang	Semilattice
67	Beijing Jingxiu Nongye Shichang	Semilattice
68	Liuzhou Shengchang Zhiliao Shichang	Lattice
69	Guangzhou Yuzhu Jiancai Cheng	Lattice
70	Qingdao Shucai Pifa Shichang	Semilattice
71	Shangqiu Nongchangpin Zhongxin	Tree
72	Suzhou Nanhuan Shichang	Lattice
73	Wuxi Guolian Jinsu Cailiao Shichang	Tree
74	Tongxiang Yangmao Shichang	Semilattice
75	Liaocheng Dadong Gangguang Shichang	Lattice
76	Shenzhen Haiji Nongmao Shichang	Lattice
77	Ningbo Meitang Jiaoyi Shichang	Semilattice
78	Beijing Chengbei Shangpin Shichang	Lattice
79	LuoYang Guanlin Shichang	Tree
80	Xuzhou Bali Gangtie Shichang	Semilattice
81	Chongqin Julong Gangcai Shichang	Lattice
82	Wuhan Baisha Nongmao Shichang	Lattice
83	Kunming Luoshi Guoji Cheng	Tree
84	Shanghai Changqiao Gangcai Shichang	Semilattice
85	Shenzhen Huanan Cheng	Lattice
86	Shanghai Gaoqiao Zhongbiao Zhongxin	Tree
87	Ningbo Haiye Huagong Shichang	Semilattice
88	Chongqin Lengchu Wuliu Zhongxin	Lattice
89	Hefei Zhougu Du Nongmao Zhongxin	Lattice
90	Tianjin Wangding Pifa Shichang	Lattice
91	Jimo Fuzhuang Pifa Shichang	Tree
92	Fuzhou Haixia Changpin Shichang	Lattice
93	Xinjiang Jiuding Shenghe Shichang	Semilattice
94	Guanghou Shengdi Pige cheng	Lattice
95	Yantai Sanzhan Shichang	Semilattice
96	Xuzhou Xuanwu Jituan Shichang	Semilattice
97	Changsha Zhongnan Qiche Zhongxin	Lattice
98	Changzhou Suliao Huagong Shichang	Tree
99	Huzhou Jili Tongzhuang Shichang	Lattice
100	Hangzhou Fangzhi Caigou Cheng	Lattice

) and classified into semi-grid structures, grid structures, and tree-shaped structures [32].

The semi-grid structure is a commercial cluster based on the gradual expansion of the completed urban area, extending the original urban street network structure. The grid structure is a commercial cluster integrated into the urban master plan and developed according to the attributes of the urban planning site, relying on the existing urban planning street. The tree structure is similar to the grid cluster, which belongs to the same form of integrated planning, but its street system is self-contained and merges into the urban traffic through specific entrances and exits.

Samples with similar construction scales, relatively good operation and complete street network data were selected from the three forms of commercial clusters for the study. Considering the impact of the global system on the local system, local street network analysis generally focuses on the global street network system. Thus, with reference to the street network traveling time and coverage area in the “Code for Transport Planning on Urban street”, finally, it was determined that the study sample would be the center, and the model boundary and buffer area of the urban traffic axis within a 5 km radius were established as the street network analysis range (

Table 1. Topology forms and study sample.

Forms	Topology Graphics	Sample	Street Network(5 km Buffer)
Semilattice		UF1, Guangzhou Zhongda
Lattice		UF2, Yiwu Guoji Shangmao
Tree		UF3, Shenzhen Huanan Cheng

) [33].

2.2. Data

Street Network data were obtained from the Open Street map on 1 January 2021, the street medians were extracted using ArcGIS, and the extracted street medians were analyzed using space syntax [34].

As for the actual measurement survey of the traffic flow, there were a total of 54 investigators who were divided into 3 groups and assigned to 18 street sections of UF1, 15 street sections of UF2, and 21 street sections of UF3 (the location of the surveyed street sections were as follows: the outer-circulation urban main and secondary arterial streets, and the inner-circulation cluster arterial and major feeder streets). From 18 January 2021 to 24 January 2021 (Monday to Sunday), we recorded video of the traffic flow on the corresponding roadway during business hours (9 am–2 am, 2 pm– pm) daily (8 times daily, including 4 times in the morning and 4 times in the afternoon, amounting to 8 business hours, with a single time duration of 15 min and a total of 120 min per day) (see Appendix B, Figure A1). The video was uploaded to OpenCV (a loading package), and the actual measured traffic flow of the three samples amounted to a total of 1,180,997. The traffic volume of the sub-sections was uploaded to Arc GIS to build a database of traffic information, and the distribution of the traffic volume in the region was simulated by spatial interpolation using the natural neighborhood method [35] (Figure 2).

3. Results

3.1. Descriptive Statistics Analysis

The spatial syntax covers the following four indicators of street network connectivity: depth, which is generally the topological grouping relationship of any one space with other spaces and is evaluated by the average depth value Mn; integration, where the global integration degree Dn is used to measure the integration degree of the study area, which reflects the closeness of the connection with other spaces centered on a certain space and its attractiveness with respect to traffic; accessibility, expressed by Rn, which reflects the ease of access of space in relation to other spaces, and connectivity, expressed by Cn, which is the number of other street sections linked to a certain street section, according to which the greater the number of linked sections is, the better the connectivity will be.

Table 2. Central tendency and statistical dispersion of the metrics.

Forms (Street Segments)	Metrics	Min	Max	Average	Std Dev	Median
UF1 (n = 145)	Mn	2.471	6.155	4.282	0.734	4.313
	Dn	0.901	3.162	1.491	0.363	1.427
	Rn	0.142	7.603	1.007	0.831	0.862
	Cn	1.000	27.00	3.642	2.643	3.000
UF2 (n = 123)	Mn	2.601	6.932	4.231	0.713	4.414
	Dn	0.757	2.752	1.441	0.335	1.412
	Rn	0.144	4.381	1.002	0.663	0.798
	Cn	1.000	18.000	3.891	2.312	3.000
UF3 (n = 84)	Mn	2.021	4.872	3.014	0.492	3.015
	Dn	1.012	3.813	2.054	0.525	1.962
	Rn	0.112	7.903	1.008	1.582	0.721
	Cn	1.000	31.000	5.471	5.752	4.000

depicts the descriptive statistics of the abovementioned metrics, and it can be seen that there are differences in each characteristic between the different forms of clusters. These differences are as follows below.

The UF1 semi-grid cluster has average location conditions, being adjacent to the urban core (Dn_Average = 1.491) and closely connected to the traffic space (Mn_Average = 4.282), and has average connectivity (Cn_Average = 3.642), an excessive concentration of traffic diversion pressure, poor external and internal system circulation, and mainly pedestrian traffic supplemented with vehicular traffic. However, it has the best accessibility (Rn_Median = 0.862), and the commercial vitality radiates from the cluster center, which will gradually decrease its commercial value. The UF2 grid-shaped cluster has average location conditions, being adjacent to the urban core area (Dn_Average = 1.441) and containing the area with the closest traffic spatial linkage (Mn_Max = 6.93). The connectivity is average, (Cn_Average = 3.891), and the traffic diversion capacity is relatively balanced. The system works through both external circulation and internal circulation and is dominated by vehicular traffic and supplemented with pedestrian traffic. The accessibility is average (Rn_Median = 0.798), with a poor radiation ability in terms of its commercial vitality and a balanced distribution of commercial values. The UF3 tree cluster has homogeneous traffic spatial links and good locational conditions (Dn_Average = 2.054) and is located in the urban nucleus, but it has weak spatial links (Mn_Std Dev = 0.49, Mn_Average = 3.01), and the best connectivity (Cn_Average = 5.471), together with the best traffic dispersal capacity. The system is dominated by internal circulation and supplemented with external circulation, where vehicular traffic and pedestrian traffic operate together, but the accessibility is mediocre (Rn_Median = 0.721), the distribution of commercial vitality is uneven, and the conformity ability is poor.

Meanwhile, comprehensibility reflects users’ degree of difficulty in establishing the overall spatial structure through local spatial experience, i.e., the ability to perceive the place in a minor way. Comprehensibility influences the commercial traffic and pedestrian flow, enabling one to establish a cognitive map, quickly locate the location, and create efficient commercial radiation. The space syntax theory specifies that the comprehensibility index can be measured by the correlation between global and local variables. The Rn-Cn scatter plots (Figure 3) of the three samples were plotted separately to establish the fitting equation and obtain the regression coefficient R², which reflects the degree of coupling between the global and local variables in each space of the whole system, i.e., the spatial comprehensibility. The theoretical definition of the comprehensibility range is as follows: the R² < 0.5 local space fails to reflect the global space; the 0.5 < R² < 0.7 local space reasonably reflects the global space, and the 0.7 < R² < 1.0 local space relatively effectively reflects the global space. The UF1 semi-grid cluster and UF2 grid cluster space are simple to understand (UF1: R² = 0.518, UF2: R² = 0.582). Users can establish a global spatial cognitive map by traveling through the local space and clearly need to access the commercial area, where it is not easy to become lost during travel, which is conducive to the radiation of commercial vitality. The UF3 tree cluster space is more difficult to understand (R² = 0.362), and it is difficult for travelers to comprehend the space in its entirety. It is difficult for travelers to perceive the space they inhabit, which causes confusion in terms of spatial perception. Therefore, it is easy to feel lost during travel, which is not conducive to the spread of commercial, the distribution of commercial vitality is uneven, and the conformity ability is poor.

3.2. Construct Mode

3.2.1. Correlation Analyses

Table 3. Correlation analysis of each metric.

Forms	Metrics	Dn	Rn	Mn	Cn	Traffic Flow
UF1	Dn	1	−0.884 **	0.973 ***	0.994 **	0.407
	Rn	−0.884 **	1	−0.949 **	−0.907 **	−0.109
	Mn	0.973 ***	−0.949 **	1	0.987	0.079
	Cn	0.994 **	−0.907 **	0.987	1	0.062
	Traffic Flow	0.407	−0.109	0.079	0.062	1
UF2	Dn	1	0.898 ***	−0.979 **	0.721 *	0.510
	Rn	0.898 ***	1	−0.406	0.900 **	0.344
	Mn	−0.979 **	−0.406	1	−0.600	−0.504
	Cn	0.721 *	0.900 **	−0.600	1	0.279
	Traffic Flow	0.510	0.344	−0.504	0.279	1
UF3	Dn	1	0.783 *	−0.916 ***	0.358	0.613 *
	Rn	0.783 *	1	0.369	0.910 **	−0.577
	Mn	−0.916 ***	0.369	1	−0.451	0.112
	Cn	0.358	0.910 **	−0.451	1	−0.652 *
	Traffic Flow	0.613 *	−0.577	0.112	−0.652 *	1

* p < 0.05, ** p < 0.01, *** p < 0.001

shows the correlations between the variables, and it can be seen that Dn, Rn, Mn, and Cn show strong correlations among the three forms. Thus, it is necessary to downscale these variables in the prediction model. Dn shows a significant negative correlation with Rn in UF1 and significant positive correlations with Mn and Cn. Dn shows significant positive correlations with Rn and Cn in UF2 and a significant negative correlation with Mn. Dn shows significant positive correlations with Rn and Cn in UF3 and a significant positive correlation with Cn in UF3. In UF2, Dn shows significant positive correlations with Rn and Cn and a significant negative correlation with Mn. In UF3, Dn shows a significant positive correlation with Rn, a weak positive correlation with Cn, and a significant negative correlation with Mn. In addition, compared to Rn, Mn, and Cn, Dn correlates more strongly with the traffic flow in all three forms. Therefore, Dn can be selected as a proxy variable to measure the street network connectivity, and because Dn is moderately correlated with the vehicle flowrate and is not considered significant, two variables, the traffic class and operational cycle, are introduced to improve the accuracy of the prediction in order to improve the reliability of the prediction.

3.2.2. Multiple Regression

According to the results of the correlation analysis, the street attribute metrics can be introduced as variables to build the regression model. According to the “Code for Transport Planning on Urban Roads” [33], there are four levels of UF1, with values ranging from 0 to 3, and three levels of UF2 and UF3, with values ranging from 0 to 2. The different retail outlets in the professional clusters have different operation cycles. Thus, their ability to attract consumers also varies greatly. The operation cycles of professional retail centers can be divided into four stages: the formation period, cultivation period, growth period, and maturity period. The operational cycle of the professional clusters is introduced to correct the integration values of the closely linked axes using the field research samples UF1, UF2, and UF3 to measure the commercial operations within the clusters and assign them values ranging from 0 to 3.

Based on the correlation analysis and the results of the proxy variable study, the three independent variables, the integration degree (X1), traffic level (X2), and operation period (X3) were selected to establish a multiple linear regression model based on the dependent variable, the traffic flow (Y). The regression models of the three forms show that the significant coefficient p-values are less than 0.05, and all of them have good statistical significance, among which the fit of samples UF1 and UF3 is higher, with R² = 0.880 and R² = 0.834, while the fit of sample UF2 is within a reasonable range, with R2 = 0.748. For the different types of business clusters, there are certain differences in the influence weight of the three independent variables. The influence weight of integration (4.273) is more significant in the semi-networked clusters, the influence weights of the traffic level (3.778) and operation period (3.619) are comparable and both are better than the integration degree in the grid clusters, and the influence weight of the operation period (3.698) is more significant in the tree clusters (

Table 4. Multiple regression results.

	Variable	Coefficient	Weight	p	Intercept	R2
UF1	X1	17.617	4.273	0.001	−14.428	0.880
	X2	2.236	2.703	0.017
	X3	3.051	3.517	0.003
UF2	X1	20.996	2.412	0.033	60.214	0.748
	X2	7.582	3.778	0.003
	X3	6.755	3.619	0.004
UF3	X1	36.155	2.316	0.033	−35.590	0.834
	X2	4.372	2.385	0.280
	X3	5.085	3.698	0.002

).

The results of the multiple regression equation indicate that the space syntax global integration degree variable can be used as a proxy variable for the structure of the traffic system in the spatial study of the professional clusters and can be used for the effective prediction of the traffic volume on the regional scale by taking into account factors such as the road system level and commercial operating conditions.

4. Discussion

This paper reports on statistical modeling research on clustered commercial traffic flow based on urban road networks. In the study, we established a statistical traffic flow prediction model by introducing the global integration degree, traffic level, operation period, etc. This technical method has the characteristics of high efficiency, ideal accuracy, and easy implementation. Unlike road monitoring detection, CNN, and other big-data-driven traffic forecasting, the main differences of this research method are as follows: (1) In terms of the forecasting time, this statistical model is suitable for the forecasting of long-term traffic distribution trends, unlike short-term traffic forecasting, which relies on real-time traffic data acquisition in urban areas. (2) In terms of the forecasting application, it is applicable to pre-traffic planning and has positive effects. (3) In terms of the implementation difficulty, this research technique is suitable for non-traffic information disciplines and reduces the data barriers caused by the difficulty in obtaining diversified data.

This study has certain shortcomings. Firstly, the space syntax method is based on topological relationships that are used to calculate the variables related to road network connectivity, which does not fully reflect the actual traffic situation, while the real traffic aggregation problem is affected by the natural environment and weather elements, representing a complex problem. Therefore, the accuracy of this research method for short-term traffic flow prediction needs to be improved. Secondly, it is not only the global integration degree, traffic level, and operation cycle that determine the traffic flow of commercial clusters. Rather, the organic integration of various other elements and functions should be considered when planning commercial clusters. Furthermore, the data statistics reported in this paper are limited by the research scope, and it is not possible to obtain comprehensive statistics on the best domestic operation of 100 samples to verify the statistical model and further improve the prediction accuracy of the model.

This study focused only on traffic forecasting for commercial clusters, and the research’s accuracy could be improved by supplementing and verifying the research cases. To enhance the applicability of the study, future research could be extended to other functional clusters in the city so as to introduce more widely applicable parameters, allowing one to further control the research errors.

5. Conclusions

This paper established a typology with commercial clusters as the research object, explored the radiating capacity of the commercial vitality of each type based on easily accessible street network data, and used multiple regression models built with three indicators: the global integration degree, traffic level, and operation period. These provided relatively good prediction results for the traffic flow of three different forms of commercial clusters (UF1: R² = 0.880; UF2: R² = 0.748; UF3: R² = 0.834). The main conclusions of the study are as follows:

(1)

The traffic flow of commercial clusters can be predicted using the following method: the study objectives are classified and matched according to the classification results in order to construct multiple regression models for the prediction using three easily accessible metrics: integration (Dn), the traffic class, and the operation cycle (Table 4).

(2)

Compared with the average depth, accessibility, and connectivity, global integration is more strongly correlated with traffic flow, while global integration shows a positive correlation with traffic flow.

(3)

Based on the comparison of street network connectivity indexes, the urban location conditions of the three types of commercial clusters are comparable, the semi-grid and grid types have better traffic accessibility, and the radiation energy of commercial vitality is better than that of the tree clusters. Moreover, it is easier to establish the spatial orientation of the customer flow.

The results of this paper offer the possibility of evaluating commercial cluster planning in countries or regions with unsound socio-economic data and have a certain guidance value for the planning of urban street networks and the layout of commercial areas. On the one hand, when carrying out commercial land use planning, consideration needs to be given to the selection of areas with high global integration so as to prioritize the commercial layout. Additionally, for areas with high structural attributes and low commercial vitality, the road networks or layouts of ancillary functions should be planned and adjusted in accordance with the actual situation. On the other hand, the spatial differences in traffic flow between different forms should be combined to improve the planning of the street network layout.