Urban Complexity and the Dynamic Evolution of Urban Land Functions in Yiwu City: A Micro-Analysis with Multi-Source Big Data

Abstract

The diversification of business forms leads to functional and spatial complexity in cities. The efficient determination of the complexity of an urban system is the basis for the scientific monitoring of the multi-functional aggregation within cities. Previous studies on the urban spatial structure were limited by the difficulty of collecting micro-data and the high time cost, and they focused on the macro-spatial structure, lacking fine-grained investigations of the micro-spatial structure. Additionally, high-resolution remote sensing images, which mainly rely on the textural characteristics of the spectrum of ground objects, cannot detect the social and economic functions of ground objects. Thus, it is difficult to meet the actual needs of urban planning and management. The purpose of this paper is to automatically identify the spatial heterogeneity and temporal variation of urban land use functions in the context of complex urban systems. The TF-IDF (term frequency–inverse document frequency) algorithm, a machine learning classification algorithm, and other methods are applied to identify the urban functions and distribution characteristics of the main urban area based on the POI (point of interest) data and urban form data. The results show the following: (1) From 2012 to 2022, all types of land use in Yiwu city grew at different rates, with logistics and warehousing space growing the fastest, which is in line with Yiwu’s goal of building a national logistics center for trade and services. (2) The residential area has a spatial structure with a dense central circle and a scattered periphery extending from northeast to southwest and from east to west. (3) The commercial service sector shows clear spatial differentiation between the core and the periphery. The commercial functional areas of Niansanli, Houzhai, and Chengxi, where the number of commercial POIs is relatively small, are located at the intersection of the administrative subdistricts near the city center, indicating that the commercial economic activities of the downtown subdistrict have a certain spillover effect on adjacent subdistricts. (4) The public facilities of each subdistrict are generally located in the core of each subdistrict, which ensures better convenience and accessibility. (5) Industrial land with a large total area that is scattered and mixed with urban residential land gradually tends to be centralized, forming an industrial belt around the city. This study comprehensively considers the aggregation relationship between urban buildings and land use and improves the accuracy of land identification and functional zoning.

1. Introduction

Over the past 10 years, although China’s economic growth has slowed down, urbanization still shows a rapid growth trend. China’s urbanization rate rose from 51.3% in 2011 to 65.2% in 2022, an increase of 13.9%. With China’s rapid urbanization, the urban development model with “urban construction land expansion or urban spatial expansion” as the main feature is difficult to maintain, and “urban renewal or urban smart growth” has become the inevitable trend of China’s future urban development. As the basic unit of urban development, an urban functional area is an important component of urban spatial planning. Therefore, it is very important to clarify the status quo of the layout of urban functional areas to optimize the structure of urban land use.

City function zoning is to arrange various material elements (such as factories, warehouses, houses, etc.) in the city according to their functional requirements to form an organic whole with mutual connections and a reasonable layout and to create a good environment and conditions for various activities in the city. It is an important method of determining the form of land use and space layout according to the principles of functional zoning. In previous studies, the identification of urban functional areas was mostly based on horizontal scale and single land use [1,2,3,4]. However, in the process of rapid urbanization in developing countries, the trend of vertical, multi-functional, and mixed urban development has become increasingly prominent. Mixed land use (MLU) is still one of the most recommended indicators for successful urban planning and urban regeneration. MLU may play a great role in solving urban problems such as pollution and the overconsumption of land. It can usually save commuting costs and enhance urban vitality [5], provide more local jobs and business options to enhance the economy of the community, and add vibrancy to the environment [6]. However, on the one hand, MLU may have a negative impact on land use compatibility (LUC) [5], because it is difficult for urban planners to find the balance between MLU and LUC. On the other hand, MLU also increases the difficulty of urban functional area identification. The diversification of business forms and the intensification of land use have led to the emergence of highly aggregated and increasingly complex types of land use in urban areas. The question of how the complexity of the urban system can be captured efficiently is therefore the basis for the scientific monitoring of multi-functional urban aggregation.

In OECD countries, population density is used to identify urban cores, and travel-to-work flows are used to identify the hinterlands where the labor market is highly integrated with the cores [7]. By introducing the concept of latent activity trajectory (LAT), which captures socioeconomic activities conducted by citizens at different locations in a chronological order, Yuan et al. [8] proposed a data-driven framework to discover urban functional zones in a city, developed a topic modeling-based approach to cluster the segmented regions into functional zones, and identified the intensity of each functional zone using a Kernel density estimation. Based on building-level social media data, Chen et al. [9] used a dynamic time warping (DTW) distance-based k-medoids method to group buildings with similar social media activities into urban functional areas. Zhang et al. [10] used hierarchical semantic cognition (HSC) to classify urban functional zones, which relies on geographic cognition and considers four semantic layers, i.e., visual features, object categories, spatial object patterns, and zone functions, as well as their hierarchical relations. Their experimental results indicate that this method can produce more accurate results than Support Vector Machines (SVMs) and Latent Dirichlet Allocations (LDAs). In addition, some scholars evaluated the polycentric index of a city and its degree according to the urban functional area [11]; others studied the social cohesion of different functional urban areas [12].

Delineating urban functional zones is critical to understanding urban dynamics, evaluating planning strategies, and formulating supportive policies. However, initial studies using point of interest (POI) data did not adequately consider the relationship between POI categories in a spatial context, nor could they provide a direct way to classify urban functional areas. If the POI data, TF-IDF algorithm, and machine learning classification algorithm are integrated, the following work objectives can be achieved: (1) The spatial relationship between the POI data and urban form data can be established by using the geographic information system (GIS) spatial connection method, and the POI data can be given spatial entity attributes. Therefore, in recent years, the POI data with accurate geographical locations and detailed attributes have become the main data source for exploring urban functional zoning from a bottom–up perspective [13,14,15,16]. (2) The TF-IDF (term frequency–inverse document frequency) algorithm can be used to identify and extract important words in text data and to classify urban functions. Combining POI data with the TF-IDF algorithm to analyze the functional distribution and activity characteristics of different regions can improve the accuracy of urban function classification results [17,18,19,20]. (3) Machine learning classification algorithms can automatically learn rules based on the training data provided to identify and classify city functions [21,22].

The automatic identification of urban land via a variety of big data, such as mobile phone signaling data, social media data, bus swipe card data, remote sensing images, and business POI data, is commonly carried out for quantitative processing. This method of comprehensively using multi-source data can improve the accuracy and comprehensiveness of urban land identification. Using the Boston metropolitan area as an example, Jing et al. [23] estimated and classified the land use by developing and using VGI (volunteered geographic information) online POI data to identify the types, development, and use of urban functional areas. Chi et al. [24] reclassified the POI data and quantitatively identified urban single-functional areas and mixed-functional areas. For underground systems, Cao et al. [25] reconstructed the temporal distribution of passenger flow at underground stations based on spatio-temporal data from swiping the underground map. On this basis, a cluster analysis was conducted to identify the working and living spaces around metro stations, and spatial visualization analysis was conducted using a GIS.

The purpose of land use identification is to accurately determine future urban development plans. The accurate and rapid identification of urban functional areas holds great significance for urban planning and resource allocation. Some scholars use social network check-in data as an information source to characterize the dynamic characteristics of urban land use and propose a grid-based aggregation method to divide urban land functional areas [26].

The heterogeneous characteristics inherent in urban environments pose difficulties in discerning and demarcating the boundaries of urban functional zones [27]. Chin et al. [28] used open big data to identify the scope of urban activity spaces and determine the land use nature of each urban plot scale within the identified urban activity spaces. Krause et al. [29] combined GPS data, land use data, and POI data on user trips to predict short-term travel behavior, aiming to help researchers understand the travel characteristics of residents in cities and to analyze the cities’ regional economic development levels. Based on an analysis of the classification and aggregation of urban industrial land data, Tian et al. [30] and Liu et al. [31] proposed a new method, the fruit fly optimization algorithm (FOA), to dynamically obtain the SVM parameters. Some scholars constructed a classification system of urban land use functions, and they used the frequency density, the proportion of types, and a kernel density estimation to identify the urban land use functions in five districts in Jinan city [32,33,34]. Other scholars proposed a framework combining heterogeneous data from multiple sources to identify functional area types and to draw functional area portraits [35,36,37]. Li et al. [38] and Guo et al. [39] conducted a spatial division of regional land use risk based on production–life–ecology analysis, and they found that the overall layout of functional zones in Zhengzhou exhibited a spatial pattern of single and multiple coordinated development. Based on POI big data on architectural forms and business forms, Yang et al. [40] adopted the TF-IDF transformation and database construction to identify urban land of different types, sizes, and locations by using supervised deep learning methods.

Overall, although previous studies have made significant progress, they still have the following shortcomings: (1) Due to the difficulty in collecting micro-data and the high time cost, scholars tend to study the urban spatial structure at the macro-scale, and detailed research on the micro-spatial structure is lacking. (2) Although high-resolution remote sensing images can be used for the rapid classification of urban land use, this method mainly relies on the spectral textural characteristics of ground features and cannot detect the socio-economic functions of ground features, which makes it difficult to meet the actual needs of urban planning and management. (3) Most studies focus on the analysis of the characteristics of the urban functional structure at a single point in time, and a dynamic change analysis of these characteristics is lacking. To solve increasingly complex urban problems, timely and accurate urban land use information urgently needs to be obtained. In this context, the integration of open “big data” and urban planning data has become a popular topic in current research on urban spaces. The identification of urban land is developing in the direction of automation and manual checking to adapt to the trends of complex mixing and high-frequency changes in land use functions.

The purpose of this paper is to automatically identify the spatial heterogeneity and temporal variation of urban land use functions in the context of complex urban systems. The research features and innovations of this paper are as follows: (1) The coupling relationship between urban buildings and land use is comprehensively considered; the complexity of the urban spatial structure is identified at the micro scale; and the accuracy of plot attribute identification and functional zoning is improved. (2) By integrating POI and building form data, the TF-IDF algorithm and a machine learning classification algorithm are used to detect the social and economic functions of urban features and to reveal the diversity and complexity of urban business forms. (3) The dynamic characteristics of urban functional land use are quantitatively revealed.

2. Study Area and Data

2.1. Overview of the Study Area

Yiwu city, a county-level city, is centrally located in Zhejiang Province, China. It shares borders with Dongyang city to the east, Yongkang city and Wuyi county to the south, the Jindong District and Lanxi city to the west, and Pujiang county and Zhuji city to the north. As one of the four major regional central cities in Zhejiang, alongside Hangzhou, Ningbo, and Wenzhou, Yiwu covers an area of 1105 square kilometers. It governs eight subdistricts (i.e., Futian, Choucheng, Beiyuan, Choujiang, Jiangdong, Houzhai, Chengxi, and Niansanli) and six towns (Dachen, Shangxi, Yiting, Fotang, Chian, and Suxi) (Figure 1).

Yiwu ranks among the top 100 counties in China and is a well-known business city. It is the largest wholesale market for small commodities in China. Yiwu is also a model city of China’s reform and opening up. In 2011, Yiwu became the first and only county-level city in China to be listed as a state-level comprehensive reform pilot city, which also means that Yiwu had the first pilot right to transform its development mode of international trade, encouraging Yiwu to enter a new period of transformation and development.

Yiwu is also a global commercial city. In 2022, the city had a licensed market construction area of 6.11 million square meters, market operating households of 84,800, market employees of 230,000, operating more than 1.8 million kinds of commodities, and a total turnover of CNY 232.283 billion, an increase of 6.9%. Yiwu maintains stable trade relations with 219 countries and regions in the world, and 28,554 foreign investors entered the city in 2022. The total volume of imports and exports was CNY 478.80 billion in 2022, up 22.7% year-on-year. As part of this, the export value was CNY 431.64 billion, up 18.0% year-on-year, and imports reached CNY 47.16 billion, up 93.5% year-on-year. The turnover of cross-border e-commerce was CNY 108.35 billion. It can be seen that Yiwu’s urbanization model is the typical global urban development model of industry and trade integration, which combines market clusters and industrial development.

In the past ten years, Yiwu’s urban functions have been changing in the direction of a service circulation center. The life service industry has developed vigorously, and knowledge-intensive services such as finance and securities, trade offices, and financial and legal consulting have continued to increase. These service industries gather in the urban center and participate in the adjustment and reorganization of the urban center system, and the spatial agglomeration of the city has entered a new stage.

2.2. Data Sources and Processing

2.2.1. Basic Geographic Data

The administrative zoning map of Yiwu city used in this paper was derived from the China Geographic Information Resource database (https://www.webmap.cn/. Online access: 15 July 2023).

2.2.2. POI Data

Point of interest (POI) data are a kind of geographical big data, including the name, category, address, and latitude and longitude coordinates of geographical entities. Such data are characterized by a large sample size and rich information and can effectively reflect the functional characteristics of urban areas. In this paper, Python 3.10 software (Python Software Foundation, 2021) was used to scrape POI data on the main urban area of Yiwu city published by Amap, and less influential data were eliminated according to need. Ultimately, 20,290 and 75,712 POI data points on the main urban area of Yiwu city in 2012 and 2022, respectively, were obtained.

First, urban functional zoning was divided into five categories: residential, commercial, public services, industry, and logistics and warehousing. Then, according to the “Urban Land Classification and Planning and Construction Land Standard” and the types of POI data obtained, the POI data were reclassified. The reclassified POI data were used for the analysis of urban functional density and the identification of urban functional areas. The collection and reclassification results of the POI data are shown in Figure 2.

2.2.3. Urban Form Data

When using POI data to identify land properties, errors may occur because the correlation between urban buildings and land use is ignored. To reduce such errors, this paper introduced urban form data, which reflects the three-dimensional physical–spatial information of a city. Different types of land use usually have different building layout forms, which are reflected in various form indicators. In this research, it was necessary to collect all kinds of building form index data (such as the highest building height, average building height, average building base area, plot ratio, etc.) to establish the database. The architectural form index of each plot will change with the change of land use function. It was combined with the business weighted attribute to form a complete weighted attribute. Finally, the combination relationship of various morphological indicators in different land use types was found through machine learning to improve the accuracy of identification.

In October 2022, building form data (including attributes such as the height, geographical location, floor area, etc.) and road network data (including attributes such as the name and length of roads) were extracted from the open-source map. These two kinds of data were superimposed in space to form urban form data. To avoid data errors caused by coordinate deviations, the WGS1984 coordinate system was adopted in this paper. By introducing urban form data, the relationship between urban buildings and land use was comprehensively considered, and the accuracy of land property identification was improved.

3. Methods

3.1. Urban Functional Area Identification Process

The identification of urban land functions based only on the business form POI data ignores the correlation between urban building form and urban land use. Coupled with the single data dimension, relying on the business form data classification will still lead to considerable errors in the final identification results and cannot achieve the practical accuracy of identifying urban land use subcategories. Therefore, this study combined the business form POI data with the building form data and realized the transformation from the business form data classification to the fine identification of urban land use subcategories through artificial intelligence technology.

The process of identifying urban land functional zones is shown in Figure 3 and includes the following steps:

(1) Spatial calibration of the POI mode point layer and plot unit layer was performed. The number of all types of POI format points of each plot spatial unit was calculated; that is, the format characteristic frequency of each plot was obtained, and a plot attribute table of the associated format points was then generated.

(2) The TF-IDF algorithm was applied to the completed land plot database of associated operational points. This algorithm reweights each type of characteristic, considering the frequency and importance of its occurrence, and obtains the land feature attribute database containing weighted type characteristics.

(3) The building layer in the building spatial data is spatially connected with the plot unit layer. On the basis of this connection, a number of building form indicators for each plot were calculated, and these indicators were added to the plot feature attribute database containing weighted type characteristics to form an “urban database”.

(4) Finally, supervised classification learning was used to identify urban land use functions.

3.2. Spatial Processing of POI Point Data

In this study, the spatial relationship between the POI data and urban form data was established via the GIS spatial connection method, so that the POI data were endowed with spatial entity attributes and the POI spatial processing was completed. Due to the difference between the sources of POI data and urban form data, a slight misalignment may occur between these two kinds of data, and the expansion method was used to correct this deviation before connection. After considering the data deviation in this paper, the boundary of the morphological data was extended to 5 m, and the POI data lacking spatial entity information after spatialization were deleted. This processing method helps to solve the problem of dislocation caused by different data sources and ensures that accurate spatial information can be obtained in the subsequent analysis.

In this study, four morphological indicators, namely, the average building height, maximum building height, average building base area, and plot ratio of each plot, were calculated, and the weighted POI data were spatially correlated on the plot to construct a database containing weighted commercial and architectural form characteristics. These data were then incorporated into the training module of the machine learning algorithm, with the aim of improving the accuracy of the property recognition of the plot. By combining the type characteristics and architectural form characteristics, the characteristics of each plot can be described more comprehensively, providing more information for the machine learning model and thus improving the accuracy and precision of the plot attributes (Figure 4).

3.3. TF-IDF Algorithm

The TF-IDF (term frequency–inverse document frequency) algorithm is an algorithm that is commonly used for information retrieval and text mining. The algorithm measures the weight of a word in a collection of texts by evaluating its importance in the document. Term frequency (TF) is a measure of how often a word appears in a document. Inverse document frequency (IDF) is a measure of the importance of a word to the entire set of texts. In this study, each plot unit was regarded as a single file, and the category of POI in a single file was regarded as a word. Therefore, the analysis of the category of each spatial research unit was transformed into the calculation of the weight of each word in the file set, and, finally, the format weighting attribute was assigned to the plot.

$$TF=\frac{n_{i,j}}{\sum _K n_{i,j}}$$

(1)

$$IDF=\log \frac{\left|D\right|}{\left\{j:t_i\in D_j\right\}}$$

(2)

$$TF-IDF=TF\times IDF$$

(3)

In Formula (1), TF is the proportion of Class i POI data in each plot unit; i is the type of POI data; j is the plot number; n_i,j is the number of occurrences of type I POI data in the unit of plot j; K is the dimension of format characteristics; and $\sum _K n_{i,j}$ is the sum of all POI occurrences in plot unit j. In Formula (2), IDF is the logarithm of the percentage of the times of occurrence of Class i POI in all plot units; |D| is the total number of plot units in the city; and {j:t_i$\in$D_j} is the total number of block units of Class i POI. In Formula (3), TF-IDF is the weighted value of class i POI in the Block j unit.

3.4. Supervised Classification Learning Algorithm

3.4.1. Algorithm Performance Comparison and Selection

In machine learning, common supervised learning classification algorithms include the logistic regression, KNN, support vector machine (SVM), and decision tree algorithms. Among them, decision trees and ensemble learning play a dominant role. Considering the numerous categories of urban land labels and the feature dimensions of plots, which include weighted business format feature dimensions and architectural form feature dimensions, experiments were conducted on four models—the decision tree, random forest (RF), AdaBoost, and GBDT models—to improve the accuracy of model classification. The classification accuracy of the models was assessed and compared (Figure 5). Additionally, a comparative analysis was conducted on the performance evaluation indicators of the different machine learning models across the various types of functional land use (

Table 1. Comparison of the accuracy of various functional land use identification results for the main urban area of Yiwu city in 2022.

	Methods
	DT			AdaBoost			RF			GBDT
Functional land use types	Precision	Recall	F1 score	Precision	Recall	F1 score	Precision	Recall	F1 score	Precision	Recall	F1 score
Transportation land	0	0	0	0	0	0	1	0.09	0.17	0.9	0.41	0.56
Parks and green spaces	0	0	0	0	0	0	1	0.12	0.22	1	1	1
Public facility land	0	0	0	0	0	0	1	0.14	0.25	0.81	0.71	0.83
Rural homestead	0	0	0	0.53	0.12	0.19	0.79	0.49	0.61	0.68	0.48	0.61
Commercial land	0.15	0.99	0.27	0.43	0.69	0.53	0.7	0.7	0.76	0.65	0.68	0.7
Urban residential land	0	0	0	0.55	0.78	0.65	0.67	0.9	0.77	0.65	0.85	0.74
Industrial land	0	0	0	0.66	0.67	0.67	0.8	0.86	0.83	0.81	0.82	0.81
Land for organizations	0	0	0	0	0	0	0	0	0	1	0.24	0.39
Science, education, culture, and health land	0	0	0	0	0	0	0	0	0	0.85	0.6	0.71
Logistics storage land	0	0	0	0	0	0	0.89	0.43	0.58	1	1	1

). For each model, the following performance indicators were provided: precision, recall, and the F1 score.

After comparing the results, the precision rates of the random forest algorithm and the GBDT algorithm were relatively close, with each showing strengths and weaknesses in their classification results. However, the GBDT performed better in terms of the F1 score, which is used to measure the overall model performance. Therefore, this study adopts the optimized GBDT (gradient boosting decision tree) model for classification. During the optimization process, the model was trained using the training set to determine parameters such as the number of learners, the maximum depth, the minimum sample split, subsamples, the maximum number of features, and the learning rate, ensuring the optimal performance on the subsequent validation set (

Table 2. Optimal Parameters of GBDT Model.

n_Estimators	Max_Depth	Min_Samples_Split	Subsample	Max_Features	Learning_Rate
50	3	2	1	Sqrt	0.05

).

3.4.2. Supervised Classification Learning

In supervised classification learning, the dataset is segmented to separate the training set and the test set. Twenty percent of the data are selected as the training set, and the remaining 80% are selected as the test set. The urban land use data of the learning sample were used as the learning labels for the supervised classification learning. The data input into the machine learning model included POI data, plot data, and building form data. In the training process, the correlations between the business type, business name, land form indicators, the other data for each plot, and the actual land use type of the plot were determined to generate a machine learning classification model.

3.4.3. Automatic Identification of Land Types

The classification model that was completed during the machine learning training was used to determine the type characteristics, land parcel distribution, and architectural form characteristics of the land to be identified in Yiwu city by setting the parameters. In this way, the system can automatically note the corresponding classification model.

By using the database, the collected type characteristics and architectural form characteristics of the target urban plots were obtained and inputted into the corresponding classification model as feature vectors. Finally, the types of possible land use properties of each block in the target city area were generated. For the land use properties obtained from each plot, the plots with the same land use properties were filled in with a color, and the confidence level of each land use identification result was automatically marked. The nature of marked land usually follows the grading standards of the urban construction land classification in the “Urban Land Classification and Planning and Construction Land Standard” (GB50137-2011) [41] and can also be adjusted according to the urban classification land standards of various regions and cities. Such an automatic identification system can provide urban land information efficiently and accurately.

4. Results

4.1. Identification Results of Urban Functional Areas

According to the supervised classification learning algorithm, ten major urban land types—transportation service stations; parks and green spaces; public facilities; rural housing land; urban housing land; commercial service facility land; industrial land; press and publication land for government organizations; science, education, culture, and health land; and logistics and warehousing land—in the main urban area of Yiwu city were identified (Figure 6 and Figure 7). This study explored and analyzed the characteristics of the change in the distribution patterns of the five functional elements of housing, commerce, public services, industry, and logistics and warehousing in 2012 and 2022. As shown in the identification results, all kinds of functional land use in Yiwu city increased to different degrees over the past ten years (

Table 3. Different types of urban land area and changes in the main urban area of Yiwu city in 2012 and 2022.

Land Use Type	Urban Residential Land	Industrial Land	Rural Homestead	Commercial Land	Logistics Storage Land	Science, Education, Culture, and Health Land	Parks and Green Spaces	Transportation Land	Public Facility Land	Land for Organizations	Total Area
Area in 2012 (ha)	1293.91	875.55	192.43	321.17	10.27	117.70	36.67	1.08	3.39	35.65	2887.83
Area in 2022 (ha)	2486.54	1505.24	707.70	500.90	251.46	243.68	56.38	52.03	21.81	15.46	5841.18
Area change (ha)	1192.63	629.69	515.27	179.73	241.19	125.98	19.71	50.95	18.42	−20.19	2953.35
2012 Structure (%)	44.8	30.3	6.7	11.1	0.4	4.1	1.3	0	0.1	1.2	100
2022 Structure (%)	42.6	25.8	12.1	8.6	4.3	4.2	1.0	0.8	0.4	0.2	100
Structural changes (%)	−2.2	−4.5	5.4	−2.5	3.9	0.1	−0.3	0.8	0.3	−1.0	0

).

As shown in Table 3, we can note the following: (1) From the perspective of the change in land use scale, except for the decrease in land use for government agencies and organizations, other types of urban land use show a trend towards scale expansion. Among them, urban residential land increased the most, followed by industrial land and rural residential land; next was land for logistics and warehousing and commercial services. The results show that urban and rural housing construction and industrialization are the dominant forces of Yiwu’s urban expansion, followed by the spatial expansion of trade logistics. (2) From the perspective of the land use structure, the proportion of rural housing land and logistics and warehousing land increased greatly, and the proportion of transportation land also increased to a certain extent. The proportion of industrial land decreased the most, followed by commercial services and urban residential land, and the proportion of government agencies and organizations also declined to a certain extent. The proportion of land used for science, education, culture, and health; parks and green spaces; and public facilities did not change much. This finding is closely related to the transformation and development of Yiwu city.

4.2. Spatial Pattern Changes in Urban Functional Areas

4.2.1. Changes in Residential Spatial Patterns

Living space bears the key responsibility for people’s living and social stability and is a crucial component of the urban structure. Housing demand is rising rapidly, especially in the current context of urbanization. According to the identification results of functional areas, there were 253 residential functional units in the main urban area of Yiwu city in 2012, 220 of which were urban residential land and 33 were rural residential land. By 2022, there were 661 residential functional units in the main urban area of Yiwu city, 542 of which were urban residential land and 119 were rural residential land. From the perspective of the connotation and characteristics of functional areas, they include many types of housing such as the more common residential communities, residential buildings, villa areas, urban villages, and dormitories. From the perspective of spatial distribution, residential space shows the characteristics of a dense central layer and a scattered peripheral distribution, being distributed from northeast to southwest and from east to west. The northeast-to-southwest trend of residential areas in the Futian subdistrict was relatively sparse in 2012, but, by 2022, the Futian subdistrict had also formed a dense layout. In 2012, the distribution from east to west was manifested only in the central area of the Jiangdong subdistrict, and, by 2022, the distribution trend had been extended, almost running throughout the entire Jiangdong subdistrict. These residential spaces are generally distributed in the core areas of various subdistricts, such as the Chouzhou subdistrict, Jiangdong subdistrict, Futian subdistrict, and Beiyuan subdistrict. The distribution of some residential spaces outside the main urban area is relatively scattered, such as the Houzhai subdistrict, Chengxi subdistrict, and Niansanli subdistrict. As shown in Figure 8, from 2012 to 2022, the number of POIs of residences in each subdistrict increased to varying degrees. The Choucheng, Choujiang, Jiangdong, and Futian subdistricts showed significant increases.

In terms of the number of POIs in residential communities in the main urban area of Yiwu city in 2022 (

Table 4. Comparison of the quantity of POIs of urban functional land in various subdistricts in the main urban area of Yiwu city.

Subdistrict	Number of POIs	Residential Land	Commercial Land	Public Facility Land	Industrial Land	Logistics Storage Land
Choucheng	Number of POIs in 2012	77	5333	604	352	44
	Number of POIs in 2022	172	7829	466	685	61
Choujiang	Number of POIs in 2012	222	2328	406	393	25
	Number of POIs in 2022	306	9663	553	2046	213
Futian	Number of POIs in 2012	175	1969	166	220	39
	Number of POIs in 2022	264	10,404	379	2191	478
Jiangdong	Number of POIs in 2012	158	2143	391	155	206
	Number of POIs in 2022	257	9467	586	1701	776
Houzhai	Number of POIs in 2012	80	391	124	130	6
	Number of POIs in 2022	94	3031	154	667	140
Chengxi	Number of POIs in 2012	36	196	87	73	2
	Number of POIs in 2022	66	2272	108	656	263
Niansanli	Number of POIs in 2012	76	511	93	129	2
	Number of POIs in 2022	98	3507	181	1025	93
Beiyuan	Number of POIs in 2012	260	1977	243	406	50
	Number of POIs in 2022	297	10,292	479	2203	522

), the distribution of POIs in the Choujiang subdistrict and the Beiyuan subdistrict is the most intensive, and the sum of these two subdistricts accounts for 38.80% of the total POIs in the main urban area of Yiwu city. In contrast, the number of POIs in the residential communities of Houzhai, Chengxi, and Niansanli on the urban fringe is relatively small, accounting for only 16.60% of the total in the main urban area of Yiwu city. The Chengxi subdistrict is positioned as a modern logistics and leisure tourism area, focusing on the development of modern logistics, tourism services, and ecological agriculture functions. The Niansanli subdistrict mainly develops industrial parks and is an important part of provincial high-tech parks. These two subdistricts are far from the city center and are relatively independent; they are mainly industrial enterprise gathering areas, and the development of residential areas has occurred relatively late. The housing quantities in the Choucheng, Futian, and Jiangdong subdistricts are more balanced, accounting for 11.07%, 16.99% and 16.54%, respectively, of the residential space in the main urban area of Yiwu city.

4.2.2. Changes in Commercial Spatial Patterns

As one of the most important economic activity spaces in modern cities, commercial space undertakes the important functions of people’s shopping, catering, recreation, and other social activities. According to the statistics of the commercial POI data on the administrative subdistricts in Yiwu city, the results show that the distribution of commercial space is mainly concentrated in the Choujiang and Futian subdistricts, which include popular commercial areas such as the Binwang business circle, Futian financial town business circle, and Huangyuan business circle. The distribution of commercial space in other regions is relatively balanced. These findings show that the Choujiang and Futian subdistricts play a central role in the city in terms of commercial activities, with thriving business circles providing people with a wealth of shopping, entertainment, and office options.

According to the identification results of functional areas, there were 143 commercial service facilities in the main urban area of Yiwu city in 2012; by 2022, there were 193 (Figure 9), revealing a clear momentum of increase. In 2012, there were 5333 commercial POIs in the Choucheng subdistrict, far ahead of several subdistricts, and the numbers in the Choujiang, Jiangdong, Beiyuan, and Futian subdistricts were not much different. The number of commercial POIs in the Niansanli, Houzhai, and Chengxi subdistricts was very small. By 2022, the number of commercial POIs in the Futian subdistrict jumped to first place, and the number of commercial POIs in the Beiyuan, Choujiang, and Jiangdong subdistricts increased rapidly. Although the commercial POIs in the Niansanli, Houzhai, and Chengxi subdistricts also increased, the numbers were still significantly lower than those in other subdistricts.

Judging from the total distribution of commercial POIs in 2022 (Table 4), there are significant differences in the total commercial volume of different subdistricts in Yiwu city. The Futian, Beiyuan, Choujiang, Jiangdong, and Choucheng subdistricts have a greater total commercial volume, because Futian is the location of the Yiwu small commodity city and the total number of shopping malls involved is relatively large. The Choucheng subdistrict is a traditional concentrated area of commercial functions, with the traditional Embroidery Lake business circle and Binwang business circle. The Beiyuan and Jiangdong subdistricts, which are areas of urban modern life and industrial upgrading, are Yiwu’s e-commerce cluster blocks. In the past ten years, Yiwu’s e-commerce industry has developed rapidly, and many related stores have been established. The Choujiang subdistrict is a key headquarters of the business circle and a quality of life area built by Yiwu city, and it has introduced large complex urban projects such as Wanda Plaza. The population and living space of Niansanli, Houzhai, and Chengxi is relatively small, resulting in relatively limited commercial facilities; thus, the number of commercial POIs accounts for a low proportion of the total number of commercial POIs.

From the perspective of the spatial differentiation of the total amount of commerce (Table 4), the total amount of shopping malls presents an obvious “centre-periphery” spatial differentiation, indicating that shopping malls are clustered mainly in the central area of the city, with the strong attribute of a commercial center. It is worth mentioning that the commercial functional plots of Niansanli, Houzhai, and Chengxi, where the number of commercial POIs is relatively low, are located at the junction of the administrative subdistricts near the city center, indicating that the commercial economic activities in the city center have certain spillover effects. For example, there are commercial plots on the west side of the Chengxi subdistrict near Shangxi Town; there are also commercial plots near the Longhui business circle on the east side of the Chengxi and Choujiang subdistricts.

4.2.3. Changes in the Spatial Pattern of Public Facilities

The land for urban public service facilities includes the land for public facilities; the land for science, education, culture, and health; the land for government agencies and organizations; the land for the press and publications; and the land for parks and green spaces and transportation service stations. These plots exist to meet the basic needs of urban residents, providing educational, medical, cultural, recreational, and social services. Adequate public service facilities are an important support for sustainable urban development, and they play a key role in improving the quality of life of urban residents and promoting urban economic and social development.

According to the identification results of the functional areas, there were 40 public service facilities in the main urban area of Yiwu city in 2012. By 2022, there were 73 units of public service facilities in the main urban area of Yiwu city, and the number of public service facilities has grown significantly over the past decade.

From 2012 to 2022, the number of POIs in each subdistrict in the main urban area of Yiwu city also significantly increased (Table 4). In 2012, there were 604 public service POIs in the Choucheng subdistrict, the largest number, because the Choucheng subdistrict is the core area of life services in the old city and is related to the living and service links on both sides of the Yiwu River. It is followed by the Choujiang subdistrict and Jiangdong subdistrict, which are urban living and residential places, respectively. Then, there are the Beiyuan and Futian subdistricts, with the Houzhai, Niansanli, and Chengxi subdistricts ranking behind them. By 2022, the Jiangdong subdistrict had the most public service facility POIs, because this subdistrict is a livable living and cultural education area of Yiwu city; has improved Yiwu’s exhibition and expo, e-commerce, and cultural and sports center functions; and has built parks and green spaces such as Nanshan Park. Additionally, the Yiwu International Expo Centre, Yiwu City Stadium, and Yiwu City Planning Exhibition Hall are located in the Jiangdong subdistrict. Many new educational facilities in Yiwu city, such as the Yiwu Art School and Yiwu University for the Elderly, are also located in the Jiangdong subdistrict. The number of public service facility POIs in the Beiyuan and Futian subdistricts increased significantly. There are large public service facilities, such as the Yiwu 365 Administrative Convenience Centre and Yiwu Airport, in the Beiyuan subdistrict. The Futian subdistrict was only established in 2014. Although it was established only recently, the Silk Road New District and the Science and Innovation New District, among the three new functional districts, are located mainly in Futian, which includes Yiwu Port and the International Business City Passenger Transportation Centre. There are eleven primary and secondary schools and four hospitals (the Fourth Affiliated Hospital of Zhejiang University, Yiwu Third People’s Hospital, Yiwu Mental Health Center, and Heyetang Branch of Fuyuan Hospital). The public service POIs in the Niansanli, Houzhai, and Chengxi subdistricts have also increased.

Judging from the spatial differentiation of the public service facilities (Figure 10), the public service facilities of each subdistrict are generally located in the center of each administrative subdistrict. Considering the convenience and accessibility of people’s travel, the new large-scale public service facilities are centrally arranged, and green parks are built around them to facilitate residents’ entertainment and rest. The green space of Yiwu Park is arranged according to its natural base conditions, such as natural water bodies, mountains, or riversides. In terms of the layout of the transportation facilities, due to the continuous expansion of the city and changes in transportation modes, the stations and urban commercial functions originally located in the city center are expected to relocate to the outer blocks of the city. For example, Jiangdong Passenger Station and Southern Joint Operation have been relocated to the Yiwu International Trade City Passenger Transport Centre.

4.2.4. Changes in Industrial Spatial Patterns

Although industry plays a core role in urban development, it is not the most prominent core element in the modern urban development of Yiwu. According to the results of the functional zone identification, there were 124 industrial spatial units in the main urban area of Yiwu city in 2012; by 2022, there were 280.

In 2012, the industrial land in the main urban area of Yiwu city was mainly distributed in the Choujiang, Beiyuan, and Houzhai subdistricts, and other small industrial land was scattered in the urban area. The total amount of industrial land is large, the distribution is scattered, and the layout is mixed with other urban construction land. Among them, the Beiyuan and Choujiang subdistricts have a better industrial foundation (Figure 11).

With the gradual relocation of industrial land in the central urban area, scattered industry has been cancelled, industrial land with a concentrated scale and a good development foundation has been retained, and the scale has been controlled. By 2022, the number of industrial POIs in the Beiyuan subdistrict, Futian subdistrict, and Choujiang subdistrict accounted for nearly 20% (Table 4). Nearly 10% were in the Niansanli subdistrict. The industrial agglomeration effect in the Futian and Niansanli subdistricts is outstanding. Yiwu is building an industrial transformation and upgrading the belt around the city through the linkage of scientific research, intelligent manufacturing, and trade functions. This belt runs through the three subdistricts of Beiyuan, Futian, and Niansanli. In the past ten years, the Niansanli Industrial Park has been integrated into the overall development framework of provincial high-tech industrial parks, introducing high-tech industries such as electronic appliances, the Internet of Things, and medical equipment. The Heyetang Industrial Park in the Futian subdistrict has been transformed into a high-end modern service industry, developing creative research and development, business upgrading, and other related functions. The number of industrial POIs in the Jiangdong subdistrict also accounts for 15.22%, and these POIs are distributed mainly in the eastern Qingkou area of the Jiangdong subdistrict and include e-commerce, jewelry, and other industries.

4.2.5. Changes in Logistics and Storage Spatial Patterns

Yiwu is the meeting point of the Yiwu–Xinjiang–Europe express and the Yiwu–Yong (Ningbo)–Zhoushan express, which have always played an important role in Yiwu’s import and export trade by opening up the dry port channel. In recent years, the “Yiwu-Xinjiang-Europe” express has expanded from one route to fifteen routes, connecting 49 countries and regions on the Eurasian continent. With an annual growth rate of 103%, it has become a golden freight channel on the southern wing of the Yangtze River Economic Belt. In 2019, Yiwu became the only city in the province with the linkage of “four ports”, namely, the land port, seaport, airport, and information port.

In 2012, the main urban area of Yiwu city was distributed mainly within the inland port station of Yiwu Port, the international logistics center and the Jiangdong freight market; these areas clustered around the market and lacked effective connections with the external transportation system. At the same time, there are a large number of private logistics and storage enterprises leasing low-cost, four-and-a-half floor housing on a small scale and experiencing traffic chaos, which seriously interferes with residents’ lives and urban traffic. From the perspective of the quantity distribution, there are more logistics and storage points in the Jiangdong subdistrict, Beiyuan subdistrict, and Choucheng subdistrict. In 2022, the logistics land of Yiwu city was distributed on the periphery of the main urban area, the layout was centralized in blocks, and the storage function was guided by fragments. In the core area of the city and important structural corridors, four-and-a-half floor rental storage functions are strictly prohibited, and the logistics enterprises on the four-and-a-half housing of the central city gather in the west logistics park (Figure 12).

The Choucheng subdistrict is positioned in planning as the area where the city’s administrative, commercial, and trade functions are concentrated, focusing on the development of municipal administrative offices, international trade, business finance, life services, quality of life, and other functions. Thus, the number of warehouse logistics POIs in the Choucheng subdistrict has reduced from the most to the least. The Jiangdong and Beiyuan subdistricts are involved in e-commerce development, are one of the birthplaces of China’s Taobao village, and are known for gathering a large number of e-commerce practitioners. The Jiangdong subdistrict has a convenient traffic network and is the exit point to Hangjinqu and Yongjin, two high-speed, convenient traffic networks. There are many e-commerce parks in the Beiyuan subdistrict. These parks have a strong atmosphere of innovation and entrepreneurship, support and inspire a large number of e-commerce entrepreneurs, and promote the development of the e-commerce industry and warehousing and logistics industry. The Futian subdistrict is the location of Yiwu International Trade City and is an important part of the Yiwu Northeast logistics group. The Chengxi subdistrict has an international dry port hub, a comprehensive free-trade zone, and Yiwu West Station; is building a global import and transit trade hub; and is an important part of the Yi southwest logistics group. The logistics and warehousing in the Choujiang, Houzhai, and Niansanli subdistricts also exhibited an increasing trend (Table 4).

5. Discussion

In this study, the POI, TF-IDF algorithm, and machine learning classification algorithm are comprehensively used to solve the following three problems: (1) Identification of urban function diversification: The POI data can provide a variety of city function information. The TF-IDF algorithm and machine learning classification algorithm can help us accurately classify city functions. (2) Difficulty of data collection: Data such as POI data and remote sensing images are easy to collect, but accurately classifying them functionally is a challenge. The use of machine learning classification algorithms can improve the effective use of these data. (3) Adapt to urban planning and management needs: The combination of the POI, TF-IDF algorithm, and machine learning classification algorithm can automatically identify the business characteristics, plot distribution, and building form characteristics of urban land; reveal the evolution of urban land function and spatial structure; and provide data support for urban planning and management.

This study compared the identification results of urban construction land in the main urban area of Yiwu city with the actual land use status and revealed that the land use function identification method proposed in this paper can achieve different accuracy rates for different land use functions. A comparison of the identification results of residential functional land with the status quo of land use in Yiwu city reveals that the average identification accuracy rate of residential functional land is 73%, that of commercial functional land is 68%, that of industrial land is 81%, that of public service facilities land is 95%, and that of logistics and warehousing land is 100%. There are various reasons for the differences between the different land use function identification results and the land use status. In addition to the errors caused by the research method itself, the phenomenon of mixed utilization of residential space and commercial space may exist, and the single main body function is merged in the actual statistics.

Moreover, the identification results of urban functional areas can be applied to individual properties. As long as there are enough sample data for the machine to learn, the data set can be classified into functional areas through algorithm selection and model parameter adjustment.

The innovation of this research is mainly reflected in the following aspects:

(1) This paper provides a new perspective for understanding subtle changes in urban space. Compared with traditional urban land use research, this methodology can more accurately capture the details of urban spatial changes, contributing to a deeper understanding of the complexity of urban development. Based on POI data and urban form data, the TF-IDF algorithm and a machine learning classification algorithm are used to identify the land use functions and distribution characteristics of the main urban area of Yiwu city. This combination of multi-source data and methods contributes to a deeper understanding of the urban spatial structure and land use. Li et al. [42] and Liu and Long [43] used a random forest algorithm to analyze point of interest (POI) data, aiming to identify the functional zoning of the central urban area of Chongqing, and they combined POI data with OpenStreetMap (OSM) data to quantify the degree of functional mixing. However, the method adopted was relatively simple, and there was no comparative analysis of multiple methods. Huang et al. [44] combined POI data, the OpenStreetMap (OSM) dataset, and high-resolution remote sensing images to construct a functional intensity index to identify urban functional areas. Barlacchi et al. [45] presented a framework for performing an automatic analysis of land use zones based on location-based social networks (LBSNs). Since there was no combination of three-dimensional spatial data, such as architectural form data, and since the research scope covered only the plots enclosed by the road network, the precision was not high. Some scholars [46,47,48] proposed an urban functional area identification method based on the ratio of frequency density to POI functions, and they selected a 250 m grid as the basic unit. Because the data source was relatively simple and the grid was used as the classification unit of functional areas, there was an inconsistency with the irregular edge of the actual plot, affecting the functional discrimination effect of the actual plot. Based on POI big data on building forms, Yang et al. [40] adopted the TF-IDF transformation and database construction to identify urban land of different types, sizes, and locations by using supervised deep learning methods; however, they compared only the overall accuracy of machine learning algorithms. This study references their methods and adds a comparison of specific precision parameters between the various algorithms (Table 1). This study selected four model algorithms, namely, the decision tree, random forest (RF), AdBoost, and GBDT algorithms, and found that the accuracy of the random forest (RF) and GBDT models was similar, but there were differences in the recognition accuracy of specific land functions.

(2) The fine-grained identification of land use functions in the whole area of Yiwu’s main urban area was achieved. Most previous studies focused on macroscopic spatial structure analysis, while fine-grained analysis of the microscopic spatial structure was insufficient. In addition, most scholars use the land surrounded by a road network as the research unit for analysis, which is prone to identification errors.

(3) This study added an analysis of social and economic functions. High-resolution remote sensing images cannot capture the socio-economic functions of ground objects; thus, they cannot meet the actual needs of urban planning and management. To compensate for this deficiency, this study used POI data and site and building form data, which helped to analyze the functional nature of land use more comprehensively. The findings of this study hold great significance for urban planning practice. Urban planners should pay close attention to the dynamic changes in urban land use and adopt more flexible planning strategies to respond to a rapidly changing market and social needs. For example, land use planning can be adjusted to promote the rational allocation of commercial and residential land to meet the needs of urban development.

Although this study has obtained some results in the fine-grained identification of construction land functions in the main urban area of Yiwu city, it still faces some challenges. First, the research is limited to construction land in the main urban area of Yiwu city, which does not fully cover non-construction land. The classification and analysis of non-construction land will be strengthened in the future. Second, due to the limited data collection, the functions of urban land in only a specific period can be identified, and a historical dynamic analysis of urban land cannot be performed. In the future, we plan to gradually collect multi-source data at different points in time to deepen the comparative study of urban functional space. Social media data and traffic flow data can also be introduced to further enrich the understanding of the evolution of urban space. Third, the accuracy of this research in identifying rural residential land, urban residential land and commercial service facility land is relatively low, which is related to the distribution of POIs. In Yiwu city, urban residential land and commercial service facility land are often intertwined, and their functions are relatively mixed. Moreover, the sample size of the main urban area of Yiwu city is relatively small; thus, the model cannot be fully trained.

6. Conclusions

Currently, urban development makes the layout of urban land increasingly complicated, which also increases the difficulty and time cost of measuring the nature of urban land. In the current environment of increasingly abundant big data, it is possible to efficiently and scientifically identify large-scale spatial land types and conduct fine-grained quantitative analyses of cities. This paper discusses a method for comprehensively extracting the distribution characteristics of urban POI data, urban spatial plots, and building form data to form multi-dimensional features that automatically identify urban land use via machine learning. Compared with the traditional field survey method, this study can effectively avoid problems such as the time and effort associated with reconnaissance and the easy misjudgment of complex plots. This method can address the complex task of urban land identification more efficiently and improve the effectiveness of urban planning and management.

This study uses machine learning automatic identification technology to identify the urban functional land in the main urban area of Yiwu city in 2012 and 2022. The main urban area of Yiwu city included traffic service station land; parks and green spaces; public facility land; rural housing land; urban housing land; commercial service facility land; industrial land; government organization land; science, education, culture, and health land; and logistics and warehousing land. In total, ten major urban land types were identified. The five functional elements of residential, commercial, public services, industry, and logistics warehousing in 2012 and 2022 were analyzed. The main conclusions are as follows: (1) From 2012 to 2022, all kinds of functional land use in Yiwu city experienced varying degrees of growth, but the proportions of different types of functional land use changed. (2) Residential space has a spatial structure with a dense central layer and a scattered outer layer, being distributed from northeast to southwest and from east to west. (3) The commercial service functional area shows obvious “centre-periphery” spatial differentiation. The commercial functional plots of the Niansanli, Houzhai, and Chengxi subdistricts, where the number of commercial POIs is relatively low, are located at the junction of the administrative subdistricts near the downtown area, indicating that the commercial economic activities of the downtown subdistricts have certain spillover effects. (4) The public service facilities of each subdistrict are generally located in the center of each administrative subdistrict, with better convenience and accessibility. (5) Industrial land is large, scattered, and mixed with urban living construction land and tends to follow the current centralized layout; this land is mainly distributed in the Beiyuan, Houzhai, Futian, and Niansanli subdistricts and forms an industrial belt around the city. (6) Logistics and warehousing land accounted for 0.4% of land use types in 2012 compared to 4.3% in 2022, which was the largest change in the proportion of all types of functional land. This finding is also consistent with the target path of Yiwu city, which is vigorously building a national logistics hub of trade and service.

Since the late 1970s, Yiwu has always adhered to the development strategy of “thriving commerce and establishing a city”, having successfully completed the leap from a regional commercial city to a national commercial city and then to a global commercial city, becoming a unique example among global cities. In addition to a group of leading entrepreneurs and advancing institutional innovations, the main reasons for Yiwu’s success lie in the diversification of industries and products, the complexity of the economy, and the wide connectivity of capacity networks. Yiwu is integrated into a global urban network that links its local industries to global economic activities and the flow of resources, capital, knowledge, talent, and information. However, Yiwu also faces some practical challenges: (1) The land use layout is highly mixed, and the urban space quality is not high. The “four-and-a-half story” houses (i.e., living upstairs, renting downstairs, and use as factories or warehouses) in the downtown area of Yiwu cover an area of about 10 square kilometers, and the functions of commerce, office, residence, logistics, warehousing, and production are highly mixed in the internal spaces. Excessive functional mixing has led to a series of problems such as freight transportation through the city, logistics throughout the city, traffic interference, lack of public space and service supporting facilities, etc., affecting the overall image and environmental quality of Yiwu. (2) The commercial space is saturated, and the centrality is insufficient. The total commercial construction area of Yiwu city is 912,000 square meters, 0.48 square meters per capita, and the total volume tends to be saturated. From the perspective of the spatial distribution, each business district basically has only one large commercial complex, which is difficult to form the agglomeration effect of commerce. (3) There is also a scattered and inefficient distribution of industrial land.

According to the theory of complexity science, cities are complex systems formed by the interaction of agents such as people, institutions, markets, and networks [49,50,51]. Moreover, cities interact with other cities at the regional, national, and global levels, which always results in various emergent outcomes. The change of trade patterns and economic conditions can affect Yiwu’s market hub status. In the face of challenges such as the impact of global climate change, security threats, epidemics, and emergencies, the competition between residential land, commercial and retail space, industrial land, transportation land, green land, recreational land, and other types of land in Yiwu will be further intensified. The development trend will be moderately compact residential space, high-quality and efficient commercial space, the reduction and intensive use of industrial space, and the expansion and quality improvement of public space. In order for Yiwu city to continue to occupy the market hub position in this emergent network, it must quickly and flexibly seize the first-mover advantage, actively connect the global networks, properly manage the complexity of the economy, ensure the balance between various types of land, and make the market gathering place become the gathering place of highly skilled talents and innovative capabilities. In addition, with the rapid development of digital technology, big data continues to emerge and present diversified sources, and artificial intelligence technology will also be more widely used. Yiwu can make full use of big data and artificial intelligence technology to carefully study complex urban systems; simulate and plan future scenarios of the city; improve the efficiency of urban development, service efficiency, and spatial quality; and make Yiwu more vibrant, more harmonious, more resilient, and more livable.