From Clustered to Networked: Multi-Dimensional and Multi-Scale Performance Evaluation of Polycentric Urban Structure Evolution in Shenzhen, China

Abstract

Polycentric urban structures are increasingly advocated to enhance economic efficiency, promote spatial equity, and support environmental sustainability. However, their roles in shaping urban performance remain insufficiently understood, particularly regarding the contributions of clustered and networked patterns to performance across different dimensions and scales. This study adopts a longitudinal framework that distinguishes the evolutionary stages of polycentric urban structures and evaluates multi-dimensional, multi-scale performance, taking Shenzhen, China as a case study. The results show that Shenzhen’s polycentric structure evolved nonlinearly in its clustered pattern and linearly in its networked characteristics, with steady improvements in spatial and economic performance contrasting with the inverted U-shaped trajectory in social performance. Clustered and networked polycentric structures contribute differently: improvements in spatial performance are driven by multiple indicators, major economic indicators (at constant prices) increase with strengthened networked characteristics, and social performance benefits from clustered patterns only when public service provision is coordinated. This research provides new evidence for the co-evolution of polycentric structures and urban performance, suggesting that the effectiveness of polycentric development lies not in choosing between clustered and networked forms, but in strategically integrating them—optimizing scale–distance coordination among centers to enhance the clustered pattern, while differentiating center functions to strengthen networked characteristics.

1. Introduction

Urbanization has become an irreversible global trend. In 2007, the global urban population surpassed the rural population, and more than 66.7% of the world’s population is projected to live in urban areas by 2050 [1,2]. This rapid concentration of population in cities has fueled continuous urban expansion, which, in turn, has exposed the agglomeration diseconomies of monocentric structures, such as traffic congestion, land scarcity, and environmental degradation. In response, polycentric urban structures have been promoted as a bridging concept, which are expected to enhance economic efficiency, reduce spatial inequity, and support environmental sustainability [3,4,5,6]. Although a growing body of literature has evaluated the promised benefits of polycentric structures, whether polycentricity is a panacea, placebo, or pathogen remains a subject of debate [6,7,8,9].

This divergence stems largely from two unresolved issues. First, most studies have employed polycentricity indices to measure urban spatial structures and explore their correlations with urban performance [7,8,10,11,12,13,14,15]. This approach has two limitations: (1) different polycentric structures may result in similar polycentricity index values but different real-world impacts (Figure 1 1) [16]; and, (2) more importantly, because these indices often ignore the functional attributes of centers, they cannot capture the clustered and networked patterns in the evolution of polycentric structures that affect urban performance.

Second, performance evaluations are often narrow in scope. (1) Most studies have adopted single-dimensional evaluation frameworks, focusing on economic, social, or environmental performance [10,11,12,13,14,15]. However, conclusions reached under different evaluation objectives may conflict; for instance, polycentric structures may reduce economic performance while improving social performance, posing a dilemma for policymakers. Although recent studies have examined the impacts of polycentricity on integrated performance, the methods used to compute integrated performance values (e.g., the non-radial directional distance function) often fail to account for the relative importance of indicators in planning practice [7,8]. (2) Evaluations are typically conducted at the regional or city level. However, overlooking intra-urban variations can mask local declines behind city-level improvements. The governance perspective on polycentric cities emphasizes the need for cross-scale cooperation and assessments to reflect how such cities actually function [17].

To address these gaps, this study adopts a longitudinal and descriptive analytical framework that (1) distinguishes the evolutionary stages of polycentric urban structures; (2) develops a multi-dimensional, multi-scale performance evaluation system; and (3) investigates of the co-evolution between polycentric structures and urban performance. The performance evaluation system integrates spatial, economic, and social dimensions across center/cluster, inter-center, and citywide scales, with indicator weights determined using a hybrid Analytic Hierarchy Process (AHP)–entropy weighting method approach. We selected Shenzhen as a case study. On one hand, Shenzhen, which has largely developed in accordance with successive master plans, is widely regarded as a representative example of polycentric urban development in China [18]. On the other hand, it faces intense urban tensions, including rapid population growth, uneven spatial development, and pressures on public services. This dual rationale—success and tension—makes Shenzhen a valuable case for polycentric studies. Therefore, this study aims to address the following three research questions:

(1) How has Shenzhen’s polycentric urban structure evolved?

(2) How has its integrated performance evolved, and how have the individual spatial, economic, and social performances evolved? What factors have influenced these trajectories?

(3) How can planners optimize polycentric urban structures to improve urban performance?

Building on these questions, we propose the following hypothesis: Shenzhen’s clustered and networked polycentric structures have developed concurrently, and its integrated performance has continued to improve. The increase in spatial performance results from growth in the number of cluster centers and the development of clustered polycentric structures; the improvement in economic performance stems from the broader context of Shenzhen’s economic development and the strengthening of networked polycentric structures; and the growth of social performance is conditional, with the positive effects of clustered polycentric structures being realized only when appropriately matched public service facilities are provided.

The remainder of this study is structured as follows: Section 2 presents the theoretical framework; Section 3 introduces the study area, data, and methodology; Section 4 traces the co-evolution of the polycentric structure and urban performance; and Section 5 and Section 6 present the discussion and conclusions.

2. Theoretical Framework

2.1. Evolutionary Stages of Polycentric Urban Structures

The evolution of polycentric urban structures typically proceeds through four stages: monocentric structures, initial polycentric structures, clustered polycentric structures, and networked polycentric structures [19].

Monocentric cities—the most common urban spatial form—are characterized by a single center, typically located near the city’s geographic center, with residential peripheral areas. In such cities, the center performs comprehensive functions, including administration, commerce, employment, and public services, while peripheral areas heavily rely on it. Accessibility to those functions decreases significantly as the distance from the center increases. With population growth and urban expansion, issues such as traffic congestion and land scarcity gradually emerge, leading to the development of new centers (Figure 2a).

Cities evolve into initial polycentric stages when one or more new centers form outside the main center, while the main center continues to dominate. The newly developed centers may function as industrial subzones or residential hubs but lack comprehensive daily life services and cannot serve as cluster centers. Most commuting and service flows are still directed toward the main center. Inter-center linkages are weak, with few networked characteristics (Figure 2b).

In the clustered polycentric stage, some mature centers form clusters that are capable of serving local residents. Each cluster provides employment, commerce, and public services, allowing residents to satisfy most of their daily needs locally and thereby reducing reliance on the main center. Although these cluster centers are connected by transport networks, industrial collaboration remains weak and their networked characteristics are still limited (Figure 2c).

The networked polycentric stage is characterized by a shift from centers operating independently to cross-center collaboration. People, goods, and information can flow rapidly among centers. Collaboration replaces coexistence as centers develop functional specializations and industrially complement each other. Such economic linkages may generate borrowed size effects [20], allowing polycentric structures to approach or match the economic efficiency of monocentric structures (Figure 2d).

It is important to emphasize that, although the evolution of polycentric structures is often depicted as a gradual progression from monocentric to networked polycentric forms, this process should not be regarded as strictly linear or universally applicable. In practice, many cities exhibit hybrid or non-sequential trajectories, with certain stages being skipped or coexisting. For example, many German cities have consistently maintained a relatively stable monocentric structure since the beginning of the 21st century, such as Berlin, Hamburg, and Munich [21]. Many cities have experienced decentralization, without forming polycentric structures, like Syracuse and Seattle in the U.S., and Kumasi in Ghana [21,22]. The study on Cape Town indicates that between 1991 and 2011 the city had evolved into a clustered polycentric structure; however, it did not progress toward a networked form [23]. A similar phenomenon is also observed in many cities across China [24]. In 2000–2007, there were 36 European cities that experienced population growth in the main center alongside shrinkage in the subcenters [25]. Although this trend does not represent a complete transition from polycentric to monocentric structures, it signals a strengthening of the main center within otherwise complex urban systems. These evidence underlines that polycentric evolution is not a uniform linear sequence but is influenced by contextual factors, including institutional frameworks, urban planning, technological innovation, and economic shocks [21,22,23,24,25]. Accordingly, the four-stage model should be treated as a heuristic framework rather than a universal law. Its applicability is conditioned by socioeconomic development, institutional arrangements, and spatial contexts. Explicitly acknowledging these limits not only enhances the theoretical coherence of the framework but also underscores the diversity and complexity of polycentric trajectories worldwide, offering a more inclusive perspective on urban development paths.

2.2. Impacts of Different Evolutionary Stages on Urban Performance

In monocentric cities, production factors such as capital, technology, and labor tend to be concentrated in the urban center. Spatially, the concentration of enterprises reduces physical distance, facilitates information exchange, and improves factor mobility, which enhances spatial efficiency. Economically, enterprise agglomeration generates positive externalities, such as technological innovation and industrial synergy, driving productivity gains and economic growth [26]. However, monocentric cities also face several challenges. From social and transport perspectives, the limited land supply in the urban center intensifies competition and raises costs. These rising costs are ultimately borne by residents, compelling trade-offs between central accessibility and affordability. Environmentally, the concentration of activities in the urban center results in localized pollution hotspots—such as high emissions from dense industrial activity and traffic congestion—exacerbating urban environmental problems.

In the initial polycentric stage, spatial efficiency depends on scale–distance coordination between centers: large, closely spaced centers struggle to reduce agglomeration diseconomies, while small, widely dispersed centers suffer from weak connectivity due to distance-decay effects. Moreover, transport networks influence inter-center factor mobility. An efficient transport network improves inter-center connectivity by reducing travel time. Economically, although initial polycentric structures may alleviate agglomeration-related negative externalities in the main center, the newly developed centers—being relatively small and weakly specialized—may also weaken positive externalities. From social and transport perspectives, while some residents gain local access to basic services, many activities still require cross-center travel due to the incomplete provision of daily life services in the new centers. Environmentally, dispersing activities across multiple centers may help to dilute pollutants through natural diffusion [27].

As the new centers mature in terms of scale and services provided, cities evolve into clustered polycentric structures. Spatially, clustered polycentric structures generally serve a larger proportion of the population. At this stage, rational coordination among clusters is essential to prevent excessive competition caused by service overlaps and structural imbalances due to service blind spots—that is, areas not serviced by any center. In social and transport aspects, clustered patterns can reduce reliance on the main center by enabling residents to meet their daily needs locally. Another important social outcome is spatial equity. It is often assumed that polycentric structures inherently enhance equity [28]. However, theoretical support for this idea remains limited and outcomes are context-dependent [7,15]. Environmentally, shorter average trip distances can lower transport emissions [29]. However, polycentric structures can hinder the development of high-capacity transit systems, thereby increasing reliance on private vehicles and consequently leading to higher emissions [30]. Additionally, a larger per capita living space in secondary centers could increase energy demand [31,32].

As mentioned above, the characteristic of networked polycentric cities is a structural shift in the economic dimension. Centers form collaborative networks through functional specialization and industrial complementarity, generating borrowed size effects that enhance economic performance, which potentially enable polycentric structures to approach or match the economic efficiency of monocentric structures.

2.3. Summary, Definition, and Operational Approaches

Synthesizing these insights reveals that urban performance is closely tied to stage-specific characteristics in the evolution of polycentric urban structures. The changes in urban performance differ significantly from monocentric to networked polycentric structures. This framework supports the subsequent empirical analysis, which traces the co-evolution of the polycentric structure and urban performance, taking Shenzhen as a case study. As concepts have different definitions across contexts and scales, we provide explicit definitions for the concepts used in this study to ensure clarity. There are two main perspectives for defining urban centers: morphological and functional [33]. This study adopts the morphological perspective.

• Urban center: a spatial unit within the city whose employment density significantly exceeds that of its surrounding areas [34].
• Polycentric urban structure: an urban spatial form characterized by multiple spatially separated yet functionally linked centers, which collectively promote urban development.
• Clustered polycentric structure: an urban spatial form where multiple centers provide most daily needs locally, including employment, retail, and public services, thus reducing residents’ reliance on the main center.
• Networked polycentric structure: an urban spatial form where multiple centers are linked through functional complementarity, forming a cohesive network.
• Urban performance: the comprehensive performance of a city in economic, social, environmental, and other aspects, reflecting its resource utilization capacity, residents’ well-being, and sustainable development.
• Integrated performance: urban performance in spatial, economic, and social dimensions across center/cluster, inter-center, and citywide scales.

The concept of urban performance is widely applicable in urban studies, providing a general framework for evaluating urban systems across different contexts and scales. The concept of integrated performance is defined specifically in this study and, while it can be extended to other similar studies, it may require adjustment based on their objectives and contexts.

It is important to note that there is no universally accepted benchmark for defining the evolutionary stages of polycentric urban structures. Previous studies have employed different criteria based on the institutional context and the availability of data. For example, travel self-sufficiency thresholds (typically 66.7–75%) are commonly used to identify whether a cluster can function independently [35]. Commuting flow thresholds (e.g., a minimum of 15% of inter-center commuting flows in the OECD/EU functional urban area framework) are used to determine whether centers are functionally integrated into networked systems [36]. To trace the co-evolution of Shenzhen’s polycentric structure and urban performance from 1986 to 2020, this study draws upon existing research to develop an operational framework for the classification of evolutionary stages from the perspective of center functions, as detailed below.

• Cluster center: a center whose daily life service structure similarity with respect to the benchmark center in the same period is ≥0.75. The benchmark center is the center with the most developed and diversified daily life services.
• Initial polycentric structures: two or more centers exist, but none of them (except for the benchmark center) meet the cluster center criterion.
• Partially clustered polycentric structures: more than one, but not all, centers meet the cluster center criterion.
• Fully clustered polycentric structures: all centers meet the cluster center criterion.
• Intensity of networked polycentric structures: represented by the average potential collaboration intensity among centers, calculated based on their agglomeration scales, inter-center travel times, and disparities in productive industrial structures.

3. Study Area, Data and Methods

3.1. Study Area

Shenzhen is located in southern China, backed by mountains and facing the sea, with terrain sloping from higher elevations in the east to lower ones in the west. The mountains function as natural barriers, shaping the city’s early adoption of a clustered development pattern. Subsequent urban master plans have consistently promoted polycentric urban structures. Having largely developed in accordance with its successive master plans, Shenzhen is widely regarded as an exemplar of polycentric urban development in China. Since its establishment in 1978, Shenzhen has experienced one of the fastest urbanization processes globally, with its permanent resident population growing from less than 0.3 million to approximately 17.56 million by 2020. The city faces intense socio-spatial tensions: rapid population inflows have created mounting housing and public service pressures, while disparities between the east and west, as well as between the Special Economic Zone (SEZ) and peripheral districts, highlight uneven spatial development. This dual character makes Shenzhen a valuable case for tracing the co-evolution between polycentric structures and urban performance.

The study area covers Shenzhen, excluding the Shenshan Cooperation Zone, with a total area of 1997.47 km². It comprises nine administrative districts and one functional zone (Figure 3); among these, Futian, Nanshan, Luohu, and Yantian—collectively referred to as the SEZ—were prioritized during the city’s early development. Analyses were conducted annually, and four representative years—1986, 2000, 2010, and 2020—were selected for detailed analysis as they correspond to the beginnings and ends of successive master plans.

3.2. Data Sources and Processing

3.2.1. Enterprise Big Data

Enterprise data were obtained from Qichacha, one of China’s most widely used business information platforms. Qichacha compiles and standardizes historical enterprise records from the State Administration for Market Regulation (SAMR) and other authoritative public sources and continuously updates them in synchronization with the official registration system to ensure the completeness and timeliness of enterprise records. The database provides long-term historical coverage of enterprises in Shenzhen, with more than five million records from 1970 to the present. Each record includes the enterprise’s name, registration status, registered address, size, business scope, Chinese national industry classification, and data-driven industry classification according to Qichacha, which together provide a detailed and reliable foundation for longitudinal studies of Shenzhen’s urban development. Data preprocessing involved three steps. First, the raw data were cleaned and geocoded. Second, using the DeepSeek large language model, we developed a cross between the Chinese national industry classification and Qichacha’s classification, based on the enterprises’ business scopes. The national classification comprises four levels—category, major group, middle group, and subgroup. For enterprise records missing middle group and subgroup codes, we supplemented the national classification using Qichacha’s classification based on the mapping relationship. To ensure validity, we conducted a random audit of 1000 enterprises. The results showed that the error rate of AI-assisted supplementation was below 3%, indicating that any bias introduced by this procedure is limited. It should be noted that this procedure is not essential and was introduced to increase the granularity of classification, enabling similarity and disparity analyses at finer levels (i.e., using middle group rather than major group data), thus improving accuracy. Third, we assigned employment weights based on the enterprise size, following the Statistical Classification of Large, Medium, Small, and Micro Enterprises (2017). The validity of the dataset was verified by benchmarking against official statistics. Pearson correlations between the weighted employment structure and sectoral (category-level) employment statistics from the Shenzhen Statistical Yearbook (1989–2020) are presented in Figure 4 2. The correlation coefficients were consistently above 0.8, confirming the suitability of the Qichacha data for long-term longitudinal analysis of Shenzhen’s urban development.

3.2.2. Other Datasets

To capture the spatial, economic, and social performance throughout Shenzhen’s polycentric development, we combined datasets from authoritative sources which have been widely used in academic and governmental research. Road network data for 1986, 2000, 2010, and 2020 were obtained from the Shenzhen Latest Traffic Overview Map (1985 edition), the Shenzhen Urban Master Plans (1996–2010), and OpenStreetMap (OSM). Rail transit network data for 2010 and 2020 were sourced from the Shenzhen Metro official website and Amap. Gridded GDP data were obtained from datasets published by Chen et al. [37] and Liu et al. [38]. Residential population data were taken from the LandScan dataset. PM2.5 data were sourced from the National Tibetan Plateau Data Center (China) [39,40]. Data on primary schools, hospitals, parks, urban squares, and major cultural facilities were collected from district annals, Amap, and official websites. These datasets provide appropriate temporal coverage and spatial resolution, enabling assessment of the evolution of Shenzhen’s polycentric performance. Road and rail transit networks were integrated through spatial matching and topological analysis. Average speeds were assigned according to the transport mode and network class, facilitating the calculation of travel times.

3.3. Methods

The flowchart is shown in Figure 5, which includes the identification of urban centers, clustered and networked polycentric structures, as well as the performance evaluation.

3.3.1. Two-Stage Nonparametric Approach

A two-stage nonparametric approach was employed to identify urban centers [41]. In stage 1, Weighted Kernel Density Estimation (WKDE) was applied to enterprise point data to detect spatial agglomerations, using a 1000 m bandwidth (determined through experimental tuning). Candidate centers were identified by spatially matching the WKDE results to land parcels and merging adjacent high- and medium–high-density parcels. In stage 2, candidate centers were validated against agglomeration scale and functional diversity criteria to determine the final set of centers. This validation was necessary for two reasons: (1) the nonparametric method primarily captures gradients in employment density while neglecting absolute employment scale, which may lead to misclassification of small but dense units as urban centers. To ensure that urban centers meet the dual requirements of both agglomeration scale and intensity, an additional threshold of a minimum employment size of 20,000 was introduced; (2) to avoid misidentifying labor-intensive manufacturing plants or other mono-functional employment clusters as urban centers, a functional diversity threshold greater than 1 was applied.

3.3.2. Cosine Similarity

The cosine similarity measures the directional consistency between multi-dimensional vectors. It is employed in this study to evaluate the similarity of industrial structures between centers; in particular, a value closer to 1 indicates higher similarity, while a lower value indicates greater disparity. The formula is:

$$S_{a,b}={\displaystyle \frac{V_a·V_b}{‖V_a‖·‖V_b‖}}={\displaystyle \frac{\sum _{k=1}^n{p}_{a,k}·p_{b,k}}{\sqrt{\sum _{k=1}^n{p_{a,k}}^2}·\sqrt{\sum _{k=1}^n{p_{b,k}}^2}}}$$

where $S_{a,b}$ is the industrial structure similarity between centers a and b, $p_{a,k}$ is the share of employment in medium-level industry k within center a, and n is the number of medium-level industries.

3.3.3. Potential Collaboration Intensity

The potential collaboration intensity between centers was evaluated using a modified gravity model. This model assumes that larger agglomeration scales of centers, shorter inter-center travel times, and greater productive industrial structure disparities result in stronger collaboration intensity. The formula is:

$${Pci}_{a,b}={\displaystyle \frac{{(1-S_{a,b})}^n·P_a·P_b}{{d_{a,b}}^2}}$$

where ${Pci}_{a,b}$ is the potential collaboration intensity between centers a and b, $P_a$ is the employment scale of center a, and $d_{a,b}$ is the travel time between centers a and b. As larger agglomeration scales and shorter travel times can intensify both collaboration and competition, we set the exponent n = 2 to emphasize the role of productive industrial structure disparity.

3.3.4. Performance Evaluation System for Polycentric Urban Spatial Structures

To systematically assess polycentric urban structures, this study developed an integrated performance evaluation system guided by three design principles: (1) Comprehensiveness, which covers spatial, economic, and social dimensions across center/cluster, inter-center, and citywide scales. Transportation and environmental dimensions were incorporated into the social performance because of their roles in shaping residents’ quality of life and spatial equity. (2) Policy relevance, which ensures alignment with planning practice by emphasizing core indicators such as spatial efficiency, economic growth, and social equity. (3) Operability, which requires that an indicator system is both clear and measurable, ensuring its feasibility in practical application.

Performance Evaluation Indicators

An initial indicator pool was developed, followed by a two-round Delphi survey involving 36 experts, including 20 university scholars, 10 government planners, and 6 research institution consultants [42]. The experts were selected for their expertise in urban planning, architecture, and human geography, with most having direct research or practice in Shenzhen to ensure contextual relevance. The panel size (n = 36) aligns with Delphi practice, where samples of 20–40 experts are generally sufficient to achieve stable consensus while maintaining a diverse range of perspectives [43]. In round 1, the experts rated the importance of the indicators on a 1–5 scale and provided feedback on their definitions and wording. Based on the results, we retained high-scoring indicators and revised them according to the expert’s feedback. In round 2, the experts re-rated the revised indicators, ultimately leading to the selection of 25 indicators with high consensus, as presented in

Table 1. Performance evaluation indicators and weights for polycentric urban spatial structures.

Performance	Level	Indicator	Indicator Meaning	Subjective Weight	Objective Weight	Combined Weight
Spatial performance	Center/Cluster	Employment number	It reflects the level of agglomeration scale of the center.	0.0339	0.0409	0.0360
		Employment density	It reflects the level of agglomeration intensity of the center.	0.0361	0.0160	0.0300
	Inter-center	Passenger travel-time distance	It reflects the level of passenger transport connectivity between centers.	0.0318	0.0055	0.0239
		Freight travel-time distance	It reflects the level of freight transport connectivity between centers.	0.0289	0.0155	0.0249
		Scale–distance coordination	It reflects the degree of coordination between center scale and inter-center distance.	0.0396	0.0727	0.0495
	Citywide	Proportion of residents within clusters	It reflects the proportion of residents covered by urban clusters.	0.1260	— 1	0.1260
		Proportion of residents within rational coordination clusters	It reflects the proportion of residents covered by urban rational coordination clusters.	0.0767	—	0.0767
Economic performance	Center/Cluster	Number of industries with LQ over 1.5	It reflects the degree of industrial specialization of the center.	0.0309	0.0101	0.0247
		Economic output (at constant prices)	It reflects the level of economic development of the center.	0.0413	0.0468	0.0430
		Patents per thousand enterprises	It reflects the level of innovation capacity of the center.	0.0309	0.0658	0.0414
	Inter-center	Potential collaboration intensity	It reflects the level of collaboration intensity between centers.	0.0719	0.0441	0.0636
	Citywide	Real GDP per worker	It reflects the level of economic development of the city.	0.0483	—	0.0483
		GDP growth rate	It reflects the economic development trend of the city.	0.0387	—	0.0387
		Industrial structure index	It reflects the industrial structure development trend of the city.	0.0435	—	0.0435
		Average commuting distance	It reflects the level of employment accessibility within the cluster.	0.0295	0.0272	0.0288
Social performance	Center/Cluster	Primary school supply-demand ratio	It reflects the supply-demand relationship of educational resources within the cluster.	0.0235	0.0572	0.0336
		Hospital supply-demand ratio	It reflects the supply-demand relationship of healthcare resources within the cluster.	0.0266	0.0379	0.0300
		Parks and urban squares supply-demand ratio	It reflects the supply-demand relationship of public space within the cluster.	0.0280	0.0199	0.0256
		PM2.5 concentration	It reflects the level of air quality within the cluster.	0.0267	0.0199	0.0247
	Inter-center	Accessibility of major public cultural facilities	It reflects the degree of sharing of major public cultural facilities between centers.	0.0210	0.0214	0.0211
	Citywide	Standard deviation of commuting distance	It reflects the degree of equity in commuting distance across clusters.	0.0388	—	0.0388
		Standard deviation of primary school supply-demand ratio	It reflects the degree of equity in the supply-demand relationship of educational resources across clusters.	0.0364	—	0.0364
		Standard deviation of hospital supply-demand ratio	It reflects the degree of equity in the supply-demand relationship of healthcare resources across clusters.	0.0357	—	0.0357
		Standard deviation of parks and urban squares supply-demand ratio	It reflects the degree of equity in the supply-demand relationship of public space across clusters.	0.0285	—	0.0285
		Standard deviation of PM2.5 concentration	It reflects the degree of equity in air quality across clusters.	0.0268	—	0.0268

¹ “—” indicates that the objective weight cannot be calculated because there is only one observation per year.

.

Weight Determination: A Hybrid Subjective–Objective Approach

A hybrid subjective–objective approach was adopted to determine the indicator weights. Subjective weights were derived using the Analytic Hierarchy Process (AHP) based on pairwise comparisons conducted by the 36 experts [44]. All judgment matrices passed consistency tests. The panel size (n = 36) was considered adequate for the AHP, with prior studies suggesting that panels exceeding 20 participants generally yield stable, reliable weight distributions [45]. Objective weights were calculated using the entropy weighting method, which evaluates variability across observations to determine the importance of indicators [46]. Recognizing the pivotal role of expert knowledge, we integrated subjective and objective components in a 70:30 ratio using a hierarchical method to obtain the final weights. The resulting weights are reported in Table 1. As a result, spatial, economic, and social dimensions were weighted approximately 0.37, 0.30, and 0.33, while center/cluster, inter-center, and citywide scales were weighted approximately 0.32, 0.18, and 0.50, respectively.

4. Results

4.1. Evolution of Shenzhen’s Polycentric Urban Spatial Structure

From 1986 to 2020, Shenzhen consistently maintained polycentric urban structures. The number of urban centers remained at four between 1986 and 2000 but had increased to ten by 2010 and twelve by 2020 (Figure 6). In 1986, the four centers were Futian–Luohu, Shekou, Shatoujiao, and Nantou, with the first three formed around the ports. By 2000, the Futian–Luohu and Shatoujiao centers remained largely stable, the Nanshan center was formed by merging the Shekou and Nantou centers, and the Chegongmiao center was newly developed. In 2010, the new Futian–Luohu center was formed by merging the former Futian–Luohu and Chegongmiao centers. The Nanshan center expanded eastward, the Shatoujiao center remained largely stable, and the Bao’an, Songgang, Buji, Longhua, Shajing, Fuyong, and Longgang centers were newly developed. In 2020, the Bao’an and Longhua centers expanded significantly, the Guanlan and Pinghu centers were newly developed, and the other centers remained largely stable. The evolution of Shenzhen’s polycentric urban structure is highly consistent with its successive master plans, demonstrating the guiding role of institutional frameworks and urban planning in shaping its urban spatial structure.

4.1.1. Evolution of the Clustered Polycentric Structure

The similarities in daily life service structures among centers in 1986, 2000, 2010, and 2020 are depicted in Figure 7. Over the study period, Shenzhen progressed through four stages: the initial polycentric structure, the fully clustered polycentric structure, the partially clustered polycentric structure and, finally, a return to the fully clustered polycentric structure. In 1986, although Shenzhen had already developed a polycentric urban structure, only the Futian–Luohu center met the cluster center criteria, with the similarity values of the other three centers to the Futian–Luohu center all below 0.60. By 2000, the centers had matured and all four urban centers met the cluster center criteria; as such, Shenzhen had evolved into a fully clustered polycentric structure. This evolutionary process was consistent with the objectives of the 1986 master plan, which proposed adopting an east–west linear clustered development pattern within the SEZ. As the available land for development within the SEZ was gradually exhausted, the 2000 master plan encouraged urban development outside the SEZ. From 2000 to 2010, six new urban centers were formed outside the SEZ. Due to their relatively late development, by 2010, the similarity values of the Bao’an, Songgang, Shajing, and Fuyong centers with respect to the Futian–Luohu center were all below 0.75, meaning they did not qualify as cluster centers, thus resulting in a partially clustered polycentric structure. As these centers matured, Shenzhen once again evolved into a fully clustered polycentric structure by 2020.

This study employed the gravity model to delineate the hinterlands of urban centers (Figure 6). For centers meeting the cluster center criteria, each center and its corresponding hinterland were identified as a cluster. From 2000 to 2020, the proportion of residents living within clusters increased continuously. In 2000, only 43.12% of residents lived within clusters, while more than half resided in service blind spots. In 2010, the proportion rose slightly to 46.88%. Although the number of centers increased by six from 2000, many new centers had not yet developed into cluster centers, and the proportion rose by only 3.76%. By 2020, as all centers had evolved into cluster centers, the proportion increased sharply to 61.19%. Unlike the coverage of clusters, the proportion of residents living within rationally coordinated clusters first increased and then declined, from 24.09% in 2000 to 36.13% in 2010, then to 30.22% in 2020. The decline between 2010 and 2020 was due to the failure of some urban centers to maintain rational scale–distance coordination with other centers during their development, which exacerbated spatial competition.

4.1.2. Evolution of the Networked Polycentric Structure

From 1986 to 2020, the networked characteristics of Shenzhen’s polycentric urban structure strengthened steadily. In 1986, although there were significant disparities in the productive industrial structures among centers, the small agglomeration scales of most centers—except for the Futian–Luohu center—constrained the overall collaboration intensity, leading to an average potential collaboration intensity of 2.37. By 2000, the disparities among centers narrowed while the agglomeration scales grew significantly, and the average intensity increased to 3.21. By 2010, the average intensity had risen slightly to 3.25. The reasons for this increase are complex: existing centers expanded in scale but the disparities among them decreased, while newly developed centers were small in scale but displayed significant disparities with existing centers. Together, these factors led to a slight rise in collaboration intensity. From 2010 to 2020, networked development mirrored that in the 1986–2000 period: structural disparities narrowed while scales expanded, resulting in an increase in average intensity to 3.93. The potential collaboration intensities among centers in 1986, 2000, 2010, and 2020 are presented in Figure 8. High- and medium–high-level collaboration primarily occurred between pairs of spatially proximate and/or large-scale centers, such as between the Nantou and Shekou centers in 1986, the Futian–Luohu and Chegongmiao centers in 2000, and the Nanshan and Bao’an centers in 2010 and 2020. The impacts of disparities in productive industrial structures were limited, primarily because the functional differences among centers were small. Although some planning documents have proposed the differentiated development of urban centers, this policy was not rigorously implemented in practice. Consequently, many new centers replicated the successful development paths of mature centers, resulting in a degree of functional convergence.

4.1.3. Summary

The evolution of Shenzhen’s polycentric urban structure followed a nonlinear trajectory in its clustered pattern and a linear trajectory in its networked characteristics. The clustered pattern progressed through four stages—namely, the initial polycentric structure, the fully clustered polycentric structure, the partially clustered polycentric structure and, once again, the fully clustered polycentric structure—influenced by planning and center development stages. Meanwhile the networked characteristics continually strengthened, but this was driven mainly by increases in center scale and geographic proximity rather than functional complementarity.

4.2. Performance Evolution of the Polycentric Urban Spatial Structure

Due to the unavailability of certain data prior to 2000, the analysis of urban performance evolution focused on the period from 2000 to 2020 (Figure 9). Overall, Shenzhen’s integrated performance followed a sustained upward trajectory, increasing from 0.65 in 2000 to 0.79 in 2010, and further to 0.86 in 2020. At the subsystem level, both spatial and economic performance improved, but with different temporal patterns: spatial performance increased steadily, while economic performance grew slowly from 2000 to 2010, followed by fast growth from 2010 to 2020. Social performance presented an inverted U-shaped trajectory. These findings highlight the multi-faceted nature of performance dynamics, which are analyzed in detail in the following sections.

4.2.1. Spatial Performance

Spatial performance improved from 2000 to 2020, mainly influenced by the growth in the number of cluster centers and the development of the clustered polycentric structure—an evolution that closely tracked successive master plans. The scores for the spatial performance indicators are presented in Figure 10. At the center level, Shenzhen was at an early stage in 1986, and most centers—except Futian–Luohu—were small in both scale and intensity. By 2000, driven by agglomeration economies, both indicators increased markedly. In 2010, they declined slightly with the emergence of new centers outside the SEZ but then rebounded in 2020 as those centers matured. At the inter-center level, passenger and freight travel times were longer in 2010 and 2020 than in 1986 and 2000, reflecting longer trips associated with the development of new centers outside the SEZ. Between 2010 and 2020, passenger transport scores increased more than freight scores, largely due to expansion of the rail-transit network, which improved inter-center accessibility. Scale–distance coordination was highest in 1986, followed by 2010, 2020, and 2000. In 1986, apart from Futian–Luohu, other centers were small in scale and spatial competition was minimal. By 2000, coordination between the Futian–Luohu and the Chegongmiao centers, and that between the Chegongmiao and Nanshan centers, became sub-optimal, with competition arising among large centers in close proximity. In 2010, the emergence of peripheral centers slightly alleviated overall inter-center competition. By 2020, however, inter-center coordination deteriorated again, due to some centers failing to align their scales and spacing. At the city level, the proportion of residents within clusters increased steadily. However, the proportion of residents in rationally coordinated clusters first rose and then fell. The 2010–2020 decline was attributed to the same factors that drove the decrease in scale–distance coordination.

4.2.2. Economic Performance

From 1986 to 2020, most economic performance indicators showed steady upward or downward trends, except for industrial specialization at the center level (Figure 11). Indicators such as real GDP per worker and innovation capacity increased consistently, whereas the GDP growth rate was the only indicator that decreased over time. Industrial specialization is primarily associated with the development stage of centers, often beginning with daily life services and progressively shifting toward productive functions as centers mature. These trends were largely shaped by China’s reform and opening up and by Shenzhen’s proactive integration into global markets. In addition, a series of local policies—such as the establishment of the Shenzhen Special Economic Zone in 1980, the approval and launch of the Shenzhen High-Tech Industrial Park in 1996, the launch of the Qianhai–Shekou area as part of the Guangdong Pilot Free Trade Zone in 2015, and the continued implementation of an innovation-driven development strategy—provided institutional and financial support for industrial upgrading and technological innovation. Together with the evolving polycentric urban structure, these policy measures facilitated the redistribution of economic activities across centers and supported the sustained growth of Shenzhen’s economic performance. To test the role of networked polycentricity, we examined the relationships between the average potential collaboration intensity and (1) GDP per worker (1986–2020) and (2) the economic output of centers (1992–2020) at constant prices. The results (Figure 12) reveal strong positive correlations (R² = 0.89 and 0.92), indicating that greater potential collaboration intensity is associated with higher values of major economic indicators. These results support the view that networked polycentric structures can enhance economic outcomes by fostering inter-center industrial collaboration.

4.2.3. Social Performance

From 2000 to 2020, social performance followed an inverted U-shaped trajectory, peaking in 2010. This result is unexpected: despite Shenzhen’s considerable efforts to improve social performance, the partially clustered polycentric structure in 2010 outperformed the fully clustered polycentric structure in 2020. The scores for the social performance indicators are presented in Figure 13. In 2010, the newly developed centers contributed to an overall improvement in residents’ quality of life. The new centers were small in scale, with low enterprise density and low air pollution, and formed compact clusters with shorter commuting distances. The smaller populations in these clusters increased per capita access to public services. At that time, social equity was poor due to disparities between new and established clusters. By 2020, as these centers matured, enterprise density increased and air quality declined. The clusters expanded spatially, and commuting distances lengthened. During the period 2010–2020, these clusters consistently attracted population inflows due to their comparative advantages; however, the provision of public services lagged, worsening supply–demand relationships for education, healthcare, and public amenities. For example, from 2010 to 2020, the Longhua cluster’s population grew by 2.16 million, while its primary schools increased by only 43, reducing the number of primary schools per 10,000 residents from 0.58 to 0.21. This indicates that the decline in social performance after 2010 was not merely the result of polycentric structure evolution but rather reflected underlying governance challenges. The mismatch between demographic growth and infrastructure expansion limits the ability of polycentric structures to deliver sustained improvements in residents’ quality of life [17,47]. Although spatial equity among clusters in 2020 was improved relative to 2010, this was achieved at the cost of lower quality of life. These dynamics suggest that achieving social performance under rapid urbanization requires not only clustered polycentric structures, but also stronger institutional frameworks and policy coordination to balance demographic growth with the provision of services.

5. Discussion

5.1. Discussion on the Co-Evolution of Polycentric Structures and Urban Performance

Shenzhen provides a representative case for tracing the co-evolution between polycentric structures and urban performance. Its development trajectory has largely followed a sequential path, shaped by successive master plans that guided its expansion through clear spatial blueprints. Moreover, Shenzhen has faced intense urban tensions resulting from rapid population inflows, uneven spatial development, and pressures on housing and public services. This dual character makes Shenzhen a valuable case: it not only demonstrates how polycentric cities evolve under strong institutional frameworks and planning contexts but also reveals the trade-offs and tensions inherent in the process. Some findings are consistent with previous research, while others may help to explain the conflicting findings reported in prior studies.

First, our results indicate that increasing the number of centers within clustered polycentric structures helps to expand service coverage, aligning with the view that polycentric development enhances spatial efficiency [48]. However, when scale–distance coordination among centers is inappropriate, intensified competition may offset these benefits, lending support to critiques that polycentric structures without rational spatial coordination often fail to achieve the promised improvements [48]. Existing research and discussions on the scale–distance coordination between centers are primarily based on theoretical assumptions or empirical standards, rather than systematic validation, partly because this relationship is influenced by a wide range of factors such as city size, population density, and transport conditions. Consequently, there is no fixed or universal optimal value. In the planning practices of major cities in China, the optimal distance between main centers is typically 15–18 km, while that between secondary centers is about 10–12 km [48]. In this context, the main centers and the secondary centers in the western part of Shenzhen are relatively densely clustered, while the secondary centers in the eastern part are loosely distributed. This uneven polycentric spatial distribution constrains urban spatial performance.

Second, the Shenzhen case provides evidence for strong positive correlations between networked polycentric structures and major economic indicators, suggesting that the borrowed size effect—previously documented primarily at the inter-city scale [49,50]—also operates at the intra-city scale. The prior debate about the association between polycentricity indices and economic outcomes may result from the failure of such indices to capture the networked characteristics of polycentric urban structures [10,11,14,51]. Moreover, the development of Shenzhen’s urban centers has obvious path dependence—a phenomenon also observed in other cities. This has resulted in relatively small disparities in industrial structure among centers, which hinders the networked development of polycentric structures.

Third, our findings indicate that although clustered polycentric structures have the potential to enhance social performance, their benefits are conditional—not only on the structure itself, but on the coordinated development of public service facilities [52]. This finding lends some support to the view that polycentric structures are not strongly associated with social performance [6,7]. While polycentric development is often expected to enhance residents’ quality of life and equity through spatial redistribution, peripheral or newly developed clusters frequently receive insufficient public service investment in practice, resulting in a deterioration of social performance. This decline reflects not only a lack of spatial or functional coordination, but also deeper constraints rooted in the political–institutional framework and national policy priorities that, amid rapid urbanization, prioritize economic growth and spatial expansion at the expense of social inclusion and equity [17,47]. As a result, in rapidly growing clusters, residents often face service shortages and reduced quality of life, highlighting the limitations of polycentric frameworks amid accelerated urbanization.

Although polycentric urban structures provide context-dependent benefits, it is important to recognize the inherent tensions among spatial, economic, and social performance. Spatial performance improvement is rooted in clustered polycentric structures, while economic performance growth is associated with the strength of networked polycentric structures. Meanwhile, social performance depends not only on clustered polycentric structures but also on the adequacy of supporting public service facilities. These dimensions impose different requirements on polycentric urban structures: spatial and social performance demand the development of clustered polycentric structures and the rational scale-distance coordination between centers, whereas economic performance requires larger agglomeration scales, shorter inter-center distances, and stronger industrial complementarity, all within a governance system that ensures the adequate provision of public services. A rational polycentric structure requires striking a balance among these three dimensions. A ‘rational’ urban spatial structure should be understood as a dynamic framework, rather than a fixed blueprint, as its effectiveness lies not in choosing between cluster or network patterns but in strategically integrating the two. The case of Shenzhen demonstrates that a rational polycentric structure can be achieved by enhancing the clustered pattern through optimizing scale–distance coordination between centers, while simultaneously promoting networked characteristics via functional specialization and industrial complementarity. Furthermore, the long-term sustainability and equity of polycentric development depend on collaborative governance and active citizen participation, which are essential to the widespread sharing of polycentricity’s benefits.

Shenzhen’s experience demonstrates that a rational polycentric structure can indeed bring improvements in spatial and economic performance; however, these gains do not automatically translate into higher quality of life for residents or greater social equity. This suggests that although the polycentric framework is important, its outcomes still depend on institutional frameworks, governance capacity, and policy coordination. Moreover, as China’s first Special Economic Zone, Shenzhen developed its polycentric structure within a unique institutional context. This structure did not arise from organic growth but was instead shaped by a series of strategic master plans with clear spatial and functional objectives. These plans enabled the city to follow an approximating sequential trajectory, which should not be regarded as universally applicable. Polycentric development in other contexts may be hybrid and nonlinear (

Table 2. Hybrid and nonlinear pathways of polycentric development: international case comparison.

City/Region	Evolutionary Pathway	Nonlinear/Hybrid Features	Influencing Factors
Kumasi City-Region (Ghana)	Shift from monocentric to a deconcentrated, dispersive pattern.	- Monocentricity is declining but does not transform into a polycentric structure. - Urban growth is becoming amorphous. - Stochastic fractal urban growth process suggests potential coalescence.	- Rapid urbanization - Spontaneous urban development - Inefficient and unsustainable spatial development - Existing urban geography models do not fully explain the city’s evolution.
Beijing (China)	From monocentric to polycentric with dispersed sub-centers	- Emergence of multiple sub-centers. - Hybrid development of central and peripheral zones	- Technological advancements - Economic development - Government policies on transportation and land use.
Mexico City (Mexico)	Multiple sub-centers without a clear transition to full polycentricity.	- Polycentric features emerge but with incomplete and uneven development.	- Economic factors - Migration patterns - Urban infrastructure development
Houston (USA)	Diffusion and coalescence in metropolitan expansion	- Nonlinear growth with fractal patterns. - Growth is neither strictly monocentric nor fully polycentric.	- Economic factors - Transport and infrastructure development - Suburbanization.

) [22,53,54,55]. Shenzhen’s experience is conditionally transferable: it is most applicable to cities with strong institutional frameworks, robust governance capacity, and a planning-led, sequential development path. While insights gained from Shenzhen may be informative for other types of cities, anticipated spatial, economic, and social outcomes should be contextualized to the urban setting to ensure their applicability across diverse global contexts.

5.2. Methodological Reflections and Global Applicability

The gravity model is an effective and widely used method for delineating the hinterlands of urban centers. Relying only on factors such as central mass, distance, and a gravitational constant—without requiring extensive detailed data—the model can produce results that closely match the observed conditions. In this study, to transform hinterlands into clusters, it is essential to establish a reasonable criterion for defining cluster centers. Conceptually, a cluster center should locally satisfy the majority of residents’ daily needs, including employment, retail, and public services. Therefore, we defined cluster centers from a functional perspective: a secondary center is considered a cluster center when its daily life services meet or exceed 75% of those of the main center. This threshold was chosen because it corresponds to the cluster’s travel self-sufficiency rate (typically 66.7–75%), representing the probability of meeting daily needs within the cluster. We assumed that when a center provides more than 75% of the required functions, a proportional share of trips will occur within the corresponding cluster. Given the long-term nature of polycentric urban structure evolution and data availability, we examined the relationships between the cluster center similarity value to the main center and the trip share within the cluster for 2017 and 2024 (using mobile signaling data from China Unicom), to validate the appropriateness of this threshold. The R² values of 0.71 and 0.67 indicate a moderate degree of correlation between cluster center functions and cluster’s travel self-sufficiency rates, which supports the plausibility of the 75% cutoff (Figure 14a). The travel self-sufficiency rates of the Bao’an and Buji clusters significantly deviate from the cluster center similarity values, lowering the R² values. This is mainly due to the high overlap of these clusters with nearby clusters, such as the Nanshan and Futian clusters. This suggests that the method may overestimate the travel self-sufficiency rate of clusters of small-scale centers surrounding clusters of large-scale centers. In 2017, the minimum intra-cluster travel self-sufficiency rate was 68.13%, and in 2024, the minimum rate was 70.64%, meeting the criteria for clusters. Evidence from Shenzhen indicates that when a secondary center’s daily life services reach 75% of the main center’s level, it can fulfill the role of a cluster center, enabling residents to meet their daily needs locally. However, this threshold may require adjustment depending on the context when applied across diverse global settings.

Unlike studies of networked polycentricity at the urban-agglomeration scale, city-level research requires more granular data, such as inter-firm investment flows and communication (e-mail) data. However, such data are often difficult to obtain, especially for multi-decade longitudinal studies. Similarly, we estimated the potential collaboration intensity between centers from a functional perspective, using a modified gravity model. We hypothesize that when there is industrial complementarity between centers, their potential for collaboration is greater. To validate its effectiveness, we performed a linear regression analysis between the potential collaboration intensity and the non-commuting trips between centers, with R² values of 0.94 and 0.95 (Figure 14b), supporting the construct validity of our approach. Although networked characteristics strengthened over time, these changes were primarily driven by the agglomeration scale and geographic proximity of its centers, rather than by industrial complementarity. To improve cross-context transferability, the study attempts to propose benchmarks to define the networked polycentric stage based on the theoretical framework and the results of this research. First, functional differentiation should be clearly evident: the structural disparity between centers should exceed 0.4, ensuring that centers perform complementary rather than redundant roles. Second, spatial distribution should be balanced within practical interaction ranges: the average distance between main centers should be approximately 15 km, while that between secondary centers should be around 10 km. Third, the efficiency of transport and communication linkages should guarantee timely interactions, such that the maximum inter-center travel time remains below 60 min under normal conditions. Taken together, these thresholds indicate that a polycentric structure may be regarded as ‘networked’ when its centers are functionally differentiated, spatially balanced, and tightly integrated through efficient connectivity. Furthermore, the existence of cross-level linkages among centers should be acknowledged; this aspect was not addressed in our study. In cross-level contexts, this research approach remains applicable: centers should first be identified at each level, after which potential collaboration intensity can be computed. When data conditions allow, inter-center investment and financing data are superior indicators for characterizing networked polycentric structures, as they more accurately capture the economic linkages generated by firms’ productive activities. When the intensity of investment and financing between centers—whether at the same or different levels—is significantly higher than that observed in other regions, it indicates that a networked polycentric structure has emerged. Within cross-level networked structures, when a lower-level center has a functional structure that differs substantially from that of a higher-level center and therefore occupies an important position within the overall networked polycentric structure, it may expand and upgrade to a higher-level center. Conversely, when a lower-level center is functionally similar to higher-level centers, competition persists. When the lower-level center is located far from functionally similar higher-level centers, it may persist because it assumes a functional role within the overall networked polycentric structure; if it is located close to functionally similar higher-level centers, it may be absorbed or naturally disappear. According to the borrowed size effect, cross-level networked characteristics should likewise enhance urban economic performance, although their effects are typically weaker than those of networked characteristics among higher-level centers.

Using Shenzhen—a representative polycentric city—as a basis, we developed a multi-dimensional, multi-scale performance evaluation system with substantial expert input, providing a useful reference that may be transferred to other contexts. However, applying this evaluation system elsewhere may require adjusting it to each city’s development stage and specific objectives. Expert panels should include individuals specialized in urban planning, architecture, and human geography, most of whom have research or practical experience in the local context. If experts are scarce, the panel can be smaller and/or participation can be facilitated online.

Data substitution is also critical for the method’s transferability. In data-scarce regions, business census data and local surveys can serve as viable alternatives. Additionally, artificial intelligence (AI) techniques can further enhance the method’s adaptability in a global context. Researchers can train models in data-rich environments to learn mappings between business scopes and industry classifications, then apply them in regions lacking standardized classifications to infer enterprise attributes. It should be noted that contextual differences may introduce prediction bias, and safeguards are needed to prevent poorly designed or applied AI tools from contaminating datasets, which could compromise the validity of the conclusions of such research.

6. Conclusions

The polycentric urban structure is not a goal in and of itself but is promoted as a bridging concept to enhance economic efficiency, improve spatial equity, and support environmental sustainability. However, the roles of such structures in shaping urban performance remain insufficiently understood, particularly regarding the contributions of clustered and networked patterns to performance across different dimensions and scales. To address these gaps, this study traced the co-evolution between Shenzhen’s polycentric structure and its multi-dimensional, multi-scale performance. The results revealed the following: (1) The evolution of Shenzhen’s polycentric structure followed a nonlinear trajectory in its clustered pattern and a linear trajectory in its networked characteristics. The clustered pattern progressed through four stages—the initial polycentric structure, the fully clustered polycentric structure, the partially clustered polycentric structure, and, once again, the fully clustered polycentric structure—while the networked characteristics continuously strengthened. (2) Throughout the study period, the integrated performance—as well as spatial and economic performance individually—improved, whereas social performance followed an inverted U-shaped trajectory. Performance dynamics are closely linked to polycentric structure evolution. Improvements in spatial performance are the result of the combined effects of multiple indicators, with an increase in the number of cluster centers and the development of the clustered polycentric structure playing pivotal roles. Economic performance, aside from the historical context, improved largely due to the strengthening of the networked polycentric structure. Although social performance benefited from the clustered polycentric structure, these gains were conditional, being realized only when supported by appropriately matched public service facilities.

6.1. Planning Implications

According to the research results, three suggestions are proposed to promote the coordinated development of clustered and networked polycentric structures. First, the coordination between center scales and inter-center distances should be optimized to prevent spatial inefficiency. When large centers are in close proximity, overlapping service catchments intensify competition. In such cases, further expansion of centers should be capped, and industry relocation should be promoted selectively based on their development goals, to alleviate spatial competition. Conversely, when small centers are widely dispersed, service coverage gaps may emerge. In areas with large surrounding populations, planning authorities should use policy instruments to strategically expand the capacities of centers. Tailoring interventions to these scenarios can help to reduce competition, close service gaps, and enhance overall spatial performance. Second, networked polycentric structures should be strengthened to improve economic performance. Inter-center linkages in polycentric cities can be fostered through productive specialization and industrial complementarity, thereby leveraging the borrowed size effect such that the economic efficiency of polycentric structures approach or match that of monocentric structures. To this end, the plan should clearly define the functions of each center and promote cross-center collaboration through industrial complementarity, thereby fostering networked polycentric development. Third, unlike spatial and economic performance, the social benefits of polycentric structures depend not only on the structure itself but also on the appropriate provision of public service facilities. Planning authorities should concurrently develop urban spatial structure and public service facility plans and dynamically allocate public service facilities based on the specific conditions of each cluster. This approach is not only expected to ensure the effective utilization of public service facilities, but also to enhance residents’ well-being and spatial equity.

6.2. Contributions, Limitations, and Future Research

Compared to previous studies, this study makes two significant contributions. First, it proposes an analytical framework based on evolutionary stages that clarifies the different roles of clustered and networked patterns of polycentric structures in shaping urban performance. Second, it develops a multi-dimensional, multi-scale performance evaluation system for polycentric cities, with indicator weights that are grounded in both expert judgment and data variability. Nonetheless, this study has several limitations. First, the analysis was based on a single case—Shenzhen—which limits the generalizability of the findings. Although Shenzhen is a valuable case for studying the co-evolution between polycentric structures and urban performance, future research should empirically test whether these insights apply to cities with different institutional environments, cultural contexts, and development trajectories. Second, due to data limitations, informal economic sectors were not incorporated into this study. These sectors likewise play an important role in shaping urban economic dynamics, especially in developing regions, and their exclusion may lead to biases in the identification of clustered and networked patterns. Additionally, the reliance on a private business information platform as the primary data source may restrict the generalizability of the findings, as such platforms often vary in coverage, data quality, and accessibility across different countries and regions. This reliance may therefore limit the cross-context transferability of the proposed benchmarks. Future research should conduct international or multi-city comparisons to reveal the diversity of polycentric development pathways across different institutional and cultural contexts, validate the transferability of the cluster center threshold, and develop benchmarks for networked polycentric structures. Additionally, econometric methods should be employed to test the causal relationships between clustered and networked polycentric structures and urban performance, thereby examining the robustness and transferability of the research findings.