Industrial park industrial ecological map system and method based on industrial chain knowledge graph

Abstract

The application discloses a park industry ecological map system and method based on an industry chain knowledge graph, and relates to the technical field of industry data analysis and knowledge graphs.The system comprises a data acquisition module, a four-dimensional ontology construction module, a niche calculation engine, a dynamic evolution deduction module and a visualization module;by constructing a four-dimensional fusion graph ontology of an industry chain-innovation chain-fund chain-space chain, a multi-dimensional improved niche model is used to quantize the ecological niche and adaptation degree of an enterprise, and an improved ecological gravity model based on the fusion niche characteristics is used to realize ecological evolution deduction and link prediction.The application can accurately depict the ecological structure of a park industry, automatically identify chain supplement and strong chain target points, predict ecological evolution trends, and provide intelligent support for fine management of a park industry.

Description

Technical Field

[0001] This invention relates to the fields of industrial data analysis, knowledge graph technology application, and smart park management technology, and in particular to a park industrial ecosystem graph system and method based on industrial chain knowledge graph. Background Technology

[0002] As industrial parks shift from attracting investment to operating industrial ecosystems, using knowledge graph technology to characterize industrial structures and assist in park governance has become the mainstream direction in the industry. However, existing technologies have significant limitations in terms of ecosystem characterization, quantitative assessment, and dynamic projection.

[0003] As disclosed in CN114491068A, a method and system for constructing an industrial park knowledge graph that integrates multi-source heterogeneous data is proposed. This solution constructs an industrial park knowledge graph by integrating multi-source data such as industry and commerce and policy, and realizes the extraction of enterprise relationship and basic industry analysis. However, it has the following technical defects: (1) The graph ontology is only constructed around three basic entities: enterprises, industries and policies. It does not establish a systematic ecological framework of industrial chain-innovation chain-capital chain-space chain, and cannot fully represent the multi-dimensional symbiotic structure of industrial ecosystem. Its essence is still the association aggregation of enterprise information, rather than the digital mapping of industrial ecosystem; (2) It lacks a dedicated quantitative evaluation model for industrial ecosystem. It can only realize basic association query and statistical analysis. It cannot accurately identify weak links in the industrial chain, nor can it quantitatively evaluate the degree of fit between enterprises and park ecosystem. Investment matching relies on manual rules; (3) It does not have dynamic evolution and deduction capabilities. It only supports static graph display. It cannot predict the development trend of industrial ecosystem, nor can it simulate the impact of different policies on the ecosystem. It is difficult to support forward-looking governance decisions. The existing technology does not provide technical inspiration for the construction of multi-chain fusion ontology, nor does it mention the relevant technical means of niche quantification and ecological gravity inference. For example, the document "Design and Implementation of Industrial Chain Big Data Platform Based on Knowledge Graph" published in "Information Technology and Informatization" designed an industrial chain big data platform for industrial parks, which realizes the visualization of industrial chain and investment promotion assistance functions based on knowledge graph. However, it has the following technical defects: (1) The core of the platform focuses on the supply and demand relationship dimension of the industrial chain, and does not include the core elements of industrial ecosystem such as innovation resources, financial elements, and spatial carriers into the graph system. The depiction of industrial ecosystem is obviously one-sided and cannot reveal the coupling effect between ecological elements; (2) The investment matching evaluation relies on keyword matching and manual rule filtering. It lacks a systematic quantitative evaluation model. The matching accuracy is highly dependent on the rules set by humans. The adaptability and objectivity are insufficient and it is difficult to support accurate investment promotion; (3) It does not involve the dynamic evolution analysis of industrial ecosystem. It can only present the current static industrial structure and cannot reveal the evolution law and potential risks of industrial ecosystem from the time dimension. It cannot provide effective support for the medium and long-term industrial planning of the park. This existing technology also fails to provide technical insights into combining ecological theory with knowledge graphs for the quantification and extrapolation of industrial ecology.

[0004] In summary, the common technical bottlenecks of existing technologies are: (1) Single dimension and insufficient systemicity: Most of them construct graphs with the industrial chain or enterprise as the core, without realizing the ecological-level representation of multi-chain integration, and cannot fully reflect the complex system structure of the park's industrial ecosystem; (2) Lack of quantification and insufficient objectivity: There is a lack of exclusive quantitative models for the park's industrial ecosystem, and core businesses such as ecological assessment and investment matching rely heavily on human experience, with strong subjectivity and insufficient accuracy; (3) Static limitations and insufficient foresight: All are static graph construction and query, without establishing a dynamic evolution and deduction mechanism, and cannot predict the ecological development trend and policy effects; (4) Poor scenario adaptability: Most are simple applications of general knowledge graph technology in park scenarios, without exclusive algorithm design for the core needs of park industrial governance, and the technical solution lacks obviousness. At present, no existing technology provides technical inspiration for combining the four-dimensional fusion ontology, the improved ecological niche quantitative model, and the improved ecological gravity deduction model to construct the park's industrial ecosystem graph. Summary of the Invention

[0005] In view of the above-mentioned defects of the existing technology, the technical problem to be solved by the present invention is: (1) to break through the limitations of the existing single-dimensional and fragmented map, to construct a multi-chain deep integration park industrial ecosystem exclusive ontology framework, and to realize the complete digital representation of the industrial ecosystem; (2) to propose an original industrial niche quantitative model adapted to the park scenario, to realize the automated and quantitative calculation of enterprise ecological positioning, adaptability assessment and chain supplement target identification, and to replace the traditional manual experience assessment; (3) to construct a dynamic evolution and deduction algorithm that integrates niche theory, to realize the technological leap from static query to dynamic prediction, and to support policy simulation and risk warning.

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0007] The present invention provides a park industrial ecosystem graph system based on industrial chain knowledge graph, including a data acquisition module, a four-dimensional ontology construction module, an ecological niche calculation engine, a dynamic evolution inference module, and a visualization interaction module.

[0008] The data acquisition module is used to collect data on the industrial chain, innovation chain, capital chain, and spatial chain within the park, and to perform data cleaning, attribute standardization, and cross-data source entity alignment preprocessing.

[0009] The four-dimensional ontology construction module is used to construct a four-dimensional integrated ecological graph ontology system that includes industrial chain dimension, innovation chain dimension, capital chain dimension and spatial chain dimension. It defines the entity type, relationship type and attribute field under each dimension, and performs entity extraction, relationship extraction and attribute weighted entity disambiguation based on the preprocessed data to generate a structured park industrial ecosystem knowledge graph.

[0010] The niche calculation engine has built-in multi-dimensional weighted improved Levins index, multi-resource dimension improved Piaka index and niche fit calculation formula. It is used to calculate the multi-dimensional integrated niche width of each enterprise in the park, the multi-dimensional niche overlap between any enterprise, and the niche fit of the enterprise to be introduced with the existing ecosystem of the park based on the knowledge graph of the park's industrial ecosystem. It identifies the gaps in the industrial chain and the targets for strengthening the chain based on the external dependence of the link and the average niche width threshold.

[0011] The dynamic evolution and deduction module has a built-in improved ecological gravity model that integrates niche characteristics. The improved ecological gravity model uses enterprise niche quality, niche distance and environmental correction coefficient as core input parameters to predict links in the park's industrial ecosystem knowledge graph, deduce the evolution of enterprise cooperation relationships and industrial chain extension trends, and simulate changes in industrial ecosystem health under different policy scenarios.

[0012] The visualization and interaction module is used to provide hierarchical visualization and human-computer interaction query of the park's industrial ecosystem knowledge graph, ecological niche quantitative calculation results, and evolutionary inference results.

[0013] Furthermore, in the aforementioned four-dimensional fused ecological map ontology system:

[0014] The entity types in the industrial chain dimension include industrial categories, links in the industrial chain, production enterprises, and products / services; the relationship types include upstream and downstream supply, OEM cooperation, industry competition, technology dependence, and supporting services.

[0015] The entity types in the innovation chain include universities and research institutes, R&D institutions, patents, technical standards, and industry talents; the relationship types include technology transfer, collaborative R&D, talent mobility, patent citation, and standard co-construction.

[0016] The types of entities involved in the capital chain include investment institutions, government industry funds, financial products, and support and subsidy projects; the types of relationships include equity investment, debt financing, policy subsidies, guarantees and credit enhancement, and venture capital.

[0017] The entity types of spatial chains include park carriers, industrial plants, industrial communities, supporting facilities, and geographic grids; the relationship types include spatial occupancy, geographic proximity, shared facilities, area planning, and commuting association.

[0018] Cross-dimensional connections are established between different dimensions through corporate entities, forming an integrated ecological graph network.

[0019] Furthermore, the niche calculation engine includes a niche width calculation unit, a niche overlap calculation unit, and a niche fit calculation unit;

[0020] The niche width calculation unit adopts a multi-dimensional weighted improved Levins index. After calculating the single-dimensional width from three dimensions—industry chain links, technology fields, and market scope—it is weighted and integrated to obtain the comprehensive niche width of the enterprise.

[0021] The niche overlap calculation unit adopts a multi-resource dimension improved Pianka index, integrating three resource dimensions: product, technology, and market, to calculate the comprehensive niche overlap of any two companies, representing the intensity of competition in the same industry.

[0022] The niche suitability calculation unit is used to calculate the overall matching degree between the enterprise to be introduced and the existing industrial ecosystem of the park by comprehensively considering three factors: industrial chain complementarity, competitive substitutability, and resource matching.

[0023] Furthermore, the niche suitability calculation unit calculates the niche suitability of the enterprise to be introduced using the following formula. ;

[0024] ;

[0025] In the formula, For companies to be introduced The complementarity with the existing industrial chain in the park, This represents the total number of links in the industrial chain. For the first in the industrial chain of the park The weights of each stage are determined by the entropy weight method based on the stage's output value share and strategic importance. For enterprises With the The functional correlation strength of each link is defined, with a value range of [0,1].

[0026] For companies to be introduced The degree of niche overlap with the average niche of existing peer companies in the park; For companies to be introduced The overall resource matching degree with the park's innovation resources, spatial carriers, and policy guidance is calculated by weighting three types of sub-indicators; , , The weighting coefficients are determined adaptively based on the park's historical investment attraction data using the entropy weighting method, and satisfy the following conditions: + + =1.

[0027] Furthermore, the specific calculation formula for the improved ecological gravity model of the dynamic evolution inference module is as follows:

[0028] ;

[0029] In the formula, For enterprises With enterprises The ecological attraction value between them represents the potential for them to form a cooperative relationship; is the ecological gravity constant, and is an empirical coefficient calibrated based on the dominant industry type of the park, with a value range of [0.8, 1.2]. For enterprises ,enterprise The normalized niche quality is calculated by weighting the normalized value of enterprise revenue, the normalized value of patent quantity, and the coreness of the industrial chain, with a value range of [0,1].

[0030] For enterprises With enterprises ecological niche distance. ,in This represents the degree of niche overlap between the two. To enhance the complementarity of their industrial chain functions, , The weighting coefficients for the niche distance dimension satisfy the following conditions: + =1; The environmental correction coefficient is calculated by weighting the normalized value of spatial distance between enterprises, the degree of policy matching, and the degree of capital chain correlation, and its value range is [0,1].

[0031] Furthermore, the dynamic evolution simulation module is also used to set different investment promotion policies, industrial support policies, and infrastructure construction policies as input variables, transform the policy effects into parameter increments in the corresponding dimensions, simulate the evolution path of the park's industrial ecosystem under multiple scenarios based on the improved ecological gravity model, and output the ecological health change curve and policy effectiveness ranking results under the corresponding scenarios.

[0032] A method for constructing a park industrial ecosystem graph based on a supply chain knowledge graph, implemented based on the park industrial ecosystem graph system based on a supply chain knowledge graph as described in any one of claims 1-6, includes the following steps:

[0033] S1. Multi-source data collection and preprocessing: Collect multi-source heterogeneous data from four dimensions within the park: industrial chain, innovation chain, capital chain, and spatial chain, and perform data cleaning, attribute standardization, and entity alignment preprocessing.

[0034] S2. Construction of a four-dimensional fusion graph: Construct a four-dimensional ecological graph ontology specification that integrates the industrial chain dimension, innovation chain dimension, capital chain dimension, and spatial chain dimension. Adopt a remote supervised extraction scheme that integrates industry domain dictionaries to perform entity extraction and relation extraction on the preprocessed data. Complete entity disambiguation through attribute weighted similarity calculation to generate a structured park industrial ecosystem knowledge graph.

[0035] S3. Quantitative Calculation of Industrial Niche: Based on the knowledge graph of the industrial ecosystem of the park, the Levins index with multi-dimensional weighting is used to calculate the comprehensive niche width of enterprises, the Piana index with multi-resource dimensions is used to calculate the niche overlap between enterprises, the niche fit formula is used to calculate the matching degree between the enterprises to be introduced and the park, and the gaps in the industrial chain and the targets for strengthening the chain are identified based on the external dependence of the links and the average niche width threshold.

[0036] S4. Ecological Dynamic Evolution Simulation: Based on the improved ecological gravity model that integrates ecological niche characteristics, the park's industrial ecosystem knowledge graph is used to predict links, deduce the evolution of enterprise cooperation relationships and industrial chain development trends, simulate changes in industrial ecosystem health under different policy scenarios, and identify potential industrial risks.

[0037] S5. Visualized Interactive Output: The industrial ecosystem knowledge graph, ecological niche quantitative calculation results, and evolutionary inference results are displayed in a layered and visual manner, supporting interactive queries and result export.

[0038] Furthermore, the remote supervised extraction scheme for integrating the industry domain dictionary in step S2 is as follows:

[0039] A pre-built domain dictionary for the target industry is constructed, including names of links in the industrial chain, product names, and technical terms, and coarse entity recall is performed on unstructured text.

[0040] Remote monitoring and automatic labeling are performed based on publicly available industry knowledge bases and publicly available enterprise information to construct positive and negative sample sets and generate relation labeling sample sets.

[0041] Based on the labeled sample set, fine-tuning the pre-trained language model enables automated classification and extraction of all four-dimensional relation types.

[0042] Finally, entity similarity is calculated by weighting three attributes: enterprise name text similarity, unified social credit code matching degree, and spatial coordinate distance, thus completing entity disambiguation and attribute fusion across data sources.

[0043] Furthermore, the specific process for identifying supply chain gaps in step S3 is as follows: traverse all links of the park's supply chain, calculate the number of enterprises, average niche quality, and external dependence for each link, where external dependence = (total demand of the link - total local supply of the link) / total demand of the link; identify links with external dependence higher than a preset threshold and an average niche width of existing enterprises within the link lower than the preset threshold as supply chain gaps; identify links with external dependence lower than the preset threshold but an average niche quality lower than the industry benchmark as strong supply chain targets; generate target enterprise profiles for supply chain gaps, and generate a ranking list of target investment enterprises from the candidate enterprise database using the niche suitability formula.

[0044] Furthermore, step S4 also includes an assessment of the health of the industrial ecosystem, specifically:

[0045] Construct a health index system that includes five primary dimensions: industrial structure diversity, industrial chain integrity, innovation activity, funding adequacy, and spatial matching. Each primary dimension has 3-4 secondary quantitative indicators.

[0046] The weights of each indicator are determined by the entropy weight method, and the weighted calculation yields the comprehensive score of the current industrial ecosystem health of the park.

[0047] Based on the results of evolutionary projections, predict the trend of health status changes within a set period in the future, and output risk warning information for links with excessive external dependence and excessive overlap of corporate niches.

[0048] The present invention has the following advantages over the prior art:

[0049] (1) Achieved a systematic and multi-dimensional representation of the industrial ecosystem: This invention for the first time constructed an ecological map ontology framework integrating the four dimensions of industrial chain, innovation chain, capital chain and spatial chain, breaking through the limitations of existing technologies that only construct maps around enterprises or industrial chains in a single dimension. Through enterprise entities, it achieves multi-dimensional correlation and coupling, which can fully represent the multi-dimensional symbiotic relationship and system structure of the park's industrial ecosystem. It has achieved an essential leap from "enterprise information aggregation" to "digital mapping of industrial ecosystem", which can more comprehensively and deeply depict the operation law of the park's industry.

[0050] (2) A customized and objective ecological quantitative assessment system has been established: This invention adapts the niche theory and introduces it into the industrial park governance scenario. It proposes calculation methods for niche width, overlap, and suitability adapted to the industrial park scenario, forming a complete industrial ecological quantitative assessment system. This system breaks through the limitations of traditional manual rules and keyword matching, can objectively quantify the position of enterprises in the ecosystem, accurately assess the matching degree between enterprises to be introduced and the park, automatically identify weak links in the industrial chain, overcome the subjectivity and experience dependence of traditional manual assessment, and provide a scientific quantitative basis for the park to strengthen and supplement the industrial chain and attract precise investment.

[0051] (3) Achieving a technological leap from static description to dynamic deduction: This invention pioneered an improved ecological gravity model that integrates niche theory, incorporating niche quality, niche distance, and environmental correction factors into the gravity calculation framework, and constructing a dedicated industrial ecosystem link prediction algorithm. This model breaks through the limitations of traditional link prediction that relies solely on network topology, fully integrates the exclusive attributes of the industrial ecosystem, and can accurately deduce the evolution of enterprise cooperation relationships and industrial chain development trends. It supports simulation and effect evaluation of multiple policy scenarios, and breaks through the limitations of existing technologies that can only provide static map queries, thus supporting forward-looking governance and medium- and long-term industrial planning of industrial parks.

[0052] (4) A complete technical solution with a scenario-based and systematic approach has been formed: This invention has systematically designed the core needs of industrial park governance, rather than simply applying general knowledge graph technology. It forms a complete technical loop from data collection, graph construction, quantitative calculation to evolution and deduction, which can directly support core business scenarios such as precise investment promotion, industrial chain optimization, ecological health assessment, and risk warning. It has clear technological progress and significant industrial application value.

[0053] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a schematic diagram of the overall architecture of the industrial ecosystem mapping system of the park described in this invention;

[0056] Figure 2 This is a schematic diagram of the structure of the four-dimensional fusion ecological map ontology described in this invention;

[0057] Figure 3 This is a schematic diagram of the internal architecture and computational logic of the niche computing engine described in this invention;

[0058] Figure 4 This is a flowchart illustrating the steps of the method for constructing an industrial ecosystem map of a park as described in this invention.

[0059] Figure 5 This is a schematic diagram illustrating the derivation logic and application path of the improved ecological gravity model described in this invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Example 1: Construction and Implementation of a Four-Dimensional Knowledge Graph for the Industrial Ecosystem of the Industrial Park

[0062] This embodiment uses a smart manufacturing industrial park in a national-level economic and technological development zone as an application scenario to explain in detail the complete construction process of a four-dimensional integrated industrial ecosystem knowledge graph.

[0063] Step 1: Multi-source data acquisition and preprocessing;

[0064] The data acquisition module connects to five types of data sources to complete the collection of raw data across four dimensions:

[0065] Industry chain data: Business registration information, main product information, upstream and downstream supplier and customer directories, and national economic industry classification information of 127 intelligent manufacturing enterprises in the park, covering four major industry chain links: complete machine manufacturing, core components, basic materials, and production services, with a total of 16 sub-links; data sources are the industrial and commercial enterprise information database and data submitted by enterprises in the park.

[0066] Innovation chain data: 2,136 authorized patents of enterprises in the park, 32 technical standards they participated in developing, educational background and technical field information of 8 cooperating universities and research institutes, and 12,000 industry talents; data source: State Intellectual Property Office database and park talent statistics system;

[0067] Capital chain data: 17 cooperative investment institutions in the park, 3 government industrial guidance funds, 26 support and subsidy projects at all levels, and historical financing and subsidy application records of enterprises; data source: park's financial system and publicly available investment and financing information;

[0068] Spatial chain data: latitude and longitude coordinates, building area, and information on resident enterprises for 32 industrial plants, 6 industrial communities, and 12 public supporting facilities in the park, as well as the geographical grid division data of the park at 1km×1km; the data source is the park's geographic information system.

[0069] Data preprocessing involves four standardized operations:

[0070] (1) Data cleaning: missing values ​​are filled with the average of companies of the same size in the same industry, and obviously abnormal values ​​such as revenue and number of patents are removed using the 3σ principle;

[0071] (2) Attribute standardization: The unified enterprise name is the full name registered with the Administration for Industry and Commerce, the unified time granularity is the year, the unified monetary unit is RMB 10,000, and the unified coordinate format is WGS84 latitude and longitude;

[0072] (3) Initial alignment of entities: Using the unified social credit code of the enterprise as the primary key, the attribute data of the same enterprise in different data sources are linked and integrated;

[0073] (4) Text preprocessing: For unstructured texts such as company profiles and product introductions, the jieba word segmentation tool is used for word segmentation, a custom industry domain dictionary is loaded, and stop words and meaningless symbols are removed.

[0074] Step 2: Define the four-dimensional ontology specification;

[0075] Based on the characteristics of the intelligent manufacturing industry, the four-dimensional ontology construction module defines the entity, relationship, and attribute specifications of the four-dimensional ontology. Each dimension establishes cross-dimensional relationships through the entities of production enterprises, forming an integrated network. The specific definitions are as follows:

[0076] The supply chain dimension includes four types of entities: industry categories, supply chain links, production enterprises, and products / services; five types of relationships: upstream and downstream supply, OEM cooperation, industry competition, technology dependence, and supporting services; and attributes such as enterprise revenue, establishment time, product model, and industry classification code.

[0077] Innovation chain dimensions: Entity types include five categories: universities and research institutes, R&D institutions, patents, technical standards, and industry talents; Relationship types include five categories: technology transfer, collaborative R&D, talent mobility, patent citation, and standard co-construction; Attributes include patent type, IPC classification number in the technical field, talent education level, and cooperation start time.

[0078] The funding chain is categorized into four types: investment institutions, government industry funds, financial products, and subsidy projects; four types of relationships: equity investment, debt financing, policy subsidies, and guarantee enhancement; and four attributes: investment amount, subsidy amount, financing term, and annualized interest rate.

[0079] Spatial chain dimensions: Entity types include 5 categories: park carrier, industrial plant, industrial community, supporting facilities, and geographic grid; Relationship types include 4 categories: spatial occupancy, geographic proximity, shared facilities, and area planning; Attributes include building area, latitude and longitude coordinates, occupancy rate, unit rent, etc.

[0080] Step 3: Knowledge Extraction and Graph Generation;

[0081] A remote supervised extraction scheme that integrates industry-specific dictionaries is used to perform entity and relation extraction. The specific process is as follows:

[0082] Domain dictionary construction and coarse entity recall: A pre-built intelligent manufacturing industry domain dictionary, containing more than 3,200 entries including industry chain link names, product names, and technical terms; The dictionary is loaded into the jieba word segmentation tool to perform word segmentation and entity recognition on the pre-processed text data, completing the coarse recall of industry domain entities and improving the accuracy of entity recognition.

[0083] Remote supervision annotation to build a sample set: The core principle of remote supervision is that if two entities have a known relationship in an external knowledge base, then all sentences that contain both entities can be labeled as training samples of that relationship.

[0084] Specific operations: Based on publicly available industry knowledge bases and cooperation information disclosed on corporate websites, a known relationship pair database is constructed; the text data is traversed, and if two corporate entities from the relationship pair database appear in the same sentence, it is labeled as a positive sample of the corresponding relationship; sentences containing corporate entity pairs with no known association are randomly selected as negative samples; finally, 12,000 relationship labeling samples are generated, covering all 18 types of relationships, with a positive to negative sample ratio of 1:1.

[0085] Fine-tuning of pre-trained language models: Based on the BERT-base Chinese pre-trained model, fine-tuning training is performed using labeled sample sets to achieve relation classification;

[0086] The training hyperparameters were set as follows: learning rate 2e-5, batch size 16, training epochs 3, maximum sequence length 128; cross-entropy loss function and Adam optimizer were used.

[0087] The model achieved an accuracy of 90.1% and an F1 score of 89.2% on the independent test set, meeting the requirements for engineering applications.

[0088] Attribute-weighted entity disambiguation: To address potential issues of synonymy and heteronymy across data sources, entity similarity is calculated using a weighted average of three types of attributes. The formula is as follows:

[0089] ;

[0090] In the formula: To measure the text similarity of enterprise names, edit distance is used, which is the minimum number of single-character edit operations required to convert between two strings. After normalization, the value range is [0,1]. To standardize the matching degree of social credit codes, a value of 1 is set for a perfect match and a value of 0 is set for a non-match. The value is a normalized value for spatial coordinate distance. The Euclidean distance is calculated using latitude and longitude and mapped to the [0,1] interval through min-max normalization. The closer the distance, the higher the value.

[0091] Weights set to =0.3, =0.5, =0.2. Judgment rule: Two enterprise entities with a similarity score higher than 0.7 are judged as the same entity, their attribute fields are merged, and entity disambiguation and attribute fusion are completed.

[0092] The final generated industrial ecosystem knowledge graph of the park contains 2,347 entities and 7,892 relationships, stored in the Neo4j graph database. It supports Cypher graph query statements for association queries and graph computation. Compared with the existing single-dimensional industrial chain graphs, the four-dimensional graph constructed in this embodiment has more comprehensive coverage of entity dimensions and richer relationship types. It can fully represent the multi-dimensional symbiotic structure of the industrial ecosystem and solve the shortcomings of existing technologies in terms of single dimension and insufficient systematicity.

[0093] Example 2: Implementation of Industrial Niche Quantitative Calculation and Supply Chain Complement Target Identification:

[0094] This embodiment is based on the map constructed in Embodiment 1. It completes the ecological niche quantification calculation and chain gap identification through the ecological niche calculation engine, and details the complete calculation process of the core algorithm.

[0095] 1. Improved Levins index calculation for niche breadth using multi-dimensional weighting

[0096] The traditional Levins index can only calculate the niche breadth of a single resource dimension. This invention improves upon this by calculating the niche breadth from three dimensions: industry chain links, technology fields, and market scope. Then, it uses an entropy weighting method to weight and fuse these dimensions to obtain the enterprise's niche breadth. The comprehensive niche breadth is calculated using the following formula:

[0097]

[0098] In the formula:

[0099] These are resource dimension numbers, with a total of 3 dimensions ( =1 industrial chain link, ( =2 Technical Fields, =3 Market Scope);

[0100] The weight of the k-th dimension is determined by the entropy weight method.

[0101] For enterprises In the The first dimension The resource utilization ratio of each sub-resource is calculated as the ratio of revenue / patents on that resource to the company's total revenue / total patents. For the first The total number of resources in each dimension.

[0102] Take a certain complete machine manufacturing enterprise A (referred to as Enterprise A) in the park as an example. Let's take an example to calculate:

[0103] Industrial chain segment dimension ( =1): Covering three stages: complete machine assembly, system integration, and after-sales maintenance, with each stage accounting for 0.6%, 0.3%, and 0.1% of revenue, respectively; single-dimensional width 1 / (0.6 2 +0.3 2 +0.1 2 = 1 / 0.46 ≈ 2.17;

[0104] Technical field dimension ( =2): Covering three technical fields: industrial control, mechanical design, and the Internet of Things, with patent percentages of 0.5%, 0.3%, and 0.2% respectively; single-dimensional width 1 / (0.5 2 +0.3 2 +0.2 2 =1 / 0.38≈2.63;

[0105] Market region dimension ( =3): Covering three regions: East China, Central China, and South China, with revenue share of 0.5%, 0.3%, and 0.2% respectively; Single-dimensional width 1 / (0.5 2 +0.3 2 +0.2 2 =1 / 0.38≈2.63;

[0106] The weights of the three dimensions were calculated using the entropy weight method. =0.4, =0.3, =0.3, Overall niche width: =0.4×2.17+0.3×2.63+0.3×2.63≈2.45. The calculation result indicates that the company has a wide range of resource usage and belongs to the core broad spectrum type of enterprise in the industrial ecosystem.

[0107] 2. Improve the calculation of niche overlap using the Piaka index from multiple resource dimensions.

[0108] Traditional Pianka indices calculate overlap based on a single resource dimension. This invention improves upon this by integrating three resource dimensions—product, technology, and market—and using a weighted calculation to obtain the enterprise's overlap. With enterprises The comprehensive niche overlap is calculated using the following formula:

[0109]

[0110] In the formula, For enterprises ,enterprise The proportion of usage on the m-th resource in the k-th dimension.

[0111] Two sensor companies, B (Company B) within the park C (Company) Taking [example] as an example: the resource distribution of the two [product types] highly overlaps in three dimensions: product type, technology field, and market region. The single-dimensional overlap is calculated to be 0.85, 0.82, and 0.79, respectively. The weights of the three dimensions are determined to be 0.4, 0.3, and 0.3, respectively, using the entropy weight method. The weighted fusion result shows the overall niche overlap: =0.4×0.85+0.3×0.82+0.3×0.79=0.823≈0.82; The calculation result indicates that the two businesses have a high degree of overlap and the intensity of competition in the same industry.

[0112] 3. Calculation of Niche Fit

[0113] For a certain precision reducer company X (denoted as Company X) that is to be introduced... The matching degree between the target ecosystem and the park's ecosystem is calculated using the niche fit formula, which is: ;

[0114] Complete chain complementarity ;

[0115] The formula is ;

[0116] in, This refers to the sequence number of the links in the industrial chain. The total number of links in the industrial chain; the intelligent manufacturing industrial chain in this park comprises... =16 links; the weight of each link is calculated based on the proportion of output value and strategic importance of each link using the entropy weight method, among which the precision reducer link (the first link) is the most important. Weight of each link =0.35;

[0117] enterprise Our main business is precision speed reducers. We calculate the functional relevance of a component to its description using cosine similarity between the product text and the component description. =0.9; the correlation strength of the remaining 15 links is all below 0.1, and the weighted sum is as follows: =0.86;

[0118] Average niche overlap ;

[0119] There are currently three reducer manufacturers in the industrial park. The improved Pianka index was used to calculate the costs for each manufacturer. The niche overlap with the three companies was 0.45, 0.42, and 0.39; taking the arithmetic mean yielded... =(0.45+0.42+0.39) / 3=0.42, that is, the competitive substitutability index 1- =0.58;

[0120] Overall resource matching degree The score is calculated by weighting the matching degree of innovation resources (0.88), spatial carriers (0.95), and policy guidance (0.92), with weights of 0.4, 0.3, and 0.3 respectively. =0.4×0.88+0.3×0.95+0.3×0.92=0.913≈0.92;

[0121] Weighting coefficient ;

[0122] Based on the park's investment attraction data over the past three years, an adaptive calculation using the entropy weight method is employed. The complete steps are as follows:

[0123] (1) Construct an indicator matrix, including three types of indicators: complementarity, overlap, and resource matching, totaling [number missing]. =50 historical examples of successful investment projects;

[0124] (2) The indicator data are standardized by min-max to eliminate the influence of units;

[0125] (3) Calculate the first The proportion of each sample under the s-th indicator ;

[0126] (4) Calculate the information entropy of the s-th index. ;in =1 / This is the normalization constant;

[0127] (5) Calculate the weight of the s-th indicator. =(1- ) / , where m is the total number of indicators;

[0128] Final result: ,satisfy ;

[0129] Final fit calculation: =0.5×0.86+0.3×0.58+0.2×0.92=0.43+0.174+0.184=0.788; Enterprise Its niche suitability is 0.788, which is excellent, making it a high-quality investment target.

[0130] 4. Chain gap identification

[0131] We traversed all 16 links in the industrial chain within the park and calculated the core indicators for each link:

[0132] External Dependence: The calculation formula is: External Dependence = (Total Demand for the Link - Total Local Supply for the Link) / Total Demand for the Link; Statistical Scope Explanation: Total demand for the link refers to the total annual demand for products / services of the entire industrial chain in the park for that link; Total local supply for the link refers to the total annual production capacity / service supply of enterprises in that link within the park;

[0133] The threshold for external dependence was set at 0.6, and the threshold for average niche breadth was set at 2.5. The thresholds were determined based on industry statistics of similar benchmark parks in China.

[0134] The external dependence of the precision reducer segment is 0.78, which is higher than the threshold; the average niche width of existing enterprises in the segment is 2.1, which is lower than the threshold. Therefore, it is identified as a core supply chain gap.

[0135] To address this gap, a target company profile was generated. Using a niche suitability formula, a list of 12 target companies was generated from a national database of precision reducers, sorted in descending order of suitability. This list can be directly used for investment promotion and matchmaking.

[0136] The niche quantification model proposed in this embodiment is an original design that deeply integrates ecological theory with the industrial scenario of the park. It has made multi-dimensional improvements to the classic ecological formula, rather than conventional technical means in this field. It can objectively and accurately complete ecological assessment and target identification, and has unexpected technical effects.

[0137] Example 3: Implementation of Dynamic Evolution and Health Assessment of Industrial Ecosystem:

[0138] This embodiment is based on an improved ecological gravity model to dynamically evolve and assess the health of the industrial ecosystem in the park, and details the complete implementation process of the algorithm.

[0139] 1. Improve ecological gravity model link prediction

[0140] The core formula of the improved ecological gravity model is:

[0141] ;

[0142] In the formula , They represent two different business entities.

[0143] Complete calculation rules for core parameters:

[0144] Niche quality :

[0145] The formula is

[0146] in For enterprises Normalized revenue min-max value For enterprises The min-max normalized value of the number of patents. For enterprises Coreness of the industrial chain (the node degree of an enterprise in the industrial chain map / the maximum node degree of the entire industrial chain).

[0147] Weights set to =0.4, =0.3, =0.3;

[0148] min-max normalization formula: This maps the values ​​to the [0,1] interval;

[0149] Niche distance ;

[0150] in, For enterprises With enterprises Niche overlap. To determine the complementarity of their industrial chain functions; the dimension weights are set as follows: =0.3, =0.7, which satisfies the condition. + =1; the value range is [0,1]. The smaller the distance, the more compatible the two ecological niches are, and the higher the possibility of cooperation.

[0151] Environmental correction factor ;

[0152] The formula is =

[0153] in For enterprises With enterprises The spatial distance normalized value (the closer the distance, the higher the value). For policy alignment, For the degree of correlation of the capital chain; the weight is set to =0.3, =0.4, =0.3; the value range is [0,1], and the higher the value is, the more favorable the environment.

[0154] Ecological gravity constant G: calibrated to 1.0 for the intelligent manufacturing industry, and can be adjusted within the range of 0.8-1.2 for different industry types.

[0155] Company A (the complete machine manufacturer in the park) ) and component enterprise D (enterprise) Taking (e.g.) as an example, calculate the ecological gravitational pull values ​​of the two:

[0156] Niche quality: Enterprise With revenue of 2 billion yuan, 120 patents, and a core industrial chain score of 0.9, the weighted average after normalization is calculated as follows: =0.85; Enterprise j has revenue of 300 million yuan, 32 patents, and a core industrial chain degree of 0.6. The weighted average after normalization is calculated as follows: =0.28;

[0157] Niche distance: the degree of overlap in the ecological niches of the two =0.2, functional complementarity of the industrial chain =0.8; =1-(0.3×0.2+0.7×0.8)=1-0.62=0.38; Environmental correction coefficient: the spatial distance between the two is 1.2km, the policy matching degree is 0.9, and there is no financial chain connection. The weighted calculation yields... =0.85;

[0158] Ecological gravitational constant G = 1.0.

[0159] Final ecological gravity value calculation: =1.0×(f0.85×0.28) / 0.38 2 ×0.85≈1.40;

[0160] The gravitational value is normalized to a cooperative probability using the Sigmoid function, as shown in the formula: Substitute into the calculation to obtain the probability of cooperation ≈0.802, or 80.2%, predicting that the two are highly likely to establish a supply cooperation relationship within the next 1-2 years.

[0161] The model was applied to predict link relationships among all enterprises in the industrial park, identifying 87 potential cooperative relationships. Compared to the traditional Common Neighbors link prediction algorithm, this model incorporates niche characteristics and environmental correction factors, fully integrating the unique attributes of the industrial ecosystem rather than solely relying on network topology, resulting in significantly improved prediction accuracy and demonstrating outstanding substantive features.

[0162] 2. Policy Scenario Evolution Simulation: Two policy scenarios are set up, and the policy effects are transformed into model parameter increments. Based on the improved ecological gravity model, the changes in the park's industrial ecosystem after 3 years are deduced: Scenario 1: An additional 20 million yuan industrial support fund. Policy effect transformation: The capital chain correlation of core component enterprises increases by 0.2, corresponding to an environmental correction coefficient. Scenario 1: Improving and lowering the financial threshold for cooperation between enterprises; Developed results: Increased industrial chain integrity, decreased external dependence on core components, and steady improvement in ecosystem health. Scenario 2: Policy effect transformation of building a public technology R&D platform for intelligent manufacturing: Increased overall innovation resource supply in the park, expected growth in enterprise patent output, and corresponding improvement in ecosystem quality. Improvements; The results show a significant increase in enterprise innovation activity, a greater increase in industry-academia-research collaboration, and a more substantial improvement in overall ecosystem health. By comparing the ecosystem health change curves under the two scenarios, the effectiveness of policies is ranked, providing a quantitative basis for park decision-making.

[0163] 3. The industrial ecosystem health assessment constructs a five-level health indicator system. The specific secondary indicators and quantification methods are as follows: Industrial structure diversity: number of industrial categories, balanced distribution of enterprise size (inverse of the Gini coefficient), and coverage rate of subdivided segments; Industrial chain integrity: local supporting rate of core links, average external dependence, and leading enterprise driving coefficient (number of affiliated enterprises of leading enterprises / total number of enterprises); Innovation activity: number of patents per 10,000 people, density of industry-university-research cooperation, and number of technical standards participated in; Funding adequacy: average financing amount per enterprise, coverage rate of industrial funds, and coverage rate of policy subsidies; Spatial matching: factory vacancy rate, industrial agglomeration, and walking coverage rate of supporting facilities.

[0164] The weights of each indicator were determined using the entropy weight method, and the weighted calculation yielded a current comprehensive score of 72.4 points (out of 100) for the ecological health of the park. Based on the evolutionary projection results, it is predicted that the health score will increase to 81.6 points in 3 years. At the same time, the high dependence on external sources for core components was identified as the main risk point, and early warning information was output.

[0165] This embodiment breaks through the limitation of existing technologies that can only perform static queries, and realizes dynamic prediction of industrial ecosystem and policy simulation, which can support the forward-looking governance of the park and has significant technological progress.

[0166] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A park industrial ecosystem graph system based on industrial chain knowledge graph, characterized in that, It includes a data acquisition module, a four-dimensional ontology construction module, an ecological niche computing engine, a dynamic evolution inference module, and a visualization and interaction module; The data acquisition module is used to collect data on the industrial chain, innovation chain, capital chain, and spatial chain within the park, and to perform data cleaning, attribute standardization, and cross-data source entity alignment preprocessing. The four-dimensional ontology construction module is used to construct a four-dimensional integrated ecological graph ontology system that includes industrial chain dimension, innovation chain dimension, capital chain dimension and spatial chain dimension. It defines the entity type, relationship type and attribute field under each dimension, and performs entity extraction, relationship extraction and attribute weighted entity disambiguation based on the preprocessed data to generate a structured park industrial ecosystem knowledge graph. The niche calculation engine has built-in multi-dimensional weighted improved Levins index, multi-resource dimension improved Piaka index and niche fit calculation formula. It is used to calculate the multi-dimensional integrated niche width of each enterprise in the park, the multi-dimensional niche overlap between any enterprise, and the niche fit of the enterprise to be introduced with the existing ecosystem of the park based on the knowledge graph of the park's industrial ecosystem. It identifies the gaps in the industrial chain and the targets for strengthening the chain based on the external dependence of the link and the average niche width threshold. The dynamic evolution and deduction module has a built-in improved ecological gravity model that integrates niche characteristics. The improved ecological gravity model uses enterprise niche quality, niche distance and environmental correction coefficient as core input parameters to predict links in the park's industrial ecosystem knowledge graph, deduce the evolution of enterprise cooperation relationships and industrial chain extension trends, and simulate changes in industrial ecosystem health under different policy scenarios. The visualization and interaction module is used to provide hierarchical visualization and human-computer interaction query of the park's industrial ecosystem knowledge graph, ecological niche quantitative calculation results, and evolutionary inference results.

2. The industrial park ecosystem graph system based on industrial chain knowledge graph as described in claim 1, characterized in that, In the aforementioned four-dimensional fusion ecological map ontology system: The entity types in the industrial chain dimension include industrial categories, links in the industrial chain, production enterprises, and products / services; the relationship types include upstream and downstream supply, OEM cooperation, industry competition, technology dependence, and supporting services. The entity types in the innovation chain include universities and research institutes, R&D institutions, patents, technical standards, and industry talents; the relationship types include technology transfer, collaborative R&D, talent mobility, patent citation, and standard co-construction. The types of entities involved in the capital chain include investment institutions, government industry funds, financial products, and support and subsidy projects; the types of relationships include equity investment, debt financing, policy subsidies, guarantees and credit enhancement, and venture capital. The entity types of spatial chains include park carriers, industrial plants, industrial communities, supporting facilities, and geographic grids; the relationship types include spatial occupancy, geographic proximity, shared facilities, area planning, and commuting association. Cross-dimensional connections are established between different dimensions through corporate entities, forming an integrated ecological graph network.

3. The industrial park ecosystem graph system based on industrial chain knowledge graph as described in claim 1, characterized in that, The niche calculation engine includes a niche width calculation unit, a niche overlap calculation unit, and a niche fit calculation unit. The niche width calculation unit adopts a multi-dimensional weighted improved Levins index. After calculating the single-dimensional width from three dimensions—industry chain links, technology fields, and market scope—it is weighted and integrated to obtain the comprehensive niche width of the enterprise. The niche overlap calculation unit adopts a multi-resource dimension improved Pianka index, integrating three resource dimensions: product, technology, and market, to calculate the comprehensive niche overlap of any two companies, representing the intensity of competition in the same industry. The niche suitability calculation unit is used to calculate the overall matching degree between the enterprise to be introduced and the existing industrial ecosystem of the park by comprehensively considering three factors: industrial chain complementarity, competitive substitutability, and resource matching.

4. The industrial park ecosystem graph system based on industrial chain knowledge graph as described in claim 3, characterized in that, The niche suitability calculation unit calculates the niche suitability of the enterprise to be introduced using the following formula. ; ; In the formula, For companies to be introduced The complementarity with the existing industrial chain in the park, This represents the total number of links in the industrial chain. For the first in the industrial chain of the park The weights of each stage are determined by the entropy weight method based on the stage's output value share and strategic importance. For enterprises With the The functional correlation strength of each link is defined, with a value range of [0,1]. For companies to be introduced The degree of niche overlap with the average niche of existing peer companies in the park; For companies to be introduced The overall resource matching degree with the park's innovation resources, spatial carriers, and policy guidance is calculated by weighting three types of sub-indicators; , , The weighting coefficients are determined adaptively based on the park's historical investment attraction data using the entropy weighting method, and satisfy the following conditions: + + =1.

5. The industrial park ecosystem graph system based on industrial chain knowledge graph as described in claim 1, characterized in that, The specific calculation formula for the improved ecological gravity model of the dynamic evolution inference module is as follows: ; In the formula, For enterprises With enterprises The ecological attraction value between them represents the potential for them to form a cooperative relationship; is the ecological gravity constant, and is an empirical coefficient calibrated based on the dominant industry type of the park, with a value range of [0.8, 1.2]. For enterprises ,enterprise The normalized niche quality is calculated by weighting the normalized value of enterprise revenue, the normalized value of patent quantity, and the coreness of the industrial chain, with a value range of [0,1]. For enterprises With enterprises ecological niche distance. ,in This represents the degree of niche overlap between the two. To enhance the complementarity of their industrial chain functions, , The weighting coefficients for the niche distance dimension satisfy the following conditions: + =1; The environmental correction coefficient is calculated by weighting the normalized value of spatial distance between enterprises, the degree of policy matching, and the degree of capital chain correlation, and its value range is [0,1].

6. The industrial park ecosystem graph system based on industrial chain knowledge graph as described in claim 5, characterized in that, The dynamic evolution simulation module is also used to set different investment promotion policies, industry support policies, and infrastructure construction policies as input variables, transform the policy effects into parameter increments in the corresponding dimensions, simulate the evolution path of the park's industrial ecosystem under multiple scenarios based on the improved ecological gravity model, and output the ecological health change curve and policy effectiveness ranking results under the corresponding scenarios.

7. A method for constructing a park industrial ecosystem graph based on a supply chain knowledge graph, implemented based on the park industrial ecosystem graph system based on a supply chain knowledge graph as described in any one of claims 1-6, characterized in that, Includes the following steps: S1. Multi-source data collection and preprocessing: Collect multi-source heterogeneous data from four dimensions within the park: industrial chain, innovation chain, capital chain, and spatial chain, and perform data cleaning, attribute standardization, and entity alignment preprocessing. S2. Construction of a four-dimensional fusion graph: Construct a four-dimensional ecological graph ontology specification that integrates the industrial chain dimension, innovation chain dimension, capital chain dimension, and spatial chain dimension. Adopt a remote supervised extraction scheme that integrates industry domain dictionaries to perform entity extraction and relation extraction on the preprocessed data. Complete entity disambiguation through attribute weighted similarity calculation to generate a structured park industrial ecosystem knowledge graph. S3. Quantitative Calculation of Industrial Niche: Based on the knowledge graph of the industrial ecosystem of the park, the Levins index with multi-dimensional weighting is used to calculate the comprehensive niche width of enterprises, the Piana index with multi-resource dimensions is used to calculate the niche overlap between enterprises, the niche fit formula is used to calculate the matching degree between the enterprises to be introduced and the park, and the gaps in the industrial chain and the targets for strengthening the chain are identified based on the external dependence of the links and the average niche width threshold. S4. Ecological Dynamic Evolution Simulation: Based on the improved ecological gravity model that integrates ecological niche characteristics, the park's industrial ecosystem knowledge graph is used to predict links, deduce the evolution of enterprise cooperation relationships and industrial chain development trends, simulate changes in industrial ecosystem health under different policy scenarios, and identify potential industrial risks. S5. Visualized Interactive Output: The industrial ecosystem knowledge graph, ecological niche quantitative calculation results, and evolutionary inference results are displayed in a layered and visual manner, supporting interactive queries and result export.

8. The method for constructing a park industrial ecosystem graph based on a supply chain knowledge graph according to claim 7, characterized in that, The remote supervised extraction scheme for integrating the industry domain dictionary in step S2 is as follows: A pre-built domain dictionary for the target industry is constructed, including names of links in the industrial chain, product names, and technical terms, and coarse entity recall is performed on unstructured text. Remote monitoring and automatic labeling are performed based on publicly available industry knowledge bases and publicly available enterprise information to construct positive and negative sample sets and generate relation labeling sample sets. Based on the labeled sample set, fine-tuning the pre-trained language model enables automated classification and extraction of all four-dimensional relation types. Finally, entity similarity is calculated by weighting three attributes: enterprise name text similarity, unified social credit code matching degree, and spatial coordinate distance, thus completing entity disambiguation and attribute fusion across data sources.

9. The method for constructing a park industrial ecosystem graph based on a supply chain knowledge graph according to claim 7, characterized in that, The specific process for identifying supply chain gaps in step S3 is as follows: Traverse all links of the industrial chain in the park, calculate the number of enterprises, average niche quality, and external dependence for each link, where external dependence = (total demand of the link - total local supply of the link) / total demand of the link; identify links with external dependence higher than a preset threshold and an average niche width of existing enterprises within the link lower than the preset threshold as supply chain gaps; identify links with external dependence lower than the preset threshold but an average niche quality lower than the industry benchmark as strong supply chain targets; generate target enterprise profiles for supply chain gaps, and generate a ranking list of target investment enterprises from the candidate enterprise database using the niche suitability formula.

10. The method for constructing a park industrial ecosystem graph based on a supply chain knowledge graph according to claim 7, characterized in that, Step S4 also includes an assessment of the health of the industrial ecosystem, specifically: Construct a health index system that includes five primary dimensions: industrial structure diversity, industrial chain integrity, innovation activity, funding adequacy, and spatial matching. Each primary dimension has 3-4 secondary quantitative indicators. The weights of each indicator are determined by the entropy weight method, and the weighted calculation yields the comprehensive score of the current industrial ecosystem health of the park. Based on the results of evolutionary projections, predict the trend of health status changes within a set period in the future, and output risk warning information for links with excessive external dependence and excessive overlap of corporate niches.

Images