Innovative driving type industrial park identification method and system based on multi-source data fusion

Abstract

The invention relates to the technical field of data processing, and provides an innovation-driven industrial park identification method and system based on multi-source data fusion, and the method comprises the steps: obtaining enterprise basic data, technical innovation data, supply chain dynamic data, talent ecological data, green development data, geographic space data and other multi-source data of a to-be-identified region; performing data preprocessing including data association, spatial function analysis and kernel density analysis on the multi-source data; determining a target data index of the to-be-identified area based on the preprocessed data; based on the target data index of the to-be-identified area, determining an innovation driving intensity index of the to-be-identified area; and inputting the innovation driving intensity index, the target data index, the cross-regional radiation capability and the technical reserve depth of the to-be-identified region into the innovation driving type industrial park identification model to obtain a park type output by the innovation driving type industrial park identification model. According to the invention, the identification accuracy of the innovative driving type industrial park can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for identifying innovation-driven industrial parks based on multi-source data fusion. Background Technology

[0002] Currently, existing evaluation systems for innovation-driven industrial parks mainly rely on single traditional indicators (such as the number of enterprises or output value). This approach has at least the following problems: 1. Inefficient spatial identification: Parks are divided based on administrative divisions or single economic indicators, which fails to identify the real spatial carriers of innovation elements (such as implicit innovation clusters).

[0003] 2. Fragmented industrial linkages: The lack of interconnected dynamic data on technology diffusion and supply chain collaboration among enterprises makes it difficult to assess the resilience of the industrial chain in the park.

[0004] 3. Insufficient policy adaptability: There is a lack of assessment models that match the connotation of innovation-driven intensity. Traditional indicator systems cannot support new evaluation dimensions such as "technology radiation capacity" and "technology reserve depth", making it difficult to accurately allocate policy resources to high-potential parks.

[0005] This results in low accuracy when identifying innovation-driven industrial parks. Summary of the Invention

[0006] This application aims to address at least one of the technical problems existing in the aforementioned related technologies. To this end, this application proposes a method and system for identifying innovation-driven industrial parks based on multi-source data fusion, to solve the problem of low accuracy in identifying innovation-driven industrial parks and to improve the accuracy of such identification.

[0007] The method for identifying innovation-driven industrial parks based on multi-source data fusion according to the first aspect of this application includes: Acquire multi-source data for the region to be identified; the multi-source data includes enterprise fundamentals data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data; The multi-source data is preprocessed to obtain preprocessed data; the data preprocessing includes data association, spatial function analysis, and kernel density analysis. Based on the preprocessed data, target data indicators for the region to be identified are determined; the target data indicators include data indicators for innovation-driven dimensions, industrial collaboration dimensions, green development dimensions, talent agglomeration dimensions, and spatial efficiency dimensions. Based on the target data indicators of the region to be identified, the innovation-driven intensity index of the region to be identified is determined; The innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth of the region to be identified are input into the innovation-driven industrial park identification model to obtain the park type output by the innovation-driven industrial park identification model. The innovation-driven industrial park identification model is obtained by training the model based on the multi-dimensional features of the sample region and the park type label of the sample region. The multi-dimensional features include the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth.

[0008] According to one embodiment of this application, determining the target data indicators of the region to be identified based on the preprocessed data includes: Based on the number of patent applications, patent grants, patent citations, and the duration of enterprise existence in the preprocessed data, the patent vitality index of the region to be identified is determined. Based on the company’s annual R&D expenditure and annual operating revenue in the preprocessed data, the R&D intensity of the region to be identified is determined. Based on the support and confidence of the upstream and downstream enterprise association rules in the preprocessed data, the industrial chain association degree of the region to be identified is determined. Based on the order release time of related enterprises in the industrial chain and the response time of supporting enterprises in the preprocessed data, the collaborative response speed of the region to be identified is determined. Based on the annual energy consumption and annual output value of enterprises in the preprocessed data, the energy consumption per unit output value of the region to be identified is determined. Based on the total annual carbon emissions and annual output value of enterprises in the preprocessed data, the carbon emission intensity of the area to be identified is determined. Based on the educational background information of the employees in the enterprise and the total number of employees in the enterprise in the preprocessed data, the talent density of the area to be identified is determined; Based on the trajectory entropy values ​​of personnel inflow and outflow in the area to be identified in the preprocessed data, the talent mobility vitality of the area to be identified is determined; the trajectory entropy value is used to measure the dispersion and activity of talent mobility. Based on the distribution vector of points of interest and the industrial planning function vector of the region to be identified in the preprocessed data, the industrial function matching degree of the region to be identified is determined. Based on the enterprise spatial density and spatial weight matrix in the preprocessed data, the industrial spatial agglomeration degree of the region to be identified is determined.

[0009] According to one embodiment of this application, determining the innovation-driven intensity index of the region to be identified based on target data indicators of the region to be identified includes: Each data indicator in the target data indicators of the region to be identified is multiplied by its corresponding indicator weight; wherein, the indicator weight corresponding to each data indicator is determined based on the first weight and the second weight of the corresponding data indicator; the first weight is determined based on the entropy method, and the second weight is determined based on the comparison results of the importance between data indicators; The results of each multiplication are summed to obtain the innovation-driven intensity index of the region to be identified.

[0010] According to one embodiment of this application, the initial model corresponding to the innovation-driven industrial park identification model is a fusion model based on graph convolutional neural network and long short-term memory network.

[0011] According to one embodiment of this application, the data association is used to link an enterprise's business registration data, patent data, logistics data and talent data to form a relationship network of enterprise-patent-logistics-talent.

[0012] According to one embodiment of this application, the process of preprocessing the multi-source data is performed by one or more edge computing nodes; each edge computing node transmits the preprocessed data obtained from the data preprocessing to the cloud; the cloud determines the target data indicators of the region to be identified based on the preprocessed data.

[0013] The innovation-driven industrial park identification system based on multi-source data fusion according to the second aspect of this application includes: The acquisition module is used to acquire multi-source data of the area to be identified; the multi-source data includes enterprise fundamentals data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data; The preprocessing module is used to preprocess the multi-source data to obtain preprocessed data; the data preprocessing includes data association, spatial function analysis, and kernel density analysis. The first determining module is used to determine the target data indicators of the region to be identified based on the preprocessed data; the target data indicators include data indicators of innovation-driven dimension, industrial collaboration dimension, green development dimension, talent agglomeration dimension and spatial efficiency dimension; The second determining module is used to determine the innovation-driven intensity index of the region to be identified based on the target data indicators of the region to be identified. The identification module is used to input the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technological reserve depth of the region to be identified into the innovation-driven industrial park identification model, and obtain the park type output by the innovation-driven industrial park identification model. The innovation-driven industrial park identification model is obtained by training the model based on the multi-dimensional features of the sample region and the park type label of the sample region. The multi-dimensional features include the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technological reserve depth.

[0014] An electronic device according to a third aspect of this application includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the innovation-driven industrial park identification method based on multi-source data fusion as described above.

[0015] According to the fourth aspect of the present application, the storage medium is a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the innovation-driven industrial park identification method based on multi-source data fusion as described above.

[0016] A computer program product according to a fifth aspect of this application includes a computer program that, when executed by a processor, implements the innovation-driven industrial park identification method based on multi-source data fusion as described above.

[0017] The above-described one or more technical solutions in the embodiments of this application have at least the following technical effects: By acquiring multi-source data, including enterprise fundamentals, technological innovation, supply chain dynamics, talent ecosystem, green development, and geospatial data, for the region to be identified; and further preprocessing the multi-source data through data association, spatial function analysis, and kernel density analysis, preprocessed data is obtained. Because multi-source data is acquired and integrated, including enterprise fundamentals, technological innovation, supply chain dynamics, talent ecosystem, green development, and geospatial data, the problem of single indicators being unable to characterize core elements such as technological innovation and industrial synergy can be solved. Furthermore, because spatial function analysis and kernel density analysis are performed on the multi-source data, the clustering characteristics of innovation elements can be quantified, addressing the problem of missed identification of implicit innovation clusters caused by administrative divisions. Based on the preprocessed data after data association, the region to be identified can be determined in terms of innovation-driven development, industrial synergy, green development, and talent... Using data indicators from the agglomeration and spatial efficiency dimensions as target data indicators, the resilience of the industrial chain in the park can be assessed through the target data indicators of the industrial synergy dimension. Furthermore, based on the target data indicators of the region to be identified, the innovation-driven intensity index of the region can be accurately determined. Moreover, by pre-training a model based on the multi-dimensional characteristics of the sample region, such as the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technological reserve depth, combined with the park type labels of the sample region, an innovation-driven industrial park identification model is obtained. This allows inputting the innovation-driven intensity index of the region to be identified into the model to obtain the park type output by the model. Because an evaluation model matching the connotation of innovation-driven intensity has been created and can support new evaluation dimensions such as technological radiation capacity and technological reserve depth, policy resources can be precisely targeted to high-potential parks. Therefore, the accuracy of identifying innovation-driven industrial parks can be effectively improved.

[0018] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is one of the flowcharts illustrating the innovation-driven industrial park identification method based on multi-source data fusion provided in this application embodiment.

[0021] Figure 2This is the second flowchart of the innovation-driven industrial park identification method based on multi-source data fusion provided in the embodiments of this application.

[0022] Figure 3 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] It should be noted that the existing park evaluation system mainly relies on traditional indicators such as the number of enterprises and the scale of output value, and lacks a precise description of the core elements of innovation-driven development, such as the concentration of technological innovation, the degree of industrial chain synergy, the level of green production, and the talent agglomeration effect.

[0025] Furthermore, relying on annual statistical data makes it impossible to capture real-time dynamic changes in technological breakthroughs, talent mobility, and environmental performance within the park. Simultaneously, the lack of edge computing and real-time data access capabilities results in early warning response times exceeding 24 hours, failing to meet the agile management needs of innovation-driven industrial parks. Therefore, dynamic monitoring also suffers from lag.

[0026] Based on this, this application proposes a method and system for identifying innovation-driven industrial parks based on multi-source data fusion.

[0027] Figure 1 This is one of the flowcharts illustrating the innovation-driven industrial park identification method based on multi-source data fusion provided in this application embodiment, such as... Figure 1 As shown, the method for identifying innovation-driven industrial parks based on multi-source data fusion includes: Step 110: Obtain multi-source data for the region to be identified; multi-source data includes enterprise fundamentals data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data.

[0028] Step 120: Perform data preprocessing on the multi-source data to obtain preprocessed data; data preprocessing includes data association, spatial function analysis, and kernel density analysis.

[0029] Step 130: Based on the preprocessed data, determine the target data indicators for the region to be identified; the target data indicators include data indicators for innovation-driven dimensions, industrial collaboration dimensions, green development dimensions, talent agglomeration dimensions, and spatial efficiency dimensions.

[0030] Step 140: Based on the target data indicators of the region to be identified, determine the innovation-driven intensity index of the region to be identified.

[0031] Step 150: Input the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth of the region to be identified into the innovation-driven industrial park identification model to obtain the park type output by the innovation-driven industrial park identification model. The innovation-driven industrial park identification model is obtained by training the model based on the multi-dimensional features of the sample region and the park type label of the sample region. The multi-dimensional features include the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth.

[0032] It should be noted that the execution subject of the innovation-driven industrial park identification method based on multi-source data fusion provided in this application embodiment can be a server or computer equipment. The computer equipment can be, for example, a mobile phone, tablet computer, laptop computer, handheld computer, vehicle electronic equipment, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc.

[0033] It should be noted that all actions involving the acquisition of signals, information, or data in this application are carried out in compliance with the relevant data protection laws and regulations of the locality and with authorization from the owner of the relevant device.

[0034] It should be noted that in this application, an area that needs to be identified as an innovation-driven industrial park can be pre-defined as the area to be identified.

[0035] Furthermore, multi-source data can be acquired for the area to be identified. This multi-source data can include heterogeneous data such as fundamental data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data.

[0036] Specifically, when acquiring fundamental data about enterprises, information such as the enterprise's registered address, equity structure, and establishment time can be crawled through the enterprise credit information disclosure system; financial statement summaries of enterprises (which may include revenue and R&D investment ratios) can be obtained by relying on open data interface services; and a spatiotemporal attribute database of enterprises can be built on a quarterly basis (including unified social credit code → time series data).

[0037] When acquiring technological innovation data, the patent database application programming interface (API) can be called to extract the patent application / grant date, international patent classification number and citation relationship; the full text of the patent can be obtained through the patent search platform, and technical keywords (such as solid-state battery-sulfide electrolyte) can be extracted using pre-trained language representation models (such as bidirectional encoder representations from transformers (BERT) models), thereby constructing a three-level technology tagging system.

[0038] When acquiring dynamic supply chain data, it is possible to connect to a VAT invoice database (such as the State Taxation Administration interface) to extract transaction amounts and commodity codes between enterprises; obtain the Global Positioning System (GPS) tracks of trucks through a geographic information platform (which may include latitude, longitude, and timestamps), and use the Apriori algorithm to mine upstream and downstream association rules (support ≥ 0.3, confidence ≥ 0.5). The Apriori algorithm is an association rule mining algorithm used to discover frequent itemsets in large-scale transaction data and generate association rules in the form of "if A occurs, then B is also likely to occur".

[0039] Among them, the support of the upstream and downstream enterprise association rule can be obtained by (number of associated enterprise pairs ÷ total number of enterprise pairs), and the confidence of the upstream and downstream enterprise association rule can be obtained by (number of related transactions ÷ number of unilateral transactions).

[0040] Total number of enterprise pairs: Based on enterprises within the park with supply chain transaction potential (screened from business registration data), all possible upstream and downstream orderly enterprise pairs are generated (excluding invalid cross-industry pairs).

[0041] Related company pairs: This involves identifying upstream and downstream companies with actual transactions based on VAT invoices (seller-buyer) and logistics tracking (shipper-receiving company). VAT invoices can be obtained through authorized third-party service platforms, provided by the companies themselves, or compiled and provided by the tax authorities.

[0042] Unilateral transaction count: For a specific upstream company A, count the total number of transactions between it and all downstream companies (based on all invoices or shipping records of A).

[0043] Related transaction count: Counts the actual number of transactions between specific upstream and downstream pairs (such as A→B) (based on invoices or delivery records from A to B).

[0044] When acquiring talent ecosystem data, we can collaborate with LinkedIn to obtain anonymized professional information (such as education level and career path), and combine it with the talent pool of local human resources and social security departments to calculate the density of highly educated talents (Master's degree + talent ratio) and the talent mobility entropy value (calculated as: -Σ(p_i×lnp_i, where p_i is the industry flow ratio).

[0045] When acquiring green development data, IoT sensors in the park (such as sensors that detect indicators such as electricity and gas consumption and pollutant concentrations) can be connected to calculate carbon emissions according to the IPCC (Intergovernmental Panel on Climate Change) Greenhouse Gas Accounting Guidelines; data from the Ministry of Ecology and Environment's monitoring platform can be synchronized to construct a time series of energy consumption per unit of Gross Domestic Product (GDP) (which may be referred to as energy consumption per unit of output in this application) and carbon emission intensity.

[0046] When acquiring geospatial data, Landsat 8 remote sensing imagery (30 MHz resolution) can be downloaded from the United States Geological Survey (USGS). The Normalized Difference Vegetation Index (NDVI) and Normalized Difference Built-up Index (NDBI) can be calculated using a remote sensing image processing platform (such as Environment for Visualizing Images (ENVI)). Points of Interest (POIs) such as R&D centers and production bases can be obtained by calling the OpenStreetMap API. Spatial information is then aggregated using H3 grid coding (8 levels of resolution, 2.4 km accuracy). H3 is an open-source hexagonal hierarchical spatial indexing system.

[0047] Furthermore, one or more of the following data preprocessing techniques can be applied to multi-source data: edge cleaning, privacy and semantic processing, encrypted association, spatial function parsing, and spatial structuring, thereby obtaining preprocessed data.

[0048] It should be noted that the multi-source data in this application can be obtained through one or more different edge computing nodes. The edge computing nodes can be IoT devices, industrial sensors, edge servers, smart gateways, terminal devices (such as cameras, smart terminals), etc., which are close to the data source.

[0049] Specifically, for each edge computing node, outlier filtering can be performed on the high-frequency dynamic data it acquires (such as GPS trajectory and data collected by energy consumption sensors). Specifically, logistics speed and energy consumption data that clearly exceed reasonable ranges can be removed.

[0050] Furthermore, after completing the necessary data filtering, edge computing nodes can locally associate data using unified credit codes and H3 grid serial numbers (Identification, ID). Specifically, heterogeneous data such as enterprise business registration data (e.g., registered address and unified credit code), patent data (IPC classification number), logistics data (GPS coordinates), and talent data (migration trajectory) can be associated to form a four-dimensional relationship network of enterprise-patent-logistics-talent, thereby completing data association.

[0051] Specifically, this application can achieve data association through dual indexing, which may include: enterprise dimension: associating enterprise business registration, patent, and financial data through unified social credit code; spatial dimension: converting enterprise registration address and logistics node latitude and longitude into H3 grid IDs to achieve geospatial alignment.

[0052] It should be noted that sensitive fields in the linked data (such as revenue and salary) need to be encrypted at the field level, for example, using the AES-256 encryption algorithm. The linked results only retain the de-identified feature vectors.

[0053] Furthermore, for non-sensitive data scenarios (such as publicly available patent texts and geographic POI data), edge computing nodes can use field-level desensitization (such as hiding the last two decimal places of corporate revenue) + Laplace noise (ε=0.8) to perform privacy and semantic processing on sensitive fields in related data.

[0054] For sensitive data scenarios (such as enterprise transaction data and tax information), a federated learning architecture can be used to train cross-source models with privacy protection between each edge computing node and the cloud.

[0055] Specifically, when multi-source data comes from multiple independent data sources (such as tax authorities, logistics companies, and university talent pools), feature fusion needs to be achieved without sharing the original data. Traditional centralized training carries the risk of data leakage (such as the external transmission of corporate tax data).

[0056] Therefore, this application adopts a federated learning architecture: each edge computing node trains a model (such as a logistics trajectory classifier or a tax credit scoring model) locally using its own data for each data source (such as logistics company A or tax department B), and only uploads encrypted gradient parameters (rather than the original data) to the cloud.

[0057] In this process, after local training at each data source, the cloud can directly average the model parameters uploaded by each edge computing node to obtain global parameters and distribute them to each edge computing node.

[0058] Therefore, high-frequency dynamic data is first cleaned by edge computing nodes to reduce invalid data transmitted to the cloud; the cleaned data is indexed by a unified credit code and spatial coordinates to build a cross-domain relationship network (such as the "enterprise A-patent X-logistics node Y" association edge); each data source is trained locally on the edge computing nodes based on the association features (such as logistics enterprises training supply chain stability models and patent institutions training technology maturity models), and then aggregated in the cloud to obtain a global recognition model (such as the innovation-driven industrial park recognition model).

[0059] It should be noted that the privacy protection processing in this application can be carried out by differential privacy or blockchain notarization to process sensitive data.

[0060] Specifically, for differential privacy (Laplace noise): Privacy protection principle: By adding carefully designed noise (such as Laplace noise) to the original data, the characteristics of individual data are blurred, while the overall statistical regularity is preserved, preventing the inference of specific individual information from the statistical results.

[0061] This application is used to add Laplace noise with ε=0.8 when processing talent information (such as mobility trajectories) and supply chain transaction data. For example, when calculating "density of highly educated talent", the noise can obscure the number of talents in a specific enterprise, ensuring that the true talent structure of a particular enterprise cannot be located, while not affecting the validity of density statistics at the park level.

[0062] For blockchain-based evidence storage: Privacy protection principle: By leveraging the immutability and traceability of blockchain, the entire process of sensitive data processing (such as encryption, association, and computation operations) is recorded, enabling transparent auditing of data processing behavior and preventing data from being illegally tampered with or misused.

[0063] This application is used to record the processing of sensitive information such as patent text technical details and supply chain transaction data (e.g., encryption algorithm calls and association rule generation). Blockchain-based evidence storage allows for the traceability of the entity and time of each operation, ensuring data processing compliance, preventing unauthorized modifications, and guaranteeing data integrity and processing credibility.

[0064] It should be noted that during data preprocessing, each edge computing node in this application can also spatially structure the data. Specifically, the S2 Geometry spatial index (15 levels of resolution, 90-meter accuracy) can be used to generate an innovation element clustering heat matrix through kernel density estimation (KDE) to identify spatial hotspot areas. The S2 Geometry spatial index is an open-source hierarchical spatial indexing system based on a sphere (rather than a plane).

[0065] The technical implementation of spatial structuring in this application may include: Coordinate transformation: Use S2 Geometry (level 15, 90-meter accuracy) to convert company addresses, GPS coordinates, etc. into a unified grid ID to achieve spatial alignment; Density calculation: Generate a heat matrix of innovation element agglomeration (such as enterprise density within a grid) through kernel density estimation (KDE). Spatial correlation: Based on grid ID matching and correlation data (such as grid proximity between enterprises and logistics nodes), spatial index calculation is supported.

[0066] By structuring the space, geographic data is transformed into computable spatial features to support the calculation of indicators such as "agglomeration of innovative elements" and "spatial synergy".

[0067] Furthermore, during data preprocessing, each edge computing node in this application can also perform semantic matching on the data. Specifically, it can generate 768-dimensional text vectors based on the SciBERT model, efficiently retrieve terms from the "Strategic Emerging Industries Classification" using the FAISS engine, and enhance the cross-modal matching accuracy of enterprises and policies (threshold ≥ 0.65) by integrating the CLIP multimodal model. FAISS is an open-source, high-efficiency similarity search library specifically designed for fast retrieval tasks involving large-scale vector data. CLIP is a multimodal pre-trained model that encodes text and images into vectors separately and aligns text and image vectors in the vector space through contrastive learning training.

[0068] By performing semantic matching on the data, the semantic gap between text and categorized data can be bridged, improving the accuracy of matching enterprises with policies.

[0069] Specifically, it can improve the accuracy of indicators that rely on text parsing, such as the patent vitality index (precisely matching the technology field), the industrial function matching degree (the degree of fit between enterprises and planning functions), and the industrial chain correlation degree (assisting in exploring upstream and downstream correlations).

[0070] Furthermore, during the data preprocessing process, each edge computing node in this application can also perform spatial function analysis on the data.

[0071] The core purpose of spatial function analysis in this application is to identify spatial function labels (such as "innovation and R&D core" and "industrial development zone").

[0072] Specifically: Spatial function analysis uses the random forest algorithm to train labeled samples ("R&D area / production area" grid) and outputs function labels; at the same time, based on the identified functional zones and enterprise spatial density data, the Moran index is used to further quantify the spatial clustering characteristics of these functional areas (such as whether the enterprise distribution in the "innovation and R&D core" shows significant clustering).

[0073] The relevant indicators for spatial function analysis in this application include: Industrial spatial agglomeration: By analyzing spatial data such as the density of enterprise geographical distribution and the concentration of industries in industrial parks, the rationality of the spatial layout of regional industrial clusters can be analyzed, which can be used to identify industrial hotspots and spatial coordination bottlenecks.

[0074] Industry function matching degree: assess the degree of fit between the technology field and supply chain role of enterprises in the region and the regional functional positioning (such as the matching degree of the leading industries in the high-tech zone), and assist in the optimization of spatial planning.

[0075] It should be noted that this application also provides an industrial synergy assessment model, whose related indicators include the degree of industrial chain correlation, the speed of synergy response, and the vitality of talent mobility.

[0076] Among them, the industrial chain correlation degree: based on the supply relationship and technology correlation edge (such as patent citation and supply chain transaction data) between enterprises in the knowledge graph, the tightness of the upstream and downstream of the industrial chain is quantified, and key node enterprises are identified.

[0077] Collaborative Response Speed: By using time-series data such as supply chain logistics data and technical cooperation response time, the efficiency of a company's response to technical collaboration or supply chain fluctuations is assessed, which is used for dynamic collaborative capability evaluation.

[0078] Talent Mobility Dynamics: Analyze the flow of talent across enterprises (such as resume data and social security migration records) to quantify the efficiency of talent resource flow and the speed of technical knowledge dissemination in the industry, reflecting the talent base for industry collaboration.

[0079] The industrial synergy assessment in this application is based on the degree of industrial chain linkage and further integrates the dimension of technological linkage to form a more comprehensive assessment of industrial synergy.

[0080] The degree of industrial chain linkage is a core component of industrial synergy assessment: The "high-frequency association rules (support ≥ 0.25) of supply chain transaction data mined by the Apriori algorithm" in the industrial collaboration assessment is essentially a calculation of the degree of industrial chain correlation (based on transaction data such as value-added tax invoices and logistics trajectories).

[0081] In addition to the degree of industrial chain linkage, the industrial synergy assessment has added a dimension of data on technological linkage: An additional patent citation relationship is introduced to construct a technology association network, and a technology dependence index (TDI) (such as the proportion of times Company A's patents are cited by Company B to the total number of citations by Company B) is calculated to measure the technological synergy between companies.

[0082] Ultimately, industrial synergy is a weighted fusion of supply chain relevance (transaction level) and technology dependence index (technology level) (e.g., transaction relevance weight 60% + technology relevance weight 40%). Therefore, the data scope not only includes supply chain transaction data but also patent citation data, which differs from single supply chain relevance data.

[0083] In this application, industrial synergy is used to quantify the efficiency of resource sharing, technological complementarity, and collaboration among upstream and downstream enterprises in the industrial chain (such as patent cross-reference density and supply chain response speed), and is a core indicator for assessing the stability of the industrial chain.

[0084] The degree of synergy in the subsequent early warning triggering part is the degree of industrial synergy: the early warning focuses on the risks of the industrial chain (such as disruption, abnormal cooperation), while the degree of industrial synergy directly reflects the closeness of cooperation among enterprises within the industrial chain. Its sharp drop (such as Z-score < -1.5) is more likely to trigger an early warning; the degree of spatial synergy focuses on the matching of geographical clusters (such as the functional fit between enterprises and parks), and is more used for regional layout optimization, and is not a core early warning indicator.

[0085] The calculation method for industrial synergy is a weighted fusion of the degree of industrial chain linkage (transaction level) and the technology dependence index (technology level). The specific logic is as follows: At the transaction level: Based on the supply chain network, the Apriori algorithm is used to mine high-frequency transaction association rules (support ≥ 0.25), corresponding to the core data of the industrial chain association degree; From a technical perspective: a technology association network is constructed based on patent citation relationships, and the technology dependence index (TDI, i.e., the proportion of enterprise patent citations) is calculated.

[0086] The two are weighted and integrated (the weights can be determined by improving the entropy method, etc.) to form a comprehensive indicator of industrial synergy.

[0087] The degree of industrial chain linkage, the speed of collaborative response, and the vitality of talent mobility are supporting indicators associated with the industrial synergy assessment model. They are used to more comprehensively depict the multi-dimensional characteristics of industrial synergy (such as the tightness of the industrial chain, the efficiency of dynamic response, and the talent base), and are not direct components of the industrial synergy degree indicator.

[0088] Therefore, by performing data preprocessing, each edge computing node can obtain preprocessed data including a company's patent application volume, patent grant volume, patent citation volume, company duration, company's annual R&D expenditure and annual operating revenue, support and confidence of upstream and downstream enterprise association rules, order release time of related enterprises in the industrial chain and response time of supporting enterprises, company's annual energy consumption and annual output value, company's annual carbon emissions and annual output value, educational information of company employees and total number of company employees, trajectory entropy values ​​of personnel inflow and outflow in the area to be identified, interest point distribution vector and industrial planning function vector of the area to be identified, and company spatial density and spatial weight matrix. The preprocessed data can be encrypted and transmitted to the cloud through the Message Queuing Telemetry Transport (MQTT) protocol.

[0089] Furthermore, edge computing nodes can periodically import new data and retrain the model. Specifically, at the end of each month, newly added patent, logistics, and talent data can be aggregated, and the model can be fully trained using both the new data and historical full data. Moreover, the edge nodes only handle data transmission and do not perform caching or feature generation tasks, reducing iterative maintenance costs.

[0090] Furthermore, the cloud can determine data indicators for dimensions such as innovation-driven development, industrial collaboration, green development, talent aggregation, and spatial efficiency based on the various data in the preprocessed data, and use these as target data indicators. In this application, the target data indicators can simultaneously include data indicators for dimensions such as innovation-driven development, industrial collaboration, green development, talent aggregation, and spatial efficiency.

[0091] It should be noted that the data indicators for the innovation-driven dimension in this application may include the patent vitality index and R&D intensity; the data indicators for the industrial synergy dimension may include the degree of industrial chain linkage and the speed of collaborative response; the data indicators for the green development dimension may include energy consumption per unit of output and carbon emission intensity; the data indicators for the talent agglomeration dimension may include talent density and talent mobility; and the data indicators for the spatial efficiency dimension may include the degree of industrial function matching and the degree of industrial spatial agglomeration.

[0092] Furthermore, the cloud can perform calculations on each data indicator in the target data metrics, along with its corresponding weight, to obtain the result. Then, the results of these calculations are summed, and the sum is used as the innovation-driven strength index for the region to be identified.

[0093] It should be noted that after calculating the innovation-driven intensity index of the region to be identified, it is still necessary to train and apply the intelligent classification model. The core reason is that the innovation-driven intensity index (NPPI) of the region to be identified is only a comprehensive quantitative result, while intelligent classification can upgrade from numerical evaluation to refined feature recognition and dynamic management by deeply mining multi-dimensional dynamic features, breaking through the limitations of the single value of NPPI and capturing multi-dimensional dynamic correlation features.

[0094] Specifically, there are: 1. The NPPI is a weighted sum of 10 data indicators, which makes it difficult to reflect the deep characteristics of a park in the "enterprise-technology-geography" triad (such as the time decay of patent citations and the impact of logistics distance on the supply chain). Intelligent classification models (such as the Transformer time series graph) can capture these correlation patterns that change over time and space by constructing dynamic relationship graphs and introducing spatiotemporal decay factors (such as the weight of patent citations from 2 years ago dropping to 0.5). For example, it can identify the complementarity of innovation elements between "potential" parks and surrounding enterprises (Moran index > 0.3), which cannot be directly reflected by the NPPI value.

[0095] 2. Achieve refined classification and match differentiated evaluation standards: While NPPI values ​​only reflect overall performance, intelligent classification, through a multi-task learning framework (joint classification and indicator prediction) and combined with segmentation threshold rules (e.g., leading parks require NPPI ≥ industry average + 1.5σ and patents cited across 3 provinces, where σ represents the standard deviation among multiple NPPIs), accurately classifies parks into "leading / growth / potential" categories. This classification not only relies on NPPI but also integrates derived characteristics such as technological reserves (e.g., growth parks need to cover the first 3 generations of technology) and radiation capacity, solving the problem that NPPI cannot distinguish between parks with "same scores but different characteristics" (e.g., two parks with the same NPPI, one being technologically advanced and the other talent-intensive).

[0096] 3. Support dynamic and flexible early warning, and enhance proactive risk identification: While NPPI changes can reflect overall trends, they are difficult to pinpoint the type of risk (e.g., supply chain risk vs. technology gap risk). Intelligent classification models can trigger targeted warnings by linking derived indicators (e.g., deviation from the technology field, cluster center distance > 0.8) with warning rules. For example, a "strategic drift" warning requires both technology deviation and a coordination Z-score < -1.5. This multi-condition combination judgment relies on the classification model's deep integration of features, rather than being achieved by a single NPPI fluctuation.

[0097] It should be noted that this application can pre-designate a certain number of regions as sample areas and obtain the multidimensional characteristics and park type labels of each sample area. Among them, the multidimensional characteristics of the sample areas may include the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth.

[0098] The park type labels in this application can include leading, growth, and potential types. The park type label for each sample park can be comprehensively determined based on information such as the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and depth of technology reserves. For example, it can be determined manually or through other feasible methods.

[0099] For example, the following threshold rule can be used for determination: Leading type: NPPI ≥ industry average + 1.5σ, and core dimensions have cross-regional radiation capabilities (e.g., technology patents are cited in more than 3 provinces). Growth type: NPPI∈[industry average -0.5σ, average +0.5σ], possessing disruptive technology reserves (covering the first 3 generations of layout in the technology field); Potential type: Top 10% in the industry for a single indicator (e.g., density of highly educated talents > 40%), and complementary to the innovation elements of surrounding enterprises (Moran Index > 0.3).

[0100] Furthermore, the multidimensional features and park type labels of the sample parks are used as training data to train the model. After training, an innovation-driven industrial park identification model is obtained that can identify the type of innovation-driven industrial park based on input information and output the park type of the corresponding region. The innovation-driven industrial park identification model is then deployed in the cloud.

[0101] Therefore, the cloud can determine the cross-regional radiation capacity and technological reserve depth of the region to be identified based on the preprocessed data transmitted by each edge computing node. Then, the innovation-driven intensity index, target data indicators, and their cross-regional radiation capacity and technological reserve depth of the region to be identified can be input into the innovation-driven industrial park identification model. The innovation-driven industrial park identification model will make predictions based on the input information and finally obtain the park type output by the innovation-driven industrial park identification model.

[0102] According to the multi-source data fusion-based innovation-driven industrial park identification method of this application embodiment, multi-source data such as enterprise fundamentals data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data of the area to be identified are acquired. Further, data preprocessing, including data association, spatial function analysis, and kernel density analysis, is performed on the multi-source data to obtain preprocessed data. Because multi-source data is acquired and integrated with enterprise fundamentals data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data, the problem that a single indicator cannot characterize core elements such as technological innovation and industrial synergy can be solved. Furthermore, because spatial function analysis and kernel density analysis are performed on the multi-source data, the clustering characteristics of innovation elements can be quantified, solving the problem of missed identification of implicit innovation clusters caused by administrative divisions. Based on the preprocessed data after data association, the innovation-driven dimension of the area to be identified can be determined. Data indicators from the dimensions of industrial synergy, green development, talent agglomeration, and spatial efficiency are used as target data indicators. The industrial synergy dimension's target data indicators can be used to assess the resilience of the industrial chain within a park. Furthermore, based on these target data indicators for the region to be identified, the innovation-driven intensity index of that region can be accurately determined. Moreover, by pre-training a model based on the multi-dimensional characteristics of the sample region, including its innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technological reserve depth, combined with the park type labels of the sample region, an innovation-driven industrial park identification model is obtained. Inputting the innovation-driven intensity index of the region to be identified into this model yields the park type output by the model. Because an evaluation model matching the connotation of innovation-driven intensity has been created and can support new evaluation dimensions such as technological radiation capacity and technological reserve depth, policy resources can be precisely targeted to high-potential parks. Therefore, the accuracy of identifying innovation-driven industrial parks can be effectively improved.

[0103] In one embodiment, based on preprocessed data, target data metrics for the region to be identified are determined, including: Based on the number of patent applications, patent grants, patent citations, and the duration of enterprise existence in the preprocessed data, the patent vitality index of the region to be identified is determined. Based on the annual R&D expenditure and annual operating revenue of enterprises in the preprocessed data, the R&D intensity of the region to be identified is determined. Based on the support and confidence of the rules for linking upstream and downstream enterprises in the preprocessed data, the degree of industrial chain linkage of the region to be identified is determined. Based on the order release time of related enterprises in the industrial chain and the response time of supporting enterprises in the preprocessed data, the collaborative response speed of the area to be identified is determined. Based on the annual energy consumption and annual output value of enterprises in the preprocessed data, determine the energy consumption per unit output value of the region to be identified. Based on the total annual carbon emissions and annual output value of enterprises in the preprocessed data, the carbon emission intensity of the area to be identified is determined. Based on the educational background of employees in enterprises and the total number of employees in enterprises in the preprocessed data, the talent density of the area to be identified is determined. Based on the trajectory entropy values ​​of personnel inflow and outflow in the preprocessed data of the area to be identified, the vitality of talent mobility in the area to be identified is determined; the trajectory entropy value is used to measure the dispersion and activity of talent mobility. Based on the distribution vector of points of interest and the industrial planning function vector of the area to be identified in the preprocessed data, the industrial function matching degree of the area to be identified is determined. Based on the enterprise spatial density and spatial weight matrix in the preprocessed data, the industrial spatial agglomeration degree of the region to be identified is determined.

[0104] Specifically, after receiving the preprocessed data uploaded by each edge computing node, the cloud in this application determines the various data indicators of the target data indicators in the area to be identified in the following manner: The patent vitality index for the region to be identified can be calculated by combining the number of patent applications, patent grants, patent citations, and the duration of enterprise existence from the preprocessed data. Specifically, the patent vitality index for each enterprise can be calculated as follows, based on the number of patent applications, patent grants, patent citations, and the duration of enterprise existence from the preprocessed data: A company's patent vitality index = (Number of patent applications in the past 3 years × 1.5 + Number of patents granted in the past 3 years × 1.2 + Historical patent citations) ÷ Company's duration of existence; The patent vitality index in this application integrates patent applications (timeliness), grants (quality), and citations (influence) from the past three years, weighted by time decay, making it more accurate than the traditional method of using patent counts.

[0105] Furthermore, it is possible to obtain the patent vitality weight set for each enterprise. The patent vitality weight of each enterprise can be set according to the enterprise size (such as revenue share) or industry weight (such as a 20% increase in the weight of strategic emerging industry enterprises).

[0106] Since the above method determines the patent vitality index of an individual enterprise, which measures the number of its patents, the frequency of citations, and the technological foresight (such as the proportion of "solid-state battery" patents), it needs to be combined with patent vitality weights to form park indicators.

[0107] Therefore, the patent vitality index of each enterprise can be weighted and aggregated according to enterprise size (such as revenue share) or industry weight (such as strategic emerging industry enterprises with a 20% weight increase) to obtain the overall patent vitality index of the region to be identified.

[0108] The R&D intensity of the region to be identified can be calculated by combining the annual R&D expenditure and annual operating revenue of enterprises in the preprocessed data. Specifically, the R&D intensity of each enterprise can be calculated based on its annual R&D expenditure and annual operating revenue in the preprocessed data as follows: A company's R&D intensity = annual R&D expenditure ÷ annual operating revenue; The R&D intensity determined by the above methods in this application can reflect the sustainability of innovation investment.

[0109] Additionally, it can obtain the R&D intensity weight set for each enterprise, and the R&D intensity weight of each enterprise can be set as the proportion of the enterprise's revenue to the total revenue of the park (i.e., the area to be identified).

[0110] Therefore, the R&D intensity of each enterprise can be weighted and averaged according to the proportion of the enterprise's revenue to the total revenue of the park (e.g., if the revenue of the leading enterprise accounts for 60%, then its R&D intensity weight is 60%) to obtain the overall R&D intensity of the area to be identified.

[0111] For the industrial chain correlation of the region to be identified, the weights set for the support and confidence of the association rules for upstream and downstream enterprises are obtained respectively. The support and confidence of the association rules for upstream and downstream enterprises in the preprocessed data are combined with their weights and weighted averaged. The result is determined as the industrial chain correlation of the region to be identified.

[0112] This application uses the Apriori algorithm to mine transaction support / confidence (a classic method in supply chain management) and quantify the closeness between upstream and downstream partners; the algorithm is relatively mature.

[0113] The collaborative response speed of the region to be identified can be characterized by the average of (order release time of related enterprises in the industrial chain (specifically, the core enterprises identified according to the actual situation) - response time of supporting enterprises) in the preprocessed data.

[0114] The collaborative response speed determined by the above method in this application can directly reflect the agility of the industrial chain.

[0115] The energy consumption per unit of output value (which may be referred to as energy consumption per unit of GDP in this application) for the region to be identified can be calculated using the annual energy consumption and annual output value of enterprises in the preprocessed data. Specifically, the energy consumption per unit of output value for each enterprise can be determined as follows: Energy consumption per unit of output of an enterprise = Total annual energy consumption (electricity, gas, etc. converted to standard coal) ÷ ​​Annual output (i.e. GDP of the same period).

[0116] The energy consumption per unit of output value determined by the above method in this application is based on data from energy consumption sensors and GDP statistics, which can reflect energy efficiency.

[0117] At the same time, the weight of energy consumption per unit output value for each enterprise can be obtained, which can be represented by the proportion of the enterprise's annual output value to the total output value of the park (i.e., the area to be identified).

[0118] Furthermore, the energy consumption per unit output of each enterprise and its weight can be weighted and averaged to determine the energy consumption per unit output of the region to be identified.

[0119] The carbon emission intensity of the area to be identified can be calculated using the total annual carbon emissions and annual output value of enterprises in the preprocessed data. Specifically, the carbon emission intensity of each enterprise can be calculated based on its total annual carbon emissions and annual output value in the preprocessed data, using the following method: A company’s carbon emission intensity = total annual carbon emissions of the company ÷ annual output value (i.e., GDP in the same period).

[0120] Furthermore, the proportion of each enterprise's annual output value in the park can be obtained as the weight of the corresponding enterprise. Then, the carbon emission intensity of each enterprise can be weighted and averaged according to the corresponding weight, and the result can be used as the carbon emission intensity of the area to be identified.

[0121] The carbon emission intensity in this application is calculated by connecting to the annual carbon emission data of enterprises and based on the total carbon emissions plus an emission factor library (composed of the weights corresponding to each enterprise), without the need for additional data collection.

[0122] To determine the talent density of the region to be identified, we can preprocess the data by analyzing the educational background of employees and the total number of employees in each company. Specifically, based on the educational background of employees and the total number of employees in the preprocessed data, we can calculate the talent density of each company using the following formula: The talent density of an enterprise = the number of employees with a master's degree or above ÷ the total number of employees in the enterprise; Simultaneously, the proportion of each company's total employees to the total number of employees in the park can be obtained, and this proportion can be used as the weight of the corresponding company. Furthermore, a weighted average can be calculated based on the talent density of each company and its corresponding weights, and the result can be used as the talent density of the area to be identified.

[0123] The density of highly educated talent determined by the above methods in this application can reflect the quality of talent, and the data comes from social security records, making the statistics convenient.

[0124] The vitality of talent mobility in the region to be identified can be calculated using the trajectory entropy values ​​of personnel inflow and outflow in the preprocessed data. These trajectory entropy values ​​can measure the dispersion and activity of talent mobility. Specifically, the vitality of talent mobility in the region to be identified can be calculated based on the trajectory entropy values ​​of personnel inflow and outflow in the preprocessed data, combined with the following method: Talent mobility vitality in the region to be identified = annual inflow / outflow trajectory entropy value of talent; The entropy value of the talent outflow trajectory can be determined in the following way: -Σ(p_i×lnp_i), where p_i is the proportion of talent flowing to different industries.

[0125] This application uses information entropy (-Σ(p_i×lnp_i)) to quantify the diversity of cross-industry flows, which can reflect the efficiency of knowledge spillover; and the entropy calculation method is relatively mature.

[0126] The matching degree of industry functions for the region to be identified can be calculated based on the distribution vector of points of interest and the vector of industry planning functions in the preprocessed data. Specifically, the cosine similarity between the distribution vector of points of interest and the vector of industry planning functions in the preprocessed data can be calculated, and this cosine similarity can be used as the matching degree of industry functions for the region to be identified.

[0127] The original data for the distribution vector of points of interest in the region to be identified is OpenStreetMap's POI data (including the coordinates and type labels of facilities such as R&D centers and production bases).

[0128] Method for generating point of interest distribution vectors: Clean and normalize POI types (e.g., merge "lab" and "technology_center" into "R&D"), calculate the weighted density of each type of POI according to the H3 grid (R&D center has a higher weight than production base), and form a vector with POI type as the dimension.

[0129] The original data for the industrial planning functional vector is the park planning drawings.

[0130] The method for generating industrial planning function vectors is as follows: the park planning map is vectorized and the coordinates are calibrated. The area ratio is calculated according to the planning function zones (R&D zone, production zone, etc.) (strategic functions such as R&D zone are weighted higher), forming a vector with function type as the dimension.

[0131] For example, for a certain high-tech zone: POI distribution vector: R&D category 0.68, production category 0.42, logistics category 0.25, service category 0.31.

[0132] Industrial planning functional vectors: R&D 0.72, production 0.35, logistics 0.18, service 0.25.

[0133] The cosine similarity is 0.92 (high matching degree), which verifies the high consistency between the actual POI distribution and the planned functions of the park.

[0134] The industrial spatial agglomeration degree of the region to be identified can be calculated using the enterprise spatial density and spatial weight matrix in the preprocessed data. Specifically, the Moran index I of the region to be identified can be calculated as the industrial spatial agglomeration degree of the region to be identified based on the enterprise spatial density and spatial weight matrix in the preprocessed data, combined with the following method: ; Where, x i x represents the spatial density of the i-th enterprise (e.g., the number of enterprises in grid H3 / grid area); j This represents the spatial density of the j-th enterprise; w ij The spatial weight matrix is ​​calculated through analysis of remote sensing imagery and Geographic Information System (GIS), where adjacent grids have a weight of 1 and non-adjacent grids have a weight of 0. This represents the average spatial density of enterprises within the area to be identified; N represents the total number of H3 grids into which the region to be identified is divided.

[0135] This application uses the Moran index (a classic method of geostatistics) to quantify the quality of enterprise clusters, which is more scientific than the quantity per unit area, and the data comes from enterprise coordinates + GIS, which is a mature technology.

[0136] Therefore, the target data indicators of the area to be identified can be obtained.

[0137] This application can accurately determine the target data indicators of the area to be identified by preprocessing various information in the data. This allows for the accurate calculation of the innovation-driven intensity index of the area to be identified, and subsequently, based on the innovation-driven intensity index, the type of industrial park in the area to be identified can be accurately determined. Therefore, it can effectively improve the accuracy of identifying innovation-driven industrial parks.

[0138] In one embodiment, the innovation-driven strength index of the region to be identified is determined based on target data indicators of the region to be identified, including: Each data indicator in the target data indicators of the region to be identified is multiplied by its corresponding weight. The weight of each data indicator is determined based on the first weight and the second weight of the corresponding data indicator. The first weight is determined based on the entropy method, and the second weight is determined based on the comparison of the importance of the data indicators. The results of each multiplication are summed to obtain the innovation-driven intensity index of the region to be identified.

[0139] Specifically, after obtaining the target data indicators of the region to be identified, the cloud application can obtain the indicator weights set for each data indicator in the target data indicators.

[0140] It should be understood that, in this application, for each data indicator, an objective weight is calculated using the improved entropy method as the first weight, and a subjective weight is determined using the analytic hierarchy process as the second weight. The first weight is then corrected using the second weight, and the corrected weight is the indicator weight of that data indicator.

[0141] Among them, the Improved Entropy Method is an objective weighting method that optimizes the traditional entropy method by addressing its shortcomings such as sensitivity to extreme values, inability to handle negative values, and large weight bias. It is widely used in fields such as multi-indicator comprehensive evaluation, performance evaluation, and regional economic analysis.

[0142] It should be noted that this application uses data discretization to measure the information content of the index, which avoids the problem of the traditional entropy method being overly sensitive to extreme values.

[0143] Specifically, this application can normalize each data indicator in the target data indicators to eliminate differences in dimensions, thereby achieving data standardization.

[0144] For example, the standardized formula for the patent vitality index (a positive indicator) is: ; Among them, b ij Represents the standardized numerical value, max(x) j ) represents the maximum value of index j across all samples; min(x) j ) represents the minimum value of index j across all samples; x ij This represents the original value of sample i in index j.

[0145] The standardized formula for unit energy consumption (a negative indicator) is: ; The meanings of each symbol are the same as in the above formula.

[0146] Therefore, the entropy value can be calculated using the following formula: ; Among them, e j Let represent the information entropy of index j, and m represent the sample size.

[0147] Therefore, the objective weights can be determined using the following formula: ; in, This represents the objective weight of indicator j.

[0148] Additionally, the analytic hierarchy process (AHP) can be used to determine the second weight by comparing the importance of data indicators.

[0149] Specifically, the problem of subjective preference can be addressed by constructing a judgment matrix through pairwise comparisons of the importance of indicators by experts: Specifically, a judgment matrix can be constructed: experts score 10 indicators in the target data index using the "1-9 scale method." For example, if they believe that "patent vitality index" is 3 times more important than "R&D intensity," then matrix element a... 1,2 =3, a 2,1 =1 / 3.

[0150] Furthermore, consistency checks can be performed in the following ways: ; Where, λ max The largest eigenvalue of the judgment matrix is ​​represented by , n represents the matrix order (e.g., fixed at 10), CI represents the consistency index, RI represents the random consistency index, and CR represents the consistency ratio.

[0151] When CR < 0.1, the judgment matrix passes the consistency check.

[0152] Then, the judgment matrix is ​​decomposed into eigenvalues, normalized, and the subjective weights are obtained and used as the second weights. .

[0153] Furthermore, a linear weighting method can be used to combine the two weights, specifically: ; in, This represents the weight of data indicator j, which can also be further expressed as... α represents the linear weighting coefficient, which can be set and adjusted according to the actual situation.

[0154] Therefore, based on various data indicators and their indicator weights The Innovation-Driven Intensity Index (NPPI) for the region to be identified is determined using the following formula: NPPI .

[0155] This application can accurately determine the innovation-driven intensity index of the region to be identified based on target data indicators, and then accurately determine the park type of the region based on the innovation-driven intensity index. Therefore, it can effectively improve the accuracy of identifying innovation-driven industrial parks.

[0156] In one embodiment, the initial model corresponding to the innovation-driven industrial park identification model is a fusion model based on Graph Convolutional Network (GCN) and Long Short-Term Memory (LSTM). GCN is used to process static graph structures (such as enterprise-patent citation relationships and enterprise-park spatial proximity relationships) and extract node association features; LSTM is used to capture temporal dynamics (such as time changes in logistics trajectories and annual trends in patent applications).

[0157] Specifically, this application can train an initial model based on the multidimensional features of the sample area and the park type label of the sample area, thereby obtaining an innovation-driven industrial park identification model.

[0158] The training process includes: We labeled 5000+ enterprise samples and synthesized scarce sample balanced data using Generative Adversarial Networks (GANs). We employed cosine annealing learning rate scheduling (0.001→0.0001) and gradient normalization for stable training, improving the validation set accuracy to 85%+.

[0159] Specifically, when labeling 5000+ enterprise samples, it is necessary to conduct manual review and machine-assisted labeling based on the multidimensional characteristics of the enterprises (NPPI value, core indicator performance, industry influence, etc.) and in combination with preset classification standards (such as leading / growth / potential). The specific process and examples are as follows: The annotation process includes: Feature extraction: Extract key features of enterprises from preprocessed data, including total NPPI score, scores of various indicators (such as patent vitality and industrial chain relevance), cross-regional radiation capability (such as the number of provinces where patents are cited), and depth of technology reserves (such as the technology generations covered). Initial screening and labeling: Machine initial labeling is conducted based on the above threshold rules (e.g., companies with NPPI ≥ industry average + 1.5σ and whose patents are cited in more than 3 provinces are initially labeled as "leading type"). Manual review: Sample the initial bidding results (approximately 20%) and correct any errors in the bidding based on the company's annual report and industry report (e.g., removing companies that meet the NPPI standard but have significant compliance risks). Sample balancing: For scarce categories (such as "leading" categories which account for less than 5%), GANs are used to synthesize virtual samples that match the features to ensure a balanced proportion of samples in each category.

[0160] Self-supervised pre-training mechanism: Contrastive learning is used to generate negative samples, and node embedding is optimized using the InfoNCE loss function. The design employs a multi-scale negative sampling strategy: at the node level, some attributes of enterprise nodes are randomly replaced (e.g., business scope); at the edge level, technological cooperation edges between enterprises are randomly deleted / added; at the subgraph level, a virtual technology subgraph is generated (e.g., a set of enterprises in the "quantum computing" field). This improves the model's ability to classify long-tail enterprises (e.g., specialized and innovative "little giants").

[0161] Multi-task learning framework: The system jointly optimizes classification tasks (leading / growth / potential) and indicator prediction tasks (NPPI index, scores of various dimensions), sharing the underlying feature extraction layer. An adaptive task weight mechanism is designed to dynamically adjust weights based on the rate of change of the loss function for each task. For example, when the NPPI prediction error is large, the weight of the indicator prediction task is automatically increased; when the classification accuracy is stable, the weight of the classification task is reduced, focusing on improving indicator prediction accuracy.

[0162] It should be noted that when the sample size is sufficient (e.g., labeled samples > 10,000), this application may not use contrastive learning self-supervised pre-training and GAN sample synthesis. Instead, it can balance the data distribution by random oversampling (e.g., repeated sampling of scarce leading park samples) combined with simple data augmentation (e.g., semantic expansion of patent text, slight perturbation of logistics trajectory coordinates). The learning rate scheduling is simplified to a fixed learning rate (e.g., 0.001), thereby eliminating the need for complex parameter adjustments of cosine annealing and reducing the difficulty of operation.

[0163] In one embodiment, after determining the park type, the cloud in this application can also trigger an early warning, which can be implemented in the following way: Supply chain risk: Abnormal logistics trajectory (speed fluctuation >40% and lasting for 6 hours) + sharp drop in supplier NPPI (≥0.2 in 24 hours), orange alert; Technology gap risk: Deviation in technology field (cluster center distance > 0.8) + Z-score of synergy < -1.5, a new purple warning for strategic drift is added; The technology deviation (cluster center distance > 0.8) does not directly correspond to any specific indicator in the target data metrics mentioned above, but is a derived indicator based on technology vector space analysis. This indicator is derived from an algorithm based on semantic analysis of technical texts. It does not directly correspond to any type of statistical data, but rather generates technology vectors through semantic vectorization of patent texts (such as word embedding models); it groups industry technology vectors using clustering algorithms (such as K-means clustering), calculates the distance between the enterprise's technology vector and the center of its cluster, and is an intermediate output of the machine learning model. Green risk: Carbon emission intensity exceeds the EU's quota for the next 3 years, triggering a blue alert (aligning with international trade and economic rules).

[0164] It should be noted that the cloud can also optimize the strategy in this application. Specifically, the cloud can use industry experience to fix the early warning threshold (such as simplifying the supply chain risk triggering condition to abnormal logistics trajectory (speed fluctuation > 40% and lasting for 6 hours) + a sharp drop in the supplier innovation drive intensity index ≥ 0.2, and simplifying the green risk triggering condition to energy consumption per unit output value exceeding the industry average by 1.2 times), and push the optimized strategy back to the edge to update the detection model.

[0165] Furthermore, three-dimensional decision support and iterative optimization can be performed in the following ways: Digital twin construction: Integrating BIM models (park buildings and equipment) with remote sensing images, overlaying NPPI spatial heat maps and enterprise classification tags, constructing a three-dimensional dynamic twin that supports drill-down analysis of "space → enterprise → indicator" (such as clicking on the grid to view the list of leading enterprises and weak indicators in the region).

[0166] Policy Adaptation and Solution Generation: Connecting with local industry policy databases (subsidies, taxation, land policies), and using a rule engine to match enterprise characteristics (e.g., "high R&D intensity + green certification" → subsidy application channel), automatically generating an "Enterprise Upgrade Recommendation Report," which includes: Diagnosis of shortcomings in indicators (e.g., "R&D intensity is less than 30% of the industry average"); Policy compatibility list (e.g., "Article 3.2 of the 'Detailed Rules for Science and Technology Innovation Subsidies in XX City'"); Improve the approach (e.g., "jointly build laboratories with universities → apply for subsidies → increase R&D intensity").

[0167] Among them, the construction of digital twins and the automatic generation of policy adaptation reports are optional value-added components that can be deployed separately when users have high-level decision-making needs.

[0168] Iterative optimization mechanism: New data (patents, logistics, talent) is added monthly, automatically triggering incremental model learning (learning rate decays to 0.0005), and dynamically updating the NPPI index and enterprise classification; full retraining is performed every quarter to ensure timely evaluation.

[0169] ①Monthly dynamic optimization mechanism: 1. Data Synchronization: Automatically collects new data in areas such as patents, logistics, and talent every month to ensure that system data is aligned with the actual scenario in real time.

[0170] 2. Incremental learning trigger: Incremental learning is initiated based on new data (i.e., the entire model is not retrained, but only learned from the new data), while the learning rate is decayed to 0.0005 (by reducing the learning rate, the model is made more stable when learning new data, and the model parameters are avoided from fluctuating greatly due to the fluctuation of new data).

[0171] 3. Results Update: Real-time updates of NPPI (which may be a comprehensive evaluation index) and enterprise classification results.

[0172] ② Quarterly full retraining mechanism: 1. Periodic Retraining: The model is retrained quarterly based on all historical data (not just new data). This addresses long-term changes in data distribution (such as adjustments in industry policies or shifts in technological trends) and avoids errors accumulated from incremental learning. 2. Recalibrate model parameter weights: For example, based on quarterly industry reports, increase the weight of the "green patent" indicator in the NPPI to ensure that the assessment logic matches the current industry environment.

[0173] ③Application Objectives: 1. Balance between real-time performance and stability: Daily incremental learning quickly responds to short-term data changes, while quarterly full training adapts to long-term trends, forming a dual optimization mechanism of "dynamic fine-tuning + global calibration"; 2. Self-evolving evaluation system: Through data-driven model iteration, the NPPI index and enterprise classification results continuously align with the real industry situation, enhancing the practical application value of patented technologies.

[0174] Figure 2 This is the second flowchart illustrating the innovation-driven industrial park identification method based on multi-source data fusion provided in this application embodiment, as follows: Figure 2 As shown, in one embodiment, the method for identifying innovation-driven industrial parks based on multi-source data fusion includes: ① Multi-source data collection, which may include enterprise data, technology data, supply chain data, talent data, green data, and geographic data; ② Data preprocessing, which can be accomplished through edge cleaning + encrypted association + federated learning; ③ Construct a dynamic model cluster, which can consist of spatial analysis, industry collaboration, NPPI assessment, and other components; ④ Intelligent classification reasoning, which may include Goods and Services Tax Classification Code (GSTC) coding + LSTM time series + threshold calculation, etc. ⑤ Determine if the NPPI is greater than the threshold? ⑥ If so, then three-dimensional decision support will be provided, which can consist of digital twin + policy adaptation; ⑦ If not, optimize the model or supplement the data.

[0175] Based on the above embodiments, this application improves the evaluation accuracy through multi-source data fusion. Specifically, it solves the problem of misjudgment of single indicators by cross-domain data association modeling, and accurately portrays core elements such as technological innovation and industrial chain collaboration.

[0176] By accelerating response through dynamic monitoring, specifically based on an edge-cloud collaborative architecture, high-frequency data such as patents and logistics are captured in real time to achieve real-time early warning of innovation vitality and supply chain risks.

[0177] By using spatial analysis for precise positioning, the identification rate of implicit innovation carriers is improved. Specifically, remote sensing indices and kernel density analysis are used to quantify the spatial clustering characteristics of innovation elements, avoiding the omission of implicit clusters caused by administrative divisions.

[0178] By leveraging the NPPI index to support precise policy implementation and improve resource allocation efficiency, a four-dimensional evaluation system is constructed to distinguish between data correlation and causality, providing a scientific basis for policy formulation and avoiding misjudgments of pseudo-innovation.

[0179] The following describes the innovation-driven industrial park identification system based on multi-source data fusion provided in this application. The innovation-driven industrial park identification system based on multi-source data fusion described below can be referred to in correspondence with the innovation-driven industrial park identification method based on multi-source data fusion described above.

[0180] Furthermore, this application also provides an innovation-driven industrial park identification system based on multi-source data fusion.

[0181] The innovation-driven industrial park identification system based on multi-source data fusion includes: The acquisition module is used to acquire multi-source data of the area to be identified; the multi-source data includes enterprise fundamentals data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data; The preprocessing module is used to preprocess the multi-source data to obtain preprocessed data; the data preprocessing includes data association, spatial function analysis, and kernel density analysis. The first determining module is used to determine the target data indicators of the region to be identified based on the preprocessed data; the target data indicators include data indicators of innovation-driven dimension, industrial collaboration dimension, green development dimension, talent agglomeration dimension and spatial efficiency dimension; The second determining module is used to determine the innovation-driven intensity index of the region to be identified based on the target data indicators of the region to be identified. The identification module is used to input the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technological reserve depth of the region to be identified into the innovation-driven industrial park identification model, and obtain the park type output by the innovation-driven industrial park identification model. The innovation-driven industrial park identification model is obtained by training the model based on the multi-dimensional features of the sample region and the park type label of the sample region. The multi-dimensional features include the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technological reserve depth.

[0182] This application's innovation-driven industrial park identification system based on multi-source data fusion acquires multi-source data, including enterprise fundamentals, technological innovation, supply chain dynamics, talent ecosystem, green development, and geospatial data, for the area to be identified. It further performs data preprocessing, such as data association, spatial function analysis, and kernel density analysis, to obtain preprocessed data. Because it acquires and integrates multi-source data, including enterprise fundamentals, technological innovation, supply chain dynamics, talent ecosystem, green development, and geospatial data, it addresses the problem that single indicators cannot characterize core elements such as technological innovation and industrial synergy. Furthermore, by performing spatial function analysis and kernel density analysis on the multi-source data, it quantifies the clustering characteristics of innovation elements, resolving the problem of missed identification of implicit innovation clusters caused by administrative divisions. Based on the preprocessed data after data association, it can further determine the innovation-driven dimension and industrial synergy of the area to be identified. Using data indicators from four dimensions—simultaneous development, green development, talent aggregation, and spatial efficiency—as target data indicators, the industrial chain resilience of the park can be assessed through the target data indicators of the industrial synergy dimension. Furthermore, based on the target data indicators of the region to be identified, the innovation-driven intensity index of the region can be accurately determined. Moreover, by pre-training a model based on the multi-dimensional characteristics of the sample region, such as its innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technological reserve depth, combined with the park type labels of the sample region, an innovation-driven industrial park identification model is obtained. This allows inputting the innovation-driven intensity index of the region to be identified into the model to obtain the park type output by the model. Because an assessment model matching the connotation of innovation-driven intensity has been created and can support new evaluation dimensions such as technological radiation capacity and technological reserve depth, policy resources can be precisely targeted to high-potential parks. Therefore, the accuracy of identifying innovation-driven industrial parks can be effectively improved.

[0183] In one embodiment, the first determining module is specifically used for: Based on the company’s annual R&D expenditure and annual operating revenue in the preprocessed data, the R&D intensity of the region to be identified is determined. Based on the support and confidence of the upstream and downstream enterprise association rules in the preprocessed data, the industrial chain association degree of the region to be identified is determined. Based on the order release time of related enterprises in the industrial chain and the response time of supporting enterprises in the preprocessed data, the collaborative response speed of the region to be identified is determined. Based on the annual energy consumption and annual output value of enterprises in the preprocessed data, the energy consumption per unit output value of the region to be identified is determined. Based on the total annual carbon emissions and annual output value of enterprises in the preprocessed data, the carbon emission intensity of the area to be identified is determined. Based on the educational background information of the employees in the enterprise and the total number of employees in the enterprise in the preprocessed data, the talent density of the area to be identified is determined; Based on the trajectory entropy values ​​of personnel inflow and outflow in the area to be identified in the preprocessed data, the talent mobility vitality of the area to be identified is determined; the trajectory entropy value is used to measure the dispersion and activity of talent mobility. Based on the distribution vector of points of interest and the industrial planning function vector of the region to be identified in the preprocessed data, the industrial function matching degree of the region to be identified is determined. Based on the enterprise spatial density and spatial weight matrix in the preprocessed data, the industrial spatial agglomeration degree of the region to be identified is determined.

[0184] In one embodiment, the second determining module is specifically used for: Each data indicator in the target data indicators of the region to be identified is multiplied by its corresponding indicator weight; wherein, the indicator weight corresponding to each data indicator is determined based on the first weight and the second weight of the corresponding data indicator; the first weight is determined based on the entropy method, and the second weight is determined based on the comparison results of the importance between data indicators; The results of each multiplication are summed to obtain the innovation-driven intensity index of the region to be identified.

[0185] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions from the memory 330 to execute the following method: acquiring multi-source data of the area to be identified; the multi-source data includes enterprise fundamentals data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data. The multi-source data is preprocessed to obtain preprocessed data; the data preprocessing includes data association, spatial function analysis, and kernel density analysis. Based on the preprocessed data, target data indicators for the region to be identified are determined; the target data indicators include data indicators for innovation-driven dimensions, industrial collaboration dimensions, green development dimensions, talent agglomeration dimensions, and spatial efficiency dimensions. Based on the target data indicators of the region to be identified, the innovation-driven intensity index of the region to be identified is determined; The innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth of the region to be identified are input into the innovation-driven industrial park identification model to obtain the park type output by the innovation-driven industrial park identification model. The innovation-driven industrial park identification model is obtained by training the model based on the multi-dimensional features of the sample region and the park type label of the sample region. The multi-dimensional features include the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth.

[0186] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0187] In another aspect, embodiments of this application also provide a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program is implemented to perform the methods provided in the above embodiments, such as: acquiring multi-source data of a region to be identified; the multi-source data includes enterprise fundamentals data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data. The multi-source data is preprocessed to obtain preprocessed data; the data preprocessing includes data association, spatial function analysis, and kernel density analysis. Based on the preprocessed data, target data indicators for the region to be identified are determined; the target data indicators include data indicators for innovation-driven dimensions, industrial collaboration dimensions, green development dimensions, talent agglomeration dimensions, and spatial efficiency dimensions. Based on the target data indicators of the region to be identified, the innovation-driven intensity index of the region to be identified is determined; The innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth of the region to be identified are input into the innovation-driven industrial park identification model to obtain the park type output by the innovation-driven industrial park identification model. The innovation-driven industrial park identification model is obtained by training the model based on the multi-dimensional features of the sample region and the park type label of the sample region. The multi-dimensional features include the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth.

[0188] In another aspect, embodiments of this application also provide a computer program product, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to perform the methods provided in the above embodiments, such as: acquiring multi-source data of the area to be identified; the multi-source data includes enterprise fundamental data, technological innovation data, supply chain dynamic data, talent ecosystem data, green development data, and geospatial data; The multi-source data is preprocessed to obtain preprocessed data; the data preprocessing includes data association, spatial function analysis, and kernel density analysis. Based on the preprocessed data, target data indicators for the region to be identified are determined; the target data indicators include data indicators for innovation-driven dimensions, industrial collaboration dimensions, green development dimensions, talent agglomeration dimensions, and spatial efficiency dimensions. Based on the target data indicators of the region to be identified, the innovation-driven intensity index of the region to be identified is determined; The innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth of the region to be identified are input into the innovation-driven industrial park identification model to obtain the park type output by the innovation-driven industrial park identification model. The innovation-driven industrial park identification model is obtained by training the model based on the multi-dimensional features of the sample region and the park type label of the sample region. The multi-dimensional features include the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth.

[0189] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0190] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0191] Finally, it should be noted that the above embodiments are only used to illustrate this application and are not intended to limit this application. Although this application has been described in detail with reference to the embodiments, those skilled in the art should understand that various combinations, modifications, or equivalent substitutions of the technical solutions of this application do not depart from the spirit and scope of the technical solutions of this application.

Claims

1. A method for identifying innovation-driven industrial parks based on multi-source data fusion, characterized in that, include: Acquire multi-source data for the region to be identified; the multi-source data includes enterprise fundamentals data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data; The multi-source data is preprocessed to obtain preprocessed data; The data preprocessing includes data association, spatial function analysis, and kernel density analysis. Based on the preprocessed data, target data indicators for the region to be identified are determined; the target data indicators include data indicators for innovation-driven dimensions, industrial collaboration dimensions, green development dimensions, talent agglomeration dimensions, and spatial efficiency dimensions. Based on the target data indicators of the region to be identified, the innovation-driven intensity index of the region to be identified is determined; The innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technological reserve depth of the region to be identified are input into the innovation-driven industrial park identification model to obtain the park type output by the innovation-driven industrial park identification model; the innovation-driven industrial park identification model is obtained by training the model based on the multi-dimensional features of the sample region and the park type label of the sample region. The multidimensional features include the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and depth of technology reserves.

2. The method for identifying innovation-driven industrial parks based on multi-source data fusion according to claim 1, characterized in that, The step of determining the target data indicators for the region to be identified based on the preprocessed data includes: Based on the number of patent applications, patent grants, patent citations, and the duration of enterprise existence in the preprocessed data, the patent vitality index of the region to be identified is determined. Based on the company’s annual R&D expenditure and annual operating revenue in the preprocessed data, the R&D intensity of the region to be identified is determined. Based on the support and confidence of the upstream and downstream enterprise association rules in the preprocessed data, the industrial chain association degree of the region to be identified is determined. Based on the order release time of related enterprises in the industrial chain and the response time of supporting enterprises in the preprocessed data, the collaborative response speed of the region to be identified is determined. Based on the annual energy consumption and annual output value of enterprises in the preprocessed data, the energy consumption per unit output value of the region to be identified is determined. Based on the total annual carbon emissions and annual output value of enterprises in the preprocessed data, the carbon emission intensity of the area to be identified is determined. Based on the educational background information of the employees in the enterprise and the total number of employees in the enterprise in the preprocessed data, the talent density of the area to be identified is determined; Based on the trajectory entropy values ​​of personnel inflow and outflow in the area to be identified in the preprocessed data, the talent mobility vitality of the area to be identified is determined; the trajectory entropy value is used to measure the dispersion and activity of talent mobility. Based on the distribution vector of points of interest and the industrial planning function vector of the region to be identified in the preprocessed data, the industrial function matching degree of the region to be identified is determined. Based on the enterprise spatial density and spatial weight matrix in the preprocessed data, the industrial spatial agglomeration degree of the region to be identified is determined.

3. The method for identifying innovation-driven industrial parks based on multi-source data fusion according to claim 1, characterized in that, The process of determining the innovation-driven intensity index of the region to be identified based on target data indicators includes: Each data indicator in the target data indicators of the region to be identified is multiplied by its corresponding indicator weight; wherein, the indicator weight corresponding to each data indicator is determined based on the first weight and the second weight of the corresponding data indicator; the first weight is determined based on the entropy method, and the second weight is determined based on the comparison results of the importance between data indicators; The results of each multiplication are summed to obtain the innovation-driven intensity index of the region to be identified.

4. The method for identifying innovation-driven industrial parks based on multi-source data fusion according to claim 1, characterized in that, The initial model corresponding to the innovation-driven industrial park identification model is a fusion model based on graph convolutional neural networks and long short-term memory networks.

5. The method for identifying innovation-driven industrial parks based on multi-source data fusion according to claim 1, characterized in that, The data association is used to link a company's business registration data, patent data, logistics data, and talent data to form a relationship network of company-patent-logistics-talent.

6. The method for identifying innovation-driven industrial parks based on multi-source data fusion according to claim 1, characterized in that, The process of preprocessing the multi-source data is performed by one or more edge computing nodes; each edge computing node transmits the preprocessed data to the cloud; the cloud determines the target data indicators of the region to be identified based on the preprocessed data.

7. An innovation-driven industrial park identification system based on multi-source data fusion, characterized in that, include: The acquisition module is used to acquire multi-source data of the area to be identified; the multi-source data includes enterprise fundamentals data, technological innovation data, supply chain dynamics data, talent ecosystem data, green development data, and geospatial data; The preprocessing module is used to preprocess the multi-source data to obtain preprocessed data; The data preprocessing includes data association, spatial function analysis, and kernel density analysis. The first determining module is used to determine the target data indicators of the region to be identified based on the preprocessed data; the target data indicators include data indicators of innovation-driven dimension, industrial collaboration dimension, green development dimension, talent agglomeration dimension and spatial efficiency dimension; The second determining module is used to determine the innovation-driven intensity index of the region to be identified based on the target data indicators of the region to be identified. The identification module is used to input the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and technology reserve depth of the region to be identified into the innovation-driven industrial park identification model to obtain the park type output by the innovation-driven industrial park identification model; the innovation-driven industrial park identification model is obtained by training the model based on the multi-dimensional features of the sample region and the park type label of the sample region. The multidimensional features include the innovation-driven intensity index, target data indicators, cross-regional radiation capacity, and depth of technology reserves.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the innovation-driven industrial park identification method based on multi-source data fusion as described in any one of claims 1-6.

9. A storage medium, said storage medium being a non-transitory computer-readable storage medium, wherein a computer program is stored thereon, characterized in that, When executed by a processor, the computer program implements the innovation-driven industrial park identification method based on multi-source data fusion as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the innovation-driven industrial park identification method based on multi-source data fusion as described in any one of claims 1-6.

Images