Enterprise migration early warning method and system for safe business and stable business in park

Abstract

The invention relates to an enterprise migration early warning method and system oriented to safe business and stable business in a park, and the method comprises the steps: 1, collecting multi-source heterogeneous enterprise related data, and carrying out the processing of the data, and obtaining a key feature set; step 2, based on multi-dimensional features in the key feature set, performing clustering analysis on the key feature set to form a plurality of enterprise classifications; step 3, for each enterprise classification, respectively constructing a machine learning early warning model for outputting a migration risk score of the enterprise under the classification; 4, performing cross-classification cross validation on the plurality of machine learning early warning models, and optimizing enterprise classification division and / or model parameters according to a validation result; and step 5, acquiring feature data of the target enterprise in real time, calling the early warning model according to the classification to calculate a migration risk score, and giving out an early warning when the score exceeds a preset threshold value. According to the invention, through fusion of multi-dimensional information, objective quantitative evaluation and real-time dynamic early warning are realized, and the early warning precision and the operation efficiency of safe and stable parks are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence application technology, and in particular to a method and system for early warning of enterprise migration aimed at stabilizing business in industrial parks. Background Technology

[0002] In the current economic development process, industrial parks, as the core carriers of industrial agglomeration, are directly and significantly affected by the relocation behavior of enterprises. Timely and accurate early warning of enterprise relocation intentions, scientific assessment of relocation risks, and taking effective intervention measures have become key needs for park operators to ensure the stability of the industrial ecosystem and enhance their ability to attract and retain businesses. This is also an important issue in the context of digital operation services for smart parks.

[0003] Currently, mainstream enterprise migration early warning mechanisms in the industry mainly rely on monitoring a single data dimension. For example, they rely solely on single indicators such as monthly tax revenue and foreign investment. Operations personnel check relevant risk dynamic data through enterprise credit inquiry platforms such as Qichacha and tax systems, or internal data systems, and then judge whether the enterprise has migration risks based on personal experience. However, existing technologies have significant drawbacks: First, relying on data from only a single dimension makes judgments prone to outliers, directly affecting the accuracy of the judgment. Second, operations personnel rely on subjective judgment based on experience, lacking objective and unified evaluation standards, which inevitably leads to judgment bias and makes it difficult to adapt to the differences between enterprises in different industries, sizes, and regions. Third, the low update frequency of a single data dimension results in significant delays in risk alerts, failing to meet the park's need for timely intervention. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for early warning of enterprise migration in industrial parks, which aims to improve the accuracy of early warning and the operational efficiency of industrial parks in stabilizing and retaining businesses by integrating multi-dimensional information, achieving objective quantitative assessment and real-time dynamic early warning.

[0005] The above-mentioned objective of this invention is achieved through the following technical solutions: A method for early warning of enterprise relocation aimed at stabilizing business in industrial parks includes the following steps: Step 1: Collect multi-source heterogeneous enterprise-related data, and clean, normalize, and perform feature dimensionality reduction on the data to obtain a key feature set for enterprise migration risk assessment. Step 2: Based on the multi-dimensional features in the key feature set, perform cluster analysis on the enterprises corresponding to the key feature set to form multiple enterprise classifications; Step 3: For each of the enterprise categories, construct and train a corresponding machine learning early warning model to output the migration risk score of the enterprise under that category; Step 4: Perform cross-classification cross-validation on the multiple machine learning early warning models, and optimize the enterprise classification and / or model parameters based on the validation results; Step 5: Obtain the characteristic data of the target enterprise in real time, call the corresponding early warning model to calculate the migration risk score according to its category, and issue an early warning when the score exceeds the preset threshold.

[0006] The aforementioned technical solutions have enabled a shift from traditional early warning methods that rely on single data dimensions and human experience to an intelligent early warning model based on multi-dimensional data analysis, automatic modeling, and real-time computation. Through systematic data processing, scientific enterprise classification, targeted model construction, and rigorous cross-validation, this method can comprehensively, objectively, and promptly assess the migration risks of enterprises, providing quantitative evidence for decisions regarding business stability and security in industrial parks.

[0007] As a further technical solution of the present invention: step 1 includes: Collect enterprise data from at least two types of sources, including structured databases, semi-structured databases, and unstructured data sources. The collected data is formatted and outlier cleaned. The cleaned data is then normalized. Principal component analysis was used to reduce the dimensionality of the normalized data in order to eliminate linear coupling between features.

[0008] Based on the dimensionality-reduced data, the key feature set used for enterprise migration risk assessment is formed.

[0009] The above technical solutions enable the standardized processing of multi-source heterogeneous data. By unifying the format, cleaning outliers, normalizing, and using PCA dimensionality reduction to eliminate feature coupling, redundant information and abnormal interference are effectively removed, and a precise set of key features is selected. This lays a high-quality data foundation for subsequent enterprise classification and model building, and solves the problem of insufficient accuracy in traditional single-dimensional data judgment.

[0010] As a further technical solution of the present invention: in step 2, a density-based clustering algorithm is used for cluster analysis.

[0011] By employing the above technical solution and using density-based clustering algorithms to cluster enterprises, the natural distribution characteristics of enterprise groups can be accurately captured, forming representative enterprise classifications. This avoids the subjectivity and one-sidedness of traditional classification methods, ensuring that the migration characteristics of enterprises within the same classification are highly correlated, and providing a scientific basis for subsequent customized modeling.

[0012] As a further technical solution of the present invention: the multi-dimensional features on which the cluster analysis is based include the industry to which the enterprise belongs, its business scale and its location.

[0013] By using the above technical solutions, cluster analysis is conducted focusing on core differences such as the industry, business scale, and region of enterprises. This accurately matches the migration characteristics of different types of enterprises, avoids the problem of poor model adaptability caused by unreasonable classification dimensions, and further improves the scientificity and pertinence of enterprise classification.

[0014] As a further technical solution of the present invention: step 3 includes: The historical data for each enterprise category is divided into training set, validation set, and test set; Using the training set, the model is trained using at least two different machine learning algorithms; The validation set is used to evaluate multiple trained models, and the model parameters are adjusted based on the evaluation results. Based on the evaluation indicators, one model is selected from multiple models for each enterprise category as the early warning model for that category.

[0015] Through the above technical solution, by reasonably dividing the dataset, comparing and training multiple algorithms, and optimizing parameters, the optimal early warning model is selected for each enterprise category. This fully leverages the advantages of different machine learning algorithms, avoids the limitations of a single algorithm, ensures that the predictive performance of the early warning model for each category reaches its best, and improves the accuracy of migration risk scoring.

[0016] As a further technical solution of the present invention: step 4 includes: Each enterprise category's corresponding early warning model is cross-validated on the validation sets of its own category and other categories. If a certain early warning model performs similarly on a validation set other than its own classification as it does on its own classification, then the classifications are merged. For the merged classification, repeat steps 2 to 4.

[0017] By using the above technical solutions, enterprise classification and model parameters are dynamically optimized through cross-classification cross-validation, effectively avoiding the problem of poor cross-class adaptability of models. This ensures that each early warning model has a significant effect on predicting migration risks only for enterprises in its corresponding classification, thereby fundamentally improving the accuracy and reliability of the overall early warning solution.

[0018] As a further technical solution of the present invention: step 5 includes: Real-time collection of characteristic data of target companies and matching them to the corresponding company categories; The migration risk score is calculated by calling the early warning model corresponding to the category. When the score exceeds the set risk threshold, an early warning message is automatically generated and sent.

[0019] The above technical solution enables real-time collection, classification and matching, and risk quantification calculation of target enterprise characteristic data. When the risk score exceeds the threshold, an early warning message is automatically generated and sent, which solves the shortcomings of traditional early warning mechanisms such as low update frequency and delayed risk warning. It supports park managers to capture risks, locate causes, and take intervention measures in a timely manner.

[0020] On the other hand, the present invention also discloses an enterprise migration early warning system for attracting and retaining businesses in industrial parks, applied to the aforementioned enterprise migration early warning method for attracting and retaining businesses in industrial parks, comprising: The data acquisition and processing module is used to collect multi-source heterogeneous enterprise data and perform cleaning, normalization, and feature dimensionality reduction. The enterprise classification module is used to perform cluster analysis on historical enterprises based on multi-dimensional features, forming multiple enterprise classifications. The model building and training module is used to build and train a corresponding machine learning early warning model for each enterprise category. The model validation and optimization module is used to perform cross-classification validation on the early warning model and optimize the classification or model based on the validation results. The real-time early warning module is used to calculate the migration risk score of the target enterprise in real time and issue an early warning when the score exceeds the threshold.

[0021] The above technical solution provides a hardware and software system entity for implementing the aforementioned early warning method. Through the collaborative work of five functional modules, the system solidifies the entire process of data collection, processing, analysis, modeling, verification, and early warning, achieving automated and systematic operation of enterprise migration risk early warning, and providing the park with a stable, efficient, and implementable digital management tool.

[0022] On the other hand, the present invention also discloses an electronic device, including at least one processor and a memory communicatively connected to the at least one processor, the memory storing a computer program, which, when executed by the at least one processor, implements a method for early warning of enterprise migration aimed at stabilizing businesses in industrial parks.

[0023] The above technical solutions provide a reliable hardware platform for the early warning method, enabling the enterprise migration early warning technology to be deployed and implemented in actual smart park operation scenarios. This ensures that the method is transformed from a theoretical solution into a practical application capability, providing park managers with an efficient risk monitoring tool.

[0024] On the other hand, the present invention also discloses a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of a method for early warning of enterprise migration aimed at stabilizing businesses in industrial parks.

[0025] The above technical solution stores and reuses the enterprise migration early warning method in the form of a computer program, which facilitates the dissemination, deployment and promotion of the technical solution. This enables more parks to easily apply this refined early warning solution, effectively solve the common problems of enterprise migration early warning in the industry, and expand the scope of application and practical value of the technology.

[0026] In summary, the present invention has at least one of the following beneficial technical effects: 1. This invention discloses a method and system for early warning of enterprise migration in industrial parks, which integrates multi-dimensional data intelligent analysis and real-time early warning through the core path of "scientific classification, customized modeling, and cross-validation optimization". It breaks away from the limitations of traditional manual experience and single-dimensional data, realizes accurate quantification and automated early warning of migration risks, and efficiently supports the stability of industrial parks.

[0027] 2. This invention discloses a computer-readable storage medium for early warning of enterprise migration aimed at stabilizing businesses in industrial parks. It provides stable hardware support and standardized software carrier for the aforementioned early warning methods, promotes the formation of rapidly deployable product forms of technical solutions, lowers the application threshold, and helps to solve common early warning problems in the industry on a large scale. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall process of an early warning method for enterprise migration aimed at stabilizing businesses in industrial parks, based on the present invention.

[0029] Figure 2 for Figure 1 A flowchart of S1.

[0030] Figure 3 for Figure 1 A flowchart of the S3 process.

[0031] Figure 4 for Figure 1 A flowchart of the S4 process.

[0032] Figure 5 for Figure 1 A flowchart of the S5 process.

[0033] Figure 6 This is an architecture diagram of an enterprise migration early warning system for stabilizing businesses in industrial parks, based on the present invention. Detailed Implementation

[0034] The technical solutions in the embodiments 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, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0035] In the description of this application, it should be noted that the terms "upper," "lower," "inner," "outer," "top / bottom," etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0036] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installed," "equipped with," "sleeved / connected," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. Example

[0037] Reference Figure 1 This invention discloses a method for early warning of enterprise migration aimed at stabilizing businesses in industrial parks. It comprises five core steps and corresponding sub-steps, with the core idea being "classification first, modeling later." The specific step framework is as follows: S1 Multi-source heterogeneous enterprise data collection and feature selection; S2 Enterprise clustering and classification based on multi-dimensional features; S3 Construction and training of a classification-specific machine learning early warning model; S4 Model cross-classification cross-validation and classification optimization; S5 Real-time migration risk early warning and intervention. The specific operations of each step and sub-step are explained in detail below with reference to the accompanying drawings. Reference Figure 2 S1 multi-source heterogeneous enterprise data collection and feature filtering includes five sub-steps, each of which is performed in a progressive manner: S101 Multi-Source Heterogeneous Data Classification, Collection, and Storage: Specifically, this involves the targeted collection of at least two types of enterprise-related data from structured databases, semi-structured databases, and unstructured data sources. Structured data includes enterprise tax data, business registration data, and foreign investment data, stored on a MySQL server adapted for structured data. Semi-structured data includes enterprise recruitment information, cooperation announcements, and equity change data, stored on a MongoDB server adapted for semi-structured data. Unstructured data includes enterprise public opinion information, industry reports, and text data related to business operations, stored on an HDFS server adapted for unstructured data. This process completes the classification, aggregation, and compatible storage of multi-source data.

[0038] S102 Data Format Unification and Outlier Handling: Specifically, the datax tool was used to unify the format of collected data fields of the same type, and the unit was standardized and converted to RMB and month, ensuring that the data format was consistent across different sources. Then, box plots were generated using Python's matplotlib package. The box plots were used to visually identify abnormal data that deviated from the normal range, such as tax data that far exceeded the industry average and registration information with logical contradictions. Outliers were removed or corrected based on the industry average to ensure the validity and authenticity of the data.

[0039] S103 Data Normalization Processing: Specifically, the Z-score method is used to uniformly normalize the effective data after S102 cleaning, transforming data of different magnitudes and dimensions, such as registered capital in the tens of thousands of yuan and number of employees, into values ​​under a unified standard, or generating risk assessment indicators such as revenue growth rate and tax burden rate according to actual early warning needs, so that the data has horizontal comparability.

[0040] S104 Data Dimensionality Reduction and Decoupling: Specifically, PCA principal component analysis is used to reduce the dimensionality of the normalized dataset. By extracting the core principal components of the data, linear coupling relationships between features such as the high correlation between revenue and net profit are eliminated, reducing the interference of redundant information on subsequent classification and modeling, and improving data processing efficiency.

[0041] S105 Key Feature Set Extraction: Specifically, based on the data after dimensionality reduction in S104, core and effective features that are strongly correlated with enterprise migration risk, such as industry adaptability, regional policy matching degree, and operating cost ratio, are integrated to form a key feature set for enterprise migration risk assessment, providing high-quality and highly relevant data support for subsequent enterprise classification and model construction.

[0042] The specific operation of S2's enterprise clustering classification based on multi-dimensional features is as follows: Based on the key feature set obtained in S105, and based on historical migration and non-migration enterprise data from across the country, core features of enterprises in multiple dimensions, such as industry, position in the industrial chain, operating scale, location, life cycle stage, and equity structure, are comprehensively extracted. The core classification dimensions are industry, operating scale, and location, which align with the core driving factors of enterprise migration. Subsequently, the DBSCAN unsupervised clustering algorithm is used to perform clustering analysis on the enterprises corresponding to the key feature set. This algorithm does not require a preset number of categories and can automatically identify the natural distribution pattern of enterprise groups through data density. Enterprises with similar migration behavior characteristics, cost-driven, policy-driven, and other similar migration driving factors are grouped into one category, ultimately forming k representative enterprise subsets (k is a positive integer ≥2), ensuring that enterprises in the same category have a high correlation in migration risk characteristics, laying a scientific classification foundation for the subsequent construction of customized models.

[0043] Reference Figure 3 The construction and training of the S3 classification-specific machine learning early warning model consists of four closely linked sub-steps: S301 Dataset Partitioning: Specifically, for each enterprise category formed in S2, the historical enterprise migration-related data under that category is divided into training set, validation set, and test set in a 70:15:15 ratio, covering both migrating and non-migrating enterprise data. The training set is used for learning and training the core parameters of the model, the validation set is used for evaluating the model's performance and tuning the parameters, and the test set is used for verifying the model's final predictive performance.

[0044] S302 Multi-Algorithm Model Training: Specifically, the training set is used as input, and at least two different machine learning algorithms such as decision tree, random forest, and XGBoost are used to train the model. Through each algorithm, the mapping relationship between enterprise migration characteristics and migration behavior, such as the correlation between high operating costs and low policy support, is learned, and multiple initial early warning models are obtained.

[0045] S303 model parameter tuning: Specifically, the validation set is used to evaluate the performance of multiple initial models trained by S302. Accuracy, precision, recall, and F1-score are used as the core evaluation indicators. Based on the evaluation results, hyperparameters and structural parameters such as decision tree depth and the number of random forest decision trees are adjusted one by one to continuously optimize the model's prediction accuracy for migration risk and reduce the false positive rate.

[0046] S304 Optimal Model Selection: The specific operation involves comparing the performance of multiple models on the validation set across various evaluation metrics to select the model with the best prediction effect on the migration risk of enterprises in the current category. The model with the highest F1-score in a certain category is determined as the exclusive early warning model for that enterprise category, ensuring that each category can obtain the most suitable prediction tool.

[0047] Reference Figure 4 The S4 model's cross-classification cross-validation and classification optimization consist of three sub-steps, forming a closed-loop optimization process: S401 Cross-classification Validation Execution: Specifically, the dedicated early warning model corresponding to each enterprise category determined in S304 is subjected to cross-classification validation on the validation set of the model's own category and the validation sets of all other k-1 categories. This comprehensively evaluates the difference in prediction performance of each model on enterprises in its own category and enterprises in other categories, with a focus on the accuracy of the model in category 1 on the validation set of category 2 and its closeness to its own category.

[0048] S402 Classification Merging Judgment: The specific operation is to set a preset similarity condition such as the difference of core indicators ≤10%. If the prediction performance of a certain early warning model on the validation set of a non-self-classification is similar to its performance on the validation set of its own classification, it indicates that the migration characteristics of the corresponding two or more types of enterprises are highly similar, the original classification is too fine, and these classifications need to be merged.

[0049] S403 Classification and Model Re-optimization: The specific operation involves re-classifying the merged new categories using the K-Means algorithm or decision tree algorithm. Then, the complete process from S2 to S4 is repeated, namely, re-extracting key feature sets, constructing new enterprise categories, training new dedicated models, and performing cross-classification cross-validation until all early warning models have significant predictive effects only on enterprises in their corresponding categories, and the cross-classification prediction accuracy is lower than a preset threshold, thus achieving precise adaptation and optimization of the classification system and model.

[0050] Reference Figure 5 S5 real-time migration risk warning and intervention includes three sub-steps to achieve real-time risk perception and response: S501 Target Enterprise Data Collection and Classification Matching: The specific operation involves continuously collecting the characteristic data of target enterprises in the park in real time by connecting to data collection terminals such as the park's enterprise management system and tax system interface. This data is consistent with the feature dimensions of the key feature set in S1. Then, based on the optimized classification rules and classification system in S2, the target enterprises are accurately matched to the corresponding enterprise classifications through feature comparison.

[0051] S502 Migration Risk Quantitative Scoring: The specific operation involves calling the dedicated early warning model of the category to which the target enterprise belongs, inputting the enterprise's characteristic data collected in real time into the model, and calculating the migration risk score of the enterprise through model calculation. The migration risk score ranges from 0 to 100 points. The higher the score, the stronger the migration tendency. The scoring result quantitatively reflects the degree of migration tendency of the enterprise and supports horizontal comparison between different enterprises.

[0052] S503 Early Warning Trigger and Information Push: Specifically, users can customize risk thresholds such as 80 points based on the park's business security and stability management needs. When the migration risk score of a target enterprise exceeds this preset threshold, the system automatically generates early warning information containing key risk characteristics such as the enterprise name, unified social credit code, migration risk score, and operating cost ratio exceeding 30% of the industry average. This information is then pushed to relevant management personnel via preset notification methods such as SMS, system messages, and emails. This enables park operations managers to promptly capture enterprise migration risks, quickly locate risk triggers, and then take targeted intervention measures such as policy support and cost subsidies to ensure the stability and sustainability of the park's industries.

[0053] Reference Figure 6This invention also discloses an enterprise migration early warning system for stabilizing businesses in industrial parks. As a modular implementation of the aforementioned early warning method, it achieves automated implementation through the coordinated operation of five core modules: a data acquisition and processing module responsible for the classification, cleaning, normalization, and dimensionality reduction of multi-source data, outputting key feature sets; an enterprise classification module completing enterprise clustering based on multi-dimensional features, supporting algorithm switching during classification optimization; a model building and training module dividing the dataset according to classification, training multiple algorithm models, and selecting the optimal model; a model validation and optimization module performing cross-classification cross-validation, triggering classification merging and remodeling processes; and a real-time early warning module responsible for target enterprise data acquisition, classification matching, risk score calculation, and early warning push, all without manual intervention, improving early warning efficiency and accuracy.

[0054] This invention also discloses an electronic device, including at least one processor and a memory communicatively connected thereto. When the computer program stored in the memory is executed by the processor, it can fully implement the entire process of the aforementioned early warning method. This device provides hardware support for the early warning method, and by scheduling various tools and algorithms through the processor, it transforms steps such as data processing, classification modeling, and early warning push into feasible automated operations, ensuring that the technical solution is transformed from theory into practical application capability.

[0055] This invention also discloses a computer-readable storage medium for early warning of enterprise migration aimed at stabilizing business in industrial parks. When the computer program stored on it is executed by a processor, it can implement all the steps of the aforementioned early warning method. This storage medium solidifies the core logic of the early warning method into executable code, achieving standardized storage and reuse of the method. This facilitates rapid deployment and promotion in different industrial parks, effectively solving common problems of traditional early warning mechanisms in the industry, and expanding the applicability and practical value of the technology.

[0056] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0057] The implementation principle of this invention is as follows: It revolves around the principle of "classification first, modeling later," overcoming the shortcomings of traditional early warning systems through multi-dimensional data fusion and intelligent analysis. In the data processing stage, multi-source heterogeneous data is collected and processed through format unification, outlier handling, normalization, and PCA dimensionality reduction. This not only compensates for the limitations of single-dimensional data but also eliminates interference, redundancy, and feature coupling, providing a high-quality key feature set for subsequent processes. In the enterprise classification stage, based on multi-dimensional features such as industry, business scale, and location, the DBSCAN unsupervised clustering algorithm is used to capture the natural distribution patterns of enterprises, avoiding subjective classification bias and ensuring that the migration characteristics of similar enterprises converge, laying the foundation for customized modeling. In the model construction and optimization stage, separate models are built for enterprises in different categories. The optimal model is selected through multi-algorithm training comparison and parameter tuning, and then cross-classification cross-validation is used to verify the model's specificity. Classifications with similar features are merged to achieve accurate matching between classification and model. In the real-time early warning stage, dynamic feature data of enterprises is continuously collected, matched with the corresponding classification model to calculate a quantitative risk score, and automatic early warning is pushed when thresholds are exceeded. This solves the problems of slow updates and delays in traditional early warning systems, providing timely and quantitative decision-making basis for ensuring business stability in industrial parks.

[0058] The embodiments described herein are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for early warning of enterprise relocation aimed at stabilizing business in industrial parks, characterized in that, Includes the following steps: Step 1: Collect multi-source heterogeneous enterprise-related data, and clean, normalize, and perform feature dimensionality reduction on the data to obtain a key feature set for enterprise migration risk assessment. Step 2: Based on the multi-dimensional features in the key feature set, perform cluster analysis on the enterprises corresponding to the key feature set to form multiple enterprise classifications; Step 3: For each of the enterprise categories, construct and train a corresponding machine learning early warning model to output the migration risk score of the enterprise under that category; Step 4: Perform cross-classification cross-validation on the multiple machine learning early warning models, and optimize the enterprise classification and / or model parameters based on the validation results; Step 5: Obtain the characteristic data of the target enterprise in real time, call the corresponding early warning model to calculate the migration risk score according to its category, and issue an early warning when the score exceeds the preset threshold.

2. The enterprise relocation early warning method for stabilizing business in industrial parks according to claim 1, characterized in that, Step 1 includes: Collect enterprise data from at least two types of sources, including structured databases, semi-structured databases, and unstructured data sources. The collected data is formatted and outlier cleaned. The cleaned data is then normalized. Principal component analysis was used to reduce the dimensionality of the normalized data in order to eliminate linear coupling between features; Based on the dimensionality-reduced data, the key feature set used for enterprise migration risk assessment is formed.

3. The enterprise relocation early warning method for stabilizing business in industrial parks according to claim 1, characterized in that, In step 2, a density-based clustering algorithm is used for cluster analysis.

4. The enterprise relocation early warning method for stabilizing business in industrial parks according to claim 3, characterized in that, The multi-dimensional features used in the cluster analysis include the industry to which the enterprise belongs, its business scale, and its location.

5. The enterprise relocation early warning method for stabilizing business in industrial parks according to claim 1, characterized in that, Step 3 includes: The historical data for each enterprise category is divided into training set, validation set, and test set; Using the training set, the model is trained using at least two different machine learning algorithms; The validation set is used to evaluate multiple trained models, and the model parameters are adjusted based on the evaluation results. Based on the evaluation indicators, one model is selected from multiple models for each enterprise category as the early warning model for that category.

6. The enterprise relocation early warning method for stabilizing business in industrial parks according to claim 1, characterized in that, Step 4 includes: Each enterprise category's corresponding early warning model is cross-validated on the validation sets of its own category and other categories. If a certain early warning model performs similarly on a validation set other than its own classification as it does on its own classification, then the classifications are merged. For the merged classification, repeat steps 2 to 4.

7. The enterprise relocation early warning method for stabilizing business in industrial parks according to claim 1, characterized in that, Step 5 includes: Real-time collection of characteristic data of target companies and matching them to the corresponding company categories; The migration risk score is calculated by calling the early warning model corresponding to the category. When the score exceeds the set risk threshold, an early warning message is automatically generated and sent.

8. A business relocation early warning system for stabilizing business in industrial parks, characterized in that: The enterprise migration early warning method for stabilizing business in industrial parks, applied to any one of claims 1-7, includes: The data acquisition and processing module is used to collect multi-source heterogeneous enterprise data and perform cleaning, normalization, and feature dimensionality reduction. The enterprise classification module is used to perform cluster analysis on historical enterprises based on multi-dimensional features, forming multiple enterprise classifications. The model building and training module is used to build and train a corresponding machine learning early warning model for each enterprise category. The model validation and optimization module is used to perform cross-classification validation on the early warning model and optimize the classification or model based on the validation results. The real-time early warning module is used to calculate the migration risk score of the target enterprise in real time and issue an early warning when the score exceeds the threshold.

9. An electronic device comprising at least one processor and a memory communicatively connected to said at least one processor, characterized in that, The memory stores a computer program that, when executed by the at least one processor, implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images