AI Agent-driven credit risk early warning strategy automatic evaluation method and device, control equipment and computer readable storage medium

Abstract

The invention relates to an AI Agent-driven credit risk early warning strategy automatic evaluation method and device, control equipment and a computer readable storage medium, and the method comprises the steps: obtaining structured financial data, unstructured text data and dynamic time sequence data of an enterprise through multi-source data, carrying out the standardization processing, and outputting a standardized risk data set; based on the standardized risk data set, static financial features, dynamic derivative features and associated network features are extracted, and a high-discrimination feature set is generated through feature screening and dimensionality reduction optimization; a multi-model integration strategy is adopted, the high-discrimination feature set is input into a neural network, a Stacking integration model and a graph neural network, and enterprise risk scores and grades are output; and automatically matching an early warning rule based on the enterprise risk score and the grade, and carrying out real-time monitoring and early warning on enterprise credit according to the early warning rule.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of credit management, in particular to an AI Agent driven credit risk early warning strategy automatic evaluation method and device, a control equipment and a computer readable storage medium. BACKGROUND

[0002] Under the background of the intensification of global economic fluctuations and the rise of market uncertainty, the field of financial credit and enterprise risk management is facing the severe challenges of accelerated credit risk transmission speed and complex risk patterns. Accurate assessment of enterprise credit risk is a core link for preventing the spread of systemic financial risks and ensuring the safety and stable operation of financial institutions. Its importance is increasingly prominent.

[0003] Traditional technical means mainly rely on single financial indicators or simple statistical models for risk assessment, which has significant limitations: first, the static assessment mode cannot capture the dynamic changes of enterprises, such as ignoring the real-time impact of revenue volatility, frequency of related transactions and other time series characteristics on debt servicing capacity; second, isolated analysis of individual enterprises cannot effectively integrate the risk transmission effects of related networks such as supply chains and guarantee circles, resulting in delayed identification of related default risks. These defects make it difficult for traditional methods to meet the needs of financial institutions for early identification, early warning and early disposal of risks in complex market environments.

[0004] Based on this, the application provides an AI Agent driven credit risk early warning strategy automatic evaluation method, device, control equipment and computer readable storage medium. SUMMARY

[0005] In order to improve the problem that the traditional method is difficult to meet the needs of financial institutions for early identification, early warning and early disposal of risks in complex market environments, the application provides an AI Agent driven credit risk early warning strategy automatic evaluation method, device, control equipment and computer readable storage medium.

[0006] In a first aspect, the application provides an AI Agent driven credit risk early warning strategy automatic evaluation method, which adopts the following technical solution: comprising: Obtain structured financial data, unstructured text data and dynamic time series data of enterprises through multi-source data, and perform standardization processing to output a standardized risk data set; Based on the standardized risk data set, extract static financial features, dynamic derived features and related network features, and generate a high-discrimination feature set through feature screening and dimensionality reduction optimization; Adopt a multi-model integration strategy, input the high-discrimination feature set into a neural network, a Stacking integrated model and a graph neural network, and output an enterprise risk score and grade; Automatic matching of early warning rules based on the enterprise risk score and grade, real-time monitoring and early warning of enterprise credit according to the early warning rules.

[0007] Preferably, the structured financial data, unstructured text data and dynamic time series data of the enterprise are obtained through multi-source data fusion technology, and standardized processing is performed, and a standardized risk data set is output, including: Interface with the enterprise ERP system and the credit investigation platform API to collect balance sheets, profit and loss statements, cash flow statements and historical default records, and use OCR technology to identify financial reports to obtain the structured financial data; Based on the large language model, the implicit risk signals of the annual report of the enterprise are analyzed to obtain the unstructured text data; Using federated learning technology to collect real-time transaction records, Internet of Things production data and economic indicators, and completing data encryption fusion without sharing original data to obtain the dynamic time series data; Construct a generative adversarial network, use an LSTM time series prediction model to fill in short-term missing values, and mark long-term missing data as data anomaly features; Based on the pre-set standardized rule library, the financial indicators are quantile standardized, and the time series data is uniformly converted to the same timestamp format to output the standardized risk data set.

[0008] Preferably, based on the standardized risk data set, static financial features, dynamic derived features and associated network features are extracted, high-discrimination feature sets are generated through feature screening and dimensionality reduction optimization, including: From the standardized risk data set, indicators reflecting the basic financial condition of the enterprise are extracted, including solvency indicators, profitability indicators and operational capacity indicators, to form the static financial features; Based on the dynamic time series data in the standardized risk data set, the rate of change, volatility and trend slope of the indicators are calculated through time series analysis, and the dynamic derived features are generated in combination with the sliding window statistical method; According to the network topology structure constructed according to the enterprise correlation data, the node degree centrality, intermediate centrality and edge weight features are extracted, and the associated network features are generated in combination with the network propagation algorithm; The static financial feature subset, dynamic derived feature subset and associated network feature subset are subjected to feature importance evaluation and multicollinearity test, and after removing redundant features, the high-dimensional feature space is mapped to a low-dimensional space through a dimensionality reduction algorithm to generate a high-discrimination feature set.

[0009] Preferably, the multi-model integration strategy is used to input the high-discrimination feature set into a neural network, a Stacking integrated model and a graph neural network to output an enterprise risk score and grade, including: inputting the static financial features and the dynamic derivative features into a neural network, inputting the associated network features into a graph neural network, obtaining meta features as a Stacking integrated model; training the neural network using regularization and early stopping mechanism, optimizing node feature propagation of the graph neural network using neighborhood aggregation algorithm, screening a combination of base classifiers and meta classifiers of the Stacking integrated model through cross-validation, and comparing a model training result with a preset validation set to obtain a validation set risk prediction error; dynamically adjusting weights of each model based on the validation set risk prediction error, and generating input features of the Stacking integrated model by weighted voting fusion of a basic risk score output by the neural network and an associated conduction score output by the graph neural network based on the meta features; outputting a comprehensive risk score through the Stacking integrated model based on the input features, and mapping the comprehensive risk score to a risk level through a preset risk level division rule, wherein the risk level includes at least three levels of low risk, medium risk and high risk.

[0010] Preferably, the comparison of the model training result with the preset validation set to obtain the validation set risk prediction error comprises: dividing original sample data into a training set, a validation set and a test set, wherein the validation set is independent of the model training process; training the neural network, the Stacking integrated model and the graph neural network based on the training set, and comparing a risk score output by the trained neural network and the graph neural network with a real risk label of the sample, and calculating the validation set risk prediction error through a preset error index.

[0011] Preferably, the automatic matching of the enterprise risk score and level with the early warning rule, and the real-time monitoring and early warning of the enterprise credit according to the early warning rule comprises: presetting a plurality of dimensions of the early warning rule, wherein the early warning rule includes a risk level threshold rule, a score fluctuation rule and a feature anomaly rule; inputting the enterprise risk score and level into a rule library, automatically associating the corresponding early warning rule through a matching algorithm, and generating an early warning instruction including an early warning level, a trigger condition and a disposal suggestion; real-time collection of enterprise credit data and associated features, and updating of the risk score and level according to a preset period; if the updated score or level meets the trigger condition of the matched early warning rule, an early warning signal is pushed to a credit management system through an interface, and a risk control personnel is notified synchronously; recording the early warning response result, and feeding back the disposal effect to the rule library.

[0012] Preferably, after recording the early warning response results and feeding back the treatment effects to the rule base, it further includes: Collecting the early warning response results and the actual credit risk performance of the corresponding enterprises to generate a feedback dataset; Updating the standardized risk dataset based on the feedback dataset, and optimizing the importance evaluation threshold value through an incremental learning mechanism to dynamically adjust the weight proportion of the static financial features, the dynamic derivative features, and the associated network features; Based on the updated high-discrimination feature set, re-executing the multi-model training process, taking the actual risk results in the feedback dataset as a new validation set; Through the validation set risk prediction error correction model weight allocation strategy, realizing the adaptive iteration of neural networks, Stacking integrated models, and graph neural networks; According to the actual triggering effectiveness of the early warning rules in the feedback dataset, adjusting the threshold parameters of the early warning rule base.

[0013] In a second aspect, an AI Agent-driven credit risk early warning strategy automatic evaluation device is provided, which adopts the following technical solutions and includes: A data processing module acquires structured financial data, unstructured text data, and dynamic time series data of an enterprise through multiple data sources, and performs standardization processing to output a standardized risk dataset; A feature extraction module is configured to extract static financial features, dynamic derivative features, and associated network features based on the standardized risk dataset, and generate a high-discrimination feature set through feature screening and dimensionality reduction optimization; A model training module is configured to input the high-discrimination feature set into neural networks, Stacking integrated models, and graph neural networks using a multi-model integration strategy, and output an enterprise risk score and grade; A risk early warning module is configured to automatically match early warning rules based on the enterprise risk score and grade, and perform real-time monitoring and early warning on the enterprise credit according to the early warning rules.

[0014] In a third aspect, the present application further provides a control device, which includes: The device includes a memory and a processor, and the memory stores a computer program that can be loaded and executed by the processor to perform the AI Agent-driven credit risk early warning strategy automatic evaluation method described above.

[0015] In a fourth aspect, the present application further provides a computer-readable storage medium storing a computer program that can be loaded and executed by a processor to perform the AI Agent-driven credit risk early warning strategy automatic evaluation method described above.

[0016] In summary, the present application obtains enterprise structured financial data, unstructured text data and dynamic time series data through multi-source data fusion, constructs a risk data set through standardization processing; extracts static financial features, dynamic derived features and correlation network features based on the data set, generates a high-discrimination feature set through screening and dimension reduction; uses a multi-model integration strategy of neural network, Stacking integrated model and graph neural network to output enterprise risk score and grade; realizes real-time monitoring and early warning based on matching early warning rules of the score grade, and optimizes the whole process through early warning disposal effect feedback. The scheme improves data integrity through multi-dimensional data fusion, enhances risk characterization ability through multi-feature collaboration, improves prediction accuracy through multi-model integration, realizes adaptive optimization of features, models and rules combined with feedback mechanism, effectively improves the real-time, accuracy and dynamic adjustment ability of credit risk early warning, and provides a credit risk management scheme. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 FIG. 1 is a flowchart of an AI Agent-driven credit risk early warning strategy automatic evaluation method.

[0018] Figure 2 FIG. 2 is a structural block diagram of an AI Agent-driven credit risk early warning strategy automatic evaluation device. DETAILED DESCRIPTION

[0019] The following will be described in detail in combination with Figure 1 Figure 2 The present application is further described in detail.

[0020] Under the background of rapid development of current financial technology, traditional credit risk evaluation methods face challenges such as single data source, limited feature dimension, insufficient model adaptability and lagging early warning response, and are difficult to meet the needs of financial institutions for dynamic risk monitoring.

[0021] To solve the above problems, the present patent proposes an AI Agent-driven credit risk early warning strategy automatic evaluation method, which integrates structured financial data, unstructured text data and dynamic time series data through multi-source data fusion technology, and constructs a standardized risk data set; extracts static financial features, dynamic derived features and correlation network features based on the data set, and generates a high-discrimination feature set through screening and optimization; uses a multi-model integration strategy of neural network, Stacking integrated model and graph neural network to output enterprise risk score and grade, and automatically matches early warning rules to realize real-time monitoring and early warning, while forming a whole-process adaptive optimization mechanism through early warning disposal effect feedback.

[0022] Referring to Figure 1 , the present application embodiments at least include steps S10 to S40.

[0023] ​S10, structured financial data, unstructured text data and dynamic time series data of enterprises are acquired through multi-source data, and standardized processing is performed to output a standardized risk data set.

[0024] S20, based on the standardized risk data set, static financial features, dynamic derived features and correlation network features are extracted, and high-discrimination feature sets are generated through feature screening and dimensionality reduction optimization.

[0025] S30, a multi-model integration strategy is adopted, and the high-discrimination feature sets are input into a neural network, a Stacking integrated model and a graph neural network to output enterprise risk scores and grades.

[0026] S40, based on the enterprise risk score and grade, the early warning rules are automatically matched, and the enterprise credit is monitored and warned in real time according to the early warning rules.

[0027] Specifically, in the bank credit approval scene, the system acquires structured financial data such as balance sheet by interfacing with enterprise ERP system, uses NLP to analyze annual report text to mine implicit risks, uses federated learning to encrypt and fuse real-time transaction water and other dynamic data, and eliminates data heterogeneity through standardization processing to lay a high-quality data foundation for risk assessment; From the standardized data, static features such as solvency, dynamic features such as volatility calculated based on time series data, and centrality features extracted from enterprise correlation network, through screening and dimensionality reduction to remove redundancy, form a high-discrimination feature set that can accurately describe enterprise risks; The features are input into neural network and graph neural network to process financial and correlation data respectively, and the risk score and grade are output through Stacking integrated model weighted fusion, and the prediction accuracy is improved through multi-model collaboration; According to the score and grade, the early warning rules are automatically matched, the enterprise credit data is monitored in real time, when the risk exceeds the standard, the early warning signal is pushed and the risk control personnel is notified, and the bad debt rate is effectively reduced.

[0028] In some embodiments, step S10 specifically includes the following steps: interfacing with enterprise ERP system and credit investigation platform API, collecting balance sheet, profit table, cash flow table and historical default records, and using OCR technology to identify financial reports to obtain structured financial data; Based on the large language model, the implicit risk signals of the enterprise annual report are analyzed to obtain unstructured text data; Using federated learning technology to collect enterprise real-time transaction water, Internet of Things production data and economic indicators, completing data encryption fusion without sharing original data, and obtaining dynamic time series data; Construct a generative adversarial network, use LSTM time series prediction model to fill in short-term missing values, and mark long-term missing data as data anomaly features; Based on the pre-set standardization rule library, the financial indicators are quantile standardized, and the time series data is uniformly converted to the same timestamp format, and the standardized risk data set is output.

[0029] Specifically, by connecting the enterprise ERP system and the credit investigation platform API to obtain structured data such as balance sheets, and using OCR technology to digitize financial information from paper financial reports, the problem of low data collection efficiency is solved; using a large language model to analyze annual report management discussions and other texts, and mining market competition and other implicit risk signals to make up for the lack of structured data; using federated learning technology to encrypt and integrate transaction records, Internet of Things production data, and macroeconomic indicators without accessing the original data of the enterprise, balancing data privacy and the value of multi-source data; using generative adversarial networks and LSTM models to fill short-term data gaps and mark long-term missing data for abnormal characteristics to improve data integrity; finally, according to the standardized rule base, the financial indicators are normalized by quantile, and the time series data are unified by timestamp to eliminate data heterogeneity, laying a foundation for high-quality data for subsequent risk feature extraction.

[0030] In some embodiments, step S20 specifically includes the following steps: extracting indicators reflecting the basic financial condition of the enterprise from the standardized risk data set, including solvency indicators, profitability indicators, and operational capacity indicators, to form static financial features; based on the dynamic time series data in the standardized risk data set, calculating the change rate, volatility, and trend slope of the indicators through time series analysis, and generating dynamic derivative features in combination with sliding window statistical methods; constructing a network topology structure according to the enterprise correlation data, extracting node degree centrality, intermediate centrality, and edge weight features, and generating correlation network features in combination with network propagation algorithms; performing feature importance evaluation and multicollinearity test on the static financial feature subset, the dynamic derivative feature subset, and the correlation network feature subset, removing redundant features, and mapping the high-dimensional feature space to the low-dimensional space through dimensionality reduction algorithm to generate a high-discriminability feature set.

[0031] Specifically, the system extracts static financial features such as liquidity ratio and gross profit margin from standardized data, which intuitively reflect the current solvency, profitability, and operational capacity of the enterprise; based on dynamic time series data such as transaction records, dynamic derivative features such as revenue growth rate volatility and inventory turnover trend slope are calculated to capture the change pattern of the enterprise's operating condition; according to the network topology constructed based on the equity correlation and supply chain relationship of the enterprise, correlation network features such as core enterprise node degree centrality and guarantee chain intermediate centrality are extracted to identify the risk transmission path; through feature importance evaluation, redundant indicators such as accounts receivable turnover rate are removed, and after multicollinearity test and PCA dimensionality reduction, the high-dimensional features are mapped to the low-dimensional space to generate a feature set that balances comprehensiveness and discriminability, providing accurate input for subsequent risk prediction models and effectively improving the accuracy and robustness of credit risk assessment.

[0032] In some embodiments, step S30 specifically comprises the following steps: inputting static financial features and dynamic derivative features into a neural network, inputting associated network features into a graph neural network, obtaining meta-features as a Stacking ensemble model; training the neural network using a regularization and early stopping mechanism, optimizing node feature propagation of the graph neural network using a neighborhood aggregation algorithm, screening a combination of base classifiers and meta-classifiers of the Stacking ensemble model through cross-validation, comparing the model training result with a preset validation set to obtain a validation set risk prediction error; dynamically adjusting the weight of each model based on the validation set risk prediction error, generating input features of the Stacking ensemble model by weighted voting of the base risk score output by the neural network and the associated conduction score output by the graph neural network based on the meta-features; outputting a comprehensive risk score based on the input features through the Stacking ensemble model, mapping the comprehensive risk score to a risk level through a preset risk level division rule, and the risk level at least includes three levels of low risk, medium risk and high risk.

[0033] Specifically, the system inputs static financial and dynamic derivative features into a neural network to process time series changes, inputs associated network features into a graph neural network to capture risk conduction between enterprises, and outputs meta-features as a Stacking ensemble model; avoid overfitting of the neural network through regularization and early stopping mechanism, optimize node feature propagation of the graph neural network through neighborhood aggregation algorithm, screen a combination of base classifiers and meta-classifiers of the Stacking through cross-validation, dynamically adjust the weight of the model by comparing the validation set error, generate input features by weighted fusion of base risk score and associated conduction score, and finally output comprehensive risk score through Stacking and map to low, medium and high levels, which improves the prediction accuracy through multiple models, and provides accurate risk level reference for credit approval.

[0034] In some embodiments, step S40 specifically comprises the following steps: dividing the original sample data into a training set, a validation set and a test set, and the validation set is independent of the model training process; training the neural network, the Stacking ensemble model and the graph neural network based on the training set, and predicting the risk of the validation set through the trained neural network and graph neural network respectively, comparing the predicted risk score with the real risk label of the sample, and calculating the validation set risk prediction error through a preset error index.

[0035] Specifically, in the bank credit risk control scene, historical credit data is divided into a training set (70% of the samples), a validation set (15%), and a test set (15%). The validation set is independent of model training to ensure the objectivity of the evaluation. The neural network, graph neural network, and stacking integrated model are trained based on the training set. The trained neural network and graph neural network are used to predict the risk of the validation set. The output risk score is compared with the real default label of the enterprise. The prediction error of the validation set is calculated by RMSE (Root Mean Square Error). This error is used to dynamically adjust the weight of each model (such as reducing the weight of high error models) to avoid overfitting and optimize the integrated strategy, ensuring that the final risk score has strong generalization ability in the actual credit scene and provides a reliable model output basis for subsequent early warning rule matching.

[0036] In some embodiments, step S50 specifically includes the following steps: presetting multi-dimensional early warning rules, including risk level threshold rules, score fluctuation rules, and feature anomaly rules; inputting the enterprise risk score and level into the rule library, automatically associating the corresponding early warning rules through matching algorithms, and generating early warning instructions containing early warning levels, trigger conditions, and disposal suggestions; collecting enterprise credit data and associated features in real time, updating risk scores and levels according to the preset period; if the updated score or level meets the trigger condition of the matched early warning rule, push the early warning signal to the credit management system through the interface, and notify the risk control personnel simultaneously; record the early warning response results, and feed back the disposal effect to the rule library.

[0037] Specifically, preset multi-dimensional early warning rules: risk level threshold rules (such as directly triggering early warning for high risk levels), score fluctuation rules (such as a 50% risk score surge within 1 month), and feature anomaly rules (such as a sudden drop in the liquidity ratio to below 50% of the industry average); input the real-time risk score and level of the enterprise into the rule library, automatically associate the corresponding rules through fuzzy matching algorithms, and generate instructions containing early warning levels, trigger conditions (such as a continuous 3-month revenue fluctuation rate exceeding the threshold), and disposal suggestions (such as suspending new credit granting); collect enterprise credit data to update the score daily, if the trigger condition is met, push the early warning signal to the credit management system through the API, and notify the risk control personnel through SMS and email; record the disposal results (such as having required the enterprise to supplement the collateral), feed back the false positive rate, disposal time efficiency, and other indicators to the rule library, continuously optimize the rule threshold, and improve the risk response efficiency.

[0038] In some embodiments, step S50 further comprises the following steps: collecting early warning response results and actual credit risk performance of the corresponding enterprise, generating a feedback dataset; updating the standardized risk dataset based on the feedback dataset, and optimizing the importance evaluation threshold value through an incremental learning mechanism to dynamically adjust the weight proportion of the static financial features, the dynamic derived features and the associated network features; based on the updated high-discrimination feature set, re-executing the multi-model training process, taking the actual risk results in the feedback dataset as a new validation set; through the validation set risk prediction error correction model weight allocation strategy, realizing the adaptive iteration of the neural network, the Stacking integrated model and the graph neural network; and adjusting the threshold parameters of the early warning rule library according to the actual triggering effectiveness of the early warning rules in the feedback dataset.

[0039] Specifically, the system generates a feedback dataset by collecting early warning response results (such as customer defaults three months after early warning) and actual risk performance of the enterprise (delinquency rate, non-performing loan rate); updates the standardized risk dataset with the dataset, dynamically adjusts the feature weights through incremental learning (increases the importance proportion of the associated enterprise guarantee default times feature); re-trains the neural network and the graph neural network with the actual default cases in the feedback dataset as a new validation set, corrects the weight allocation strategy of the Stacking integrated model through the validation set error; and adjusts the early warning threshold according to the actual triggering effectiveness of the rules (such as the default prediction accuracy of the liquidity ratio sudden drop rule), continuously improves the accuracy and timeliness of risk early warning, and reduces the loss of credit assets.

[0040] The implementation principle of an embodiment of the AI Agent-driven credit risk early warning strategy automatic evaluation method is as follows: structured financial data, unstructured text data and dynamic time series data of an enterprise are obtained through multi-source data fusion, and a risk dataset is constructed through standardization processing; static financial features, dynamic derived features and associated network features are extracted based on the dataset, and a high-discrimination feature set is generated through screening and dimension reduction; a multi-model integrated strategy of a neural network, a Stacking integrated model and a graph neural network is adopted to output an enterprise risk score and grade; real-time monitoring and early warning are realized based on the matching of the score grade and early warning rules, and the whole process is optimized through early warning disposal effect feedback. This scheme improves data integrity through multi-dimensional data fusion, enhances risk characterization ability through multi-feature cooperation, improves prediction accuracy through multi-model integration, realizes adaptive optimization of features, models and rules through a feedback mechanism, effectively improves the real-time performance, accuracy and dynamic adjustment capability of credit risk early warning, and provides a credit risk management scheme.

[0041] Figure 1 A flowchart of an embodiment of the AI Agent-driven credit risk early warning strategy automatic evaluation method is shown. It should be understood that, although Figure 1The steps in the flowcharts are shown in sequence according to the arrows, but the steps are not necessarily executed in the order indicated by the arrows; unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other orders; and Figure 1 At least part of the steps in the flowcharts can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of the sub-steps or stages is not necessarily sequential, but can be alternately executed with other steps or at least part of the sub-steps or stages of other steps.

[0042] Based on the same technical concept, referring to Figure 2 The application embodiment further provides an AI Agent-driven credit risk early warning strategy automatic evaluation device, which adopts the following technical scheme, and the device comprises: A data processing module acquires structured financial data, unstructured text data and dynamic time series data of an enterprise through multi-source data, and performs standardization processing to output a standardized risk data set; A feature extraction module is configured to extract static financial features, dynamic derived features and associated network features based on the standardized risk data set, and generate a high-discrimination feature set through feature screening and dimensionality reduction optimization; A model training module is configured to input the high-discrimination feature set into a neural network, a Stacking integrated model and a graph neural network by adopting a multi-model integration strategy, and output an enterprise risk score and a grade; A risk early warning module is configured to automatically match an early warning rule based on the enterprise risk score and the grade, and perform real-time monitoring and early warning on enterprise credit according to the early warning rule.

[0043] In some embodiments, the data processing module is specifically configured to acquire structured financial data, unstructured text data and dynamic time series data of an enterprise through multi-source data, and perform standardization processing to output a standardized risk data set; The feature extraction module is configured to extract static financial features, dynamic derived features and associated network features based on the standardized risk data set, and generate a high-discrimination feature set through feature screening and dimensionality reduction optimization; The model training module is configured to input the high-discrimination feature set into a neural network, a Stacking integrated model and a graph neural network by adopting a multi-model integration strategy, and output an enterprise risk score and a grade; The risk early warning module is configured to automatically match an early warning rule based on the enterprise risk score and the grade, and perform real-time monitoring and early warning on enterprise credit according to the early warning rule.

[0044] In some embodiments, the feature extraction module is specifically used to connect the enterprise ERP system and the credit investigation platform API, collect balance sheets, profit statements, cash flow statements, and historical default records, and use OCR technology to identify financial reports to obtain structured financial data. Based on the large language model, the implicit risk signals of the annual report of the enterprise are analyzed, and unstructured text data is obtained. Using federated learning technology to collect real-time transaction records, Internet of Things production data and economic indicators, and completing data encryption and fusion without sharing raw data, dynamic time series data is obtained. A generative adversarial network is constructed, an LSTM time series prediction model is used to fill in short-term missing values, and long-term missing data is marked as data anomaly features. Based on the pre-set standardized rule library, the financial indicators are quantile standardized, and the time series data is uniformly converted to the same timestamp format, and the standardized risk data set is output.

[0045] In some embodiments, the model training module is specifically used to extract indicators reflecting the basic financial condition of the enterprise from the standardized risk data set, including solvency indicators, profitability indicators, and operational capacity indicators, to form static financial features. Based on the dynamic time series data in the standardized risk data set, the rate of change, volatility, and trend slope of the indicators are calculated through time series analysis, and dynamic derived features are generated by combining sliding window statistical methods. According to the enterprise correlation data, a network topology structure is constructed, node degree centrality, intermediate centrality and edge weight features are extracted, and correlation network features are generated by combining network propagation algorithms. The static financial feature subset, dynamic derived feature subset, and correlation network feature subset are subjected to feature importance evaluation and multicollinearity test, and after removing redundant features, the high-dimensional feature space is mapped to a low-dimensional space through dimensionality reduction algorithm to generate a high-discriminatory feature set.

[0046] In some embodiments, the risk warning module is specifically used to input static financial features and dynamic derived features into a neural network, and input correlation network features into a graph neural network to obtain meta-features as a Stacking ensemble model. The neural network is trained using regularization and early stopping mechanisms, the graph neural network is optimized using neighborhood aggregation algorithms to propagate node features, the Stacking ensemble model is combined by selecting base classifiers and meta-classifiers through cross-validation, and the model training result is compared with the pre-set validation set to obtain the validation set risk prediction error. The model weights are dynamically adjusted based on the validation set risk prediction error, and the base risk scores output by the neural network and the associated conduction scores output by the graph neural network are fused based on the meta features to generate input features of the Stacking ensemble model; Based on the input features, the Stacking ensemble model outputs a comprehensive risk score, and the comprehensive risk score is mapped to a risk level through a pre-set risk level division rule. The risk level includes at least three levels of low risk, medium risk and high risk.

[0047] In some embodiments, the risk warning module is specifically configured to divide the original sample data into a training set, a validation set and a test set, and the validation set is independent of the model training process. Based on the training set, the neural network, the Stacking ensemble model and the graph neural network are trained, and the trained neural network and the graph neural network are used to predict the risk of the validation set, respectively. The predicted risk score is compared with the real risk label of the sample, and the validation set risk prediction error is calculated through a pre-set error index.

[0048] In some embodiments, the risk warning module is specifically configured to pre-set multi-dimensional warning rules, including risk level threshold rules, score fluctuation rules and feature anomaly rules. The enterprise risk score and level are input into the rule library, and the corresponding warning rules are automatically associated through a matching algorithm to generate a warning instruction containing a warning level, a trigger condition and a disposal suggestion. Real-time collection of enterprise credit data and associated features, updating of risk scores and levels according to pre-set periods; If the updated score or level meets the trigger condition of the matched warning rule, the warning signal is pushed to the credit management system through the interface, and the risk control personnel are notified synchronously; Record the warning response results and feed back the disposal effect to the rule library.

[0049] In some embodiments, the risk warning module is specifically configured to collect warning response results and actual credit risk performance of the corresponding enterprise to generate a feedback data set. Based on the feedback data set, the standardized risk data set is updated, and the importance evaluation threshold is optimized through an incremental learning mechanism to dynamically adjust the weight proportions of static financial features, dynamic derived features and associated network features; Based on the updated high-discrimination feature set, the multi-model training process is re-executed, and the actual risk results in the feedback data set are used as a new validation set; The model weight distribution strategy is corrected through the validation set risk prediction error to realize the adaptive iteration of the neural network, the Stacking ensemble model and the graph neural network. According to actual triggering effectiveness of the early warning rules in the feedback data set, the threshold parameters of the early warning rule library are adjusted.

[0050] The application further discloses a control device.

[0051] Specifically, the control device comprises a memory and a processor, and the memory stores a computer program capable of being loaded and executed by the processor and implementing the AI Agent-driven credit risk early warning strategy automatic evaluation method.

[0052] The application further discloses a computer readable storage medium.

[0053] Specifically, the computer readable storage medium stores a computer program capable of being loaded and executed by the processor and implementing the AI Agent-driven credit risk early warning strategy automatic evaluation method, and the computer readable storage medium comprises various storage medium capable of storing program codes, for example, a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0054] The above are preferred embodiments of the application, and are not intended to limit the protection scope of the application, so that: equivalent changes made according to the structure, shape, principle of the application should be covered within the protection scope of the application.

Claims

1. An AI Agent-driven automated credit risk warning strategy assessment method, characterized by: include: Acquire the company's structured financial data, unstructured text data, and dynamic time series data through multi-source data, perform standardization processing, and output standardized risk data sets; Based on the standardized risk data set, static financial features, dynamic derivative features, and associated network features are extracted, and a high-discrimination feature set is generated through feature screening and dimensionality reduction optimization; Adopting a multi-model integration strategy, the high-discrimination feature set is input into a neural network, a stacking integration model, and a graph neural network to output the enterprise risk score and level; Automatically match warning rules based on the enterprise risk score and level, and conduct real-time monitoring and warning of enterprise credit according to the warning rules; The multi-model integration strategy is used to input the high-discrimination feature set into the neural network, stacking integration model and graph neural network to output the enterprise risk score and level, including: Inputting the static financial features and the dynamic derivative features into a neural network, and inputting the associated network features into a graph neural network to obtain meta-features as a stacking integration model; Regularization and early stopping mechanisms are used to train neural networks. Neural network features are optimized with a neighborhood aggregation algorithm. The stacking ensemble model uses cross-validation to select base classifiers and meta-classifiers. The model training results are compared with the preset validation set to obtain the validation set risk prediction error. Dynamically adjust the weights of each model based on the risk prediction error of the validation set, and generate input features of the Stacking integration model by fusing the basic risk score output by the neural network and the association conduction score output by the graph neural network through weighted voting based on the meta-features; Based on the input features, a comprehensive risk score is output through the Stacking integration model, and the comprehensive risk score is mapped to a risk level through a preset risk level classification rule. The risk level includes at least three levels: low risk, medium risk and high risk.

2. The AI ​​Agent-driven automated credit risk warning strategy assessment method according to claim 1, characterized in that: The method of acquiring the enterprise's structured financial data, unstructured text data, and dynamic time series data through multi-source data, performing standardization processing, and outputting a standardized risk data set includes: Connect to the enterprise ERP system and the credit platform API to collect balance sheets, income statements, cash flow statements and historical default records, and use OCR technology to identify financial reports to obtain the structured financial data; Analyzing implicit risk signals in corporate annual reports based on a large language model to obtain the unstructured text data; Use federated learning technology to collect real-time enterprise transaction flows, IoT production data, and economic indicators, complete data encryption and fusion without sharing the original data, and obtain the dynamic time series data; Build a generative adversarial network and use the LSTM time series prediction model to fill in short-term missing values ​​and mark long-term missing data as data anomalies. Based on a preset standardization rule library, the financial indicators are quantile-normalized, the time series data are uniformly converted into the same timestamp format, and the standardized risk data set is output.

3. The AI ​​Agent-driven automated credit risk early warning strategy assessment method according to claim 2, characterized in that: Based on the standardized risk data set, static financial features, dynamic derivative features and associated network features are extracted, and a high-discrimination feature set is generated through feature screening and dimensionality reduction optimization, including: Extracting indicators reflecting the basic financial status of the enterprise from the standardized risk data set, including debt repayment ability indicators, profitability indicators and operating ability indicators, to form the static financial characteristics; Based on the dynamic time series data in the standardized risk data set, the rate of change, volatility and trend slope of the indicator are calculated through time series analysis, and the dynamic derived features are generated in combination with the sliding window statistical method; Constructing a network topology structure based on enterprise association relationship data, extracting node degree centrality, betweenness centrality and edge weight features, and generating the association network features by combining the network propagation algorithm; The static financial feature subset, dynamic derivative feature subset and associated network feature subset are subjected to feature importance evaluation and multicollinearity test. After removing redundant features, the high-dimensional feature space is mapped to the low-dimensional space through the dimensionality reduction algorithm to generate a high-discrimination feature set.

4. The AI ​​Agent-driven automated credit risk warning strategy assessment method according to claim 3, characterized in that: Comparing the model training results with a preset validation set to obtain the validation set risk prediction error includes: Divide the original sample data into a training set, a validation set, and a test set, wherein the validation set is independent of the model training process; Based on the training set, the neural network, stacking integration model and graph neural network are trained, and the risk prediction of the validation set is performed using the trained neural network and graph neural network respectively. The risk score of the predicted output is compared with the true risk label of the sample, and the risk prediction error of the validation set is calculated using the preset error indicator.

5. The AI ​​Agent-driven automated credit risk warning strategy evaluation method according to claim 4, characterized in that: The automatic matching of early warning rules based on the enterprise risk score and level, and real-time monitoring and early warning of enterprise credit according to the early warning rules, include: Preset multi-dimensional warning rules, including risk level threshold rules, score fluctuation rules and feature anomaly rules; The enterprise risk score and level are input into the rule library, and the corresponding warning rules are automatically associated through the matching algorithm to generate warning instructions including warning level, trigger conditions and disposal suggestions; Collect corporate credit data and related features in real time, and update risk scores and levels according to preset cycles; If the updated score or grade meets the triggering conditions of the matching warning rules, a warning signal will be pushed to the credit management system through the interface, and the risk control personnel will be notified simultaneously; Record the warning response results and feed back the disposal effects to the rule base.

6. The AI ​​Agent-driven automated credit risk early warning strategy evaluation method according to claim 5, characterized in that: After recording the warning response results and feeding the disposal effects back to the rule base, it also includes: Collect the warning response results and the actual credit risk performance of the corresponding enterprises to generate a feedback data set; The standardized risk dataset is updated based on the feedback dataset, and the importance assessment threshold is optimized through an incremental learning mechanism to dynamically adjust the weight ratios of the static financial features, the dynamic derivative features, and the associated network features; Based on the updated high-discrimination feature set, re-execute the multi-model training process, and use the actual risk results in the feedback dataset as the new validation set; Adaptive iteration of neural networks, stacking integration models, and graph neural networks is achieved through the weight allocation strategy of the validation set risk prediction error correction model. According to the actual triggering effectiveness of the warning rules in the feedback data set, the threshold parameters of the warning rule library are adjusted.

7. An AI Agent-driven automated credit risk warning strategy assessment device, characterized in that: The device comprises: The data processing module acquires the enterprise's structured financial data, unstructured text data, and dynamic time series data from multiple sources, performs standardization processing, and outputs a standardized risk data set; A feature extraction module is used to extract static financial features, dynamic derivative features, and associated network features based on the standardized risk data set, and generate a high-discrimination feature set through feature screening and dimensionality reduction optimization; A model training module is used to adopt a multi-model integration strategy to input the high-discrimination feature set into a neural network, a stacking integration model, and a graph neural network to output an enterprise risk score and level; A risk warning module, which is used to automatically match warning rules based on the enterprise risk score and level, and conduct real-time monitoring and warning of enterprise credit according to the warning rules; The model training module is specifically used to input the static financial features and the dynamic derivative features into a neural network, and input the associated network features into a graph neural network to obtain meta-features as a stacking integration model; Regularization and early stopping mechanisms are used to train neural networks. Neural network features are optimized with a neighborhood aggregation algorithm. The stacking ensemble model uses cross-validation to select base classifiers and meta-classifiers. The model training results are compared with the preset validation set to obtain the validation set risk prediction error. Dynamically adjust the weights of each model based on the risk prediction error of the validation set, and generate input features of the Stacking integration model by fusing the basic risk score output by the neural network and the association conduction score output by the graph neural network through weighted voting based on the meta-features; Based on the input features, a comprehensive risk score is output through the Stacking integration model, and the comprehensive risk score is mapped to a risk level through a preset risk level classification rule. The risk level includes at least three levels: low risk, medium risk and high risk.

8. A control device, characterized in that: The device comprises: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 6.

Images