Financial institution post-loan risk control method and risk control system based on fused multi-source dynamic data

Abstract

The invention relates to a financial institution post-loan risk control method and risk control system based on fused multi-source dynamic data, and the method comprises the steps: fusing the multi-source dynamic data, carrying out the classification and feature extraction of transaction / sales / production risk monitoring data and dynamic feedback data, and updating a transaction / sales / production risk assessment index and dynamic feedback driving data. Generating a transaction / sales / production risk score and a transaction / sales / production risk weight based on the transaction / sales / production risk assessment index; the transaction / sales / production risk weight is dynamically adjusted according to the dynamic feedback driving data; and a comprehensive risk score is obtained through weighted calculation, so that a risk level and a corresponding risk control measure are obtained. According to the method, multi-dimensional data values of loan enterprise operation, market environment and the like are fully mined, more accurate risk assessment on the loan repayment risk of the loan enterprise is realized, the loan repayment risk of the loan enterprise can be comprehensively measured more reasonably and scientifically, the risk early warning accuracy is improved, and the risk early warning false alarm rate is reduced.

Description

Technical Field

[0001] This invention relates to the field of financial risk management technology, and in particular to a post-loan risk control method and system for financial institutions based on the integration of multi-source dynamic data. Background Technology

[0002] With the rapid development of fintech and the advancement of digital transformation, various financial institutions are facing unprecedented challenges and opportunities. In current financial operations, there is a prevalent tendency to prioritize pre-loan management over post-loan management. For pre-loan review, financial institutions utilize digital technologies such as big data analytics and machine learning models to assess loan applicants' creditworthiness, repayment ability, and loan purpose, and to predict risks. However, after loan disbursement, financial institutions have relatively weak post-loan management measures, relying heavily on traditional post-loan risk monitoring methods and manual analysis.

[0003] Traditional post-loan risk monitoring methods and manual analysis have certain limitations:

[0004] First, it relies heavily on the experience and judgment of loan officers, resulting in insufficient efficiency and objectivity: loan officers need to manually verify the actual purpose of each loan and the financial statements of the loan companies, making it difficult to monitor the massive number of loans, such as batch credit granting to micro and small enterprises, in real time and in full; at the same time, manual judgment is easily affected by individual professional abilities and cognitive biases, and there is a lack of standardized risk assessment framework.

[0005] Second, data fragmentation makes it difficult to gain insight into post-loan risks from multiple dimensions: internal data of financial institutions is scattered, such as the post-loan fund transactions of loan enterprises being scattered across multiple platforms such as the credit management system, core transaction system, and credit card system within the financial institution; external data is not effectively integrated, forming "data silos", such as the post-loan business registration changes, tax, and judicial information of loan enterprises.

[0006] Third, relying primarily on static data makes it difficult to effectively assess risk development trends: Financial institutions use static data as the main basis for risk assessment and monitoring, with historical static data such as corporate financial reports and credit reports accounting for over 80%; there is insufficient mining of the dynamic characteristics of loan enterprises after the loan is issued, and there is a lack of capture of dynamic data such as real-time transaction flow, upstream and downstream links in the supply chain, and industry public opinion. For example, if a manufacturing enterprise is suddenly subject to environmental penalties, the financial institution cannot promptly warn of the deterioration of its repayment ability and raise the risk level to change its response measures because it has not accessed the administrative penalty data, which increases the risk of bad debts.

[0007] Fourth, risk prediction is lagging and forward-looking prevention and control measures are lacking: financial institutions mainly rely on the ex-post control logic of "overdue triggering early warning", and risk identification lags behind the actual occurrence of risks; at the same time, they use statistical models such as logistic regression based on fixed feature weights to assess risks, lacking dynamic modeling and scenario simulation capabilities, and are unable to adapt to changes in the market environment, such as interest rate fluctuations and industry policy adjustments, and thus cannot adjust the repayment risks of loan enterprises in a timely manner.

[0008] In summary, existing post-loan risk monitoring methods cannot effectively and accurately analyze the post-loan repayment risks of loan enterprises by integrating internal and external data of financial institutions; they cannot dynamically provide forward-looking warnings of potential post-loan repayment risks of enterprises, thereby reducing the risk of bank asset losses and improving the scientific nature and effectiveness of post-loan risk control. Summary of the Invention

[0009] Based on this, the purpose of this invention is to provide a post-loan risk control method for financial institutions based on the integration of multi-source dynamic data. By establishing a risk assessment indicator generation library and a dynamic risk assessment model that integrates multi-source dynamic data, this invention aims to solve the problem that existing risk assessments cannot effectively and accurately predict the repayment risks of loan enterprises during their operations, which leads to increased asset loss risks for financial institutions.

[0010] A post-loan risk control method for financial institutions based on the fusion of multi-source dynamic data includes the following steps:

[0011] S10: Obtain internal and external data of financial institutions for loan enterprises since the implementation of the previous risk control measures, and classify the internal and external data into transaction risk monitoring data, sales risk monitoring data, production risk monitoring data and dynamic feedback data, extract risk characteristics from each category of data, and update the transaction risk assessment indicators, sales risk assessment indicators, production risk assessment indicators and dynamic feedback driven data on which the previous risk control measures were based according to the risk characteristics of each category.

[0012] S20: Generate transaction risk score, sales risk score and production risk score, as well as transaction risk weight, sales risk weight and production risk weight, based on the updated transaction risk assessment index, sales risk assessment index and production risk assessment index; dynamically adjust the transaction risk weight, sales risk weight and production risk weight based on dynamic feedback-driven data; and obtain a comprehensive risk score through weighted calculation.

[0013] S30: Assess the risk level of loan enterprises based on comprehensive risk scores, and combine the risk level with the risk preferences of financial institutions and financial regulatory policies to generate corresponding risk control measures for loan enterprises.

[0014] Furthermore, the comprehensive risk score in step S20 satisfies:

[0015] S com =α′*S jy +β′*S xs +γ′*S sc

[0016] Among them, S com S represents the overall risk score. jy S represents the trading risk score. xs S represents the sales risk score. sc α′ represents the equal distribution of production risk, β′ represents the dynamic transaction risk weight, γ′ represents the dynamic sales risk weight, and γ′ represents the dynamic production risk weight.

[0017] Furthermore, the transaction risk score, sales risk score, and production risk score are obtained through the following methods:

[0018] S21-A: Use a pre-trained trading risk scoring model to score the current trading risk assessment indicators and obtain a trading risk score, which ranges from 1 to 100 points.

[0019] S21-B: Use a pre-trained sales risk scoring model to score the current sales risk assessment indicators and obtain a sales risk score, which ranges from 1 to 100 points.

[0020] S21-C: Use a pre-trained production risk scoring model to score the current production risk assessment indicators and obtain a production risk score, which ranges from 1 to 100 points.

[0021] Among them, the transaction risk scoring model, sales risk scoring model, and production risk scoring model adopt logistic regression, random forest, or XGBoost.

[0022] Furthermore, the dynamic feedback-driven data is used as an incremental training set to adjust the parameters of the transaction risk scoring model, sales risk scoring model, and production risk scoring model.

[0023] Furthermore, the transaction risk weight, sales risk weight, and production risk weight are obtained through the following methods:

[0024] SA1: Decompose comprehensive risk into a hierarchical structure, namely: target layer - comprehensive risk assessment, criteria layer - transaction risk / sales risk / production risk, and indicator layer - transaction risk assessment indicators / sales risk assessment indicators / production risk assessment indicators.

[0025] SA2: The comparison results are quantified using the Saaty 1-9 scaling method to obtain a typical judgment matrix;

[0026] SA3: Calculate the transaction risk weight α, sales risk weight β, and production risk weight γ using the square root method or the sum-of-products method.

[0027] Furthermore, it also includes testing the calculated transaction risk weight α, sales risk weight β, and production risk weight γ:

[0028] SA41: Calculate the consistency index CI, which satisfies:

[0029] CI=(λ-k) / (k-1)

[0030] Where λ is the largest eigenvalue of the typical judgment matrix, and k is the matrix order;

[0031] SA42: Query the random consistency index (RI) corresponding to the matrix order;

[0032] SA43: Validation Consistency Ratio (CR):

[0033] If the consistency ratio CR is less than 0.1, the verification is passed, and the transaction risk weight α, sales risk weight β, and production risk weight γ calculated by SA3 are used.

[0034] If the consistency ratio (CR) is greater than or equal to 0.1, the verification fails.

[0035] The consistency ratio CR satisfies:

[0036] CR = CI / RI.

[0037] Furthermore, the transaction risk assessment indicators include, but are not limited to, transaction time, transaction amount, transaction type, counterparty, list of overdue customers, list of outstanding customers, loan disbursement time, and repayment records of operational transaction data;

[0038] The sales risk assessment indicators include sales-related indicators, dynamic fluctuation indicators, and cross-validation indicators; among them, sales-related indicators include enterprise sales indicators and industry sales indicators; the enterprise sales indicators include the average sales revenue of the loan enterprise over the past m months. Sales growth rate r xs Average month-on-month growth rate of sales over n months One or more of the following; industry sales indicators, including the year-on-year growth rate of industry sales in the industry in which the loan company operates. Average year-on-year growth rate of industry sales over the past n months One or more of the following; the dynamic volatility indicator includes the loan enterprise sales volatility R. xs Sales volatility R xs This represents the ratio of the standard deviation of sales revenue over the past m months to the mean; the cross-validation indicators include the average month-on-month growth rate of sales revenue over the past n months. Compared with the average year-on-year growth rate of industry sales over n months The difference between

[0039] The production risk assessment indicators are one or both of electricity cost indicators and output indicators; among them, the electricity cost indicator includes the year-on-year growth rate of electricity cost r. df Electricity price volatility R df One or two of these methods can be used: production indicators, which can be obtained by monitoring the number of products that have gone offline and entered the warehouse of the loan enterprise; or by installing vibration sensors, current sensors, and temperature sensors on the production lines and core production equipment of the loan enterprise's core products through the Internet of Things, and calculating the daily output based on data such as the number of equipment start-ups and shutdowns, working hours, and energy consumption fluctuations; or by verifying both methods in parallel and taking the smaller value as the production indicator.

[0040] Compared with existing technologies, the post-loan risk control method for financial institutions based on the fusion of multi-source dynamic data proposed in this invention has the following beneficial effects:

[0041] 1) By integrating multi-source dynamic data, including internal and external data from financial institutions closely related to the economy such as loan enterprises, their industries, and national policies, data silos can be broken down, and the value of multi-dimensional data such as loan enterprise operations and market environment can be fully explored, ensuring the accuracy, consistency, and criticality of the data, and providing comprehensive and reliable data support for risk assessment.

[0042] 2) By preprocessing and classifying the acquired internal and external data, scattered data can be effectively integrated to achieve a more accurate risk assessment of loan repayment risks for borrowing companies;

[0043] 3) By assigning different weights to the risk assessment data of each category, and dynamically adjusting the weights of the risk assessment data of each category based on dynamic feedback driven by factors such as the loan company's own unforeseen circumstances, industry and national policy adjustments, and changes in the international transaction environment, it is possible to more reasonably and scientifically measure the loan repayment risk of the loan company, improve the accuracy of risk warning and reduce the false alarm rate of risk warning, and provide guidance for financial institutions to take risk control measures in advance to achieve the goal of reducing losses.

[0044] 4) By continuously monitoring internal and external data at different stages of risk control measures, a closed-loop risk control system is constructed, which integrates multi-source data, feedback data to drive risk scoring model iteration, risk control measures, and feedback data. This makes each risk score more accurate. Combined with the dynamic adjustment of the weight of each risk, it can provide forward-looking warnings of potential loan repayment risks for loan companies, thereby reducing the risk of asset loss for financial institutions and improving the scientificity and effectiveness of post-loan risk control.

[0045] Meanwhile, this invention proposes a post-loan risk control system for financial institutions based on the fusion of multi-source dynamic data, comprising:

[0046] A risk assessment indicator generation library, a dynamic risk assessment model, and a risk control measure generation device that integrates multi-source dynamic data, including:

[0047] A risk assessment indicator generation library that integrates multi-source dynamic data is used to obtain internal and external data of financial institutions for loan enterprises since the implementation of the previous risk control measures. The internal and external data are classified into transaction risk monitoring data, sales risk monitoring data, production risk monitoring data and dynamic feedback data. Risk characteristics are extracted from each category of data, and the transaction risk assessment indicators, sales risk assessment indicators, production risk assessment indicators and dynamic feedback-driven data on which the previous risk control measures were based are updated according to the risk characteristics of each category.

[0048] The dynamic risk assessment model is used to generate transaction risk scores, sales risk scores, and production risk scores, as well as transaction risk weights, sales risk weights, and production risk weights, based on updated transaction risk assessment indicators, sales risk assessment indicators, and production risk assessment indicators; and to dynamically adjust the transaction risk weights, sales risk weights, and production risk weights based on dynamic feedback-driven data; and to obtain a comprehensive risk score through weighted calculation.

[0049] The risk control measures generation device is used to assess the risk level of loan enterprises based on comprehensive risk scores, and combine the risk level with the risk preferences of financial institutions and financial regulatory policies to generate corresponding risk control measures for loan enterprises.

[0050] Furthermore, the dynamic risk assessment model includes a transaction risk scoring unit, a sales risk scoring unit, a production risk scoring unit, a risk weight generation unit, a weight dynamic adjustment unit, and a comprehensive risk scoring unit; wherein:

[0051] The transaction risk scoring unit is used to score the current transaction risk assessment indicators using a pre-trained transaction risk scoring model to obtain a transaction risk score, which ranges from 1 to 100 points.

[0052] The sales risk scoring unit is used to score the current sales risk assessment indicators using a pre-trained sales risk scoring model to obtain a sales risk score, which ranges from 1 to 100 points.

[0053] The production risk scoring unit is used to score the current production risk assessment indicators using a pre-trained production risk scoring model to obtain a production risk score, which ranges from 1 to 100 points.

[0054] The risk weight generation unit is used to determine the transaction risk weight α, sales risk weight β, and production risk weight γ based on the current transaction risk assessment index, current sales risk assessment index, and current production risk assessment index using the analytic hierarchy process (AHP).

[0055] The dynamic weight adjustment unit is used to adjust the transaction risk weight α, and / or sales risk weight β, and / or production risk weight γ according to the dynamic feedback characteristic data, so as to obtain the dynamic transaction risk weight α′, the dynamic sales risk weight β′, and the dynamic production risk weight γ′.

[0056] Furthermore, the risk assessment indicator generation library that integrates multi-source dynamic data includes an internal data acquisition unit, an external data acquisition unit, a data classification unit, a risk feature extraction unit, and a risk assessment indicator generation and updating unit; wherein:

[0057] The internal data acquisition unit is used to acquire internal data generated by the loan enterprise in the financial institution since the implementation of the previous risk control measures. The sources of internal data include, but are not limited to, transaction flow database, loan account database, and deduction database.

[0058] The external data acquisition unit is used to acquire external data of the loan enterprise since the implementation of the previous risk control measures. External data sources include, but are not limited to, financial statement databases, third-party financial databases, IoT device data collection databases, industry information, business registration information, tax information, judicial information, and macroeconomic policies.

[0059] The data classification unit is used to preprocess the acquired internal and external data; and to classify transaction risk monitoring data, sales risk monitoring data, production risk monitoring data, and dynamic feedback data based on key information in the preprocessed data.

[0060] The risk feature extraction unit is used to extract key risk feature variables from transaction risk monitoring data, sales risk monitoring data, production risk monitoring data, and dynamic feedback data, respectively, to form transaction risk feature data, sales risk feature data, production risk feature data, and dynamic feedback feature data.

[0061] The risk assessment indicator generation and update unit is used to update the transaction risk assessment indicators, sales risk assessment indicators, production risk assessment indicators and dynamic feedback driving data on which the previous risk control measures were based, according to transaction risk characteristic data, sales risk characteristic data, production risk characteristic data and dynamic feedback characteristic data, respectively, to obtain the current transaction risk assessment indicators, current sales risk assessment indicators, current production risk assessment indicators and current dynamic feedback driving data used for current risk prediction.

[0062] The advantages of the financial institution post-loan risk control system based on the fusion of multi-source dynamic data and the financial institution post-loan risk control method based on the fusion of multi-source dynamic data proposed in this invention will not be elaborated here.

[0063] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the post-loan risk control system for financial institutions based on the fusion of multi-source dynamic data, as described in this invention.

[0065] Figure 2 for Figure 1 The diagram shows the workflow of a financial institution's post-loan risk control system based on the fusion of multi-source dynamic data.

[0066] Figure 3 for Figure 2 Flowchart of the sub-steps in step S20. Detailed Implementation

[0067] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings of the embodiments of the present invention.

[0068] Please also refer to Figure 1 and Figure 2 ,in, Figure 1 This is a schematic diagram of the post-loan risk control system for financial institutions based on the fusion of multi-source dynamic data constructed in this invention. Figure 2 yes Figure 1 The diagram shows the workflow of a financial institution's post-loan risk control system based on the fusion of multi-source dynamic data.

[0069] The post-loan risk control system for financial institutions based on the integration of multi-source dynamic data constructed by the present invention includes a risk assessment indicator generation library 10 that integrates multi-source dynamic data, a dynamic risk assessment model 20, and a risk control measure generation device 30.

[0070] A risk assessment indicator generation library 10, which integrates multi-source dynamic data, is used to execute step S10: acquiring internal and external data of financial institutions for loan enterprises since the implementation of the previous risk control measures, classifying the internal and external data into transaction risk monitoring data, sales risk monitoring data, production risk monitoring data, and dynamic feedback data, extracting risk characteristics from each category of data, and updating the transaction risk assessment indicators, sales risk assessment indicators, production risk assessment indicators, and dynamic feedback-driven data on which the previous risk control measures were based according to the risk characteristics of each category.

[0071] In practice, the risk assessment indicator generation library that integrates multi-source dynamic data includes an internal data acquisition unit 11, an external data acquisition unit 12, a data classification unit 13, a risk feature extraction unit 14, and a risk assessment indicator generation and updating unit 15.

[0072] The internal data acquisition unit 11 is used to perform step S11: acquire the internal data generated by the loan enterprise in the financial institution since the implementation of the previous risk control measures. The internal data sources include, but are not limited to, transaction flow database, loan account database, and deduction database.

[0073] External data acquisition unit 12 is used to perform step S12: acquire external data of the loan enterprise since the implementation of the previous risk control measures. External data sources include, but are not limited to, financial statement databases, third-party financial databases, IoT device collection databases, industry information, business registration information, tax information, judicial information and macroeconomic policies.

[0074] The data classification unit 13 is used to perform step S13: preprocessing the acquired internal and external data; classifying transaction risk monitoring data, sales risk monitoring data, production risk monitoring data and dynamic feedback data according to the key information of the preprocessed data.

[0075] Transaction risk monitoring data consists of operational transaction flow data for loan companies, including but not limited to key fields such as transaction time, transaction amount, transaction type, and counterparty.

[0076] Sales risk monitoring data includes sales data of loan companies and industry sales data, such as the month-on-month growth rate of sales of loan companies in the past 3 months, the 12-month average binning (the 12-month average sales binning is high / medium / low), and the year-on-year growth rate of the industry in the past 6 months.

[0077] Production risk monitoring data includes data on electricity bill payments made by loan companies and industry representatives, as well as production indicators collected by the Internet of Things.

[0078] Dynamic feedback data includes loan repayment data, external environment change data, and risk event data. Specifically: loan repayment data includes on-time repayment rate, overdue days, and fluctuations in repayment amount; external environment change data includes industry policy adjustments, macroeconomic fluctuations, and changes in market competition; risk event data includes the handling results after loan companies trigger warning thresholds, such as loan extensions, collection efforts, asset disposal, changes in non-performing loan ratios, changes in recovered amounts, and administrative penalties and judicial judgments.

[0079] The preprocessing in step S13 includes cleaning outlier data, transformation, noise removal, filling in missing values, and standardizing data format to provide a high-quality data foundation for subsequent data classification. For example, cleaning outlier data includes removing non-operating income and expenditure such as large transfers during holidays, inter-account circular transactions, dividends, and personal consumption.

[0080] Risk feature extraction unit 14 is used to perform step S14: extract key risk feature variables from transaction risk monitoring data, sales risk monitoring data, production risk monitoring data and dynamic feedback data respectively to form transaction risk feature data, sales risk feature data, production risk feature data and dynamic feedback feature data.

[0081] Specifically, time series analysis and large-scale AI modeling techniques are used to extract key risk characteristic variables.

[0082] Time series analysis techniques are employed to extract time-series features, such as transaction frequency and capital fluctuations, from structured data in transaction risk monitoring, sales risk monitoring, production risk monitoring, and dynamic feedback data. Time series analysis is primarily used for trend and cycle detection. For example, the STL decomposition algorithm (esonal-trend decomposition using LOESS) is used to separate the seasonality, trend, and residual components of transaction flows to identify "consecutive months of decline"; the CUSUM algorithm (Cumulative Sum) is used to detect abrupt changes in transaction amounts, such as a single-day plunge exceeding a threshold; and the Holt-Winters triple exponential smoothing algorithm is used to predict future cyclical transaction trends, such as providing early warnings of a decline in flow in the next quarter.

[0083] Large-scale AI technology is used to extract profiling features, such as basic enterprise information and industry classification, from unstructured text data in transaction risk monitoring, sales risk monitoring, production risk monitoring, and dynamic feedback data. For example, the DeepSeek-R1 model is used to parse and organize unstructured data, mapping "fined 500,000 yuan" to the "administrative penalty" event type and quantifying the risk value.

[0084] The risk assessment indicator generation and update unit 15 is used to execute step S15: based on transaction risk characteristic data, sales risk characteristic data, production risk characteristic data and dynamic feedback characteristic data, update the transaction risk assessment indicators, sales risk assessment indicators, production risk assessment indicators and dynamic feedback driving data on which the previous risk control measures were based, respectively, to obtain the current transaction risk assessment indicators, current sales risk assessment indicators, current production risk assessment indicators and current dynamic feedback driving data used for current risk prediction.

[0085] Specifically, the transaction risk characteristic data updates the transaction risk assessment indicators on which the previous risk control measures were based. Transaction risk assessment indicators include, but are not limited to, transaction time, transaction amount, transaction type, counterparty, list of overdue customers, list of outstanding customers, loan disbursement time, repayment records, etc., of operational transaction flow data.

[0086] The sales risk characteristic data is updated based on the sales risk assessment indicators used as the basis for the previous risk control measures. Sales risk assessment indicators include sales-related indicators, dynamic fluctuation indicators, and cross-validation indicators.

[0087] Sales indicators include enterprise sales indicators and industry sales indicators; among them, enterprise sales indicators include the average sales of loan-receiving enterprises over the past m months. Sales growth rate r xs Average month-on-month growth rate of sales over n months One or more of the following; industry sales indicators, including the year-on-year growth rate of industry sales in the industry in which the loan company operates. Average year-on-year growth rate of industry sales over the past n months One or more of them.

[0088] Average sales satisfy: m is greater than or equal to 2;

[0089] Where x xs-i Let m represent the sales revenue for month i, and m represent the number of months in the statistics.

[0090] Sales growth rate r xs Satisfy: r xs =(x xs-i -x xs-(i-1) ) / x xs-(i-1) ×100%

[0091] Where x xs-i Let x represent the sales revenue for month i. xs-(i-1) This represents the sales revenue for month i-1.

[0092] Average month-on-month growth rate of sales satisfy: n is greater than or equal to 2;

[0093] In the formula, r xs Let represent the month-on-month growth rate for month i, and n represent the number of months in the statistics.

[0094] Industry sales growth rate satisfy:

[0095] In the formula, y xs-i Let y′ represent the total industry sales for month i. xs-i This represents the total industry sales for the same period last year in month i.

[0096] Average year-on-year growth rate of industry sales satisfy: n is greater than or equal to 2;

[0097] In the formula, rxs-i Let represent the year-on-year growth rate of industry sales in month i, and n represent the number of months in the statistics.

[0098] Dynamic volatility indicators, including the sales volatility R of loan-receiving companies. xs Sales volatility R xs This is the ratio of the standard deviation of sales over the past m months to the mean:

[0099]

[0100] In the formula, m represents the number of months for which statistics are collected, and x xs-i μ represents the sales revenue for month i. xs This represents the average sales revenue over m months.

[0101] Cross-validation metrics, including the average month-on-month growth rate of enterprise sales over n months. Compared with the average year-on-year growth rate of industry sales over n months The difference.

[0102] The production risk characteristic data is updated based on the production risk assessment indicators used in the previous risk control measures. The production risk assessment indicators are one or both of the following: electricity cost indicators and output indicators.

[0103] In the formula, the electricity price indicator includes the year-on-year growth rate of electricity price r. df Electricity price volatility R df One or two of them.

[0104] Electricity price year-on-year growth rate r df Satisfy: r df =(y df-i -y′ df-i ) / y′ df-i ×100%

[0105] In the formula, y df-i Let y′ represent the electricity cost for month i. df-i This represents the electricity bill for the i-th month of the same period last year.

[0106] Electricity price volatility R df satisfy:

[0107] In the formula, m represents the number of months for which statistics are collected, and r df-i This represents the month-on-month growth rate of electricity costs in month i. This represents the average month-on-month growth rate of electricity costs over m months.

[0108] Production indicators can be determined by monitoring the number of products that have gone offline and entered the warehouse of the loan company; or by installing vibration sensors, current sensors, and temperature sensors on the production lines and core production equipment of the loan company's core products through the Internet of Things, and estimating the daily output based on data such as the number of equipment start-ups and shutdowns, working hours, and energy consumption fluctuations; or by verifying both methods in parallel and taking the smaller value as the production indicator.

[0109] Dynamic feedback feature data updates the dynamic feedback-driven data upon which the previous risk control measures were based. Dynamic feedback-driven data includes, but is not limited to, the current decision-making period, overdue payments, defaults, macroeconomic indicators such as PMI and / or CPI, exchange rates, supply chain stability, administrative penalties and rulings, judicial decisions and judgments, etc.

[0110] The Risk Assessment Indicator Generation Library 10, which integrates multi-source dynamic data, gathers data from multiple sources, including internal and external data from financial institutions closely related to the economy, such as loan recipients, their industries, and national policies. Through preprocessing, classifying, and extracting key risk characteristics from both internal and external data, it obtains dynamic economic risk data from multiple sources, providing a high-quality data foundation for subsequent dynamic risk assessments. The library integrates scattered data, breaks down data silos, and ensures the accuracy, consistency, and criticality of the data, providing comprehensive and reliable data support for risk assessment.

[0111] The dynamic risk assessment model 20 is used to perform step S20: generate transaction risk score, sales risk score and production risk score, as well as transaction risk weight, sales risk weight and production risk weight, based on the updated transaction risk assessment index, sales risk assessment index and production risk assessment index; dynamically adjust the transaction risk weight, sales risk weight and production risk weight based on dynamic feedback-driven data; and obtain a comprehensive risk score through weighted calculation.

[0112] The dynamic risk assessment model 20 includes a transaction risk scoring unit 21-A, a sales risk scoring unit 21-B, a production risk scoring unit 21-C, a risk weight generation unit 22, a weight dynamic adjustment unit 23, and a comprehensive risk scoring unit 24.

[0113] The transaction risk scoring unit 21-A is used to execute step S21-A: using a pre-trained transaction risk scoring model to score the current transaction risk assessment indicators and obtain the transaction risk score S. jy .

[0114] The trading risk scoring model uses logistic regression, random forest, or XGBoost to map trading risk scores from 1 to 100.

[0115] The transaction risk scoring model is trained and validated based on transaction risk assessment indicators of a sample of historical loan enterprises (labeled as overdue and non-overdue).

[0116] For example, special characteristics of historical loan enterprise samples, such as the average decline in turnover over the past 3 months, the acceleration rate of decline over the past 3 months, the number of consecutive months of decline, the maximum single-month decline, and the degree of abnormality in transaction periods, can be used as feature variables to train logistic regression, random forest, or XGBoost.

[0117] Sales risk scoring unit 21-B is used to perform step S21-B: using a pre-trained sales risk scoring model to score the current sales risk assessment indicators, and obtain the sales risk score S. xs .

[0118] The sales risk scoring model uses logistic regression, random forest, or XGBoost to map sales risk scores from 1 to 100.

[0119] The sales risk scoring model is trained and validated based on the sales risk assessment indicators of a sample of historical loan enterprises (labeled as overdue and non-overdue).

[0120] Production risk scoring unit 21-C is used to execute step S21-C: using a pre-trained production risk scoring model to score the current production risk assessment indicators, and obtain the production risk score S. sc .

[0121] The production risk scoring model uses logistic regression, random forest, or XGBoost to map production risk scores from 1 to 100.

[0122] The production risk scoring model is trained and validated based on production risk assessment indicators of a sample of historical loan enterprises (labeled as overdue and non-overdue).

[0123] The aforementioned transaction risk scoring model, sales risk scoring model, and production risk scoring model can be iteratively upgraded using decision-making measures, supply chain stability, and default information from dynamic feedback feature data. Specifically, online learning technology is used to use relevant data from the dynamic feedback feature data as an incremental training set to update the parameters of the transaction risk scoring model, sales risk scoring model, and production risk scoring model in real time.

[0124] For example, when the XGBoost model performs poorly on dynamic feedback feature data, online learning techniques can readjust the hyperparameters (such as learning rate and tree depth) of the transaction risk scoring model, sales risk scoring model, and production risk scoring model through grid search or Bayesian optimization to improve the model's generalization ability.

[0125] Risk weight generation unit 22 is used to perform step S22: using the hierarchical analysis model to determine the transaction risk weight α, sales risk weight β, and production risk weight γ based on the current transaction risk assessment index, the current sales risk assessment index, and the current production risk assessment index.

[0126] Using the Analytic Hierarchy Process (AHP), the transaction risk weight α, sales risk weight β, and production risk weight γ are quantified in the following manner.

[0127] SA1: Decompose comprehensive risk into a hierarchical structure, namely: target layer - comprehensive risk assessment, criteria layer - transaction risk / sales risk / production risk, and indicator layer - transaction risk assessment indicators / sales risk assessment indicators / production risk assessment indicators.

[0128] SA2: The results of the comparison are quantified using the Saaty 1-9 scale to obtain a typical judgment matrix; specifically, a cross-disciplinary expert group of 15-20 people is formed for comparison, including risk control experts, industry analysts, and credit managers.

[0129] SA3: Calculate the transaction risk weight α, sales risk weight β, and production risk weight γ using the square root method or the sum product method;

[0130] SA4: Test the transaction risk weight α, sales risk weight β, and production risk weight γ according to the consistency test in the Analytic Hierarchy Process (AHP); including the following sub-steps:

[0131] SA41: Calculate the consistency index CI, which satisfies:

[0132] CI=(λ-k) / (k-1)

[0133] Where λ is the largest eigenvalue of the typical judgment matrix, k is the matrix order, and in this embodiment the number of eigenvalues ​​is 3, corresponding to a matrix order of 3;

[0134] SA42: Query the random consistency index (RI) corresponding to the matrix order;

[0135] In this embodiment, the matrix order is 3, and its corresponding random consistency index RI satisfies RI = 0.58;

[0136] SA43: Validation Consistency Ratio (CR):

[0137] If the consistency ratio (CR) is less than 0.1, the verification is successful, and the transaction risk weight, sales risk weight, and production risk weight calculated using SA3 are used.

[0138] If the conformity ratio (CR) is greater than or equal to 0.1, the verification fails, and different response measures are taken depending on the CR value:

[0139] If 0.1 ≤ CR < 0.15, then the automatic correction algorithm is optimized.

[0140] If 0.15≤CR<0.20, then organize experts to quickly review and adjust the weights of different risks, and correct the algorithm optimization;

[0141] If CR ≥ 0.20, then organize an expert consultation and reconstruct the model.

[0142] The consistency ratio CR satisfies: CR = CI / RI. Example: For a 3rd order matrix, CI = 0.0365, RI = 0.58, and CR = 0.063 < 0.1. The test is passed, and the comprehensive risk scoring unit accepts the weight allocation.

[0143] Step SA4 is used to ensure that the decision-maker's subjective judgment is logically consistent and to avoid the distortion of calculation results due to contradictions in thinking.

[0144] The weight dynamic adjustment unit 23 is used to perform step S23: adjust the transaction risk weight α, and / or sales risk weight β, and / or production risk weight γ according to the dynamic feedback feature data to obtain the dynamic transaction risk weight α′, the dynamic sales risk weight β′, and the dynamic production risk weight γ′.

[0145] The weight dynamic adjustment unit 23 adjusts the risk contribution of transaction risk, sales risk, and production risk in real time based on dynamic feedback feature data, that is, it adjusts the weight of transaction risk, sales risk, and production risk.

[0146] For example: if macroeconomic indicators such as PMI and / or CPI fluctuate greatly, or if industry prosperity declines, the production risk weight will be automatically increased, such as increasing the production risk weight γ from 25% to the dynamic production risk weight γ′35%; if a consumption subsidy policy is implemented, the sales risk weight will be automatically reduced; if the international trading environment deteriorates, and there is a tariff war or currency war, the sales risk weight will be automatically increased; if there are cash loss events such as administrative fines or court-ordered compensation, the transaction risk weight will be automatically increased.

[0147] Furthermore, dynamic feedback feature data can also be applied to hierarchical analysis models, combining expert experience with dynamic feedback feature data to periodically update the typical judgment matrix.

[0148] For example: If dynamic feedback feature data shows that the impact of the "macroeconomic indicator PMI" on production risk is underestimated, then the relative importance of the indicator is adjusted by experts re-scoring it, thereby updating the AHP weight calculation results.

[0149] The comprehensive risk scoring unit 24 is used to perform step S24: Based on the transaction risk score, sales risk score, and production risk score, and the dynamic transaction risk weight α′, dynamic sales risk weight β′, and dynamic production risk weight γ′, a weighted summation is performed to obtain the comprehensive risk score S. com .

[0150] Specifically, the comprehensive risk score S com satisfy:

[0151] S com =α′*S jy +β′*S xs +γ′*S sc

[0152] Among them, S com S represents the overall risk score. jy S represents the trading risk score. xs S represents the sales risk score. sc α′ represents the equal distribution of production risk, β′ represents the dynamic transaction risk weight, γ′ represents the dynamic sales risk weight, and γ′ represents the dynamic production risk weight.

[0153] Risk control measure generation device 30 is used to perform step S30: assess the risk level of the loan enterprise based on the comprehensive risk score, combine the risk level with the risk preferences of financial institutions and financial regulatory policies, and generate corresponding risk control measures for the loan enterprise.

[0154] Specifically, the comprehensive risk score output by the dynamic risk assessment model is classified into levels, and then combined with the risk preferences of financial institutions and financial regulatory policies. The corresponding risk control measures for loan companies are generated through the rule engine and intelligent decision-making module.

[0155] Among them, those with a comprehensive risk score greater than 75 points are classified as high-risk.

[0156] A comprehensive risk score of 75 or less and 60 or more indicates a medium risk level.

[0157] If the overall risk score is less than 60, the risk level is low.

[0158] Risk control measures include early collection and adjustment of credit limits.

[0159] For example, when the comprehensive risk score of a loan company output by the dynamic risk assessment model increases and reaches a high risk level, the risk control measure generation device outputs risk control measures for early collection.

[0160] Furthermore, visual reports can be used to display specific data on the risk level of loan companies, as well as transaction risk assessment indicators, sales risk assessment indicators, and production risk assessment indicators.

[0161] The system uses a three-color warning system to indicate the risk level of loan recipients: the comprehensive risk score is divided into three levels: high, medium, and low. The system uses red, yellow, and green to visually display the comprehensive risk level of the customer. Red light represents high risk, yellow light represents medium risk, and green light represents low risk, enabling financial institution staff to quickly understand the risk profile of loan recipients.

[0162] Transaction risk assessment indicator charts: Data related to transaction risks of loan companies are presented in the form of charts such as line charts and bar charts, such as the trend of changes in operating transaction flow and the distribution of transaction risk scores of different customer groups, to help analyze the characteristics and patterns of transaction risks.

[0163] Sales risk assessment indicator charts: Charts are used to show the comparison between the loan company's sales data and industry data, the trend of sales risk scores over time, etc., so as to make it easy to intuitively grasp the dynamics of the loan company's sales risk.

[0164] Production risk assessment indicator charts: The charts present information such as the comparison of electricity cost data between loan companies and the industry, and the fluctuation of production risk scores, providing a visual basis for assessing the production and operation risks of enterprises.

[0165] Compared with the prior art, the present invention has the following beneficial effects:

[0166] 1) By constructing a risk assessment indicator generation library that integrates multi-source dynamic data, it gathers multi-source data, covering internal and external data of financial institutions that are closely related to the economy, such as loan enterprises, their industries, and national policies. This can break down data silos, fully explore the value of multi-dimensional data such as loan enterprise operations and market environment, ensure the accuracy, consistency and criticality of data, and provide comprehensive and reliable data support for risk assessment.

[0167] 2) By preprocessing and classifying the acquired internal and external data, scattered data can be effectively integrated to achieve a more accurate risk assessment of loan repayment risks for borrowing companies;

[0168] 3) By assigning different weights to the risk assessment data of each category, and dynamically adjusting the weights of the risk assessment data of each category based on dynamic feedback driven by factors such as the loan company's own unforeseen circumstances, industry and national policy adjustments, and changes in the international transaction environment, it is possible to more reasonably and scientifically measure the loan repayment risk of the loan company, improve the accuracy of risk warning and reduce the false alarm rate of risk warning, and provide guidance for financial institutions to take risk control measures in advance to achieve the goal of reducing losses.

[0169] 4) By continuously monitoring internal and external data at different stages of risk control measures, a closed-loop risk control system is constructed, which integrates multi-source data, feedback data to drive risk scoring model iteration, risk control measures, and feedback data. This makes each risk score more accurate. Combined with the dynamic adjustment of the weight of each risk, it can provide forward-looking warnings of potential loan repayment risks for loan companies, thereby reducing the risk of asset loss for financial institutions and improving the scientificity and effectiveness of post-loan risk control.

[0170] Based on the same inventive concept, this application also provides an electronic device, which can be a server, desktop computing device, or mobile computing device (e.g., laptop computing device, handheld computing device, tablet computer, netbook, etc.). This device includes one or more processors and a memory, wherein the processor includes a financial institution post-loan risk control system based on fused multi-source dynamic data, used to execute programs to implement a financial institution post-loan risk control method based on fused multi-source dynamic data; the memory is used to store computer programs executable by the processor.

[0171] Based on the same inventive concept, this application also provides a computer-readable storage medium corresponding to the aforementioned embodiments of the financial institution post-loan risk control method based on fused multi-source dynamic data. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the financial institution post-loan risk control method based on fused multi-source dynamic data described in any of the above embodiments.

[0172] This application may take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer storage media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information may be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to: phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0173] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and the present invention also intends to include these modifications and variations.

Claims

1. A post-loan risk control method for financial institutions based on the fusion of multi-source dynamic data, characterized in that: Includes the following steps: S10: Obtain internal and external data of financial institutions for loan enterprises since the implementation of the previous risk control measures, and classify the internal and external data into transaction risk monitoring data, sales risk monitoring data, production risk monitoring data and dynamic feedback data, extract risk characteristics from each category of data, and update the transaction risk assessment indicators, sales risk assessment indicators, production risk assessment indicators and dynamic feedback driven data on which the previous risk control measures were based according to the risk characteristics of each category. S20: Generate transaction risk score, sales risk score and production risk score, as well as transaction risk weight, sales risk weight and production risk weight based on the updated transaction risk assessment index, sales risk assessment index and production risk assessment index; The transaction risk weight, sales risk weight, and production risk weight are dynamically adjusted based on dynamic feedback-driven data. A comprehensive risk score is obtained through weighted calculation; S30: Assess the risk level of loan enterprises based on comprehensive risk scores, and combine the risk level with the risk preferences of financial institutions and financial regulatory policies to generate corresponding risk control measures for loan enterprises.

2. The post-loan risk control method for financial institutions based on the fusion of multi-source dynamic data as described in claim 1, characterized in that: The comprehensive risk score in step S20 satisfies: S com =α'*S jy +β'*S xs +γ'*S sc Among them, S com S represents the overall risk score. jy S represents the trading risk score. xs S represents the sales risk score. sc α′ represents the equal distribution of production risk, β′ represents the dynamic transaction risk weight, γ′ represents the dynamic sales risk weight, and γ′ represents the dynamic production risk weight.

3. The post-loan risk control method for financial institutions based on fused multi-source dynamic data according to claim 2, characterized in that: The transaction risk score, sales risk score, and production risk score are obtained through the following methods: S21-A: Use a pre-trained trading risk scoring model to score the current trading risk assessment indicators and obtain a trading risk score, which ranges from 1 to 100 points. S21-B: Use a pre-trained sales risk scoring model to score the current sales risk assessment indicators and obtain a sales risk score, which ranges from 1 to 100 points. S21-C: Use a pre-trained production risk scoring model to score the current production risk assessment indicators and obtain a production risk score, which ranges from 1 to 100 points. Among them, the transaction risk scoring model, sales risk scoring model, and production risk scoring model adopt logistic regression, random forest, or XGBoost.

4. The post-loan risk control method for financial institutions based on the fusion of multi-source dynamic data as described in claim 3, characterized in that: The dynamic feedback-driven data is used as an incremental training set to adjust the parameters of the transaction risk scoring model, sales risk scoring model, and production risk scoring model.

5. The post-loan risk control method for financial institutions based on the fusion of multi-source dynamic data according to any one of claims 1-4, characterized in that: The transaction risk weight, sales risk weight, and production risk weight are obtained through the following methods: SA1: Decompose comprehensive risk into a hierarchical structure, namely: target layer - comprehensive risk assessment, criteria layer - transaction risk / sales risk / production risk, and indicator layer - transaction risk assessment indicators / sales risk assessment indicators / production risk assessment indicators. SA2: The comparison results are quantified using the Saaty 1-9 scaling method to obtain a typical judgment matrix; SA3: Calculate the transaction risk weight α, sales risk weight β, and production risk weight γ using the square root method or the sum-of-products method.

6. The post-loan risk control method for financial institutions based on fused multi-source dynamic data according to claim 5, characterized in that: This also includes testing the calculated transaction risk weight α, sales risk weight β, and production risk weight γ: SA41: Calculate the consistency index CI, which satisfies: CI=(λ-k) / (k-1) Where λ is the largest eigenvalue of the typical judgment matrix, and k is the matrix order; SA42: Query the random consistency index (RI) corresponding to the matrix order; SA43: Validation Consistency Ratio (CR): If the consistency ratio CR is less than 0.1, the verification is passed, and the transaction risk weight α, sales risk weight β, and production risk weight γ calculated by SA3 are used. If the consistency ratio (CR) is greater than or equal to 0.1, the verification fails. The consistency ratio CR satisfies: CR = CI / RI.

7. The post-loan risk control method for financial institutions based on the fusion of multi-source dynamic data as described in claim 1, characterized in that: The transaction risk assessment indicators include, but are not limited to, transaction time, transaction amount, transaction type, counterparty, list of overdue customers, list of customers with outstanding debts, loan disbursement time, and repayment records of operational transaction data; The sales risk assessment indicators include sales-related indicators, dynamic fluctuation indicators, and cross-validation indicators; among them, sales-related indicators include enterprise sales indicators and industry sales indicators; the enterprise sales indicators include the average sales revenue of the loan enterprise over the past m months. Sales growth rate r xs Average month-on-month growth rate of sales over n months One or more of the following; industry sales indicators, including the year-on-year growth rate of industry sales in the industry in which the loan company operates. Average year-on-year growth rate of industry sales over the past n months One or more of the following; the dynamic volatility indicator includes the loan enterprise sales volatility R. xs Sales volatility R xs This represents the ratio of the standard deviation of sales revenue over the past m months to the mean; the cross-validation indicators include the average month-on-month growth rate of sales revenue over the past n months. Compared with the average year-on-year growth rate of industry sales over n months The difference between The production risk assessment indicators are one or both of electricity cost indicators and output indicators; among them, the electricity cost indicator includes the year-on-year growth rate of electricity cost r. df Electricity price volatility R df One or two of these methods can be used: production indicators, which can be obtained by monitoring the number of products that have gone offline and entered the warehouse of the loan enterprise; or by installing vibration sensors, current sensors, and temperature sensors on the production lines and core production equipment of the loan enterprise's core products through the Internet of Things, and calculating the daily output based on data such as the number of equipment start-ups and shutdowns, working hours, and energy consumption fluctuations; or by verifying both methods in parallel and taking the smaller value as the production indicator.

8. A post-loan risk control system for financial institutions based on the fusion of multi-source dynamic data, characterized in that: include: A risk assessment indicator generation library, a dynamic risk assessment model, and a risk control measure generation device that integrates multi-source dynamic data, including: A risk assessment indicator generation library that integrates multi-source dynamic data is used to obtain internal and external data of financial institutions for loan enterprises since the implementation of the previous risk control measures. The internal and external data are classified into transaction risk monitoring data, sales risk monitoring data, production risk monitoring data and dynamic feedback data. Risk characteristics are extracted from each category of data, and the transaction risk assessment indicators, sales risk assessment indicators, production risk assessment indicators and dynamic feedback-driven data on which the previous risk control measures were based are updated according to the risk characteristics of each category. The dynamic risk assessment model is used to generate transaction risk scores, sales risk scores, and production risk scores, as well as transaction risk weights, sales risk weights, and production risk weights, based on updated transaction risk assessment indicators, sales risk assessment indicators, and production risk assessment indicators; and to dynamically adjust the transaction risk weights, sales risk weights, and production risk weights based on dynamic feedback-driven data; and to obtain a comprehensive risk score through weighted calculation. The risk control measures generation device is used to assess the risk level of loan enterprises based on comprehensive risk scores, and combine the risk level with the risk preferences of financial institutions and financial regulatory policies to generate corresponding risk control measures for loan enterprises.

9. The post-loan risk control system for financial institutions based on the fusion of multi-source dynamic data as described in claim 8, characterized in that: The dynamic risk assessment model includes a transaction risk scoring unit, a sales risk scoring unit, a production risk scoring unit, a risk weight generation unit, a weight dynamic adjustment unit, and a comprehensive risk scoring unit; wherein: The transaction risk scoring unit is used to score the current transaction risk assessment indicators using a pre-trained transaction risk scoring model to obtain a transaction risk score, which ranges from 1 to 100 points. The sales risk scoring unit is used to score the current sales risk assessment indicators using a pre-trained sales risk scoring model to obtain a sales risk score, which ranges from 1 to 100 points. The production risk scoring unit is used to score the current production risk assessment indicators using a pre-trained production risk scoring model to obtain a production risk score, which ranges from 1 to 100 points. The risk weight generation unit is used to determine the transaction risk weight α, sales risk weight β, and production risk weight γ based on the current transaction risk assessment index, current sales risk assessment index, and current production risk assessment index using the analytic hierarchy process (AHP). The dynamic weight adjustment unit is used to adjust the transaction risk weight α, and / or sales risk weight β, and / or production risk weight γ according to the dynamic feedback feature data, so as to obtain the dynamic transaction risk weight α′, the dynamic sales risk weight β′, and the dynamic production risk weight γ′.

10. The post-loan risk control system for financial institutions based on fused multi-source dynamic data according to claim 8 or 9, characterized in that: The risk assessment indicator generation library, which integrates multi-source dynamic data, includes an internal data acquisition unit, an external data acquisition unit, a data classification unit, a risk feature extraction unit, and a risk assessment indicator generation and updating unit; wherein: The internal data acquisition unit is used to acquire internal data generated by the loan enterprise in the financial institution since the implementation of the previous risk control measures. The sources of internal data include, but are not limited to, transaction flow database, loan account database, and deduction database. The external data acquisition unit is used to acquire external data of the loan enterprise since the implementation of the previous risk control measures. External data sources include, but are not limited to, financial statement databases, third-party financial databases, IoT device data collection databases, industry information, business registration information, tax information, judicial information, and macroeconomic policies. The data classification unit is used to preprocess the acquired internal and external data; and to classify transaction risk monitoring data, sales risk monitoring data, production risk monitoring data, and dynamic feedback data based on key information in the preprocessed data. The risk feature extraction unit is used to extract key risk feature variables from transaction risk monitoring data, sales risk monitoring data, production risk monitoring data, and dynamic feedback data, respectively, to form transaction risk feature data, sales risk feature data, production risk feature data, and dynamic feedback feature data. The risk assessment indicator generation and update unit is used to update the transaction risk assessment indicators, sales risk assessment indicators, production risk assessment indicators and dynamic feedback driving data on which the previous risk control measures were based, according to transaction risk characteristic data, sales risk characteristic data, production risk characteristic data and dynamic feedback characteristic data, respectively, to obtain the current transaction risk assessment indicators, current sales risk assessment indicators, current production risk assessment indicators and current dynamic feedback driving data used for current risk prediction.

Images