Real-time post-loan risk assessment and early warning method and device based on reinforcement learning

Abstract

The invention discloses a real-time post-loan risk assessment and early warning method and device based on reinforcement learning and electronic equipment, and belongs to the technical field of financial risk control and information. The method comprises the following steps: for each service object, respectively obtaining corresponding law execution result data information, a credit report and industrial and commercial information, and recording the information in a corresponding table; automatically obtaining a pdf file of the bank statement, identifying text content in the pdf file, extracting key fields from the text content, and storing the key fields in a bank statement detail list; calculating a risk score by using the obtained information; and evaluating a risk level according to the risk score, and performing early warning according to the risk level. According to the method, the legal execution result data information, the credit report and the industrial and commercial registration information of the service object are automatically acquired, the OCR technology is utilized to identify the bank flow of the service object, the risk score is calculated, and the early warning mechanism is triggered in combination with the preset rule, so that real-time post-loan risk assessment and early warning are realized, and the loan risk is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the fields of financial risk control and information technology, and in particular to a real-time post-loan risk assessment and early warning method, device and electronic equipment based on reinforcement learning. Background Art

[0002] Currently, banks primarily conduct post-loan management through manual processes, regularly checking clients' legal enforcement status, credit records, and industrial and commercial changes. Relationship managers also collect and review clients' bank statements. This manual information collection process results in information lags, a high workload, and a high probability of errors, hindering post-loan risk assessment and early warning. Summary of the Invention

[0003] In order to solve the problems existing in the prior art, the present invention provides the following technical solutions.

[0004] A first aspect of the present invention provides a real-time post-loan risk assessment and early warning method, comprising:

[0005] Generate a list of service objects to be checked;

[0006] For each service object in the service object list, obtain the corresponding legal enforcement result data information, credit report and industrial and commercial information from the compliance data service agency, credit reporting agency and regulatory agency through standardized API, record it in the corresponding compliance data service agency execution record table, credit reporting record table and industrial and commercial information table, and update the data regularly through polling;

[0007] For each service object in the service object list, automatically obtain the bank statement PDF file, identify the text content in the PDF file, extract key fields from the text content, and store them in the bank statement details table;

[0008] Calculate risk scores using information from the execution record form, credit record form, industrial and commercial information form, and bank statement;

[0009] Assess the risk level based on the risk score and issue an early warning based on the risk level.

[0010] Preferably, the automatic acquisition of the bank statement PDF file includes: the service object uploading the bank statement PDF file through a web form in the electronic bank.

[0011] Preferably, the identifying text content in the PDF file is: using an OCR tool to identify the text content in the PDF file.

[0012] Preferably, the method further comprises: merging the execution record table, the credit record table, the industrial and commercial information table and the bank flow detail table into a unified database, and performing data cleaning based on SQL query.

[0013] Preferably, the fields of the service object list include: service object ID, name, ID number, contact information, registration information and / or account opening information;

[0014] The fields of the execution record table include: case number, execution target amount, execution status, case filing date and / or case closing date;

[0015] The fields of the credit record form include: overdue repayment, debt balance, credit limit and / or inquiry record;

[0016] The fields of the business information table include: time of change in registered capital, change in legal representative and / or change in business scope;

[0017] The fields of the bank statement include: transaction time, transfer amount, transfer counterparty, transfer purpose and / or transaction type of each transaction.

[0018] Preferably, the risk score is calculated using the information in the execution record form, the credit record form, the industrial and commercial information form, and the bank flow statement form, using the following formula:

[0019] Risk score = (W0×E)+Min((W1×F),Max1)+Min((W2×G),Max2)+Min((W3×H),Max3)+Min((W4×I),Max4);

[0020] Among them, W0 is the weight coefficient of the basic credit score, W1 is the execution risk coefficient, W2 is the credit record risk coefficient, W3 is the industrial and commercial information risk coefficient, W4 is the bank flow risk coefficient, E is the basic credit score, F is the score calculated based on the information in the execution record table, G is the score calculated based on the information in the credit record table, H is the score calculated based on the information in the industrial and commercial information table, I is the score calculated based on the information in the bank flow statement table, Max1 is the preset maximum value of F, Max2 is the preset maximum value of G, Max3 is the preset maximum value of H, and Max4 is the preset maximum value of I.

[0021] Preferably,

[0022] F = (number of cases × case score coefficient) + (amount of execution × score coefficient of the amount of execution);

[0023] G = (number of overdue payments × overdue payment score coefficient) + (total overdue amount × overdue amount score coefficient) + (longest overdue period × longest overdue period score coefficient) + (number of inquiries in the past six months × inquiry score coefficient);

[0024] H = (number of tax disputes × tax dispute score coefficient) + (number of changes in major shareholders × shareholder change score coefficient) + (change in registered capital × registered capital change score coefficient);

[0025] I = (number of consecutive months with tight cash flow × liquidity risk score coefficient) + (number of large expenditures at abnormal time points × abnormal transaction score coefficient) + (maximum single-day capital outflow as a percentage of total assets × capital concentration score coefficient); among which, each score coefficient in each formula is the expert experience value.

[0026] A second aspect of the present invention provides a real-time post-loan risk assessment and early warning device, comprising:

[0027] A service object list generation module is used to generate a service object list to be checked;

[0028] The first information acquisition module is used to obtain the corresponding legal enforcement result data information, credit report and industrial and commercial information from the compliance data service agency, credit reporting agency and regulatory agency through standardized API for each service object in the service object list, record it in the corresponding compliance data service agency execution record table, credit reporting record table and industrial and commercial information table, and update the data through regular polling;

[0029] The second information acquisition module is used to automatically obtain the bank statement PDF file for each service object in the service object list, identify the text content in the PDF file, extract key fields from the text content, and store them in the bank statement detailed table;

[0030] The risk score calculation module is used to calculate the risk score using the information in the execution record form, credit record form, industrial and commercial information form, and bank flow statement form;

[0031] The risk level assessment module is used to assess the risk level based on the risk score and issue early warnings based on the risk level.

[0032] A third aspect of the present invention provides a memory storing a plurality of instructions, wherein the instructions are used to implement the real-time post-loan risk assessment and early warning method as described in the first aspect.

[0033] The fourth aspect of the present invention provides an electronic device, comprising a processor and a memory connected to the processor, wherein the memory stores a plurality of instructions, which can be loaded and executed by the processor so that the processor can execute the real-time post-loan risk assessment and early warning method as described in the first aspect.

[0034] The beneficial effects of the present invention include:

[0035] 1. Improved risk assessment efficiency: By automatically collecting legal enforcement results data, credit reports, and industrial and commercial change data, and using OCR technology to automatically identify and extract key bank statement information, the post-loan inspection cycle is greatly shortened, improving the timeliness of risk management.

[0036] 2. Improved risk identification accuracy: The risk assessment model, built on multi-dimensional data, can more comprehensively reflect the credit status and potential risks of loan customers. By quantitatively analyzing various indicators, the accuracy of credit risk identification and early warning is effectively improved.

[0037] 3. Real-time monitoring and early warning: The system can monitor the risk changes of service objects in real time, and trigger early warning signals in time according to preset risk assessment rules and thresholds, helping financial institutions to detect and respond to potential credit risks in advance and reduce the probability of non-performing loans.

[0038] 4. Simplified workflows and reduced labor costs: Automated processing replaces a large amount of manual work, reducing the investment of account managers in collecting, organizing, and analyzing various types of data, thereby freeing up human resources to focus on higher-level risk management decisions and service optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of the real-time post-loan risk assessment and early warning method of the present invention;

[0040] Figure 2 This is a functional structure diagram of the real-time post-loan risk assessment and early warning device described in the present invention. DETAILED DESCRIPTION

[0041] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0042] The method provided by the present invention can be implemented in the following terminal environment, which may include one or more of the following components: a processor, a memory, and a display screen. The memory stores at least one instruction, which is loaded and executed by the processor to implement the method described in the following embodiments.

[0043] A processor can include one or more processing cores. It connects various components within the terminal using various interfaces and circuits. It executes instructions, programs, code sets, or instruction sets stored in memory, and accesses data stored in memory to perform various terminal functions and process data.

[0044] The memory may include random access memory (RAM) or read-only memory (ROM). The memory may be used to store instructions, programs, codes, code sets, or instructions.

[0045] The display is used to show the user interface of each application.

[0046] In addition, those skilled in the art will appreciate that the structure of the terminal described above does not limit the terminal. The terminal may include more or fewer components, or a combination of certain components, or a different arrangement of components. For example, the terminal may also include a radio frequency circuit, an input unit, a sensor, an audio circuit, a power supply, and other components, which will not be described in detail here.

[0047] Example 1

[0048] like Figure 1 As shown, an embodiment of the present invention provides a real-time post-loan risk assessment and early warning method, comprising:

[0049] S101, generating a list of service objects to be checked;

[0050] S102: For each service object in the service object list, obtain the corresponding legal enforcement result data information, credit report and industrial and commercial information from the compliance data service agency, credit reporting agency and corporate regulatory agency through a standardized API (Application Programming Interface), record them in the corresponding enforcement record table, credit record table and industrial and commercial information table, and regularly poll for updated data;

[0051] S103, for each service object in the service object list, automatically obtain the bank statement PDF file, identify the text content in the PDF file, extract key fields from the text content, and store them in the bank statement detail table;

[0052] S104, calculating a risk score using the information in the execution record form, the credit record form, the industrial and commercial information form, and the bank statement;

[0053] Preferably, the law enforcement result data is obtained through the RESTful API of the judicial data service platform;

[0054] Credit reports are obtained through the encrypted interface of the central bank's credit reporting system;

[0055] Industrial and commercial information is obtained through the Enterprise Information Disclosure API of the State Administration for Market Regulation.

[0056] S105, assess the risk level according to the risk score, and issue an early warning based on the risk level.

[0057] The automatic acquisition of a PDF file of a bank statement may include: the service recipient uploading the PDF file of the bank statement via a web form in an electronic bank. Specifically, a simple web form can be constructed in the electronic bank for the service recipient to upload the bank statement file, which only supports PDF format and has a file size limit, so as to obtain the bank statement data.

[0058] In one embodiment of the present invention, an OCR (Optical Character Recognition) tool can be used to recognize the text content in the PDF file of the bank statement, extract key fields, such as the transfer amount, transfer counterparty and transfer purpose in the statement, and store them in the bank statement details table.

[0059] In a preferred embodiment of the present invention, the real-time post-loan risk assessment and early warning method may further include: merging the execution record table, credit record table, industrial and commercial information table and bank flow detail table into a unified database, and performing data cleaning based on SQL (Structured Query Language) query.

[0060] In an embodiment of the present invention, the fields of the service object list may include: service object ID, name, ID number, contact information, enterprise registration information and / or account opening information;

[0061] The fields of the execution record table may include: case number, execution target amount, execution status, case filing date and / or case closing date;

[0062] The fields of the credit record table may include: overdue payments, debt balances, credit limits and / or inquiry records;

[0063] The fields of the business information table may include: time of change in registered capital, change in legal representative and / or change in business scope;

[0064] The fields of the bank statement may include: transaction time, transfer amount, transfer counterparty, transfer purpose and / or transaction type of each transaction.

[0065] Preferably, this embodiment can also extract transaction time, transfer amount, counterparty and purpose fields, etc. through natural language processing (NLP).

[0066] In a preferred embodiment of the present invention, the risk score is calculated using the information in the execution record form, the credit record form, the industrial and commercial information form, and the bank flow statement form, and the following formula can be used:

[0067] Risk score = (W0×E)+Min((W1×F),Max1)+Min((W2×G),Max2)+Min((W3×H),Max3)+Min((W4×I),Max4);

[0068] Among them, W0 is the weight coefficient of the basic credit score, W1 is the execution risk coefficient, W2 is the credit record risk coefficient, W3 is the industrial and commercial information risk coefficient, W4 is the bank flow risk coefficient, E is the basic credit score, F is the score calculated based on the information in the execution record table, G is the score calculated based on the information in the credit record table, H is the score calculated based on the information in the industrial and commercial information table, I is the score calculated based on the information in the bank flow statement table, Max1 is the preset maximum value of F, Max2 is the preset maximum value of G, Max3 is the preset maximum value of H, and Max4 is the preset maximum value of I.

[0069] Preferably, in order to make the risk score more accurate, the present invention also proposes another method for calculating the risk score, which is specifically as follows:

[0070]

[0071] Among them, R is the risk score, W k is the weight of the k-th risk item, is the kth risk item X k The risk term processing function, X k is the kth risk item, γ1 is the first adjustment factor of the risk score, and γ2 is the second adjustment factor of the risk score. The risk items include: basic credit score E, score F calculated based on the information in the execution record table, score G calculated based on the information in the credit record table, score H calculated based on the information in the industrial and commercial information table, and score I calculated based on the information in the bank transaction details table.

[0072] The kth risk term X k Risk item processing function It includes five risk item processing functions, as follows:

[0073] Risk item processing function of basic credit score E

[0074]

[0075] Among them, η1 is the first adjustment factor of the basic credit score risk, η2 is the second adjustment factor of the basic credit score risk, η3 is the third adjustment factor of the basic credit score risk, and η4 is the fourth adjustment factor of the basic credit score risk.

[0076] Risk item processing function of score F calculated based on the information in the execution record table

[0077]

[0078] in, is a first adjustment factor for the score calculated based on the information in the execution log table, is a second adjustment factor for the score calculated based on the information in the execution log table, is a third adjustment factor for the score calculated based on the information in the execution log table, A fourth adjustment factor for the score calculated based on the information in the execution log table.

[0079] The risk item processing function of score G calculated based on the information in the credit record table

[0080]

[0081] in, is the first adjustment factor of the score calculated based on the information in the credit record table, θ' is the phase offset, A second adjustment factor for the score calculated based on the information in the credit report form.

[0082] Risk item processing function of score H calculated based on information in the industrial and commercial information table

[0083]

[0084] Among them, β1 is the first adjustment factor of the score calculated based on the information in the industrial and commercial information table, β2 is the second adjustment factor of the score calculated based on the information in the industrial and commercial information table, β3 is the third adjustment factor of the score calculated based on the information in the industrial and commercial information table, and β4 is the fourth adjustment factor of the score calculated based on the information in the industrial and commercial information table.

[0085] Risk item processing function of score I calculated based on the information in the bank's bank statement

[0086]

[0087] Where n is the number of bank transactions, ΔI j is the change in the score calculated based on the information in the bank statement after the j-th bank transaction, θ1 is the first adjustment factor of the score calculated based on the information in the bank statement, θ2 is the second adjustment factor of the score calculated based on the information in the bank statement, and T is the total assets of the service object to be checked.

[0088] Specifically, the algorithm for calculating F, G, H, and I is as follows:

[0089] F = (number of cases × case score coefficient) + (amount of execution × score coefficient of the amount of execution);

[0090] G = (number of overdue payments × overdue payment score coefficient) + (total overdue amount × overdue amount score coefficient) + (longest overdue period × longest overdue period score coefficient) + (number of inquiries in the past six months × inquiry score coefficient);

[0091] H = (number of tax disputes × tax dispute score coefficient) + (number of changes in major shareholders × shareholder change score coefficient) + (change in registered capital × registered capital change score coefficient);

[0092] I = (number of consecutive months of tight cash flow × liquidity risk score coefficient) + (number of large expenditures at unusual times × abnormal transaction score coefficient) + (maximum single-day capital outflow as a percentage of total assets × capital concentration score coefficient). The score coefficients in each formula are based on expert experience.

[0093] Preferably, this embodiment calculates the risk score through a dynamic weight optimization model based on multi-dimensional data of the execution record table, credit record table, industrial and commercial information table, and bank flow statement table. The dynamic weight optimization model uses a reinforcement learning algorithm to adjust the weight coefficient in real time, as shown below:

[0094] After completing the risk score calculation, the system dynamically optimizes the weight coefficients W0, W1, W2, W3, and W4 through the reinforcement learning module. The specific steps are as follows:

[0095] 1. State space construction. Map the real-time risk characteristics of the service object into a state vector St, including:

[0096] Basic characteristics: current risk scores E, F, G, H, I;

[0097] Dynamic features: (1) Industry volatility index (such as manufacturing PMI); (2) Recent risk change rate (standard deviation of risk scores in the past 7 days); (3) External event markers (such as policy adjustments, public opinion events, represented by binary coding). External events include: 1. Policy adjustment events. Coding logic: If the model presets "major policy release" as the first bit, then the event is marked as 1 when it occurs, otherwise it is 0. For example: 100000 (assuming a total of 6 bits, only the first bit is triggered). 2. Trade friction events. Coding logic: Mark the escalation of trade friction as 1 and the easing as 0, such as 010000 (the second bit indicates the status of trade friction). 3. Industry volatility events. Coding logic: Abnormal fluctuations in the industry index are marked as 1, such as 001000 (the third bit indicates industry risk).

[0098] State vector: S t =[E,F,G,H,I,PMI,ΔRiskstd,0 / 1policy]

[0099] Among them, ΔRiskstd is the recent risk change rate, and 0 / 1policy is the binary coding representation of the external event marker.

[0100] 2. Action space definition. The intelligent agent outputs the adjustment amount ΔWi for the weight coefficient. The action space is:

[0101] At=[ΔW0,ΔW1,ΔW2,ΔW3,ΔW4],ΔWi∈[-0.05,0.05]

[0102] 3. Reward function design. Calculate the immediate reward R based on the effect after weight adjustment t :

[0103] R t =α·(Accuracy ncw -Accuracy old )-β·(FPR new -FPR old )-γ

[0104] ·∑|ΔW i |

[0105] Accuracy ncw : The accuracy of risk warning under the new weights (verified by historical data);

[0106] Accuracy old : Accuracy of risk warning under the old weights

[0107] FPR new : False positive rate of risk warning under the new weight (FalsePositiveRate);

[0108] FPR old : False alarm rate of risk warning under the old weight

[0109] α, β, γ are hyperparameters (default values: 0.6, 0.3, 0.1).

[0110] 4. Strategy training and updating. Algorithm selection: Use the Proximal Policy Optimization (PPO) algorithm to balance exploration and exploitation;

[0111] Training process:

[0112] (1) Initialize the policy network π θ and value network Vφ;

[0113] (2) Collect a batch of service object state-action-reward data (St, At, Rt) every 24 hours;

[0114] (3) Calculate the advantage function And update the network parameters:

[0115]

[0116]

[0117]

[0118] (4) Synchronize the updated weight coefficients to all branches.

[0119] 5. Risk score recalculation and early warning

[0120] According to the optimized weight W' i =W i +ΔW i Recalculate risk score:

[0121]

[0122] Trigger an alert based on the updated score and record the adjustment log.

[0123] In an embodiment of the present invention, an initial value may be set for the basic credit score, for example, the initial value may be set to 1000 points, indicating the default basic credit level of the service object in the absence of any negative records.

[0124] The basic credit score weight coefficient can be set according to the bank's risk preference and business strategy, for example, it can be set to 0.3.

[0125] The execution risk coefficient, credit record risk coefficient, business information risk coefficient, and bank flow risk coefficient can be determined based on the degree of impact of the corresponding risk factors on the possibility of default. For example, W1 can be set to 0.3, W2 to 0.45, W3 to 0.25, and W4 to 0.1. In the embodiment of the present invention, the maximum score values ​​for each risk factor are also set: Max1, Max2, Max3, and Max4, to ensure that the deduction range for each risk factor is reasonable and does not cause the total risk score to deviate from the actual risk situation. In the embodiment of the present invention, Max1 is set to 200 points, Max2 is set to 250 points, Max3 is set to 150 points, and Max4 is set to 100 points.

[0126] In one embodiment of the present invention, assessing the risk level according to the risk score includes:

[0127] When the risk score is ≤300 points, the risk level is A (low risk);

[0128] When the risk score is 300 points < ≤ 500 points, the risk level is B (lower risk);

[0129] When the risk score is 500 points < ≤ 700 points, the risk level is C (medium risk);

[0130] When the risk score is 700 points < 900 points, the risk level is D (higher risk);

[0131] When the risk score is >900 points, the risk level is E (high risk).

[0132] During the specific application process, the risk score may change, and the risk level of the service object needs to be updated based on the risk score.

[0133] Specifically, simple warning rules can be set. For example, when it is detected that the risk level changes to meet the warning conditions, a warning reminder will be sent. For example, email services, text messages, etc. can be used to send warning notifications.

[0134] In this specific embodiment, Zhang is a loan customer who has been approved for a personal business loan at the bank. Li is our bank's dedicated account manager, responsible for tracking and managing Zhang's credit business. The method provided by the present invention is used to perform real-time post-loan risk assessment and early warning, including the following steps:

[0135] Step 1: Determine the post-loan inspection list: Based on the bank's post-loan risk management strategy, the system automatically generates a service object list for service objects that need to undergo post-loan inspections in the near future. Because Zhang's loan has been issued for three months and has entered the first post-loan inspection cycle, Zhang is included in the service object list as a service object.

[0136] Step 2: External data call: The system triggers a scheduled task and initiates requests to the compliance data service agency, credit reporting center, and corporate regulatory agency through the API service object end to obtain relevant information about Zhang, including legal execution result data (no execution cases were found for Zhang), credit report (showing that Zhang had two new small overdue payments), and industrial and commercial information (it was found that the company of which Zhang was the legal representative had a record of tax disputes in the past quarter).

[0137] Step 3: The service recipient uploads the bank statement by himself: In accordance with the bank's requirements, Zhang uploaded the bank statement PDF file of the last three months on the website or mobile application, which was 2MB in size.

[0138] Step 4: OCR Recognition: Using an OCR tool, the bank statement PDF file uploaded by Zhang was recognized and converted, extracting key transaction information (such as date, counterparty, and amount) and storing it in a structured database. Cash flow statistics for two consecutive months were compiled, for example, by expenditure and income, revealing that expenditures exceeded income. Three large expenditures were also discovered at 2 a.m.

[0139] Step 5: Data analysis and integration: The database already has records of Zhang's basic information table. New data captured from various data sources include: execution record table (no records yet). In this case, no execution cases have been found for Zhang; credit record table: showing that Zhang has two new small overdue payments; industrial and commercial information table: the company of which Zhang is the legal representative has had a tax dispute record in the past quarter; bank flow statement: cash flow has been tight for two consecutive months (expenditure is greater than income), and there are three large expenditures in Zhang's account at abnormal times.

[0140] Based on the above information, Zhang’s risk score is calculated as follows.

[0141] The basic credit score E is set to 1000 points.

[0142] The weight coefficient W0 of the basic credit score is set to 0.3.

[0143] Additionally preset:

[0144] W1=0.3, W2=0.45, W3=0.25, W4=0.1, Max1=200, Max2=250, Max3=150, Max4=100.

[0145] Calculate separately:

[0146] F = (number of cases × case score coefficient) + (amount of execution × amount of execution score coefficient); since no cases involving Zhang were found to have been executed, the number of cases and the amount of execution are both 0, so F = 0;

[0147] G = (number of overdue payments × overdue payment score coefficient) + (total overdue amount × overdue amount score coefficient) + (longest overdue period × longest overdue period score coefficient) + (number of inquiries in the past six months × inquiry number score coefficient); Zhang has two new small overdue payments. The number of overdue payments is 2, the overdue number score coefficient is 10, the total overdue amount is 10,000, the overdue amount score coefficient is 0.001, the longest overdue period is 3, the longest overdue period score coefficient is 10, the number of inquiries in the past six months is 5, and the inquiry number score coefficient is 1. Therefore, G = 2*10 + 10,000*0.001 + 3*10 + 5*1 = 65;

[0148] H = (number of tax disputes × tax dispute score coefficient) + (number of changes in major shareholders × shareholder change score coefficient) + (registered capital change range × registered capital change score coefficient); the company of which Zhang is the legal representative had one tax dispute record, with a tax dispute score coefficient of 20, the number of changes in major shareholders was 0, and the registered capital change range remained unchanged, so H = 20;

[0149] I = (consecutive months of tight cash flow × liquidity risk score coefficient) + (number of large expenditures at abnormal time points × abnormal transaction score coefficient) + (maximum single-day cash outflow as a percentage of total assets × capital concentration score coefficient); there is one month in which expenditures exceed income, that is, the continuous month of tight cash flow is 1, and the liquidity risk score coefficient is 30; there are three large expenditures at 2 a.m., that is, the number of large expenditures at abnormal time points is 3, and the abnormal transaction score coefficient is 10 points; the maximum single-day cash outflow accounts for 10% of total assets, and the capital concentration score coefficient is 100; therefore, I = 1*30(liquidity risk)+3*10(abnormal transactions)+10%*100(capital concentration)=70.

[0150] Next, plug these values ​​into the risk scoring formula:

[0151] Risk score = (W0×E) + Min((W1×F),Max1) + Min((W2×G),Max2) + Min((W3×H),Max3) + Min((W4×I),Max4)

[0152] =(0.3×1000)+Min((0.3×0),200)+Min((0.45×60),250)+Min((0.25×20),150)+Min((0.1×70),100)

[0153] =300+0+27 (because 0.45×60=27, but Min(27,250)=27)+5 (because 0.25×20=5, but Min(5,150)=5)+7 (because 0.1×70=7, but Min(7,100)=7)

[0154] =339;

[0155] After completing Zhang's initial risk score (339 points), the system triggers the reinforcement learning weight adjustment module. The specific process is as follows:

[0156] 1. State space construction

[0157] The system extracts Zhang's real-time risk characteristics and environmental data and generates a state vector St:

[0158] Basic risk factors: F = 0, G = 65, H = 20, I = 70 (calculated in the above steps);

[0159] Dynamic features:

[0160] (1) Industry volatility index: The PMI index of the wholesale and retail industry to which Mr. Zhang belongs recently dropped to 48.5 (below the boom-bust line);

[0161] (2) External event flag: The tax dispute event flag is 1 (negative record exists);

[0162] (3) Historical adjustment effect: After the weight adjustment in the past 30 days, the accuracy of early warning for similar customers increased by 12%.

[0163] Final state vector: St = [0, 65, 20, 70, 48.5, 1, 0.12]

[0164] 2. Action Generation and Weight Adjustment

[0165] The pre-trained PPO policy network outputs action At according to St:

[0166] Action output: ΔW1 = +0.02, ΔW2 = +0.03, ΔW3 = +0.01, ΔW4 = -0.01 (W0 remains fixed);

[0167] Weight normalization: The adjusted weights are updated to: W1′=0.32, W2′=0.48, W3′=0.26, W4′=0.09

[0168] (Note: After adjustment, ∑Wi′=1.15>1, and normalized to W1″=0.278, W2″=0.417, W3″=0.226, W4″=0.078)

[0169] 3. Risk score recalculation

[0170] Recalculate Zhang's risk score using the new weights:

[0171] Scorenew=(0.3×1000)+Min(0.278×0,200)+Min(0.417×65,250)+Min(0.226×20,150)+Min(0.078×70,100)=300+0+27.1+4.5+5.5=337.1

[0172] Therefore, the final calculated risk score for Zhang was 337.1 points.

[0173] According to the risk classification standard, a score of 337.1 falls within the range of 300 < Risk Score ≤ 500, placing Zhang's risk level at B (low risk). However, the system generated a dynamic risk alert: "Due to a decline in the industry's PMI and the existence of tax disputes, it is recommended to increase the frequency of post-loan inspections (from quarterly to monthly)."

[0174] Step 6: Assess the risk level based on the risk score: Assume that Zhang's risk score is 660 points, which is greater than the risk threshold set by the bank (for example, 500 points), and the risk level is C-medium risk. The system automatically generates a risk warning and notifies the account manager Li in real time.

[0175] Step 7, real-time warning and notification: After receiving the warning, Li, as Zhang's account manager, can view Zhang's latest risk score and risk level assessment results through the internal management system, check Zhang's relevant information in detail, and contact Zhang to understand the specific situation and reasons. At the same time, the situation will be reported to the superior department for further evaluation and decision-making. Possible measures include but are not limited to adjusting loan conditions, increasing guarantee requirements or early collection.

[0176] In this way, through the solution provided by the present invention, the bank can timely and accurately grasp Zhang's credit dynamics to ensure the security of loan assets and the effectiveness of risk management.

[0177] Example 2

[0178] like Figure 2 As shown, another aspect of the present invention also includes a functional module architecture that is completely consistent with the aforementioned method process, that is, the embodiment of the present invention further provides a real-time post-loan risk assessment and early warning device, including:

[0179] A service object list generating module 201 is used to generate a service object list to be checked;

[0180] The first information acquisition module 202 is used to obtain, for each service object in the service object list, corresponding legal enforcement result data information, credit report, and industrial and commercial information from the compliance data service agency, credit reporting agency, and corporate regulatory agency through a standardized API interface, record the information in the corresponding enforcement record table, credit record table, and industrial and commercial information table, and update the data through regular polling;

[0181] The second information acquisition module 203 is used to automatically obtain a bank statement PDF file for each service object in the service object list, identify the text content in the PDF file, extract key fields from the text content, and store them in a bank statement detail table;

[0182] The risk score calculation module 204 is used to calculate the risk score using the information in the execution record form, the credit record form, the industrial and commercial information form, and the bank statement form;

[0183] The risk level assessment module 205 is used to assess the risk level according to the risk score and issue an early warning according to the risk level.

[0184] Furthermore, the automatic acquisition of the bank statement PDF file includes: the service object uploading the bank statement PDF file through a web form in the electronic bank.

[0185] Furthermore, identifying the text content in the PDF file includes: using an OCR tool to identify the text content in the PDF file.

[0186] Furthermore, the device also includes a data unification module for merging the execution record table, the credit record table, the industrial and commercial information table, and the bank flow detail table into a unified database, and performing data cleaning based on SQL queries.

[0187] Furthermore, the fields of the service object list include: service object ID, name, ID number, contact information, enterprise registration information and / or account opening information;

[0188] The fields of the execution record table include: case number, execution target amount, execution status, case filing date and / or case closing date;

[0189] The fields of the credit record form include: overdue repayment, debt balance, credit limit and / or inquiry record;

[0190] Fields in the business information form: time of change in registered capital, change in legal representative, and / or change in business scope;

[0191] The fields of the bank statement include: transaction time, transfer amount, transfer counterparty, transfer purpose and / or transaction type of each transaction.

[0192] Furthermore, the risk score is calculated using the information in the execution record form, credit record form, industrial and commercial information form, and bank flow statement form, using the following formula:

[0193] Risk score = (W0×E)+Min((W1×F),Max1)+Min((W2×G),Max2)+Min((W3×H),Max3)+Min((W4×I),Max4).

[0194] Among them, W0 is the weight coefficient of the basic credit score, W1 is the execution risk coefficient, W2 is the credit record risk coefficient, W3 is the industrial and commercial information risk coefficient, W4 is the bank flow risk coefficient, E is the basic credit score, F is the score calculated based on the information in the execution record table, G is the score calculated based on the information in the credit record table, H is the score calculated based on the information in the industrial and commercial information table, I is the score calculated based on the information in the bank flow statement table, Max1 is the preset maximum value of F, Max2 is the preset maximum value of G, Max3 is the preset maximum value of H, and Max4 is the preset maximum value of I.

[0195] Furthermore, F = (number of cases × case score coefficient) + (amount of execution × target score coefficient);

[0196] G = (number of overdue payments × overdue payment score coefficient) + (total overdue amount × overdue amount score coefficient) + (longest overdue period × longest overdue period score coefficient) + (number of inquiries in the past six months × inquiry score coefficient);

[0197] H = (number of tax disputes × tax dispute score coefficient) + (number of changes in major shareholders × shareholder change score coefficient) + (change in registered capital × registered capital change score coefficient);

[0198] I = (consecutive months of tight cash flow × liquidity risk score coefficient) + (number of large expenditures at abnormal time points × abnormal transaction score coefficient) + (maximum single-day capital outflow as a proportion of total assets × capital concentration score coefficient).

[0199] The device can be implemented by the real-time post-loan risk assessment and early warning method provided in the above-mentioned embodiment 1. The specific implementation method can be found in the description of embodiment 1 and will not be repeated here.

[0200] The present invention also provides a memory storing a plurality of instructions, wherein the instructions are used to implement the real-time post-loan risk assessment and early warning method as described in the first embodiment.

[0201] The present invention also provides an electronic device, comprising a processor and a memory connected to the processor, wherein the memory stores a plurality of instructions, which can be loaded and executed by the processor so that the processor can execute the real-time post-loan risk assessment and early warning method as described in Example 1.

[0202] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the invention. Thus, the present invention is intended to include such changes and modifications as fall within the scope of the claims and their equivalents.

Claims

1. A real-time post-loan risk assessment and early warning method, characterized in that: include: Generate a list of service objects to be checked; For each service object in the service object list, obtain the corresponding legal enforcement result data information, credit report and industrial and commercial information from the compliance data service agency, credit agency and regulatory agency through standardized API, record it in the corresponding execution record table, credit record table and industrial and commercial information table, and update the data regularly through polling; For each service object in the service object list, automatically obtain the bank statement PDF file, identify the text content in the PDF file, extract key fields from the text content, and store them in the bank statement details table; Calculate risk scores using information from the execution record form, credit record form, industrial and commercial information form, and bank statement; Assess the risk level based on the risk score and issue an early warning based on the risk level.

2. The real-time post-loan risk assessment and early warning method according to claim 1, characterized in that: The method of automatically obtaining the PDF file of the bank statement includes: the service object uploading the PDF file of the bank statement through a web form in the electronic bank.

3. The real-time post-loan risk assessment and early warning method according to claim 1, characterized in that: The identifying of the text content in the PDF file is: using an OCR tool to identify the text content in the PDF file.

4. The real-time post-loan risk assessment and early warning method according to claim 1, characterized in that: The method further includes: merging the execution record table, the credit record table, the industrial and commercial information table, and the bank flow detail table into a unified database, and performing data cleaning based on SQL queries.

5. The real-time post-loan risk assessment and early warning method according to claim 1, characterized in that: The fields of the service object list include: service object ID, name, ID number, contact information, registration information and / or account opening information; The fields of the execution record table include: case number, execution target amount, execution status, case filing date and / or case closing date; The fields of the credit record form include: overdue repayment, debt balance, credit limit and / or inquiry record; The fields of the business information table include: time of change in registered capital, change in legal representative and / or change in business scope; The fields of the bank statement include: transaction time, transfer amount, transfer counterparty, transfer purpose and / or transaction type of each transaction.

6. The real-time post-loan risk assessment and early warning method according to claim 1, characterized in that: The risk score is calculated using the information in the execution record form, credit record form, industrial and commercial information form, and bank flow statement form, using the following formula: Risk score = (W0×E)+Min((W1×F),Max1)+Min((W2×G),Max2)+Min((W3×H),Max3)+Min((W4×I),Max4); Among them, W0 is the weight coefficient of the basic credit score, W1 is the execution risk coefficient, W2 is the credit record risk coefficient, W3 is the industrial and commercial information risk coefficient, W4 is the bank flow risk coefficient, E is the basic credit score, F is the score calculated based on the information in the execution record table, G is the score calculated based on the information in the credit record table, H is the score calculated based on the information in the industrial and commercial information table, I is the score calculated based on the information in the bank flow statement table, Max1 is the preset maximum value of F, Max2 is the preset maximum value of G, Max3 is the preset maximum value of H, and Max4 is the preset maximum value of I.

7. The real-time post-loan risk assessment and early warning method according to claim 6, characterized in that: F = (number of cases × case score coefficient) + (amount of execution × score coefficient of the amount of execution); G = (number of overdue payments × overdue payment score coefficient) + (total overdue amount × overdue amount score coefficient) + (longest overdue period × longest overdue period score coefficient) + (number of inquiries in the past six months × inquiry score coefficient); H = (number of tax disputes × tax dispute score coefficient) + (number of changes in major shareholders × shareholder change score coefficient) + (change in registered capital × registered capital change score coefficient); I = (number of consecutive months with tight cash flow × liquidity risk score coefficient) + (number of large expenditures at abnormal time points × abnormal transaction score coefficient) + (maximum single-day capital outflow as a percentage of total assets × capital concentration score coefficient); among which, each score coefficient in each formula is the expert experience value.

8. A real-time post-loan risk assessment and early warning device, characterized in that: include: A service object list generation module is used to generate a service object list to be checked; The first information acquisition module is used to obtain the corresponding legal enforcement result data information, credit report and industrial and commercial information from the compliance data service agency, credit reporting agency and regulatory agency through standardized API for each service object in the service object list, record it in the corresponding execution record table, credit record table and industrial and commercial information table, and update the data through regular polling; The second information acquisition module is used to automatically obtain the bank statement PDF file for each service object in the service object list, identify the text content in the PDF file, extract key fields from the text content, and store them in the bank statement detailed table; The risk score calculation module is used to calculate the risk score using the information in the execution record form, credit record form, industrial and commercial information form, and bank flow statement form; The risk level assessment module is used to assess the risk level based on the risk score and issue early warnings based on the risk level.

9. A memory, characterized in that: Multiple instructions are stored, and the instructions are used to implement the real-time post-loan risk assessment and early warning method as described in any one of claims 1-7.

10. An electronic device, characterized in that: It includes a processor and a memory connected to the processor, the memory stores multiple instructions, and the instructions can be loaded and executed by the processor to enable the processor to execute the real-time post-loan risk assessment and early warning method as described in any one of claims 1-7.

Images