A method and apparatus for intelligent optimization of credit strategies based on digital twins

Abstract

This application relates to a method and apparatus for intelligent optimization of credit strategies based on digital twins. The method includes: integrating internal financial data and external related data using multi-source data fusion technology to obtain multi-source data; constructing a digital twin based on the multi-source data and knowledge graph technology; based on the digital twin, inputting real-time variable data through a pre-set credit assessment model and digital twin simulation technology to obtain risk assessment results adapted to the business scenario; constructing a risk scenario library based on the risk assessment results and the digital twin, simulating the implementation effects of different credit strategies, and generating an optimal strategy combination; based on the optimal strategy combination and pre-set credit objectives, using a particle swarm optimization algorithm to generate a dynamic adjustment plan including credit limits, interest rate pricing, collateral conditions, and post-loan management strategies; collecting feedback data from business execution, synchronizing the feedback data to the digital twin, and correcting deviations between the virtual model and the physical business.

Description

Technical Field

[0001] This application relates to the field of risk control technology, and in particular to a method and apparatus for intelligent optimization of credit strategies based on digital twins. Background Technology

[0002] With the development of the social economy, more and more small and medium-sized enterprises (SMEs) are being established, and their loan requests to banks are also increasing. Currently, when faced with loan requests from SMEs, banks make credit decisions based on reference indicators such as national policies, the company's transaction invoices, and the company's upstream and downstream suppliers and customers.

[0003] Traditional credit risk assessment mainly relies on limited structured data such as corporate financial statements and personal credit reports. It uses fixed quantitative models to statically calculate a customer's debt repayment ability and credit level, and then credit managers supplement unstructured information (such as the business owner's industry reputation, related-party transaction background, and unexpected business events) based on their experience to complete the final review.

[0004] The aforementioned credit assessment models struggle to capture the dynamic impact of external variables such as customer behavior, industry environment, and supply chain, rely excessively on manual review, and are unsuitable for loans to increasingly large enterprises. Summary of the Invention

[0005] To address the problem that credit assessment models struggle to capture the dynamic impact of external variables such as customer behavior, industry environment, and supply chain, and rely excessively on manual review, making them unsuitable for loans to increasingly large-scale enterprises, this application provides a method and apparatus for intelligent optimization of credit strategies based on digital twins.

[0006] Firstly, this application provides a digital twin-driven intelligent optimization method for credit strategies, employing the following technical solution: including:

[0007] Based on multi-source data fusion technology, internal financial data and external related data are integrated to obtain multi-source data, and a digital twin is constructed based on the multi-source data and knowledge graph technology;

[0008] Based on the digital twin, a risk assessment result adapted to the business scenario is obtained by inputting real-time variable data through a preset credit assessment model and digital twin simulation technology.

[0009] Based on the risk assessment results and the digital twin, a risk scenario library is constructed to simulate the implementation effects of different credit strategies and generate the optimal strategy combination.

[0010] Based on the optimal strategy combination and the preset credit target, a dynamic adjustment plan including credit limit, interest rate pricing, collateral conditions and post-loan management strategy is generated using the particle swarm optimization algorithm.

[0011] Collect feedback data on business execution, synchronize the feedback data to the digital twin, and correct the deviation between the virtual model and the physical business.

[0012] Preferably, the step of integrating internal financial data and external related data based on multi-source data fusion technology to obtain multi-source data, and constructing a digital twin based on the multi-source data and knowledge graph technology, includes:

[0013] The internal financial data and the external related data are cleaned and standardized. The standardization process includes missing value imputation, outlier denoising, unit unification, and data normalization to form a structured data set, thus obtaining the multi-source data.

[0014] Based on knowledge graph technology, semantic association modeling is performed on customer basic information, risk characteristics, industry impact information and business scenario information in the multi-source data to construct the digital twin. Dynamic data interaction is realized in the digital twin through the association relationship between knowledge graph nodes and edges.

[0015] Preferably, the step of obtaining a risk assessment result adapted to the business scenario based on the digital twin, through a preset credit assessment model and digital twin simulation technology, by inputting real-time variable data, includes:

[0016] Based on the digital twin, the credit assessment model is integrated with digital twin simulation technology to generate a dynamic assessment framework;

[0017] Based on the three-dimensional mapping characteristics of the digital twin, the real-time variable data is input into the dynamic assessment framework, and the initial risk evolution path is simulated and output through simulation calculation, and the risk logic preset by the dynamic assessment framework is verified.

[0018] If the dynamic assessment framework conforms to the risk logic, then the business scenario characteristics are obtained through the digital twin, and the model parameters in the dynamic assessment framework are adaptively adjusted in combination with real-time feedback data and the initial risk evolution path.

[0019] The adjusted dynamic evaluation framework is simulated and verified to output the final risk evolution path and calculate the matching degree between the final risk evolution path and the risk features in the business scenario features.

[0020] If the matching degree is greater than the preset threshold, the risk assessment result that is adapted to the business scenario will be output.

[0021] Preferably, calculating the matching degree between the final risk evolution path and the risk feature in the business scenario features includes:

[0022] The difference between the default probability distribution of the adjusted path and the historical actual default probability distribution in the business scenario is calculated using Euclidean distance.

[0023] Calculate the coverage rate of nodes in the final risk evolution path to the preset core nodes;

[0024] Based on the digital twin, select several target samples that are most similar to the characteristics of the business scenario;

[0025] Based on the digital twin, historical default data, real-time behavioral data, and industry data corresponding to the target sample are integrated to extract key features;

[0026] The actual risk level is determined based on the preset risk level quantification rules and the key characteristics mentioned above;

[0027] The predicted default probability is simulated based on the dynamic evaluation framework of the business scenario, and the predicted default probability is mapped to the predicted risk level. The difference between the actual risk level and the predicted risk level is calculated.

[0028] If the difference value, the coverage rate, and the grade difference meet the preset matching conditions, the risk assessment result adapted to the business scenario will be output.

[0029] Preferably, the step of constructing a risk scenario library based on the risk assessment results and the digital twin, simulating the implementation effects of different credit strategies, and generating an optimal strategy combination includes:

[0030] Based on the risk assessment results, the historical default data, the real-time behavioral data, and the industry data, a risk scenario library covering multiple dimensions of the risk characteristics is constructed.

[0031] Based on the objectives of credit business, multiple candidate credit strategies are defined, and the credit strategies include at least: guarantee pricing, credit line, interest rate pricing, and term structure;

[0032] Based on the simulation capabilities of the digital twin, the credit strategies of each group are input into the risk scenario library to conduct scenario stress tests at different risk levels, and the simulation results are output.

[0033] The simulation results are comprehensively evaluated based on preset credit conditions, the optimal strategy combination is selected, and the optimal strategy combination is associated with the business scenario characteristics and stored to form dynamically callable associated data.

[0034] Preferably, based on the optimal strategy combination and the preset credit target, the dynamic adjustment scheme for generating credit limits, interest rate pricing, collateral conditions, and post-loan management strategies using the particle swarm optimization algorithm includes:

[0035] Based on the optimal strategy combination, the initial parameter range of the particle swarm is defined;

[0036] All initial parameters are combined and input into the risk scenario library to simulate the performance under different risk scenarios, calculate the risk and benefit indicators of each solution, and generate an evaluation score.

[0037] Record the best solution that has performed best in the history of each particle, and also record the best optimal solution that has performed best among all particles;

[0038] Adjust the parameters of each particle based on the individual optimal solution and the global optimal solution;

[0039] When the preset number of rounds of adjustment is repeated or no better solution is found in several consecutive rounds, the optimization stops and the current global optimal solution is set as the dynamic adjustment solution.

[0040] Preferably, the feedback data collected during the service execution is synchronized to the digital twin to correct the discrepancy between the virtual model and the physical service, including:

[0041] After collecting the feedback data and performing standardized processing, it is synchronized to the database of the digital twin;

[0042] Based on the simulated predicted values ​​in the digital twin and the actual observed values ​​in the feedback data, a deviation index is calculated, which includes: risk deviation, return deviation, and risk path deviation.

[0043] The feature weights of the virtual model of the digital twin are adjusted based on the deviation index.

[0044] Secondly, this application discloses a digital twin-driven intelligent optimization device for credit strategies, which employs the following technical solution, including:

[0045] The data acquisition module is used to integrate internal financial data and external related data based on multi-source data fusion technology to obtain multi-source data, and to construct a digital twin based on the multi-source data and knowledge graph technology.

[0046] The risk assessment module, based on the digital twin, uses a preset credit assessment model and digital twin simulation technology to input real-time variable data and obtain risk assessment results adapted to the business scenario.

[0047] The strategy combination module is used to construct a risk scenario library based on the risk assessment results and the digital twin, simulate the implementation effects of different credit strategies, and generate the optimal strategy combination.

[0048] The dynamic adjustment module, based on the optimal strategy combination and the preset credit target, uses the particle swarm optimization algorithm to generate a dynamic adjustment plan including credit limit, interest rate pricing, collateral conditions and post-loan management strategies.

[0049] The physical correction module collects feedback data from business execution, synchronizes the feedback data to the digital twin, and corrects the deviation between the virtual model and the physical business.

[0050] Thirdly, this application also provides a control device, the device comprising:

[0051] It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed, such as the above-described intelligent optimization method for credit strategies based on digital twins.

[0052] Fourthly, this application also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above regarding the intelligent optimization method for credit strategies based on digital twins.

[0053] In summary, this application first integrates internal and external data through multi-source data fusion and knowledge graph technology to construct a digital twin (data cleaning, standardization, and semantic association modeling) to achieve data structuring and dynamic interaction. Next, it utilizes the digital twin to integrate credit assessment models and simulation technology, inputting real-time variable data to generate risk assessment results adapted to business scenarios (dynamic assessment framework adjustment and risk evolution path matching calculation). Then, based on the risk assessment results, it constructs a risk scenario library covering multi-dimensional risk characteristics, simulating the implementation effects of different credit strategies and selecting the optimal strategy combination. Following this optimal strategy, it uses particle swarm optimization to generate dynamic adjustment schemes for credit limits, interest rates, etc. Finally, it collects business feedback data and synchronizes it to the digital twin, correcting virtual model parameters by calculating deviation indicators such as risk and return, ensuring the model is synchronized with physical business. This facilitates the capture of the dynamic impact of external variables such as customer behavior, industry environment, and supply chain, reducing reliance on manual review and thereby improving the intelligence, accuracy, and adaptability of credit strategies for efficient credit business management. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating a digital twin-driven intelligent optimization method for credit strategies.

[0055] Figure 2 This is a structural block diagram of a credit strategy intelligent optimization device driven by digital twins. Detailed Implementation

[0056] The following combination Figure 1 - Figure 2 This application will be described in further detail.

[0057] As the digital transformation of the financial industry accelerates, credit operations are placing higher demands on the precision of risk control and the dynamism of strategy adjustments. Traditional credit strategies rely on historical data modeling, which has three main drawbacks: First, risk assessment is mostly based on static indicators, making it difficult to capture the complex impact of real-time variables such as customer behavior and industry policies. Second, strategy optimization relies on human experience, resulting in insufficient coverage of multi-dimensional risk scenarios and a potential imbalance between returns and risks. Third, the lack of an effective feedback mechanism after strategy execution makes it difficult to correct deviations between the virtual model and actual business in a timely manner, which can reduce the adaptability of the strategy in the long run. Digital twin technology, by constructing a virtual mapping of physical business operations, provides a new path for dynamic simulation, prediction, and optimization.

[0058] Reference Figure 1 The embodiments of this application include at least steps S10 to S50.

[0059] S10, based on multi-source data fusion technology, integrates internal financial data and external related data to obtain multi-source data, and constructs a digital twin based on multi-source data and knowledge graph technology.

[0060] S20, based on a digital twin, uses a pre-set credit assessment model and digital twin simulation technology to input real-time variable data and obtain risk assessment results adapted to the business scenario.

[0061] S30 constructs a risk scenario library based on risk assessment results and digital twins, simulates the implementation effects of different credit strategies, and generates the optimal strategy combination.

[0062] S40, based on the optimal strategy combination and preset credit targets, uses the particle swarm optimization algorithm to generate a dynamic adjustment plan that includes credit limit, interest rate pricing, collateral conditions and post-loan management strategies.

[0063] S50 collects feedback data from business execution, synchronizes the feedback data to the digital twin, and corrects the discrepancies between the virtual model and the physical business.

[0064] Specifically, the process begins by integrating multi-source data from both internal and external sources and constructing a digital twin using knowledge graphs, creating a virtual mapping of physical business operations. Next, based on this digital twin, credit assessment models and simulation technology are integrated, and real-time variable data is used to dynamically assess business scenario risks. Then, relying on the risk assessment results and digital twin simulation capabilities, a risk scenario library is built to simulate the implementation effects of different credit strategies (such as credit limits and interest rate pricing) and select the optimal combination. Based on the optimal strategy, a particle swarm optimization algorithm is used to generate a dynamic adjustment plan including credit limits, interest rates, and collateral conditions. Finally, business execution feedback data is collected and synchronized to the digital twin to correct deviations between the virtual model and actual business operations. This forms a complete chain of "data fusion - risk assessment - strategy optimization - dynamic adjustment - closed-loop correction," enabling precise, dynamic, and intelligent management of credit strategies and improving risk control and return balance capabilities.

[0065] In some embodiments, step S10 specifically includes the following steps: performing data cleaning and standardization on internal financial data and external related data, including missing value imputation, outlier denoising, unit unification, and data normalization to form a structured data set and obtain multi-source data; based on knowledge graph technology, performing semantic association modeling on customer basic information, risk characteristics, industry impact information, and business scenario information in the multi-source data to construct a digital twin, in which dynamic data interaction is achieved through the association relationship between knowledge graph nodes and edges.

[0066] Specifically, the foundation for building a digital twin is achieved through data cleaning and standardization, as well as knowledge graph modeling. First, internal financial data and related external data undergo standardization processes such as missing value imputation, outlier denoising, unit unification, and data normalization to eliminate noise and inconsistencies in the original data, forming a structured multi-source data set and ensuring data quality and usability. Then, based on knowledge graph technology, customer basic information, risk characteristics, industry impact information, and business scenario information from the multi-source data are semantically linked through modeling, establishing dynamic connections in the form of nodes (entities) and edges (relationships), thus constructing a digital twin that reflects the true state of physical business operations. Its role is to improve data reliability through data cleaning and standardization, and to achieve dynamic interaction of multi-dimensional information through the semantic connections of the knowledge graph, providing a precise virtual mapping foundation for subsequent risk assessment and strategy optimization based on the digital twin.

[0067] In some embodiments, step S20 specifically includes the following steps: Based on a digital twin, the credit assessment model is integrated with digital twin simulation technology to generate a dynamic assessment framework; based on the three-dimensional mapping characteristics of the digital twin, real-time variable data is input into the dynamic assessment framework, and the initial risk evolution path is simulated and output through simulation calculation, and the dynamic assessment framework is verified to conform to the preset risk logic; if the dynamic assessment framework conforms to the risk logic, business scenario characteristics are obtained through the digital twin, and the model parameters in the dynamic assessment framework are adaptively adjusted in combination with real-time feedback data and the initial risk evolution path; the adjusted dynamic assessment framework is simulated and verified, the final risk evolution path is output, and the matching degree between the final risk evolution path and the risk characteristics in the business scenario characteristics is calculated; if the matching degree is greater than a preset threshold, a risk assessment result adapted to the business scenario is output.

[0068] Specifically, the process first combines a credit assessment model with simulation technology based on a digital twin to generate a dynamically adjustable assessment framework. Utilizing the three-dimensional mapping characteristics of the digital twin, real-time variable data is input into the framework. Simulation calculations are then used to simulate the initial risk evolution path and verify whether the framework conforms to the preset risk logic. If it conforms, the model parameters in the framework are adaptively adjusted based on the business scenario characteristics obtained from the digital twin, real-time feedback data, and the initial path. The adjusted framework is then simulated and verified again, outputting the final risk evolution path and calculating its matching degree with the business scenario risk characteristics. When the matching degree exceeds a threshold, a risk assessment result adapted to the business scenario is output. Its purpose is to ensure that the risk assessment results not only conform to the preset logic but also dynamically adapt to changes in the business scenario through dynamic framework construction, real-time data input, adaptive parameter adjustment, and multiple rounds of verification, thereby improving the accuracy and scenario adaptability of risk assessment.

[0069] In some embodiments, step S20 further includes the following steps: calculating the difference between the default probability distribution of the adjusted path and the historical actual default probability distribution of the business scenario using Euclidean distance; calculating the coverage rate of nodes in the final risk evolution path to preset core nodes; selecting several target samples that are most similar to the characteristics of the business scenario from the digital twin; integrating historical default data, real-time behavioral data, and industry data corresponding to the target samples according to the digital twin, and extracting key features; determining the actual risk level according to preset risk level quantification rules and key features; simulating the predicted default probability of the business scenario characteristics according to the dynamic evaluation framework, mapping the predicted default probability to the predicted risk level, and calculating the level difference between the actual risk level and the predicted risk level; if the difference value, coverage rate, and level difference meet preset matching conditions, then outputting a risk assessment result adapted to the business scenario.

[0070] Specifically, the algorithm first uses Euclidean distance to calculate the difference between the default probability distribution of the adjusted path and the historical actual distribution, quantifying the degree of matching at the probability level. Next, it calculates the coverage rate of nodes in the final risk evolution path to preset core nodes, ensuring the path covers key risk points. Then, it selects target samples from the digital twin that are most similar to the current business scenario, integrates their historical default data, real-time behavioral data, and industry data, and extracts key features. Based on risk level quantification rules and key features, it determines the actual risk level, and simultaneously uses a dynamic assessment framework to simulate and predict the default probability, mapping it to the predicted risk level, and calculates the level difference between the two. If the difference, coverage rate, and level difference all meet preset matching conditions, it outputs a risk assessment result adapted to the business scenario. Its purpose is to ensure a high degree of fit between the risk assessment results and the actual business scenario through multi-dimensional (probability distribution, node coverage, sample comparison, level verification) quantitative verification, thereby improving the scientific rigor and accuracy of the assessment.

[0071] In some embodiments, step S30 specifically includes the following steps: constructing a risk scenario library covering multi-dimensional risk characteristics based on risk assessment results, historical default data, real-time behavioral data, and industry data; defining multiple sets of candidate credit strategies based on credit business objectives, wherein the credit strategies include at least: guarantee pricing, credit line, interest rate pricing, and term structure; inputting each set of credit strategies into the risk scenario library for scenario stress testing at different risk levels based on the simulation capabilities of digital twins, and outputting simulation results; comprehensively evaluating the simulation results based on preset credit conditions, selecting the optimal strategy combination, and storing the optimal strategy combination in association with business scenario characteristics to form dynamically callable associated data.

[0072] Specifically, based on risk assessment results, historical default data, real-time behavioral data, and industry data, a risk scenario library covering multi-dimensional risk characteristics is constructed. Then, according to credit business objectives, multiple candidate strategies, including guarantee pricing, credit limits, interest rate pricing, and term structure, are defined. Utilizing the simulation capabilities of digital twins, each strategy is input into the risk scenario library for stress testing at different risk levels, outputting the simulation results. Finally, based on preset credit conditions, the simulation results are comprehensively evaluated to select the optimal strategy combination, which is then linked and stored with business scenario characteristics, forming dynamically accessible associated data. Its function is to simulate the implementation effects of different strategies through multi-dimensional risk scenarios, accurately select the optimal strategy combination suitable for the business scenario, improve the scientific nature and dynamic adaptability of credit strategy formulation, and provide data support for actual business decisions.

[0073] In some embodiments, step S40 specifically includes the following steps: defining the initial parameter range of the particle swarm based on the optimal strategy combination; inputting all initial parameter combinations into the risk scenario library to simulate the performance under different risk scenarios, calculating the risk index and return index of each scheme, and generating an evaluation score; recording the best-performing scheme in the history of each particle, and simultaneously recording the best-performing scheme among all particles; adjusting the parameters of each particle according to the individual best scheme and the global best scheme; stopping optimization when no better scheme appears after repeated adjustments for a preset number of rounds or after multiple consecutive rounds, and setting the current global best scheme as the dynamically adjusted scheme.

[0074] Specifically, the algorithm first defines the initial parameter range of the particle swarm (such as the value range of parameters like credit limit and interest rate) based on the optimal strategy combination. This initial parameter combination is then input into a risk scenario library to simulate performance under different risk scenarios. Risk indicators (such as default probability) and return indicators for each solution are calculated, generating a comprehensive evaluation score. The algorithm records the historical best solution for each particle and the "best solution" among all particles globally. Particle parameters are adjusted based on the individual best and global best solutions to guide the search towards a better solution. Optimization stops when a preset number of iterations is reached or when no better solution is found after multiple consecutive iterations, and the current global best solution is set as the dynamically adjusted solution. Its function is to efficiently select dynamically adjusted solutions that balance risk control and return objectives through swarm intelligence algorithms combined with risk scenario simulation, thereby improving the accuracy and adaptability of credit strategies.

[0075] In some embodiments, step S50 specifically includes the following steps: collecting feedback data and performing standardization processing, then synchronizing it to the database of the digital twin; calculating deviation indicators based on the simulated prediction values ​​in the digital twin and the actual observed values ​​of the feedback data, the deviation indicators including: risk deviation, return deviation, and risk path deviation; and adjusting the feature weights of the virtual model of the digital twin based on the deviation indicators.

[0076] Specifically, the process involves collecting and standardizing feedback data from business operations, then synchronizing this data to the digital twin's database. Next, the simulated predictions of the digital twin (such as risk level and returns) are compared with the actual observed values ​​from the feedback data. Multi-dimensional deviation indicators, including risk deviation, return deviation, and risk path deviation, are calculated to quantify the differences between the virtual model and the physical business. Finally, the feature weights of the virtual model are adjusted based on these deviation indicators (i.e., the degree of influence of each risk / return factor in the model is corrected). Its purpose is to dynamically calibrate the virtual model of the digital twin through real-time feedback data, reducing the discrepancy between the virtual and real worlds, ensuring the accuracy and reliability of subsequent risk assessments and strategy optimizations, and making the model more closely aligned with actual business scenarios.

[0077] The implementation principle of a digital twin-driven intelligent optimization method for credit strategies, as described in this application, is as follows: First, a digital twin is constructed by integrating internal and external data through multi-source data fusion and knowledge graph technology (data cleaning, standardization, and semantic association modeling) to achieve data structuring and dynamic interaction. Next, the digital twin is used to integrate credit assessment models and simulation technology, inputting real-time variable data to generate risk assessment results adapted to business scenarios (dynamic assessment framework adjustment, risk evolution path matching degree calculation). Then, based on the risk assessment results, a risk scenario library covering multi-dimensional risk characteristics is constructed to simulate the implementation effects of different credit strategies and select the optimal strategy combination. Based on the optimal strategy, a particle swarm optimization algorithm is used to optimize and generate dynamic adjustment schemes for credit limits, interest rates, etc. Finally, business feedback data is collected and synchronized to the digital twin. The virtual model parameters are corrected by calculating deviation indicators such as risk and return to ensure that the model is synchronized with the physical business. This facilitates the capture of the dynamic impact of external variables such as customer behavior, industry environment, and supply chain, reducing reliance on manual review and thus improving the intelligence, accuracy, and adaptability of credit strategies for efficient credit business management.

[0078] Figure 1 This is a flowchart illustrating a digital twin-driven intelligent optimization method for credit strategies in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows; unless explicitly stated otherwise, there is no strict order requirement for the execution of these steps, and they can be executed in other orders; and Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0079] Based on the same technical concept, referring to Figure 2 This application also provides a digital twin-driven intelligent optimization device for credit strategies, which adopts the following technical solution: The device includes:

[0080] The data acquisition module is used to integrate internal financial data and external related data based on multi-source data fusion technology to obtain multi-source data, and to build a digital twin based on multi-source data and knowledge graph technology.

[0081] The risk assessment module, based on a digital twin, uses a pre-set credit assessment model and digital twin simulation technology to input real-time variable data and obtain risk assessment results adapted to the business scenario.

[0082] The strategy combination module is used to build a risk scenario library based on risk assessment results and digital twins, simulate the implementation effects of different credit strategies, and generate the optimal strategy combination.

[0083] The dynamic adjustment module, based on the optimal strategy combination and preset credit targets, uses the particle swarm optimization algorithm to generate dynamic adjustment plans including credit limits, interest rate pricing, collateral conditions and post-loan management strategies.

[0084] The physical correction module collects feedback data from business execution, synchronizes the feedback data to the digital twin, and corrects the deviation between the virtual model and the physical business.

[0085] In some embodiments, the data acquisition module is specifically used to perform data cleaning and standardization processing on internal financial data and external related data. The standardization processing includes missing value imputation, outlier denoising, unit unification and data normalization to form a structured data set and obtain multi-source data.

[0086] Based on knowledge graph technology, semantic association modeling is performed on customer basic information, risk characteristics, industry impact information and business scenario information from multi-source data to construct a digital twin. Dynamic data interaction is achieved in the digital twin through the association relationship between knowledge graph nodes and edges.

[0087] In some embodiments, the risk assessment module is specifically used to integrate the credit assessment model with digital twin simulation technology based on the digital twin to generate a dynamic assessment framework;

[0088] Based on the three-dimensional mapping characteristics of digital twins, real-time variable data is input into the dynamic assessment framework. The initial risk evolution path is simulated and output through simulation calculation, and the risk logic preset by the dynamic assessment framework is verified.

[0089] If the dynamic assessment framework conforms to the risk logic, then the business scenario characteristics are obtained through the digital twin, and the model parameters in the dynamic assessment framework are adaptively adjusted in combination with real-time feedback data and the initial risk evolution path.

[0090] The adjusted dynamic assessment framework is simulated and verified to output the final risk evolution path and calculate the matching degree between the final risk evolution path and the risk characteristics in the business scenario.

[0091] If the matching degree is greater than the preset threshold, the risk assessment result that is adapted to the business scenario will be output.

[0092] In some embodiments, the risk assessment module is also used to calculate the difference between the default probability distribution of the adjusted path and the historical actual default probability distribution of the business scenario using Euclidean distance;

[0093] Calculate the coverage of nodes in the final risk evolution path to the preset core nodes;

[0094] Select several target samples from the digital twin that are most similar to the characteristics of the business scenario;

[0095] Based on the integration of historical default data, real-time behavioral data, and industry data corresponding to the target sample using the digital twin, key features are extracted.

[0096] The actual risk level is determined based on the preset risk level quantification rules and key characteristics;

[0097] Based on the dynamic evaluation framework, the predicted default probability is simulated based on the characteristics of the business scenario, and the predicted default probability is mapped to the predicted risk level. The difference between the actual risk level and the predicted risk level is calculated.

[0098] If the difference value, coverage rate, and grade difference meet the preset matching conditions, the risk assessment result adapted to the business scenario will be output.

[0099] In some embodiments, the strategy combination module is specifically used to construct a risk scenario library covering multiple dimensions of risk characteristics based on risk assessment results, historical default data, real-time behavioral data, and industry data.

[0100] Based on the objectives of credit business, multiple candidate credit strategies are defined. Each credit strategy includes at least: guarantee pricing, credit line, interest rate pricing, and term structure.

[0101] Based on the simulation capabilities of digital twins, each group of credit strategies is input into a risk scenario library to conduct scenario stress tests at different risk levels, and the simulation results are output.

[0102] The simulation results are comprehensively evaluated based on preset credit conditions to select the optimal strategy combination. The optimal strategy combination is then associated with business scenario characteristics and stored to form dynamically accessible related data.

[0103] In some embodiments, the dynamic adjustment module is specifically used to define the initial parameter range of the particle swarm based on the optimal strategy combination;

[0104] All initial parameters are combined and input into the risk scenario library to simulate the performance under different risk scenarios, calculate the risk and benefit indicators of each solution, and generate an evaluation score.

[0105] Record the best solution that has performed best in the history of each particle, and also record the best optimal solution that has performed best among all particles;

[0106] Adjust the parameters of each particle based on the individual optimal solution and the global optimal solution;

[0107] When the preset number of rounds has been repeatedly adjusted or no better solution has been found in several consecutive rounds, optimization stops and the current global optimal solution is set as the dynamic adjustment solution.

[0108] In some embodiments, the physical correction module is specifically used to collect feedback data, perform standardization processing, and then synchronize it to the database of the digital twin;

[0109] Based on the simulated predicted values ​​in the digital twin and the actual observed values ​​in the feedback data, deviation indicators are calculated, including: risk deviation, return deviation, and risk path deviation.

[0110] Based on the deviation index, the feature weights of the virtual model of the digital twin are adjusted.

[0111] This application also discloses a control device.

[0112] Specifically, the control device includes a memory and a processor, with the memory storing a computer program that can be loaded and executed by the processor to perform the aforementioned intelligent optimization method for credit strategies based on digital twins.

[0113] This application also discloses a computer-readable storage medium.

[0114] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed, such as the aforementioned intelligent optimization method for credit strategies based on digital twins. The computer-readable storage medium includes, for example, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0115] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for intelligent optimization of credit strategies based on digital twins, characterized in that, include: Based on multi-source data fusion technology, internal financial data and external related data are integrated to obtain multi-source data, and a digital twin is constructed based on the multi-source data and knowledge graph technology. The multi-source data includes: customer basic information, risk characteristics, industry impact information and business scenario information. Based on the digital twin, a risk assessment result adapted to the business scenario is obtained by inputting real-time variable data through a preset credit assessment model and digital twin simulation technology. Based on the risk assessment results and the digital twin, a risk scenario library is constructed to simulate the implementation effects of different credit strategies and generate the optimal strategy combination. Based on the optimal strategy combination and the preset credit target, a dynamic adjustment scheme including credit limit, interest rate pricing, collateral conditions and post-loan management strategy is generated using the particle swarm optimization algorithm. Collect feedback data on business execution, synchronize the feedback data to the digital twin, and correct the deviation between the virtual model and the physical business. Specifically, based on the digital twin, through a preset credit assessment model and digital twin simulation technology, real-time variable data is input to obtain risk assessment results adapted to the business scenario, including: Based on the digital twin, the credit assessment model is integrated with digital twin simulation technology to generate a dynamic assessment framework; Based on the three-dimensional mapping characteristics of the digital twin, the real-time variable data is input into the dynamic assessment framework, and the initial risk evolution path is simulated and output through simulation calculation, and the dynamic assessment framework is verified to conform to the preset risk logic. If the dynamic assessment framework conforms to the risk logic, then the business scenario characteristics are obtained through the digital twin, and the model parameters in the dynamic assessment framework are adaptively adjusted in combination with real-time feedback data and the initial risk evolution path. The adjusted dynamic evaluation framework is simulated and verified to output the final risk evolution path and calculate the matching degree between the final risk evolution path and the risk features in the business scenario features. If the matching degree is greater than the preset threshold, the risk assessment result that is adapted to the business scenario is output. The calculation of the matching degree between the final risk evolution path and the risk features in the business scenario features includes: The difference between the default probability distribution of the adjusted path and the historical actual default probability distribution in the business scenario is calculated using Euclidean distance. Calculate the coverage rate of nodes in the final risk evolution path to the preset core nodes; Based on the digital twin, select several target samples that are most similar to the characteristics of the business scenario; Based on the digital twin, historical default data, real-time behavioral data, and industry data corresponding to the target sample are integrated to extract key features; The actual risk level is determined based on the preset risk level quantification rules and the key characteristics mentioned above; The predicted default probability of the business scenario characteristics is simulated according to the dynamic evaluation framework, and the predicted default probability is mapped to the predicted risk level. The difference between the actual risk level and the predicted risk level is calculated. If the difference value, the coverage rate, and the grade difference meet the preset matching conditions, the risk assessment result adapted to the business scenario will be output.

2. The intelligent optimization method for credit strategy based on digital twin-driven according to claim 1, characterized in that, The method involves integrating internal financial data and external related data using multi-source data fusion technology to obtain multi-source data, and constructing a digital twin based on the multi-source data and knowledge graph technology, including: The internal financial data and the external related data are cleaned and standardized. The standardization process includes missing value imputation, outlier denoising, unit unification, and data normalization to form a structured data set, thus obtaining the multi-source data. Based on knowledge graph technology, the customer basic information, risk characteristics, industry impact information and business scenario information in the multi-source data are semantically associated and modeled to construct the digital twin. Dynamic data interaction is realized in the digital twin through the association relationship between knowledge graph nodes and edges.

3. The intelligent optimization method for credit strategy based on digital twin as described in claim 2, characterized in that, The process of constructing a risk scenario library based on the risk assessment results and the digital twin, simulating the implementation effects of different credit strategies, and generating an optimal strategy combination includes: Based on the risk assessment results, the historical default data, the real-time behavioral data, and the industry data, a risk scenario library covering multiple dimensions of the risk characteristics is constructed. Based on the objectives of credit business, multiple candidate credit strategies are defined, and the credit strategies include at least: guarantee pricing, credit line, interest rate pricing, and term structure; Based on the simulation capabilities of the digital twin, the credit strategies of each group are input into the risk scenario library to conduct scenario stress tests at different risk levels, and the simulation results are output. The simulation results are comprehensively evaluated based on preset credit conditions, the optimal strategy combination is selected, and the optimal strategy combination is associated with the business scenario characteristics and stored to form dynamically callable associated data.

4. The intelligent optimization method for credit strategy based on digital twin as described in claim 3, characterized in that, Based on the optimal strategy combination and the preset credit target, the particle swarm optimization algorithm is used to generate a dynamic adjustment plan including credit limit, interest rate pricing, collateral conditions, and post-loan management strategies, including: Based on the optimal strategy combination, the initial parameter range of the particle swarm is defined; All initial parameters are combined and input into the risk scenario library to simulate the performance under different risk scenarios, calculate the risk and benefit indicators of each solution, and generate an evaluation score. Record the best solution that has performed best in the history of each particle, and also record the best optimal solution that has performed best among all particles; Adjust the parameters of each particle based on the individual optimal solution and the global optimal solution; When the preset number of rounds of adjustment is repeated or no better solution is found in several consecutive rounds, the optimization stops and the current global optimal solution is set as the dynamic adjustment solution.

5. The intelligent optimization method for credit strategy based on digital twin as described in claim 1, characterized in that, The feedback data collected during the service execution is synchronized to the digital twin to correct discrepancies between the virtual model and the physical service, including: After collecting the feedback data and performing standardized processing, it is synchronized to the database of the digital twin; Based on the simulated predicted values ​​in the digital twin and the actual observed values ​​in the feedback data, a deviation index is calculated, which includes: risk deviation, return deviation, and risk path deviation. The feature weights of the virtual model of the digital twin are adjusted based on the deviation index.

6. A credit strategy intelligent optimization device based on digital twin, characterized in that, The device includes: The data acquisition module is used to integrate internal financial data and external related data based on multi-source data fusion technology to obtain multi-source data, and to construct a digital twin based on the multi-source data and knowledge graph technology. The multi-source data includes: customer basic information, risk characteristics, industry impact information and business scenario information. The risk assessment module, based on the digital twin, uses a preset credit assessment model and digital twin simulation technology to input real-time variable data and obtain risk assessment results adapted to the business scenario. The strategy combination module is used to construct a risk scenario library based on the risk assessment results and the digital twin, simulate the implementation effects of different credit strategies, and generate the optimal strategy combination. The dynamic adjustment module, based on the optimal strategy combination and the preset credit target, uses the particle swarm optimization algorithm to generate a dynamic adjustment plan including credit limit, interest rate pricing, collateral conditions and post-loan management strategies. The physical correction module collects feedback data from business execution, synchronizes the feedback data to the digital twin, and corrects the deviation between the virtual model and the physical business. The risk assessment module is specifically used to integrate the credit assessment model with digital twin simulation technology based on the digital twin to generate a dynamic assessment framework; Based on the three-dimensional mapping characteristics of the digital twin, the real-time variable data is input into the dynamic assessment framework, and the initial risk evolution path is simulated and output through simulation calculation, and the dynamic assessment framework is verified to conform to the preset risk logic. If the dynamic assessment framework conforms to the risk logic, then the business scenario characteristics are obtained through the digital twin, and the model parameters in the dynamic assessment framework are adaptively adjusted in combination with real-time feedback data and the initial risk evolution path. The adjusted dynamic evaluation framework is simulated and verified to output the final risk evolution path and calculate the matching degree between the final risk evolution path and the risk features in the business scenario features. If the matching degree is greater than the preset threshold, the risk assessment result that is adapted to the business scenario is output. The risk assessment module is also used to calculate the difference between the default probability distribution of the adjusted path and the historical actual default probability distribution of the business scenario using Euclidean distance. Calculate the coverage rate of nodes in the final risk evolution path to the preset core nodes; Based on the digital twin, select several target samples that are most similar to the characteristics of the business scenario; Based on the digital twin, historical default data, real-time behavioral data, and industry data corresponding to the target sample are integrated to extract key features; The actual risk level is determined based on the preset risk level quantification rules and the key characteristics mentioned above; The predicted default probability of the business scenario characteristics is simulated according to the dynamic evaluation framework, and the predicted default probability is mapped to the predicted risk level. The difference between the actual risk level and the predicted risk level is calculated. If the difference value, the coverage rate, and the grade difference meet the preset matching conditions, the risk assessment result adapted to the business scenario will be output.

7. A control device, characterized in that, The device includes: It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 5.

Images