Non-performing asset disposal optimization system and method based on intelligent data analysis

Abstract

The application discloses a non-performing asset disposal optimization system and method based on intelligent data analysis, and relates to the technical field of financial management; the application integrates multi-source heterogeneous data, constructs a dynamically updated knowledge graph, extracts credit, market and disposal risk characteristics; multi-dimensional risk sub-models are constructed, historical data and optimization algorithms are used to dynamically adjust model parameters, and the prediction accuracy and generalization ability in complex scenarios are improved; based on a weighted scoring mechanism, multiple risk values are fused, and a preset threshold is combined to realize automatic determination of non-performing assets; through a genetic algorithm, a multi-objective optimization is performed on a disposal scheme, an optimal strategy is generated and executed, and a 'evaluation-disposal-feedback' closed loop is formed; the application solves the problems of data island, static model rigidity, high artificial dependence and lack of quantitative optimization of strategies in traditional systems, improves the non-performing asset identification accuracy and recovery rate, compresses the disposal period, and provides a full-process intelligent management scheme.

Description

Technical Field

[0001] This invention belongs to the field of financial management technology, and specifically relates to an optimized system and method for the disposal of non-performing assets based on intelligent data analysis. Background Technology

[0002] In the financial industry, non-performing asset disposal is a core aspect of risk management for financial institutions, directly impacting capital recovery efficiency and systemic risk prevention. However, traditional non-performing asset disposal methods have significant shortcomings in terms of technological implementation and decision-making efficiency.

[0003] Chinese patent CN109658228B discloses a full lifecycle non-performing asset management system and a non-performing asset management method. This invention encompasses a comprehensive non-performing asset management system, covering four core modules: asset issuance, pricing, distribution, and collection. The asset issuance module seamlessly integrates non-performing assets with the system to achieve information consolidation. The pricing module performs standardized valuations of assets based on market dynamics and business characteristics. The distribution module uses intelligent strategies to accurately match assets to suitable processing institutions. The collection module is responsible for subsequent collection work. Furthermore, this invention proposes a corresponding management method, collectively constituting comprehensive management of the entire lifecycle of non-performing assets.

[0004] Existing technologies for non-performing asset management face challenges such as fragmented multi-source data leading to delayed risk assessment, static models struggling to adapt to dynamic market changes, excessive reliance on manual labor resulting in low disposal efficiency, and a lack of quantitative optimization in disposal strategies leading to an imbalance between recovery rate and risk. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the problems in related technologies, this invention provides an optimized method for non-performing asset disposal based on intelligent data analysis. This invention solves the problems faced by non-performing asset management, such as risk assessment lag due to fragmented multi-source data, static models being unable to adapt to dynamic market changes, excessive reliance on manual labor resulting in low disposal efficiency, and the imbalance between recovery rate and risk caused by the lack of quantitative optimization of disposal strategies.

[0007] (II) Technical Solution

[0008] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0009] S1. Collect asset data and real-time dynamic data, and construct a dynamic knowledge graph; extract credit risk characteristics, market risk characteristics, and disposal feasibility characteristics from the dynamic real-time graph;

[0010] S2. Construct a credit risk sub-model, a market risk sub-model, and a disposal feasibility sub-model; use historical data to train the credit risk sub-model, the market risk sub-model, and the disposal feasibility sub-model to obtain the final credit risk sub-model, the final market risk sub-model, and the final disposal feasibility sub-model.

[0011] S3. Input the credit risk characteristics, market risk characteristics, and disposal feasibility characteristics into the final credit risk sub-model, the final market risk sub-model, and the final disposal feasibility sub-model, respectively, to obtain the credit risk value, market risk value, and disposal risk value; input the credit risk value, market risk value, and disposal risk value into the asset scoring function to determine whether the asset is a non-performing asset;

[0012] S4. For non-performing assets, construct a set of disposal schemes and an objective function, and combine them with an optimization algorithm to obtain the optimal processing scheme; use the optimal processing scheme to process the non-performing assets.

[0013] Preferably, step S1 includes the following steps:

[0014] S11. Collect internal bank data and external data sources through legal and compliant means to obtain asset data; collect real-time dynamic data;

[0015] S12. Clean the asset data and real-time dynamic data to obtain cleaned data; establish a dynamic knowledge graph based on the cleaned data, and automatically trigger knowledge graph reconstruction when a new policy is released or market fluctuations are detected; the dynamic knowledge graph includes entity recognition and relationship mapping between entities;

[0016] S13. Extract numerical features from the structured data in the dynamic knowledge graph; parse the unstructured data in the dynamic knowledge graph to generate text features; extract credit risk features, market risk features, and disposal feasibility features from the numerical features and text features;

[0017] The above steps ensure the comprehensiveness of asset data by collecting various internal bank data and information from external data sources, and update dynamic data in real time to maintain data timeliness. The collected asset data and real-time dynamic data are cleaned, and a dynamic knowledge graph is constructed based on the cleaned data. This knowledge graph has an intelligent update function, automatically triggering reconstruction when new policies are released or market fluctuations occur, ensuring the accuracy and timeliness of information. The knowledge graph includes entity recognition and mapping of relationships between entities, providing a structured foundation for subsequent analysis. The system extracts numerical features from the structured data of the dynamic knowledge graph, while simultaneously parsing unstructured data to generate text features, and then extracts credit risk, market risk, and disposal feasibility features from them. This solves the problem of data silos in traditional systems and provides strong support for the management and decision-making of non-performing assets.

[0018] Preferably, step S2 includes the following steps:

[0019] S21. Collect historical dynamic knowledge graphs, extract features from the historical dynamic knowledge graphs to obtain historical credit risk features, historical market risk features, and historical disposal feasibility features; collect historical credit risk values, historical market risk values, and historical disposal feasibility values ​​from the historical dynamic knowledge graphs; the historical credit risk features, historical market risk features, historical disposal feasibility features, historical credit risk values, historical market risk values, and historical disposal feasibility values ​​together constitute historical data.

[0020] S22. Construct a credit risk sub-model, a market risk sub-model, and a disposal feasibility sub-model; train the credit risk sub-model, the market risk sub-model, and the disposal feasibility sub-model using historical data; obtain the final credit risk sub-model, the final market risk sub-model, and the final disposal feasibility sub-model.

[0021] The above steps involve collecting previously generated dynamic knowledge graphs and extracting historical credit risk characteristics, historical market risk characteristics, and historical disposal feasibility characteristics. The corresponding historical credit risk values, market risk values, and disposal feasibility values ​​are also collected. These features and values ​​together constitute a complete historical dataset, providing a foundation for subsequent model training. Based on this historical data, credit risk sub-models, market risk sub-models, and disposal feasibility sub-models are constructed. Each sub-model performs specialized analysis and prediction for different aspects of non-performing asset management. By training these three sub-models using historical data, the system continuously optimizes model parameters, improving model accuracy and predictive ability. The system then obtains the final trained credit risk sub-model, final market risk sub-model, and final disposal feasibility sub-model. These models address the problem of static and rigid models, providing more accurate and effective decision support for non-performing asset management.

[0022] Preferably, step S22 includes the following steps:

[0023] S221. Construct a credit risk sub-model, a market risk sub-model, and a disposal feasibility sub-model; set the credit model parameters, market model parameters, and disposal model parameters for the credit risk sub-model, market risk sub-model, and disposal feasibility sub-model, respectively.

[0024] S222. Set the credit training error threshold and credit training error of the credit risk sub-model; train the credit risk sub-model using historical credit risk features and historical credit risk values; during the training process, combine optimization algorithms to find the credit model parameters of the credit risk sub-model to obtain the optimal credit model parameters; use the optimal credit model parameters as the credit model parameters of the credit risk sub-model to obtain the final credit risk sub-model.

[0025] Set the market training error threshold and market training error for the market risk sub-model; train the market risk sub-model using historical market risk characteristics and historical market risk values; during the training process, combine optimization algorithms to find the optimal market model parameters for the market risk sub-model; use the optimal market model parameters as the market model parameters for the market risk sub-model to obtain the final market risk sub-model.

[0026] Set the disposal training error threshold and disposal training error for the disposal feasibility sub-model; use historical disposal risk features and historical disposal risk values ​​to train the disposal feasibility sub-model; during the training process, combine optimization algorithms to find the disposal model parameters of the disposal feasibility sub-model to obtain the optimal disposal model parameters; use the optimal disposal model parameters as the disposal model parameters of the disposal feasibility sub-model to obtain the final disposal feasibility sub-model.

[0027] The above steps involve setting specific model parameters for each sub-model, which are key factors affecting the model's predictive performance. Training error thresholds and errors are set for each sub-model to monitor the training process and quality. Taking the credit risk sub-model as an example, the system uses historical credit risk features and corresponding historical credit risk values ​​for training. During training, the system continuously adjusts the credit model parameters using optimization algorithms to find the optimal parameters that minimize training error. Once the optimal credit model parameters are found, they are used as the final parameters for the credit risk sub-model, resulting in the final credit risk sub-model. The same process applies to the market risk sub-model and the disposal feasibility sub-model, resulting in optimized final market risk sub-model and final disposal feasibility sub-model. Each sub-model undergoes meticulous parameter adjustment and optimization, ensuring the accuracy and effectiveness of the model in predicting credit risk, market risk, and disposal feasibility, providing strong data support and decision-making for the management of non-performing assets.

[0028] Preferably, the steps in S222 for obtaining optimal credit model parameters by combining optimization algorithms to find the credit model parameters of the credit risk sub-model, obtaining optimal market model parameters by combining optimization algorithms to find the market model parameters of the market risk sub-model, and obtaining optimal disposal model parameters by combining optimization algorithms to find the disposal feasibility sub-model include the following steps:

[0029] S2221. Construct a credit particle population, a market particle population, and a disposal particle population; set the maximum number of credit optimization iterations, the maximum number of market optimization iterations, and the maximum number of disposal optimization iterations.

[0030] S2222: Set the initial positions of the credit particle population, market particle population, and disposal particle population according to the credit model parameters, market model parameters, and disposal model parameters to obtain the initial position set of the credit particle population, the initial position set of the market particle population, and the initial position set of the disposal particle population.

[0031] S2223. Define a credit fitness function based on the credit training error threshold and the credit training error; define a market fitness function based on the market training error threshold and the market training error; define a disposal fitness function based on the disposal training error threshold and the disposal training error.

[0032] S2224. Perform iterative operations on the initial position sets of the credit particle population, the initial position sets of the market particle population, and the initial position sets of the disposal particle population. Based on the credit fitness function, the market fitness function, and the disposal fitness function, update the positions of each particle in the initial position sets of the credit particle population, the initial position sets of the market particle population, and the initial position sets of the disposal particle population. In each iteration, obtain the best individual particle position and the global best particle position in the credit particle population, the market particle population, and the disposal particle population.

[0033] S2225, repeat S2224. When the maximum number of credit optimization iterations is reached, stop the iteration and obtain the optimal credit model parameters; when the maximum number of market optimization iterations is reached, stop the iteration and obtain the optimal credit model parameters; when the maximum number of disposal optimization iterations is reached, stop the iteration and obtain the optimal credit model parameters.

[0034] The above steps employ particle swarm optimization (PSO) to find the optimal model parameters for each sub-model. The system constructs corresponding particle populations for the credit risk sub-model, market risk sub-model, and disposal feasibility sub-model, setting a maximum number of optimization iterations for each particle population to control the iteration process. Based on the initial model parameters of each sub-model, the initial positions of the corresponding particle populations are set, forming initial position sets for the credit, market, and disposal particle populations. These initial position sets represent initial guesses of the model parameters. Credit fitness functions, market fitness functions, and disposal fitness functions are defined, based on their respective training error thresholds and training errors, to measure... The system evaluates the performance of model parameters on training data; it iterates through the initial position set; in each iteration, the system updates the position of each particle in the particle population according to the fitness function, and records the best individual particle position and the global best particle position in each iteration, which represent the currently found optimal model parameters; the above iterative process is repeated until the maximum number of optimization iterations is reached; the system obtains the optimal credit model parameters, optimal market model parameters, and optimal disposal model parameters; these optimal parameters are used to construct the final credit risk sub-model, market risk sub-model, and disposal feasibility sub-model, overcoming the overfitting problem of traditional static models in complex scenarios such as interest rate fluctuations and changes in the judicial environment.

[0035] Preferably, step S3 includes the following steps:

[0036] S31. By using the credit risk characteristics, market risk characteristics, and disposal feasibility characteristics respectively through the credit risk sub-model, market risk sub-model, and disposal feasibility sub-model, the credit risk value, market risk value, and disposal risk value are obtained.

[0037] S32. Construct an asset score function; calculate the asset score based on the credit risk value, market risk value, and disposal risk value.

[0038] Set a threshold for non-performing asset scores. If the asset score is less than or equal to the threshold, the asset is considered a non-performing asset; otherwise, it is considered a non-performing asset and is continuously monitored.

[0039] The above steps utilize previously trained credit risk, market risk, and disposal feasibility sub-models to evaluate the input credit risk, market risk, and disposal feasibility characteristics, respectively, yielding corresponding credit risk, market risk, and disposal risk values. These values ​​represent the quantitative assessment results of non-performing assets across different risk dimensions. An asset score function is constructed, comprehensively considering the credit risk, market risk, and disposal risk values ​​to calculate a comprehensive asset score, reflecting the overall risk level of the non-performing asset. A non-performing asset score threshold is set as the standard for determining whether an asset is non-performing. If the calculated asset score is less than or equal to this threshold, the system classifies the asset as non-performing and requires corresponding management measures. If the asset score is higher than the threshold, it is classified as non-performing, but the system continuously monitors the asset's status to ensure timely detection of any potential risk changes. In this way, the system achieves accurate identification and effective management of non-performing assets, automates their classification, reduces the risk of human error, improves the accuracy of non-performing asset identification, and shortens the decision-making cycle.

[0040] Preferably, step S4 includes the following steps:

[0041] S41. Set constraints and construct disposal plans based on the non-performing assets to obtain a set of disposal plans;

[0042] S42. Construct an objective function that combines returns, time efficiency, and risk controllability;

[0043] S43. By combining the optimization algorithm with the objective function, the set of disposal schemes is iteratively processed to obtain the optimal disposal scheme;

[0044] S44. Process non-performing assets through the optimal disposal plan and monitor it in real time. When the dynamic knowledge graph is updated, the system automatically adjusts the optimal disposal plan until the non-performing assets are disposed of.

[0045] The above steps involve setting specific constraints for non-performing assets and constructing a set of possible disposal solutions based on these constraints. An objective function is then built, combining three key factors: return, time efficiency, and risk controllability, aiming to find an optimal disposal solution that balances the interests of all parties. To select the optimal solution from the set, an optimization algorithm is used to iteratively optimize the solution set in conjunction with the objective function. Through continuous iteration, the system ultimately determines an optimal disposal solution that achieves the best balance in terms of return, time efficiency, and risk controllability. Non-performing assets are processed according to the optimal disposal solution, and the disposal process is monitored in real time. When the dynamic knowledge graph is updated, i.e., when the market environment or asset condition changes, the system automatically adjusts the optimal disposal solution to adapt to the new situation. This process continues until the non-performing assets are completely disposed of. In this way, the system ensures the efficiency, flexibility, and risk controllability of non-performing asset disposal.

[0046] Preferably, step S43 includes the following steps:

[0047] S431. Construct a chromosome population, where each chromosome in the chromosome population represents a disposal scheme from the disposal scheme set, and set a maximum number of optimization iterations;

[0048] S432. Perform iterative operations on the chromosome population; according to the objective function, perform selection, crossover, and mutation operations on the chromosomes in the chromosome population to obtain the chromosome population under operation;

[0049] S433. Repeat S432. When the maximum number of optimization iterations is reached, stop the iteration and obtain the optimal solution.

[0050] The above steps use a genetic algorithm to find the optimal disposal solution; construct a chromosome population, where each chromosome represents a specific disposal solution from the set of disposal solutions; set a maximum number of optimization iterations to control the algorithm's iteration process; perform iterative operations on the chromosome population; in each iteration, select, crossover, and mutate chromosomes in the chromosome population according to the objective function; these operations simulate the natural selection and gene recombination processes in biological evolution, aiming to generate new, potentially better disposal solutions; repeat the above iterative process until the set maximum number of optimization iterations is reached; select the chromosome with the best performance from the final chromosome population, which is the optimal disposal solution; utilizing the powerful search capability of the genetic algorithm, the optimal disposal solution is found from many possible disposal solutions, which performs best in terms of benefit, time efficiency, and risk controllability.

[0051] The non-performing asset disposal optimization system based on intelligent data analysis is used to implement the above-mentioned non-performing asset disposal optimization method based on intelligent data analysis. It includes a multi-source data integration and knowledge graph construction module, a multi-dimensional risk sub-model training module, an asset risk assessment and non-performing asset determination module, and a disposal strategy optimization and execution module.

[0052] The multi-source data integration and knowledge graph construction module is used to collect data from internal and external banks, clean heterogeneous data, and construct a dynamically updated knowledge graph. Through entity recognition technology, entity nodes are used to establish a cross-data source association network. At the same time, credit risk characteristics, market risk characteristics, and disposal feasibility characteristics are generated by combining structured data extraction and text analysis.

[0053] The multi-dimensional risk sub-model training module is used to construct credit risk sub-model, market risk sub-model, and disposal feasibility sub-model, and uses historical data combined with optimization algorithms for training; thus obtaining the final credit risk sub-model, final market risk sub-model, and final disposal feasibility sub-model.

[0054] The asset risk assessment and non-performing asset determination module is used to weight and fuse the risk values ​​output by each sub-model, and comprehensively calculate the asset score through the asset scoring function formula; the system automatically classifies the asset level according to the preset threshold, and triggers the non-performing asset marking when the score exceeds the risk threshold.

[0055] The disposal strategy optimization and execution module is used to find a balance point of recovery rate, cycle and risk among various disposal schemes using a multi-objective optimization mechanism; select the optimal disposal scheme that adapts to the real-time market status through a genetic algorithm, and use the optimal disposal scheme to dispose of non-performing assets, and monitor in real time. When the dynamic knowledge graph is updated, the system automatically adjusts the optimal disposal scheme until the non-performing assets are disposed of.

[0056] The storage medium for optimizing the disposal of non-performing assets based on intelligent data analysis stores a program that, when executed by a processor, is used to implement the aforementioned optimization method for the disposal of non-performing assets based on intelligent data analysis.

[0057] (III) Beneficial Effects

[0058] This invention breaks down the barriers between structured and unstructured data within and outside banks through dynamic knowledge graph construction technology, enabling automated mapping and updating of multi-dimensional entity relationships. Combined with data cleaning and feature extraction, it significantly improves data utilization efficiency, ensures that risk characteristics are updated during policy adjustments or market fluctuations, and solves the problems of data silos and delayed response in traditional systems.

[0059] This invention adapts to credit, market, and disposal scenarios by constructing credit risk sub-models, market risk sub-models, and disposal feasibility sub-models; it enhances the model's generalization ability and reduces prediction errors through a historical data-driven training mechanism; and it overcomes the overfitting problem of traditional static models in complex scenarios such as interest rate fluctuations and changes in the judicial environment by dynamically adjusting model parameters through a particle swarm optimization algorithm.

[0060] This invention achieves automated classification of non-performing assets by using a weighted scoring mechanism based on multi-dimensional risk values ​​and a preset threshold, thereby reducing the risk of human error, improving the accuracy of non-performing asset identification, and shortening the decision-making cycle.

[0061] This invention uses a genetic algorithm to optimize the recovery rate, cycle, and risk of disposal solutions, generating the optimal strategy that conforms to the real-time market conditions. The system directly connects to external platforms such as judicial auctions for automated execution and dynamically monitors the execution effect, triggering strategy re-optimization to form a closed loop of "assessment-disposal-feedback". This improves the asset recovery rate, shortens the disposal cycle, and reduces the risk.

[0062] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0063] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0064] Figure 1 This is a flowchart illustrating the optimized method for non-performing asset disposal based on intelligent data analysis according to the present invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, and not all embodiments. Based on the embodiments of the invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the invention.

[0066] In the description of this invention, it should be understood that the terms "opening", "upper", "lower", "top", "middle", "inner", etc., which indicate orientation or positional relationship, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the invention.

[0067] Example 1:

[0068] Please see Figure 1 This invention discloses an optimized method for the disposal of non-performing assets based on intelligent data analysis, comprising the following steps:

[0069] S1. Collect asset data and real-time dynamic data, and construct a dynamic knowledge graph; extract credit risk characteristics, market risk characteristics, and disposal feasibility characteristics from the dynamic real-time graph;

[0070] S1 includes the following steps:

[0071] S11. Collect internal bank data (such as user credit limits, repayment behavior data, and multiple borrowing records) and external data sources (such as the central bank's credit reporting interface, third-party credit scoring platforms, and public opinion monitoring systems) through legal and compliant channels to obtain asset data;

[0072] Obtain real-time dynamic data (such as real-time user transaction records, data from the Black Cat Complaint Platform, and announcements from the State Financial Regulatory Commission) through API interfaces;

[0073] S12. Clean the asset data and real-time dynamic data to obtain cleaned data; establish a dynamic knowledge graph using the cleaned data, which will automatically trigger knowledge graph reconstruction when a new policy is released or market fluctuations are detected; the dynamic knowledge graph includes entity recognition and relationship mapping; entity recognition includes nodes such as labeled users, collateral, and repayment behavior; the relationship mapping defines the associations between entities;

[0074] S13. Extract numerical features from the structured data in the dynamic knowledge graph;

[0075] The unstructured data in the dynamic knowledge graph is analyzed using NLP technology to generate text features;

[0076] Extract credit risk features, market risk features, and disposal feasibility features from numerical and textual features;

[0077] S2. Construct a credit risk sub-model, a market risk sub-model, and a disposal feasibility sub-model; train the credit risk sub-model, market risk sub-model, and disposal feasibility sub-model using historical data; obtain the final credit risk sub-model, the final market risk sub-model, and the final disposal feasibility sub-model.

[0078] S2 includes the following steps:

[0079] S21. Collect historical dynamic knowledge graphs, extract features from the historical dynamic knowledge graphs to obtain historical credit risk features, historical market risk features, and historical disposal feasibility features; collect historical credit risk values, historical market risk values, and historical disposal feasibility values ​​from the historical dynamic knowledge graphs; the historical credit risk features, historical market risk features, historical disposal feasibility features, historical credit risk values, historical market risk values, and historical disposal feasibility values ​​together constitute historical data.

[0080] S22. Construct a credit risk sub-model, a market risk sub-model, and a disposal feasibility sub-model; train the credit risk sub-model, the market risk sub-model, and the disposal feasibility sub-model using historical data; obtain the final credit risk sub-model, the final market risk sub-model, and the final disposal feasibility sub-model.

[0081] S22 includes the following steps:

[0082] S221. Construct a credit risk sub-model, a market risk sub-model, and a disposal feasibility sub-model; set the credit model parameter as c1, the market model parameter as c2, and the disposal model parameter as c3 for the credit risk sub-model, the market risk sub-model, and the disposal feasibility sub-model, respectively.

[0083] S222. Set the credit training error threshold of the credit risk sub-model to b1 and the credit training error to b2. Use historical credit risk features and historical credit risk values ​​to train the credit risk sub-model. During the training process, combine the optimization algorithm to find the credit model parameters of the credit risk sub-model to obtain the optimal credit model parameters. Use the optimal credit model parameters as the credit model parameters of the credit risk sub-model to obtain the final credit risk sub-model.

[0084] The market training error threshold of the market risk sub-model is set as d1, and the market training error is set as d2. The market risk sub-model is trained using historical market risk characteristics and historical market risk values. During the training process, the market model parameters of the market risk sub-model are found by combining optimization algorithms to obtain the optimal market model parameters. The optimal market model parameters are used as the learning rate of the market risk sub-model to obtain the final market risk sub-model.

[0085] Set the disposal training error threshold as e1 and the disposal training error as e2 for the disposal feasibility sub-model; use historical disposal risk features and historical disposal risk values ​​to train the disposal feasibility sub-model; during the training process, combine optimization algorithms to find the disposal model parameters of the disposal feasibility sub-model to obtain the optimal disposal model parameters; use the optimal disposal model parameters as the disposal model parameters of the disposal feasibility sub-model to obtain the final disposal feasibility sub-model.

[0086] The steps in S222 for obtaining optimal credit model parameters by combining optimization algorithms to find the credit model parameters of the credit risk sub-model, obtaining optimal market model parameters by combining optimization algorithms to find the market model parameters of the market risk sub-model, and obtaining optimal disposal model parameters by combining optimization algorithms to find the disposal feasibility sub-model include:

[0087] S2221. Construct a credit particle population, a market particle population, and a disposal particle population; set the maximum number of credit optimization iterations, the maximum number of market optimization iterations, and the maximum number of disposal optimization iterations.

[0088] S2222: Set the initial positions of the credit particle population, market particle population, and disposal particle population according to the credit model parameters, market model parameters, and disposal model parameters to obtain the initial position set of the credit particle population, the initial position set of the market particle population, and the initial position set of the disposal particle population.

[0089] S2223. Based on the credit training error threshold and the credit training error, define a credit fitness function; based on the market training error threshold and the market training error, define a market fitness function; based on the disposal training error threshold and the disposal training error, define a disposal fitness function; the formulas for the credit fitness function, the market fitness function, and the disposal fitness function are as follows.

[0090]

[0091] S2224. Perform iterative operations on the initial position sets of the credit particle population, the initial position sets of the market particle population, and the initial position sets of the disposal particle population. Based on the credit fitness function, the market fitness function, and the disposal fitness function, update the positions of each particle in the initial position sets of the credit particle population, the initial position sets of the market particle population, and the initial position sets of the disposal particle population. In each iteration, obtain the best individual particle position and the global best particle position in the credit particle population, the market particle population, and the disposal particle population.

[0092] S2225, repeat S2224. When the maximum number of credit optimization iterations is reached, stop the iteration and obtain the optimal credit model parameters; when the maximum number of market optimization iterations is reached, stop the iteration and obtain the optimal credit model parameters; when the maximum number of disposal optimization iterations is reached, stop the iteration and obtain the optimal credit model parameters.

[0093] S3. Input the credit risk characteristics, market risk characteristics, and disposal feasibility characteristics into the final credit risk sub-model, the final market risk sub-model, and the final disposal feasibility sub-model, respectively, to obtain the credit risk value, market risk value, and disposal risk value; input the credit risk value, market risk value, and disposal risk value into the asset scoring function to obtain the non-performing assets;

[0094] S3 includes the following steps:

[0095] S31. By using the credit risk characteristics, market risk characteristics, and disposal feasibility characteristics respectively through the credit risk sub-model, market risk sub-model, and disposal feasibility sub-model, the credit risk value, market risk value, and disposal risk value are obtained.

[0096] S32. Construct the asset score function; calculate the asset score based on the credit risk value, market risk value, and disposal risk value; the asset score function is as follows.

[0097] G=β1·j1+β2·j2+β3·j3;

[0098] Where G represents the asset fractional function, β1, β2 and β3 represent the weighting coefficients of the set credit risk, market risk and disposal risk values ​​respectively, and j1, j2 and j3 represent the credit risk value, market risk value and disposal risk value respectively;

[0099] Set a threshold for non-performing asset scores. If an asset score is less than or equal to the threshold, the asset is considered a non-performing asset; otherwise, it is considered a non-performing asset and is continuously monitored.

[0100] S4. Based on the non-performing assets, construct a set of disposal schemes and an objective function, and combine them with an optimization algorithm to obtain the optimal processing scheme; use the optimal processing scheme to process the non-performing assets.

[0101] S4 includes the following steps:

[0102] S41. Set constraints and construct a disposal plan based on the non-performing assets, resulting in a disposal plan h = {h1, h2, ..., h...} i ,...,h m}, where h i This indicates the i-th disposal plan that is concentrated on the disposal of non-performing assets, and m represents the total number of disposal plans in the disposal plan concentration; if the disposal plan is based on commercial real estate as collateral, it will be given priority to package and issue ABS (predicted recovery rate 92%, cycle 120 days);

[0103] S42. Construct an objective function that combines returns, time efficiency, and risk controllability; the formula for the objective function is as follows.

[0104]

[0105] Where F(x) represents the objective function, x represents the disposal plan in the disposal plan set, α1, α2 and α3 represent the weights of recovery rate, disposal period and risk respectively, satisfying α1+α2+α3=1, R represents the theoretical maximum recovery rate (usually 100%, or adjusted according to asset type), T represents the maximum allowed disposal period (such as the disposal time limit required by regulations), Z represents the upper limit of risk coefficient, and r, t and z represent the expected recovery rate, disposal period and risk coefficient respectively.

[0106] S43. By combining the optimization algorithm with the objective function, the set of disposal schemes is iteratively processed to obtain the optimal disposal scheme;

[0107] S43 includes the following steps:

[0108] S431. Construct a chromosome population, then the chromosome population is represented as n = {n1, n2, ..., n}. i ,...,n m Each chromosome in the chromosome population represents a disposal scheme from the disposal scheme set, and a maximum number of optimization iterations is set;

[0109] S432. Perform iterative operations on the chromosome population; according to the objective function, perform selection, crossover, and mutation operations on the chromosomes in the chromosome population to obtain the chromosome population under operation;

[0110] S433. Repeat S432. When the maximum number of optimization iterations is reached, stop the iteration and obtain the optimal solution.

[0111] S44. Process non-performing assets through the optimal disposal plan and monitor it in real time. When the dynamic knowledge graph is updated, the system automatically adjusts the optimal disposal plan until the non-performing assets are disposed of.

[0112] Taking a certain internet consumer credit asset pool as an example:

[0113] The system identified a certain internet consumer credit asset pool (scale of 80 million yuan, average user credit score of 680) due to tightening industry regulatory policies (such as the reduction of the upper limit of loan interest rates). After the market risk sub-model, the market risk score increased from the original 62 points to 89 points. Further, the asset score function was used to calculate and determine that the asset pool was a non-performing asset.

[0114] The strategy recommendations generated after combining the objective function with the optimization algorithm are as follows:

[0115] Within 3 months, the asset pool will be packaged and issued as ABS (senior priority with an expected recovery rate of 91% and a cycle of 60 days) + the remaining debt will be transferred to a licensed financial institution (expected recovery rate of 87%).

[0116] Ten days after implementation, the dynamic knowledge graph detected a weekly increase of 4% in the delinquency rate of similar assets. The system automatically triggered a strategy adjustment, changing to "batch transfer to the financial technology debt trading platform at an 8% discount". The final recovery rate was 85%, avoiding a deterioration in liquidity.

[0117] Example 2:

[0118] The non-performing asset disposal optimization system based on intelligent data analysis is used to implement the above-mentioned non-performing asset disposal optimization method based on intelligent data analysis. It includes a multi-source data integration and knowledge graph construction module, a multi-dimensional risk sub-model training module, an asset risk assessment and non-performing asset determination module, and a disposal strategy optimization and execution module.

[0119] The multi-source data integration and knowledge graph construction module is used to collect data from internal and external banks, clean heterogeneous data, and construct a dynamically updated knowledge graph. Through entity recognition technology, entity nodes are used to establish a cross-data source association network. At the same time, credit risk characteristics, market risk characteristics, and disposal feasibility characteristics are generated by combining structured data extraction and text analysis.

[0120] The multi-dimensional risk sub-model training module is used to construct credit risk sub-model, market risk sub-model, and disposal feasibility sub-model, and uses historical data combined with optimization algorithms for training; thus obtaining the final credit risk sub-model, final market risk sub-model, and final disposal feasibility sub-model.

[0121] The asset risk assessment and non-performing asset determination module is used to weight and fuse the risk values ​​output by each sub-model, and comprehensively calculate the asset score through the asset scoring function formula; the system automatically classifies the asset level according to the preset threshold, and triggers the non-performing asset marking when the score exceeds the risk threshold.

[0122] The disposal strategy optimization and execution module is used to find a balance point of recovery rate, cycle and risk among various disposal schemes using a multi-objective optimization mechanism; select the optimal disposal scheme that adapts to the real-time market status through a genetic algorithm, and use the optimal disposal scheme to dispose of non-performing assets, and monitor in real time. When the dynamic knowledge graph is updated, the system automatically adjusts the optimal disposal scheme until the non-performing assets are disposed of.

[0123] Example 3:

[0124] The storage medium for optimizing the disposal of non-performing assets based on intelligent data analysis stores a program that, when executed by a processor, is used to implement the aforementioned optimization method for the disposal of non-performing assets based on intelligent data analysis.

[0125] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0126] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for non-performing asset disposal optimization based on intelligent data analysis, characterized in that, Includes the following steps: S1. Collect asset data and real-time dynamic data, and construct a dynamic knowledge graph; Credit risk characteristics, market risk characteristics, and feasibility of resolution characteristics are extracted from the dynamic real-time graph. S1 includes the following steps: S11. Collect internal bank data and external data sources through legal and compliant means to obtain asset data; collect real-time dynamic data; S12. Clean the asset data and real-time dynamic data to obtain cleaned data; establish a dynamic knowledge graph based on the cleaned data, and trigger knowledge graph reconstruction when a new policy is released or market fluctuations are detected; the dynamic knowledge graph includes entity recognition and relationship mapping between entities; S13. Extract numerical features from the structured data in the dynamic knowledge graph; parse the unstructured data in the dynamic knowledge graph to generate text features; extract credit risk features, market risk features, and disposal feasibility features from the numerical features and text features; S2. Construct a credit risk sub-model, a market risk sub-model, and a disposal feasibility sub-model; use historical data to train the credit risk sub-model, the market risk sub-model, and the disposal feasibility sub-model to obtain the final credit risk sub-model, the final market risk sub-model, and the final disposal feasibility sub-model. S3. Input the credit risk characteristics, market risk characteristics, and disposal feasibility characteristics into the final credit risk sub-model, the final market risk sub-model, and the final disposal feasibility sub-model, respectively, to obtain the credit risk value, market risk value, and disposal risk value; input the credit risk value, market risk value, and disposal risk value into the asset scoring function to determine whether the asset is a non-performing asset; S4. For non-performing assets, construct a set of disposal schemes and an objective function, and combine them with an optimization algorithm to obtain the optimal processing scheme; use the optimal processing scheme to process the non-performing assets.

2. The method for optimizing the disposal of non-performing assets based on intelligent data analysis according to claim 1, characterized in that, S2 includes the following steps: S21. Collect historical dynamic knowledge graphs, extract features from the historical dynamic knowledge graphs to obtain historical credit risk features, historical market risk features, and historical disposal feasibility features; collect historical credit risk values, historical market risk values, and historical disposal feasibility values ​​from the historical dynamic knowledge graphs; the historical credit risk features, historical market risk features, historical disposal feasibility features, historical credit risk values, historical market risk values, and historical disposal feasibility values ​​together constitute historical data. S22. Construct a credit risk sub-model, a market risk sub-model, and a disposal feasibility sub-model; train the credit risk sub-model, the market risk sub-model, and the disposal feasibility sub-model using historical data; obtain the final credit risk sub-model, the final market risk sub-model, and the final disposal feasibility sub-model. 3.The smart data analysis based non-performing asset disposal optimization method according to claim 2, characterized in that, S22 includes the following steps: S221. Construct a credit risk sub-model, a market risk sub-model, and a disposal feasibility sub-model; set the credit model parameters, market model parameters, and disposal model parameters for the credit risk sub-model, market risk sub-model, and disposal feasibility sub-model, respectively. S222. Set the credit training error threshold and credit training error of the credit risk sub-model; train the credit risk sub-model using historical credit risk features and historical credit risk values; during the training process, combine optimization algorithms to find the credit model parameters of the credit risk sub-model to obtain the optimal credit model parameters; use the optimal credit model parameters as the credit model parameters of the credit risk sub-model to obtain the final credit risk sub-model. Set the market training error threshold and market training error for the market risk sub-model; train the market risk sub-model using historical market risk characteristics and historical market risk values; during the training process, combine optimization algorithms to find the optimal market model parameters for the market risk sub-model; use the optimal market model parameters as the market model parameters for the market risk sub-model to obtain the final market risk sub-model. Set the disposal training error threshold and disposal training error for the disposal feasibility sub-model; use historical disposal risk features and historical disposal risk values ​​to train the disposal feasibility sub-model; during the training process, combine optimization algorithms to find the disposal model parameters of the disposal feasibility sub-model to obtain the optimal disposal model parameters; use the optimal disposal model parameters as the disposal model parameters of the disposal feasibility sub-model to obtain the final disposal feasibility sub-model.

4. The method for optimizing the disposal of non-performing assets based on intelligent data analysis according to claim 3, characterized in that, The steps in S222 for obtaining optimal credit model parameters by combining optimization algorithms to find the credit model parameters of the credit risk sub-model, obtaining optimal market model parameters by combining optimization algorithms to find the market model parameters of the market risk sub-model, and obtaining optimal disposal model parameters by combining optimization algorithms to find the disposal feasibility sub-model include: S2221. Construct a credit particle population, a market particle population, and a disposal particle population; set the maximum number of credit optimization iterations, the maximum number of market optimization iterations, and the maximum number of disposal optimization iterations. S2222. Set the initial positions of the credit particle population, market particle population, and disposal particle population according to the credit model parameters, market model parameters, and disposal model parameters to obtain the initial position set of the credit particle population, the initial position set of the market particle population, and the initial position set of the disposal particle population. S2223. Define a credit fitness function based on the credit training error threshold and the credit training error; define a market fitness function based on the market training error threshold and the market training error; define a disposal fitness function based on the disposal training error threshold and the disposal training error. S2224. Perform iterative operations on the initial position sets of the credit particle population, the initial position sets of the market particle population, and the initial position sets of the disposal particle population. Based on the credit fitness function, the market fitness function, and the disposal fitness function, update the positions of each particle in the initial position sets of the credit particle population, the initial position sets of the market particle population, and the initial position sets of the disposal particle population. In each iteration, obtain the best individual particle position and the global best particle position in the credit particle population, the market particle population, and the disposal particle population. S2225. Repeat S2224. When the maximum number of credit optimization iterations is reached, stop the iteration and obtain the optimal credit model parameters; when the maximum number of market optimization iterations is reached, stop the iteration and obtain the optimal credit model parameters; when the maximum number of disposal optimization iterations is reached, stop the iteration and obtain the optimal credit model parameters.

5. The method for optimizing the disposal of non-performing assets based on intelligent data analysis according to claim 1, characterized in that, S3 includes the following steps: S31. By using the credit risk characteristics, market risk characteristics, and disposal feasibility characteristics respectively through the credit risk sub-model, market risk sub-model, and disposal feasibility sub-model, the credit risk value, market risk value, and disposal risk value are obtained. S32. Construct an asset score function; calculate the asset score based on the credit risk value, market risk value, and disposal risk value. Set a threshold for non-performing asset scores. If the asset score is less than or equal to the threshold, the asset is considered a non-performing asset; otherwise, it is considered a non-performing asset and is continuously monitored.

6. The smart data analysis based non-performing asset disposal optimization method according to claim 1, characterized in that, S4 includes the following steps: S41. Set constraints and construct disposal plans based on the non-performing assets to obtain a set of disposal plans; S42. Construct an objective function that combines returns, time efficiency, and risk controllability; S43. By combining the optimization algorithm with the objective function, the set of disposal schemes is iteratively processed to obtain the optimal disposal scheme; S44. Process non-performing assets through the optimal disposal plan and monitor it in real time. When the dynamic knowledge graph is updated, the system automatically adjusts the optimal disposal plan until the non-performing assets are disposed of.

7. The smart data analysis based non-performing asset disposal optimization method according to claim 6, characterized in that, S43 includes the following steps: S431. Construct a chromosome population, where each chromosome in the chromosome population represents a disposal scheme from the disposal scheme set, and set a maximum number of optimization iterations; S432. Perform iterative operations on the chromosome population; according to the objective function, perform selection, crossover, and mutation operations on the chromosomes in the chromosome population to obtain the chromosome population under operation; S433. Repeat S432. When the maximum number of optimization iterations is reached, stop the iteration and obtain the optimal solution.

8. A non-performing asset disposal optimization system based on intelligent data analysis, used to implement the non-performing asset disposal optimization method based on intelligent data analysis as described in any one of claims 1-7; the system includes a multi-source data integration and knowledge graph construction module, a multi-dimensional risk sub-model training module, an asset risk assessment and non-performing asset determination module, and a disposal strategy optimization and execution module.

9. A storage medium for optimization of disposal of non-performing assets based on intelligent data analysis, characterized in that, It stores a program that, when executed by a processor, implements the non-performing asset disposal optimization method based on intelligent data analysis as described in any one of claims 1-7.

Images