Method and device for processing bank customer credit risk based on federated learning

Abstract

The application discloses a kind of based on the processing method and device of bank customer credit risk of federal learning.There is related to distributed computing technical field, the method includes: in local security environment, by each branch node, the multi-source heterogeneous business data of target enterprise stored locally is acquired, and the feature vector set of target enterprise is obtained according to multi-source heterogeneous business data;The feature vector set of target enterprise is input into local credit risk prediction model, and the credit risk enterprise information is output by local credit risk prediction model, and the credit risk vector of the credit risk enterprise corresponding to credit risk enterprise information of credit risk enterprise information;According to the risk level of credit risk enterprise, the credit risk vector corresponding to credit risk enterprise is determined to handle the suggestion of credit risk enterprise, and according to the risk disposal of credit risk enterprise by each branch node according to processing suggestion.It is solved that the credit risk enterprise cannot be accurately identified in the related art, and the problem that the security of bank credit assets is poor.

Description

Technical Field

[0001] This application relates to the field of distributed computing technology, and more specifically, to a method and apparatus for processing bank customer credit risk based on federated learning. Background Technology

[0002] Currently, banks primarily rely on traditional credit scoring models and centralized machine learning methods in credit risk management. Traditional credit scoring models, using linear weighting, struggle to capture complex, non-linear risk signals. Centralized machine learning methods typically require branches to centrally upload corporate credit data to the head office data center, where predictive models are built on a unified platform. This centralized data model has two main drawbacks. First, it's difficult to meet regulatory compliance requirements in the financial industry; sensitive data from different branches may not be fully shared due to privacy concerns, leading to severe data silos. Second, existing solutions often train local models independently based on historical data from individual branches, lacking cross-institutional data fusion mechanisms. This hinders the utilization of the complementarity of heterogeneous data to improve risk identification accuracy, resulting in an inability to accurately identify high-risk enterprises and implement effective control measures in the early stages of risk, ultimately leading to poor overall bank security.

[0003] There is currently no effective solution to the problem that the relevant technologies cannot accurately identify enterprises with credit risks, resulting in poor security of bank credit assets. Summary of the Invention

[0004] The main purpose of this application is to provide a method and apparatus for handling bank customer credit risk based on federated learning, so as to solve the problem that the inability to accurately identify enterprises with credit risk in related technologies leads to poor security of bank credit assets.

[0005] To achieve the above objectives, according to one aspect of this application, a method for handling bank customer credit risk based on federated learning is provided. The method includes: in a local secure environment, acquiring multi-source heterogeneous business data of a target enterprise stored locally through each branch node, and obtaining a set of feature vectors for the target enterprise based on the multi-source heterogeneous business data; inputting the set of feature vectors of the target enterprise into a local credit risk prediction model, and outputting credit risk enterprise information and the corresponding credit risk vector of the credit risk enterprise through the local credit risk prediction model, wherein the local credit risk prediction model is a model obtained by aggregating encrypted gradient parameters obtained by each branch node through federated learning; inputting the credit risk vector of the credit risk enterprise into a rule matching engine, determining the risk level of the credit risk enterprise according to a preset rule set in the rule matching engine, determining a handling suggestion for the credit risk enterprise based on the risk level and the corresponding credit risk vector, and handling the risk of the credit risk enterprise through each branch node according to the handling suggestion.

[0006] Optionally, in a local security environment, acquiring multi-source heterogeneous business data of the target enterprise stored locally through each branch node, and obtaining the feature vector set of the target enterprise based on the multi-source heterogeneous business data includes: in a local security environment, acquiring multi-source heterogeneous business data of the target enterprise stored locally through each branch node; dividing the multi-source heterogeneous business data into data within multiple time intervals according to a preset sliding time window; extracting features from the data within each time interval to obtain the feature vector corresponding to each time interval; and obtaining the feature vector set of the target enterprise based on the feature vector corresponding to each time interval.

[0007] Optionally, the feature vector set of the target enterprise is input into the local credit risk prediction model, and the local credit risk prediction model outputs credit risk enterprise information and the credit risk vector of the corresponding credit risk enterprise. This includes: inputting the feature vector set of the target enterprise into the local credit risk prediction model; the local credit risk prediction model determines the credit risk vector of each enterprise based on the feature vector corresponding to each enterprise; determining the enterprise risk value of each enterprise based on the credit risk vector of each enterprise; determining enterprises whose enterprise risk value is greater than a preset threshold as credit risk enterprises; and outputting credit risk enterprise information and the credit risk vector of the corresponding credit risk enterprise.

[0008] Optionally, determining the enterprise risk value for each enterprise based on its credit risk vector includes: determining the risk value corresponding to each credit risk vector in the enterprise's credit risk vector; and determining the enterprise risk value for each enterprise based on the risk value corresponding to each credit risk vector and the weight corresponding to each credit risk vector.

[0009] Optionally, the steps for generating the local credit risk prediction model include: receiving the basic model configuration file issued by the head office through each branch node, obtaining the initial model parameters in the basic model configuration file for training local training data, obtaining the gradient parameters of the trained model, encrypting the gradient parameters of the trained model to obtain encrypted gradient parameters; uploading the encrypted gradient parameters to the coordinator node through each branch node, so that the coordinator node can aggregate the encrypted gradient parameters uploaded by all branch nodes to obtain global gradient parameters; sending the global gradient parameters to each branch node, and obtaining the local credit risk prediction model through each branch node based on the global gradient parameters.

[0010] Optionally, each branch node uploads encrypted gradient parameters to the coordinator node, and the coordinator node aggregates the encrypted gradient parameters uploaded by all branch nodes to obtain global gradient parameters. This process includes: each branch node uploads encrypted gradient parameters to the coordinator node via a secure channel, and the coordinator node aggregates the encrypted gradient parameters uploaded by all branch nodes to obtain aggregated gradient parameters; differential privacy noise is injected into the aggregated gradient parameters to obtain global gradient parameters.

[0011] Optionally, after each branch node performs risk management on credit risk enterprises according to the processing suggestions, the method further includes: each branch node optimizes the corresponding local credit risk prediction model based on the risk management results of the credit risk enterprises to obtain optimized gradient parameters; each branch node uploads the optimized gradient parameters to the coordinator node, so that the coordinator node can aggregate the optimized gradient parameters to obtain optimized global gradient parameters; the optimized global gradient parameters are sent to each branch node, and each branch node obtains the optimized local credit risk prediction model based on the optimized global gradient parameters.

[0012] To achieve the above objectives, according to another aspect of this application, a bank customer credit risk processing apparatus based on federated learning is provided. The apparatus includes: an acquisition unit, configured to acquire locally stored multi-source heterogeneous business data of a target enterprise through each branch node in a local secure environment, and obtain a set of feature vectors of the target enterprise based on the multi-source heterogeneous business data; an input unit, configured to input the set of feature vectors of the target enterprise into a local credit risk prediction model, and output credit risk enterprise information and the corresponding credit risk vector of the credit risk enterprise through the local credit risk prediction model, wherein the local credit risk prediction model is a model obtained by aggregating encrypted gradient parameters obtained by each branch node through local training using federated learning; and a determination unit, configured to input the credit risk vector of the credit risk enterprise into a rule matching engine, determine the risk level of the credit risk enterprise according to a preset rule set in the rule matching engine, determine a processing suggestion for the credit risk enterprise based on the risk level and the corresponding credit risk vector of the credit risk enterprise, and perform risk disposal on the credit risk enterprise through each branch node according to the processing suggestion.

[0013] According to another aspect of this application, a computer-readable storage medium is provided, which includes a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform any federated learning-based bank customer credit risk processing method.

[0014] According to another aspect of this application, an electronic device is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include a processing method for performing any federated learning-based bank customer credit risk.

[0015] According to another aspect of this application, a computer program product is provided, including computer instructions that, when executed by a processor, implement the steps of the federated learning-based bank customer credit risk processing method described above.

[0016] In this embodiment, by acquiring multi-source heterogeneous business data of the target enterprise stored locally through each branch node in a local secure environment, and obtaining a set of feature vectors of the target enterprise based on the multi-source heterogeneous business data, the feature vector set of the target enterprise is input into a local credit risk prediction model. The local credit risk prediction model outputs credit risk enterprise information and the corresponding credit risk vector of the credit risk enterprise. The local credit risk prediction model is obtained by aggregating the encrypted gradient parameters obtained by each branch node through federated learning. The credit risk vector of the credit risk enterprise is input into a rule matching engine. The risk level of the credit risk enterprise is determined according to the preset rule set in the rule matching engine. Based on the risk level and the corresponding credit risk vector of the credit risk enterprise, a handling suggestion for the credit risk enterprise is determined. Each branch node handles the risk of the credit risk enterprise according to the handling suggestion, thus solving the technical problem of poor security of bank credit assets due to the inability to accurately identify credit risk enterprises.

[0017] In this application, each branch node acquires multi-source heterogeneous business data and generates feature vectors in a local secure environment, satisfying data privacy compliance requirements. Utilizing a federated learning mechanism, the encrypted gradient parameters trained locally by each branch are securely aggregated at the coordinator to obtain a local credit risk prediction model. This model outputs information on credit-risk enterprises and their corresponding credit risk vectors, improving the accuracy of identifying such enterprises. The credit risk vectors are then input into a rule matching engine, which, combined with a preset rule set, determines the risk level of each enterprise and provides processing recommendations. Each branch node then promptly addresses the risks based on these recommendations, thereby enhancing the security of bank credit assets. Attached Figure Description

[0018] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0019] Figure 1 A hardware block diagram of a computer terminal for implementing a federated learning-based method for handling bank customer credit risk is shown.

[0020] Figure 2 This is a flowchart of a method for handling bank customer credit risk based on federated learning, according to an embodiment of this application.

[0021] Figure 3 This is a schematic diagram of the federated training of the local credit risk prediction model provided in the embodiments of this application;

[0022] Figure 4 This is a schematic diagram of a method for handling bank customer credit risk based on federated learning, provided in an embodiment of this application.

[0023] Figure 5 This is a schematic diagram illustrating the generation of early warnings for enterprises with credit risk, provided in the embodiments of this application.

[0024] Figure 6 This is a schematic diagram of a bank customer credit risk processing device based on federated learning, provided according to an embodiment of this application.

[0025] Figure 7 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0027] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] It should be noted that all relevant information (including at least one of the following: customer information, bank information, etc.) and data (including at least one of the following: data used for display, data used for analysis, etc.) involved in this application are information and data authorized by the user or fully authorized by all parties. For example, this system has an interface with the relevant user or institution. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or institution through the interface, and obtain the relevant information after receiving consent from the aforementioned user or institution.

[0029] Example 1

[0030] According to an embodiment of this application, a method embodiment for processing bank customer credit risk based on federated learning is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a federated learning-based method for processing bank customer credit risk is shown. Figure 1As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include at least one of the following: a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0032] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0033] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the federated learning-based bank customer credit risk processing method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned federated learning-based bank customer credit risk processing method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include at least one of the following: the Internet, corporate intranet, local area network, mobile communication network, and combinations thereof.

[0034] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0035] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0036] Under the aforementioned operating environment, this application provides the following: Figure 2 The example shown is a method for handling bank customer credit risk based on federated learning. Figure 2 This is a flowchart of a method for handling bank customer credit risk based on federated learning, according to Embodiment 1 of this application.

[0037] Step S201: In the local security environment, obtain the multi-source heterogeneous business data of the target enterprise stored locally through each branch node, and obtain the feature vector set of the target enterprise based on the multi-source heterogeneous business data.

[0038] Optionally, the target enterprise's multi-source heterogeneous business data can be obtained through each branch node and stored locally. The multi-source heterogeneous business data may include the target enterprise's structured business feature data, semi-structured time-series data, and unstructured text data, which can ensure that the data in each branch node is always in the local storage environment, meeting the requirements of data security and privacy protection.

[0039] Step S202: Input the feature vector set of the target enterprise into the local credit risk prediction model, and output the credit risk enterprise information and the credit risk vector of the corresponding credit risk enterprise through the local credit risk prediction model. The local credit risk prediction model is a model obtained by aggregating the encrypted gradient parameters obtained by each branch node in local training through federated learning.

[0040] Optionally, each branch node can independently perform forward and backward propagation calculations using local training data to generate local gradient parameters. Then, a homomorphic encryption algorithm can be used to encrypt the local gradient parameters. After receiving the encrypted gradient parameters uploaded by each branch, the coordinator node directly performs gradient aggregation operations in the ciphertext domain using the additive homomorphic property of homomorphic encryption. The aggregated encrypted gradient is decrypted and the global model parameters are updated. The global model parameters are then distributed back to each branch node to obtain the local credit prediction model, which can accurately identify and output credit risk enterprise information and the credit risk vector of the corresponding credit risk enterprise.

[0041] Step S203: Input the credit risk vector of the credit risk enterprise into the rule matching engine, determine the risk level of the credit risk enterprise according to the preset rule set in the rule matching engine, determine the processing suggestion of the credit risk enterprise according to the risk level and the corresponding credit risk vector of the credit risk enterprise, and carry out risk disposal of the credit risk enterprise through each branch node according to the processing suggestion.

[0042] Optionally, after receiving the credit risk vector, the rule matching engine can determine the risk level of a credit risk enterprise through dynamic threshold judgment and multi-rule combination logic. It can generate different handling suggestions for credit risk enterprises with different risk levels. For example, for high-risk credit risk enterprises, it can trigger an immediate freeze on credit limits or stop credit granting; for medium-risk credit risk enterprises, it can output a risk vector summary and generate a manual due diligence work order; and for low-risk credit risk enterprises, it can only record logs without triggering warnings. This enables refined control of risk management and improves the security of bank credit assets.

[0043] The method for handling bank customer credit risk based on federated learning provided in this application embodiment obtains multi-source heterogeneous business data and generates feature vectors in a local secure environment by each branch node, thus meeting data privacy compliance requirements. Utilizing a federated learning mechanism, the encrypted gradient parameters trained locally by each branch are securely aggregated at a coordinator to obtain a local credit risk prediction model. This model outputs information on credit-risk enterprises and their corresponding credit risk vectors, improving the accuracy of identifying such enterprises. The credit risk vectors are input into a rule matching engine, and the risk level of each enterprise is determined by combining it with a preset rule set. This determines the appropriate handling recommendations, and each branch node promptly addresses these recommendations to mitigate the risk of credit-risk enterprises. This solves the technical problem of inaccurate identification of credit-risk enterprises, leading to poor security of bank credit assets, and ultimately improves the security of bank credit assets.

[0044] To accurately identify enterprises with credit risk, optionally, in the bank customer credit risk processing method based on federated learning provided in this application embodiment, in a local secure environment, multi-source heterogeneous business data of the target enterprise is obtained through each branch node, and the feature vector set of the target enterprise is obtained based on the multi-source heterogeneous business data, including:

[0045] The first step is to obtain the multi-source heterogeneous business data of the target enterprise from the local storage through each branch node in the local security environment, and divide the multi-source heterogeneous business data into data within multiple time intervals according to the preset sliding time window.

[0046] Optionally, branch nodes can divide continuous time series data into multiple time intervals based on preset business logic or fixed durations (such as T+1 day, T+7 day or T+30 day), which helps to analyze the dynamic evolution trend of the target enterprise's risk status.

[0047] The second step is to extract features from the data within each time interval to obtain the feature vector corresponding to each time interval.

[0048] Optionally, feature extraction can be performed on the data within each time interval. For numerical data, statistical features such as mean, variance, extreme values, and rate of change can be calculated. For categorical data, one-hot encoding can be performed. For text-based unstructured data, natural language processing techniques can be used to extract key risk entities. Finally, features from all dimensions are concatenated or fused to obtain the feature vector corresponding to each time interval.

[0049] The third step is to obtain the set of feature vectors for the target company based on the feature vectors corresponding to each time interval.

[0050] Optionally, after obtaining the feature vector corresponding to each time interval, the feature vector corresponding to each time interval can be standardized, and an attention mechanism can be used to weight and filter the feature vector. After obtaining the processed feature vector corresponding to each time interval, the processed feature vector corresponding to each time interval of the target company can be constructed into a time series matrix in chronological order to form the feature vector set of the target company.

[0051] In summary, through the above steps, the multi-source heterogeneous business data is divided into multiple time intervals using a sliding window mechanism, preserving time-dependent characteristics. Feature extraction is performed on the data within each time interval to obtain the feature vector corresponding to each time interval, thereby obtaining the feature vector set of the target enterprise. This provides a data foundation for subsequent identification of enterprises with credit risks and helps improve the accuracy and efficiency of credit risk identification.

[0052] To improve the security of bank credit assets, optionally, in the method for processing bank customer credit risk based on federated learning provided in this application embodiment, the feature vector set of the target enterprise is input into the local credit risk prediction model, and the local credit risk prediction model outputs credit risk enterprise information and the credit risk vector of the corresponding credit risk enterprise, including:

[0053] The first step is to input the set of feature vectors of the target companies into the local credit risk prediction model. The local credit risk prediction model determines the credit risk vector of each company based on the feature vectors corresponding to each company.

[0054] Optionally, the local credit risk prediction model can adopt an algorithm architecture based on deep neural networks or gradient boosting decision trees. The input layer receives the feature vector set of the target enterprise, and after nonlinear transformation and feature interaction through multiple hidden layers, the output layer generates a credit risk vector.

[0055] The second step is to determine the enterprise risk value for each enterprise based on its credit risk vector.

[0056] Optionally, the credit risk vector of each enterprise can be weighted by summing, pooling the maximum value, or using an attention mechanism to obtain an enterprise risk value that can comprehensively reflect the overall credit status of the enterprise. The closer the enterprise risk value is to the preset maximum value (e.g., 1.0 or 100%), the higher the probability of the enterprise defaulting.

[0057] The third step is to identify enterprises whose risk values ​​exceed a preset threshold as credit risk enterprises, and output credit risk enterprise information and the credit risk vector of the corresponding credit risk enterprise.

[0058] Optionally, the preset threshold can be dynamically adjusted according to the bank's risk appetite. After determining the credit risk enterprises based on the preset threshold, the output credit enterprise risk information and credit risk vector can provide the bank with a basis for subsequent risk management.

[0059] In summary, by taking the steps described above and determining the enterprise risk value for each company, those companies with risk values ​​exceeding a preset threshold are identified as credit risk enterprises. This improves the flexibility of credit risk identification, and the output credit enterprise risk information and credit risk vector can provide banks with a basis for subsequent management and control, thereby enhancing the security of credit assets.

[0060] To accurately identify enterprises with credit risk, optionally, in the federated learning-based bank customer credit risk processing method provided in this application embodiment, determining the enterprise risk value of each enterprise based on its credit risk vector includes:

[0061] The first step is to determine the risk value corresponding to each credit risk vector in the credit risk vector of each enterprise.

[0062] Optionally, each credit risk vector can correspond to a risk dimension, such as liquidity risk, operational stability risk, industry cyclical risk, etc. The risk value corresponding to each credit risk vector represents the probability of credit default or the degree of potential economic loss for the enterprise under the corresponding risk dimension.

[0063] The second step is to determine the enterprise risk value for each enterprise based on the risk value corresponding to each credit risk vector and the weight corresponding to each credit risk vector.

[0064] Optionally, the weight of each credit risk vector can be dynamically calculated using an attention mechanism to assess which risk-dominant mode the enterprise is currently in. Then, based on the weight of each credit risk vector, the risk values ​​corresponding to each credit risk vector are weighted and summed to obtain the enterprise risk value for each enterprise.

[0065] In summary, by taking the steps described above, determining the risk value corresponding to each credit risk vector, and then weighting and summing the risk values ​​corresponding to each credit risk vector according to their respective weights to obtain the enterprise risk value, the risk situation of each enterprise is accurately quantified, which helps to accurately identify enterprises with credit risks.

[0066] To improve the prediction accuracy of the local credit risk prediction model, optionally, in the federated learning-based bank customer credit risk processing method provided in this application embodiment, the step of generating the local credit risk prediction model includes:

[0067] The first step is to receive the basic model configuration file issued by the head office through each branch node, obtain the initial model parameters in the basic model configuration file for local training data, obtain the gradient parameters of the trained model, and encrypt the gradient parameters of the trained model to obtain encrypted gradient parameters.

[0068] Optionally, the basic model configuration file issued by the head office may include initial model parameters, initial weight initialization strategy, global hyperparameters for federated learning, etc. After receiving the basic model configuration file, each branch node loads the initial model in its local secure environment, trains the initial model parameters using local training data, and obtains the gradient parameters of the trained model. In order to protect data privacy, the branch node can use a homomorphic encryption algorithm to encrypt the gradient parameters and generate encrypted gradient parameters.

[0069] The second step involves uploading the encrypted gradient parameters from each branch node to the coordinator node. The coordinator node then aggregates the encrypted gradient parameters uploaded by all branch nodes to obtain the global gradient parameters.

[0070] Optionally, the coordinator node can adopt a ciphertext field aggregation strategy and take advantage of the additive homomorphic property of homomorphic encryption to directly perform addition operations on multiple encrypted gradient parameters without decrypting the encrypted gradient parameters uploaded by the branch nodes, so as to obtain encrypted global gradient parameters.

[0071] The third step is to send the global gradient parameters to each branch node, and each branch node obtains a local credit risk prediction model based on the global gradient parameters.

[0072] Optionally, the coordinator node distributes the global gradient parameters to each branch node through a secure channel. After receiving the parameters, each branch node can use its local optimizer to update its local model parameters under the guidance of the global gradient parameters, thereby obtaining its local credit risk prediction model.

[0073] In summary, through the above steps, each branch node trains locally based on the initial model parameters and obtains encrypted gradient parameters after encryption processing. These encrypted gradient parameters are then uploaded to the coordinator node, which aggregates all encrypted gradient parameters to obtain global gradient parameters, which are then distributed to each branch node. Each branch node then obtains its local credit risk prediction model based on the global gradient parameters, thereby improving the prediction accuracy of the local credit risk prediction model.

[0074] To protect enterprise privacy data, optionally, in the method for processing bank customer credit risk based on federated learning provided in this application embodiment, each branch node uploads encrypted gradient parameters to the coordinator node, so that the coordinator node aggregates the encrypted gradient parameters uploaded by all branch nodes to obtain global gradient parameters, including:

[0075] The first step is to upload the encrypted gradient parameters to the coordinator node through a secure channel by each branch node. The coordinator node then aggregates the encrypted gradient parameters uploaded by all branch nodes to obtain the aggregated gradient parameters.

[0076] Optionally, the gradient parameters of the trained model can be encrypted. Homomorphic encryption algorithms can be used to encrypt the gradient parameters element by element or block by block, so that the coordinator node can directly perform aggregation operations on the encrypted gradient parameters from different branch nodes in the ciphertext state to obtain the aggregated gradient parameters.

[0077] The second step is to inject differential privacy noise into the aggregated gradient parameters to obtain global gradient parameters.

[0078] Optionally, random noise conforming to a Laplace or Gaussian distribution can be added to the aggregated gradient parameters. The scale of the noise can be dynamically calculated based on the privacy budget or the sensitivity of the gradient parameters. The sensitivity of the gradient parameters can be determined by limiting the maximum norm of the gradient parameters, which can not only protect data privacy but also preserve the validity of the gradient parameters.

[0079] In summary, by injecting differential privacy noise into the aggregated gradient parameters through the above steps, the risk of leaking corporate privacy data through shared gradient parameters can be avoided, thus meeting the data privacy protection requirements in the financial industry.

[0080] Optionally, such as Figure 3 The diagram illustrates the federated training of a local credit risk prediction model. First, the head office distributes the basic model configuration file and assigns an encryption key. Each branch node uses local training data to train the basic model parameters in the configuration file, obtaining the gradient parameters of the trained model. This meets the requirement that data remains within its domain. The gradient parameters are then encrypted using the encryption key, resulting in encrypted gradient parameters. Each branch node uploads these encrypted gradient parameters to the coordinator node. The coordinator node aggregates the encrypted gradient parameters uploaded by all branch nodes to obtain global gradient parameters. These global gradient parameters are then distributed to each branch node, supporting version control of the model on each branch node and enabling rollback in case of anomalies.

[0081] To improve the prediction accuracy of the local credit risk prediction model, optionally, in the federated learning-based bank customer credit risk processing method provided in this application embodiment, after each branch node performs risk disposal on the credit risk enterprise according to the processing suggestions, the method further includes:

[0082] The first step is for each branch node to optimize its local credit risk prediction model based on the risk disposal results of the credit risk enterprises, thereby obtaining the optimized gradient parameters.

[0083] Optionally, each branch node can use the actual results of risk disposal of credit risk enterprises (i.e., risk disposal results) as new label data, such as whether the credit risk enterprise ultimately defaults, the number of overdue days, and the disposal recovery rate. The label data corresponding to the risk disposal results is combined with the credit risk vector of the credit risk enterprise to construct a validation set or test set. The loss value of the current local credit risk prediction model on the risk disposal results is calculated according to the loss function. The gradient of the current local credit risk prediction model parameters relative to the error of the risk disposal results is calculated through the backpropagation algorithm, thereby obtaining the optimized gradient parameters.

[0084] The second step involves uploading the optimized gradient parameters to the coordinator node through each branch node, so that the coordinator node can aggregate the optimized gradient parameters to obtain the optimized global gradient parameters.

[0085] Optionally, each branch node first encrypts the optimized gradient parameters and then uploads them to the coordinator node through a secure channel. After receiving the optimized encrypted gradient parameters uploaded by each branch node, the coordinator node can aggregate them using a federated averaging algorithm or its weighted variant to obtain the optimized global gradient parameters.

[0086] The third step is to send the optimized global gradient parameters to each branch node, and each branch node obtains the optimized local credit risk prediction model based on the optimized global gradient parameters.

[0087] Optionally, the coordinator node distributes the optimized global gradient parameters to all branch nodes through a secure channel. After receiving the parameters, each branch node can use its local optimizer to update its local model parameters, thereby achieving iterative optimization of the model and obtaining the optimized local credit risk prediction model.

[0088] In summary, through the above steps and an iterative optimization mechanism, each branch node optimizes its local credit risk prediction model based on the risk management results, obtaining optimized gradient parameters. The coordinator node then obtains optimized global gradient parameters based on the optimized encrypted gradient parameters of each branch node and sends them to each branch node. Each branch node then receives its optimized local credit risk prediction model, enabling the local credit risk prediction model to continuously iterate and adaptively adjust based on the actual risk management results, thereby continuously improving the prediction accuracy of the local credit risk prediction model in each branch.

[0089] Optionally, such as Figure 4The diagram illustrates a method for handling bank customer credit risk based on federated learning. This system is applied to credit risk management, which includes a data acquisition layer, a federated learning layer, a model service layer, and an application layer. The data acquisition layer comprises multiple branch nodes. Each branch node retrieves multi-source heterogeneous business data of the target enterprise from its local database, extracts features from this data to obtain a set of feature vectors for the target enterprise, and trains encrypted gradient parameters using local training data. These encrypted gradient parameters are then sent to the federated computing layer. The coordinator node in the federated computing layer receives and aggregates the encrypted gradient parameters from each node to obtain global gradient parameters, which are then sent to each branch node. The dynamic weight model in the model service layer calculates and assigns different aggregation weights in real time based on the performance of each branch node during the federated learning process. The model service layer also manages model versions for each branch node. Each branch node adjusts its model based on the global gradient parameters to obtain a local credit risk prediction model. Each local credit risk prediction model can identify credit risk enterprises, and each branch node performs risk management for these enterprises. The application layer visualizes information about credit risk enterprises and provides early warnings, thereby improving the security of bank credit assets.

[0090] Optionally, such as Figure 5 The diagram illustrates the generation of early warnings for enterprises with credit risk. First, multi-source heterogeneous business data is collected. Then, this data is processed, dividing it into multiple time intervals according to a preset sliding time window. Feature extraction is performed on the data within each time interval to obtain a feature vector, resulting in a set of feature vectors for the target enterprise. Next, a local credit risk prediction model outputs information about the enterprise with credit risk and its corresponding credit risk vector. The risk rule base in the rule matching engine determines the enterprise's risk value. Based on the risk level of the enterprise, corresponding handling suggestions are determined. This process helps to promptly identify enterprises with credit risk and to address credit risks in a timely manner based on the suggestions, thereby improving the security of bank credit assets.

[0091] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0092] Example 2

[0093] This application also provides a device for processing bank customer credit risk based on federated learning. It should be noted that this device can be used to execute the method for processing bank customer credit risk based on federated learning provided in this application. The following describes the device for processing bank customer credit risk based on federated learning provided in this application.

[0094] According to an embodiment of this application, an apparatus for implementing the above-described method for processing bank customer credit risk based on federated learning is also provided, such as... Figure 6 As shown, the device includes: an acquisition unit 601, an input unit 602, and a determination unit 603.

[0095] The acquisition unit 601 is used to acquire multi-source heterogeneous business data of the target enterprise stored locally through each branch node in a local security environment, and obtain the feature vector set of the target enterprise based on the multi-source heterogeneous business data.

[0096] Input unit 602 is used to input the feature vector set of the target enterprise into the local credit risk prediction model, and output credit risk enterprise information and credit risk vector of the credit risk enterprise corresponding to the credit risk enterprise information through the local credit risk prediction model. The local credit risk prediction model is a model obtained by aggregating the encrypted gradient parameters obtained by each branch node in local training through federated learning.

[0097] The determination unit 603 is used to input the credit risk vector of the credit risk enterprise into the rule matching engine, determine the risk level of the credit risk enterprise according to the preset rule set in the rule matching engine, determine the processing suggestion of the credit risk enterprise according to the risk level and the corresponding credit risk vector of the credit risk enterprise, and carry out risk disposal of the credit risk enterprise through each branch node according to the processing suggestion.

[0098] The bank customer credit risk processing device based on federated learning provided in this application embodiment solves the technical problem of poor security of bank credit assets due to the inability to accurately identify credit risk enterprises, which is achieved by the acquisition unit 601 acquiring multi-source heterogeneous business data of target enterprises stored locally through each branch node in a local secure environment, and obtaining a set of feature vectors of target enterprises based on the multi-source heterogeneous business data; the input unit 602 inputs the set of feature vectors of target enterprises into a local credit risk prediction model, and outputs credit risk enterprise information and the credit risk vector of the credit risk enterprise corresponding to the credit risk enterprise information through the local credit risk prediction model, wherein the local credit risk prediction model is obtained by aggregating the encrypted gradient parameters obtained by each branch node in local training through federated learning; the determination unit 603 inputs the credit risk vector of the credit risk enterprise into a rule matching engine, determines the risk level of the credit risk enterprise according to the preset rule set in the rule matching engine, and determines the processing suggestions for the credit risk enterprise based on the risk level and the corresponding credit risk vector of the credit risk enterprise. Each branch node performs risk disposal on the credit risk enterprise according to the processing suggestions, thereby solving the technical problem of poor security of bank credit assets due to the inability to accurately identify credit risk enterprises and achieving the technical effect of improving the security of bank credit assets.

[0099] Optionally, in the bank customer credit risk processing device based on federated learning provided in this application embodiment, the acquisition unit 601 includes: an acquisition module, used to acquire multi-source heterogeneous business data of the target enterprise stored locally through each branch node in a local secure environment, and divide the multi-source heterogeneous business data into data in multiple time intervals according to a preset sliding time window; an extraction module, used to extract features from the data in each time interval to obtain the feature vector corresponding to each time interval; and a first processing module, used to obtain the feature vector set of the target enterprise according to the feature vector corresponding to each time interval.

[0100] Optionally, in the bank customer credit risk processing device based on federated learning provided in this application embodiment, the input unit 603 includes: a first determining module, used to input the feature vector set of the target enterprise into the local credit risk prediction model, wherein the local credit risk prediction model determines the credit risk vector of each enterprise based on the feature vector corresponding to each enterprise; a second determining module, used to determine the enterprise risk value of each enterprise based on the credit risk vector of each enterprise; and a third determining module, used to determine that enterprises with enterprise risk values ​​greater than a preset threshold are credit risk enterprises, and output credit risk enterprise information and the credit risk vector of the credit risk enterprise corresponding to the credit risk enterprise information.

[0101] Optionally, in the bank customer credit risk processing device based on federated learning provided in the embodiments of this application, the second determining module includes: a first determining submodule, used to determine the risk value corresponding to each credit risk vector in the credit risk vector of each enterprise; and a second determining submodule, used to determine the enterprise risk value of each enterprise based on the risk value corresponding to each credit risk vector and the weight corresponding to each credit risk vector.

[0102] Optionally, in the bank customer credit risk processing apparatus based on federated learning provided in this application embodiment, the apparatus further includes: a training unit, configured to receive a basic model configuration file issued by the head office through each branch node, obtain local training data to train the initial model parameters in the basic model configuration file, obtain the gradient parameters of the trained model, and encrypt the gradient parameters of the trained model to obtain encrypted gradient parameters; a first aggregation unit, configured to upload the encrypted gradient parameters to the coordinator node through each branch node, so that the coordinator node can aggregate the encrypted gradient parameters uploaded by all branch nodes to obtain global gradient parameters; and a first sending unit, configured to send the global gradient parameters to each branch node, so that each branch node can obtain a local credit risk prediction model based on the global gradient parameters.

[0103] Optionally, in the bank customer credit risk processing device based on federated learning provided in the embodiments of this application, the first aggregation unit includes: an aggregation module, used to upload encrypted gradient parameters to the coordinator node through a secure channel through each branch node, so that the coordinator node can aggregate the encrypted gradient parameters uploaded by all branch nodes to obtain aggregated gradient parameters; and a second processing module, used to inject differential privacy noise into the aggregated gradient parameters to obtain global gradient parameters.

[0104] Optionally, in the bank customer credit risk processing apparatus based on federated learning provided in this application embodiment, the apparatus further includes: an optimization unit, used to optimize the corresponding local credit risk prediction model of each branch node based on the risk disposal result of the credit risk enterprise after the branch node performs risk disposal according to the processing suggestions, and obtain optimized gradient parameters; a second aggregation unit, used to upload the optimized gradient parameters to the coordinator node through each branch node, so that the coordinator node can aggregate the optimized gradient parameters to obtain optimized global gradient parameters; and a second sending unit, used to send the optimized global gradient parameters to each branch node, so that each branch node can obtain the optimized local credit risk prediction model based on the optimized global gradient parameters.

[0105] It should be noted that the acquisition unit 601, input unit 602, and determination unit 603 mentioned above correspond to steps S201 to S203 in Embodiment 1. The three units and their corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules can also be part of a device and run in the computer terminal 10 provided in Embodiment 1.

[0106] Example 3

[0107] Embodiments of this application may provide a computer terminal, which may be any computer terminal device in a group of computer terminals. Optionally, in this embodiment, the aforementioned computer terminal may also be replaced with a mobile terminal or an electronic device, etc.

[0108] Optionally, in this embodiment, the computer terminal may be located in at least one of a plurality of network devices in a computer network.

[0109] In this embodiment, the aforementioned computer terminal can execute the program code for the following steps in the method for handling bank customer credit risk based on federated learning: In a local secure environment, the multi-source heterogeneous business data of the target enterprise stored locally is obtained through each branch node, and a set of feature vectors of the target enterprise is obtained based on the multi-source heterogeneous business data; the set of feature vectors of the target enterprise is input into a local credit risk prediction model, and the local credit risk prediction model outputs credit risk enterprise information and the credit risk vector of the credit risk enterprise corresponding to the credit risk enterprise information, wherein the local credit risk prediction model is a model obtained by aggregating the encrypted gradient parameters obtained by each branch node in local training through federated learning; the credit risk vector of the credit risk enterprise is input into a rule matching engine, the risk level of the credit risk enterprise is determined according to the preset rule set in the rule matching engine, and the handling suggestions for the credit risk enterprise are determined according to the risk level of the credit risk enterprise and the corresponding credit risk vector of the credit risk enterprise, and the credit risk enterprise is handled by each branch node according to the handling suggestions.

[0110] Optionally, the aforementioned computer terminal can execute program code for the following steps in the method for processing bank customer credit risk based on federated learning: In a local secure environment, acquiring multi-source heterogeneous business data of the target enterprise stored locally through each branch node, and obtaining the feature vector set of the target enterprise based on the multi-source heterogeneous business data includes: In a local secure environment, acquiring multi-source heterogeneous business data of the target enterprise stored locally through each branch node, dividing the multi-source heterogeneous business data into data within multiple time intervals according to a preset sliding time window; extracting features from the data within each time interval to obtain the feature vector corresponding to each time interval; and obtaining the feature vector set of the target enterprise based on the feature vector corresponding to each time interval.

[0111] Optionally, the aforementioned computer terminal can execute the program code for the following steps in the federated learning-based bank customer credit risk processing method: inputting the feature vector set of the target enterprise into the local credit risk prediction model, and outputting credit risk enterprise information and the credit risk vector of the corresponding credit risk enterprise through the local credit risk prediction model, including: inputting the feature vector set of the target enterprise into the local credit risk prediction model; the local credit risk prediction model determining the credit risk vector of each enterprise based on the feature vector corresponding to each enterprise; determining the enterprise risk value of each enterprise based on the credit risk vector of each enterprise; determining enterprises with enterprise risk values ​​greater than a preset threshold as credit risk enterprises, and outputting credit risk enterprise information and the credit risk vector of the corresponding credit risk enterprise.

[0112] Optionally, the aforementioned computer terminal can execute program code for the following steps in the federated learning-based bank customer credit risk processing method: determining the enterprise risk value of each enterprise based on each enterprise's credit risk vector, including: determining the risk value corresponding to each credit risk vector in each enterprise's credit risk vector; and determining the enterprise risk value of each enterprise based on the risk value corresponding to each credit risk vector and the weight corresponding to each credit risk vector.

[0113] Optionally, the aforementioned computer terminal can execute the program code for the following steps in the method for processing bank customer credit risk based on federated learning: The steps for generating the local credit risk prediction model include: receiving the basic model configuration file issued by the head office through each branch node, obtaining the initial model parameters in the basic model configuration file for training local training data, obtaining the gradient parameters of the trained model, encrypting the gradient parameters of the trained model to obtain encrypted gradient parameters; uploading the encrypted gradient parameters to the coordinator node through each branch node, so that the coordinator node can aggregate the encrypted gradient parameters uploaded by all branch nodes to obtain global gradient parameters; sending the global gradient parameters to each branch node, and obtaining the local credit risk prediction model through each branch node based on the global gradient parameters.

[0114] Optionally, the aforementioned computer terminal can execute the following steps in the method for handling bank customer credit risk based on federated learning: Uploading encrypted gradient parameters to the coordinator node through each branch node, and then aggregating the encrypted gradient parameters uploaded by all branch nodes through the coordinator node to obtain global gradient parameters, including: uploading encrypted gradient parameters to the coordinator node through a secure channel through each branch node, and then aggregating the encrypted gradient parameters uploaded by all branch nodes through the coordinator node to obtain aggregated gradient parameters; injecting differential privacy noise into the aggregated gradient parameters to obtain global gradient parameters.

[0115] Optionally, the aforementioned computer terminal can execute the program code for the following steps in the federated learning-based bank customer credit risk processing method: After each branch node performs risk disposal on the credit risk enterprise according to the processing suggestions, the method further includes: each branch node optimizes the corresponding local credit risk prediction model based on the risk disposal result of the credit risk enterprise to obtain optimized gradient parameters; each branch node uploads the optimized gradient parameters to the coordinator node so that the coordinator node can aggregate the optimized gradient parameters to obtain optimized global gradient parameters; the optimized global gradient parameters are sent to each branch node so that each branch node obtains the optimized local credit risk prediction model based on the optimized global gradient parameters.

[0116] Optionally, Figure 7 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 7 As shown, the electronic device may include: one or more ( Figure 7 (Only one is shown) Processor 702, memory 704, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0117] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the federated learning-based bank customer credit risk processing method and apparatus in this application embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned federated learning-based bank customer credit risk processing method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include at least one of the following: the Internet, corporate intranet, local area network, mobile communication network, and combinations thereof.

[0118] The processor can access the information and application programs stored in the memory via the transmission device to execute the steps described above in the federated learning-based bank customer credit risk processing method.

[0119] This application provides a solution for handling bank customer credit risk based on federated learning. In a secure local environment, each branch node acquires locally stored multi-source heterogeneous business data of the target enterprise and obtains a set of feature vectors for the target enterprise based on this data. This set of feature vectors is then input into a local credit risk prediction model, which outputs information about the enterprise at risk and its corresponding credit risk vector. The local credit risk prediction model is obtained by aggregating encrypted gradient parameters trained locally by each branch node using federated learning. The credit risk vector of the enterprise at risk is then input into a rule matching engine. The engine determines the risk level of the enterprise based on a preset rule set and provides processing recommendations based on the risk level and the corresponding credit risk vector. Each branch node then handles the risk according to these recommendations. This solution addresses the technical problem of inaccurate identification of enterprises at risk, leading to poor security of bank credit assets, and improves the security of bank credit assets.

[0120] Those skilled in the art will understand that Figure 7 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 7 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 7The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 7 The different configurations shown.

[0121] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0122] Example 4

[0123] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the federated learning-based bank customer credit risk processing method provided in Embodiment 1.

[0124] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0125] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing steps of a method for processing bank customer credit risk based on federated learning.

[0126] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0127] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0128] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between units or modules may be electrical or other forms.

[0129] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0130] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0131] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0132] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for handling bank customer credit risk based on federated learning, characterized in that, include: In a local security environment, multi-source heterogeneous business data of the target enterprise is obtained through each branch node, and the feature vector set of the target enterprise is obtained based on the multi-source heterogeneous business data. The feature vector set of the target enterprise is input into the local credit risk prediction model, and the local credit risk prediction model outputs the credit risk enterprise information and the credit risk vector of the credit risk enterprise corresponding to the credit risk enterprise information. The local credit risk prediction model is a model obtained by aggregating the encrypted gradient parameters obtained by each branch node in local training through federated learning. The credit risk vector of the credit risk enterprise is input into the rule matching engine. The risk level of the credit risk enterprise is determined according to the preset rule set in the rule matching engine. The processing suggestion for the credit risk enterprise is determined according to the risk level and the corresponding credit risk vector of the credit risk enterprise. The risk is handled by each branch node according to the processing suggestion.

2. The method according to claim 1, characterized in that, In a local security environment, multi-source heterogeneous business data of the target enterprise is obtained through each branch node, and the feature vector set of the target enterprise is obtained based on the multi-source heterogeneous business data, including: In a local security environment, multi-source heterogeneous business data of the target enterprise is obtained through each branch node and the multi-source heterogeneous business data is divided into data within multiple time intervals according to a preset sliding time window. Feature extraction is performed on the data within each time interval to obtain the feature vector corresponding to each time interval; Based on the feature vectors corresponding to each time interval, the feature vector set of the target enterprise is obtained.

3. The method according to claim 1, characterized in that, The feature vector set of the target enterprise is input into the local credit risk prediction model, and the local credit risk prediction model outputs credit risk enterprise information and the credit risk vector of the credit risk enterprise corresponding to the credit risk enterprise information, including: The feature vector set of the target enterprise is input into the local credit risk prediction model, and the local credit risk prediction model determines the credit risk vector of each enterprise based on the feature vector corresponding to each enterprise. Determine the enterprise risk value for each enterprise based on its credit risk vector. Enterprises with risk values ​​greater than a preset threshold are identified as credit risk enterprises, and the credit risk enterprise information and the credit risk vector of the corresponding credit risk enterprise are output.

4. The method according to claim 3, characterized in that, The enterprise risk value for each enterprise is determined based on its credit risk vector, including: Determine the risk value corresponding to each credit risk vector in the credit risk vector of each enterprise; The enterprise risk value for each enterprise is determined based on the risk value corresponding to each credit risk vector and the weight corresponding to each credit risk vector.

5. The method according to claim 1, characterized in that, The steps for generating the local credit risk prediction model include: Each branch node receives the basic model configuration file issued by the head office, obtains local training data to train the initial model parameters in the basic model configuration file, obtains the gradient parameters of the trained model, and encrypts the gradient parameters of the trained model to obtain encrypted gradient parameters. Each branch node uploads the encrypted gradient parameters to the coordinator node, and the coordinator node aggregates the encrypted gradient parameters uploaded by all branch nodes to obtain the global gradient parameters. The global gradient parameters are sent to each branch node, and each branch node obtains the local credit risk prediction model based on the global gradient parameters.

6. The method according to claim 5, characterized in that, Each branch node uploads the encrypted gradient parameters to the coordinator node, which then aggregates the uploaded parameters to obtain global gradient parameters, including: Each branch node uploads the encrypted gradient parameters to the coordinator node via a secure channel, and the coordinator node aggregates the encrypted gradient parameters uploaded by all branch nodes to obtain the aggregated gradient parameters. Differential privacy noise is injected into the aggregated gradient parameters to obtain the global gradient parameters.

7. The method according to claim 5, characterized in that, After handling the risk of the credit-risk enterprise through each branch node according to the processing recommendations, the method further includes: Each branch node optimizes its local credit risk prediction model based on the risk disposal results of the credit risk enterprise, and obtains the optimized gradient parameters. The optimized gradient parameters are uploaded to the coordinator node through each branch node, so that the coordinator node can aggregate the optimized gradient parameters to obtain the optimized global gradient parameters. The optimized global gradient parameters are sent to each branch node, and each branch node obtains the optimized local credit risk prediction model based on the optimized global gradient parameters.

8. A device for processing bank customer credit risk based on federated learning, characterized in that, include: The acquisition unit is used to acquire multi-source heterogeneous business data of the target enterprise stored locally through each branch node in a local security environment, and to obtain the feature vector set of the target enterprise based on the multi-source heterogeneous business data. The input unit is used to input the feature vector set of the target enterprise into the local credit risk prediction model, and output the credit risk enterprise information and the credit risk vector of the credit risk enterprise corresponding to the credit risk enterprise information through the local credit risk prediction model. The local credit risk prediction model is a model obtained by aggregating the encrypted gradient parameters obtained by each branch node in local training through federated learning. The determining unit is used to input the credit risk vector of the credit risk enterprise into the rule matching engine, determine the risk level of the credit risk enterprise according to the preset rule set in the rule matching engine, determine the processing suggestion of the credit risk enterprise according to the risk level of the credit risk enterprise and the corresponding credit risk vector, and carry out risk disposal of the credit risk enterprise through each branch node according to the processing suggestion.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the computer-readable storage medium is located to perform the bank customer credit risk processing method based on any one of claims 1 to 7.

10. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, executes the method for processing bank customer credit risk based on any one of claims 1 to 7.

11. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method for processing bank customer credit risk based on any one of claims 1 to 7.

Images