Data generation method and system for financial risk detection model

Abstract

The invention discloses a data generation method and system for a financial risk detection model. Multidimensional financial time series data is subjected to standardization processing, and then a forward diffusion process is executed based on a continuous time stochastic differential equation framework. In the process, dynamic drifting and noise intensity changes in different market stages, fluctuation levels and risk states are adaptively learned in a data driving mode by training a score function, then constraint generation is carried out through a controllable back diffusion process fused with financial condition variables, and finally enhanced financial time series data are output. The method can effectively maintain the fluctuation structure of the asset income sequence, the tail risk characteristics and the correlation between cross-factors, thereby providing more real and reliable training data for a subsequent financial risk detection model.

Description

Technical Field

[0001] This application belongs to the field of information technology, and in particular relates to a data generation method and system for financial risk detection models. Background Technology

[0002] Existing diffusion models, generative models, and time series synthesis techniques have achieved some success in certain data augmentation scenarios. For example, the general diffusion generation framework proposed in patent US20230109379A1 is mainly aimed at images, tables, and general structured data. Its modeling method relies on directly adding noise forward and removing noise backward in the original data space. Although this method has universality, it is not specifically designed for the high correlation, strong non-stationarity, and factor structure dominance of financial time series. In addition, this technology lacks mechanisms for simulating extreme events, tail distribution calibration, and enhancing risk-sensitive features. The generated data may exhibit statistical bias or structural deficiencies in key risk dimensions. For example, the statistical model and parametric generation method used in patent US20200012902A1 rely on fixed distribution assumptions or shallow model structures, making it difficult to capture complex phenomena common in financial data such as nonlinear jumps, volatility clusters, and leverage effects. The data generated by this type of method is structurally smooth and lacks the extreme volatility and peak-and-fat-tail characteristics of real markets, which may prevent it from effectively covering risk scenarios in risk modeling or stress testing.

[0003] In summary, existing financial risk early warning and assessment methods have the following main shortcomings: 1) They lack a dedicated modeling mechanism for the factor structure of financial time series, resulting in the generated samples failing to fully reflect the intrinsic correlations and common factor-driven structures among assets; 2) They lack the ability to enhance performance against extreme risk events, making it difficult to generate rare samples for default prediction, black swan simulation, or stress testing; 3) They lack an adaptive correction mechanism for the generated distribution, causing the generated data to deviate from the actual market distribution in terms of statistical characteristics; 4) They do not fully consider the dynamic and multi-scale characteristics of financial data, which is not conducive to accurately reproducing the evolution of market risks. Summary of the Invention

[0004] The purpose of this application is to provide a data generation method and system for financial risk detection models, which can solve at least one of the problems existing in the prior art.

[0005] According to a first aspect of this application, this application provides a data generation method for a financial risk detection model, comprising the following steps:

[0006] S1: Obtain multidimensional financial time series data and perform standardization processing to obtain standardized financial time series data;

[0007] S2: Based on the continuous-time stochastic differential equation framework, a forward diffusion process is performed on standardized financial time series data to gradually evolve it into a noise distribution. At the same time, the scoring function is trained to learn the distribution gradient information of standardized financial time series data on the diffusion path.

[0008] S3: Construct a conditional scoring function based on preset financial condition variables and scoring functions, and use the conditional scoring function to perform a controllable back diffusion process on the noise distribution to generate synthetic financial time series data that conforms to financial condition variables;

[0009] S4: Validate the synthetic financial time series data and correct it based on the validation results to obtain enhanced financial time series data;

[0010] S5: The enhanced financial time series data is fused with the standardized financial time series data to obtain a dataset for training the financial risk detection model.

[0011] In some embodiments, in step S3, the controllable reverse diffusion process is achieved through a controllable reverse nonlinear diffusion mechanism;

[0012] Among them, the controllable reverse nonlinear diffusion mechanism is achieved by explicitly embedding financial condition variables into the drift and fluctuation terms of the stochastic differential equation corresponding to the controllable reverse diffusion process.

[0013] In some embodiments, the controllable reverse nonlinear diffusion mechanism includes a conditional drift function and a conditional wave function;

[0014] Among them, the conditional drift function and the conditional fluctuation function both take the potential state, time and financial condition variables in the controllable back diffusion process as inputs, and their parameters are learned in a data-driven manner.

[0015] In some embodiments, the value of the conditional drift function is determined by subtracting the product of the square of a diffusion intensity function and the conditional score function from the value of the basic drift term defined in the forward diffusion process.

[0016] In some embodiments, the conditional score function is configured to learn the log probability density gradient of the underlying state in a controlled backdiffusion process of standardized financial time series data, given financial condition variables.

[0017] In some embodiments, financial condition variables include at least one variable that represents market conditions, volatility levels, macroeconomic indicators, or risk level labels.

[0018] In some embodiments, in step S2, the forward diffusion process is constructed based on a continuous-time stochastic differential equation framework, and the noise distribution is controlled by a time-dependent drift term and a noise intensity scheduling function.

[0019] In some embodiments, the scoring function trained in step S2 is configured to take the time of the forward diffusion process as input and can adaptively capture the dynamic changes in the distribution of standardized financial time series data, thereby implicitly modeling the time-dependent dynamics of the forward diffusion process.

[0020] In some embodiments, the verification of the synthetic financial time series data in step S4 includes:

[0021] To verify the consistency of statistical distribution characteristics and / or time series dynamic characteristics of synthetic financial time series data.

[0022] In some embodiments, in step S4, correcting the synthetic financial time series data based on the verification result includes:

[0023] The synthetic financial time series data was reconstructed based on the verification results;

[0024] The verification results are then fed back to the forward diffusion process or the controlled reverse diffusion process to adjust the subsequent data generation process.

[0025] In some embodiments, after step S1 and before step S2, the method further includes:

[0026] Standardized financial time series data are subjected to latent factor decomposition, which transforms them into low-dimensional latent factor time series data expressed by a few common factors, while preserving the covariance structure of multidimensional financial time series data.

[0027] Step S2 performs a forward diffusion process based on latent factor time series data.

[0028] In some embodiments, multidimensional financial time series data include asset return series, trading volume, volatility indicators, interest rate factors, and / or information spreads.

[0029] A second aspect of this application provides a data generation system for a financial risk detection model, the data generation system comprising:

[0030] The data preprocessing module is used to acquire multidimensional financial time series data and perform standardization processing to obtain standardized financial time series data.

[0031] The SDE modeling module is used to provide a framework for continuous-time stochastic differential equations;

[0032] The diffusion process construction module is used to perform a forward diffusion process on standardized financial time series data based on the continuous-time stochastic differential equation framework, so that it gradually evolves into a noise distribution, while training the scoring function to learn the distribution gradient information of standardized financial time series data on the diffusion path.

[0033] The conditional sampling and generation module is used to construct a conditional scoring function based on preset financial conditional variables and scoring functions, and to use the conditional scoring function to perform a controllable back diffusion process on the noise distribution to generate synthetic financial time series data that conforms to the financial conditional variables.

[0034] The data verification and reconstruction module is used to verify the synthetic financial time series data and correct it based on the verification results to obtain enhanced financial time series data.

[0035] The fusion module is used to merge enhanced financial time series data with standardized financial time series data to obtain a dataset for training financial risk detection models.

[0036] A third aspect of this application provides an electronic device including a memory and a processor coupled to each other, the processor being configured to execute program instructions stored in the memory to implement the data generation method for a financial risk detection model described above.

[0037] A fourth aspect of this application provides a computer-readable storage medium having program instructions stored thereon, which, when executed by a processor, implement the data generation method for a financial risk detection model described above.

[0038] The data generation method for financial risk detection models provided in this application first standardizes multidimensional financial time series data, and then performs a forward diffusion process based on a continuous-time stochastic differential equation framework. During this process, a scoring function is trained to adaptively learn dynamic drift and noise intensity changes under different financial market stages, volatility levels, and risk states in a data-driven manner. Then, a controlled back diffusion process incorporating financial condition variables is used for constrained generation, ultimately outputting enhanced financial time series data. This method effectively preserves the volatility structure, tail risk characteristics, and cross-factor correlations of asset return series, thereby providing more realistic and reliable training data for subsequent financial risk detection models.

[0039] This method endows multidimensional financial time series data with differentiability, continuity, and interpretability along the diffusion time axis, better reflecting the dynamic behavior of price fluctuations, risk transmission, and extreme events in real financial markets. The enhanced financial time series data can fully cover extreme market scenarios, tail risk, volatility clustering characteristics, and factor correlation structures, providing a solid and reliable data foundation for financial risk detection models. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating an embodiment of the data generation method for the financial risk detection model used in this application.

[0042] Figure 2 yes Figure 1 A flowchart illustrating step S4.

[0043] Figure 3 This is a flowchart illustrating Embodiment 2 of the data generation method for the financial risk detection model used in this application.

[0044] Figure 4 This is a schematic diagram of an embodiment of the data generation system used in the financial risk detection model of this application.

[0045] Figure 5 This is a schematic diagram of the hardware structure of an embodiment of the electronic device of this application.

[0046] Figure 6 This is a schematic diagram of the structure of an embodiment of the computer-readable storage medium of this application. Detailed Implementation

[0047] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0048] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0049] It should be noted that, in this document, the term "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0050] The following description, in conjunction with the accompanying drawings, details the enhancement method, system, device, and storage medium for the investment advice suitability review intelligent agent provided in this application, through specific embodiments and application scenarios.

[0051] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the data generation method for a financial risk detection model according to this application. The data generation method for the financial risk detection model includes:

[0052] S1: Obtain multidimensional financial time series data and perform standardization processing to obtain standardized financial time series data;

[0053] Specifically, multidimensional financial time series data are financial time series data from multiple dimensions of the financial market, including asset return series, trading volume, volatility indicators, interest rate factors and / or information spreads, which characterize the financial market situation from different dimensions.

[0054] Standardize the multidimensional financial time series data to ensure that it can meet the data requirements of any of steps S2-S5.

[0055] S2: Based on the continuous-time stochastic differential equation framework, a forward diffusion process is performed on standardized financial time series data to gradually evolve it into a noise distribution. At the same time, the scoring function is trained to learn the distribution gradient information of standardized financial time series data on the diffusion path.

[0056] Specifically, in step S2, the forward diffusion process is based on a continuous-time stochastic differential equation framework. A geometric Brownian motion model is constructed and introduced to guide the diffusion process. The geometric Brownian motion model defines a time-related drift term and a noise intensity scheduling function, so that the noise injection process conforms to the mechanism of stochastic evolution of financial asset prices.

[0057] The forward diffusion process is used to inject controlled noise and model the probability distribution evolution of standardized financial time series data. This forward diffusion process is no longer limited to the traditional discrete Markov chain noise addition method, but is based on a continuous-time stochastic differential equation framework to construct the diffusion process, enabling the standardized financial time series data to gradually evolve into an analytical noise distribution over time, thus providing a stable data foundation for subsequent steps.

[0058] Specifically, by explicitly modeling the drift and fluctuation terms during the forward diffusion process, the noise injection process conforms to the basic mechanism of the stochastic evolution of financial assets.

[0059] Its continuous form can be expressed as:

[0060]

[0061] in, This indicates that standardized financial time series are in diffusion time. The potential state below, For time-related drift terms, For noise intensity scheduling function, This is a standard Brownian motion process.

[0062] Explicitly defined drift terms are used to simulate the long-term trend or expected return component of financial asset prices or returns under noise-free conditions. This is achieved by correlating it with time (i.e.,...). This can model the overall drift characteristics of different stages of the financial market (such as the upward trend of a bull market and the downward trend of a bear market), thus making the noise injection process more consistent with the evolution of financial assets at the trend level.

[0063] Fluctuation term (usually determined by noise intensity scheduling function) The noise directly corresponds to the volatility of the financial market. By designing it to be time-dependent or state-dependent, it can simulate the core characteristics of the financial market, such as volatility clustering (high volatility is often followed by high volatility) and heteroscedasticity (volatility changes over time). This makes the injected noise conform to the uncertainty pattern of risk evolution in the real financial market in terms of intensity variation.

[0064] In practical implementation, the above forward diffusion process can be equivalently represented by a noise intensity scheduling function. The progressive noise injection mechanism allows the distribution of standardized financial time series data to smoothly transition to an approximately standard Gaussian distribution along the diffusion time axis. Compared to diffusion methods that rely solely on independent and identically distributed Gaussian noise, this forward diffusion process effectively avoids uneconomically reasonable anomalous patterns in the generated data samples by introducing stochastic evolution constraints that conform to financial statistics.

[0065] Furthermore, to overcome the limitations of traditional geometric Brownian motion models, which typically assume the drift and fluctuation terms to be constants and are difficult to characterize the heteroscedasticity and volatility clustering features of real financial markets, a diffusion learning mechanism based on a score function is introduced into the forward diffusion process.

[0066] During the forward diffusion process, a scoring function is trained to learn the distribution gradient information of standardized financial time series data along the diffusion path. This scoring function is configured to take the time state of the diffusion process as input and adaptively capture the dynamic changes in the distribution of standardized financial time series data under different financial market conditions, thus implicitly modeling the time-dependent dynamics of the forward diffusion process. Here, the dynamics refer to the drift term and the noise intensity scheduling function.

[0067] Specifically, by constructing a scoring function The data distribution at different time states during the diffusion process is modeled, and its objective is approximated as follows:

[0068]

[0069] in, This indicates that standardized financial time series data are in diffusion time. The probability distribution is as follows. This differs from the explicitly fixed drift term parameters in traditional geometric Brownian motion. And fluctuation term parameters Unlike other functions, this scoring function learns the drift and volatility parameters adaptively during training using a data-driven approach, adapting to different financial market stages, volatility levels, and risk states.

[0070] S3: Construct a conditional scoring function based on preset financial condition variables and scoring functions, and use the conditional scoring function to perform a controllable back diffusion process on the noise distribution to generate synthetic financial time series data that conforms to financial condition variables;

[0071] Specifically, in step S3, the controllable reverse diffusion process is achieved through a controllable reverse nonlinear diffusion mechanism;

[0072] Among them, the controllable reverse nonlinear diffusion mechanism is achieved by explicitly embedding financial condition variables into the drift and fluctuation terms of the stochastic differential equation corresponding to the controllable reverse diffusion process.

[0073] After completing the forward diffusion process, step S3 further performs a controlled back diffusion process on the noise distribution after the forward diffusion process. This controlled back diffusion process is nonlinear and is used to achieve conditional control over the generated data samples during the controlled back diffusion process. This controlled back diffusion process generates a nonlinear back diffusion model by explicitly embedding preset financial condition variables into the drift and fluctuation terms of the SDE equation framework and fusing them with geometric Brownian motion. Algorithmically, this ensures that the generation process is continuously constrained at the diffusion dynamics level, rather than performing post-generation screening.

[0074] The pre-defined financial condition variables include at least one variable representing market conditions, volatility levels, macroeconomic indicators, or risk level labels.

[0075] Furthermore, the controllable reverse nonlinear diffusion mechanism includes a conditional drift function and a conditional fluctuation function; both the conditional drift function and the conditional fluctuation function take the potential state, time, and financial condition variables in the controllable reverse diffusion process as inputs, and their parameters are learned through a data-driven approach.

[0076] The mathematical form of the above process is:

[0077]

[0078] in, This represents the potential state of a financial time series during a controllable reverse diffusion process; This represents a pre-defined financial condition variable; To conditionally change the drift function, Both are conditional wave functions, and are determined by parameters. Control and learn through a data-driven approach.

[0079] Furthermore, the conditional drift function is constructed based on the trained conditional score function;

[0080] Conditional drift function Used to characterize under given financial conditions Below, the expected evolution direction of standardized financial time series data along the time dimension during controlled back diffusion is determined. Unlike traditional geometric Brownian motion models that use fixed or linear drift term parameters, the value of the conditional drift function is determined by subtracting the product of the square of a diffusion intensity function and the conditional score function from the value of the basic drift term defined during forward diffusion.

[0081] The mathematical form of the conditional drift function is:

[0082]

[0083] in, This is the basic drift term defined during the forward diffusion process. Let be the diffusion intensity function. This is a conditional scoring function.

[0084] Furthermore, the conditional score function is configured to learn the log probability density gradient of the underlying state in a controlled backdiffusion process of standardized financial time series data, given financial conditional variables.

[0085] The mathematical form of the conditional score function is:

[0086]

[0087] From a unified stochastic process perspective, the controllable backdiffusion process employed in this application can be viewed as a nonlinear generalization of the traditional geometric Brownian motion model. Its drift and fluctuation terms are extended from static constants to dynamic functions dependent on time, state, and financial conditions. This allows the process to characterize complex financial risk dynamics that are difficult to describe using traditional methods, including frequent extreme events, risk state transitions, and distortions in distribution structures. In this way, synthetic financial time series data conforming to financial condition variables are generated.

[0088] Through the methods in steps S2 and S3 above, this application achieves high-fidelity and controllable data expansion of standardized financial time series data while maintaining the consistency of the random evolution mechanism of financial assets, generating synthetic financial time series data that conforms to financial condition variables, providing more accurate and reliable data support for the training and optimization of subsequent financial risk detection models in scenarios of sample imbalance, small sample size and tail risk.

[0089] S4: Validate the synthetic financial time series data and correct it based on the validation results to obtain enhanced financial time series data;

[0090] Specifically, such as Figure 2 As shown, the validation of synthetic financial time series data includes:

[0091] S401: Verify the consistency of the statistical distribution characteristics and / or time series dynamic characteristics of the synthesized financial time series data.

[0092] Optionally, consistency tests are performed on the statistical properties of synthetic financial time series data, such as mean, volatility, skewness, and kurtosis. Quantile distribution and tail risk indicators are used to compare and analyze the synthetic financial time series data with real financial market data. Secondly, the time dependence structure and volatility clustering characteristics of the synthetic financial time series data are verified using autocorrelation functions, partial autocorrelation functions, and conditional heteroscedasticity tests to ensure that it conforms to the dynamic behavior of real financial markets.

[0093] For details, please refer to [link / reference]. Figure 2 The correction of synthetic financial time series data based on the verification results includes:

[0094] S402: Reconstruct the synthetic financial time series data based on the verification results;

[0095] S403: And feed the verification result back to the forward diffusion process or the controllable reverse diffusion process to adjust the subsequent data generation process.

[0096] Optionally, if deviations are found in the synthetic financial time series data in terms of drift trend, noise structure, or risk distribution during the verification process, a reconstruction mechanism will be triggered to perform drift correction, noise smoothing, or feature reparameterization on the synthetic financial time series data. The verification results will be fed back to the forward diffusion process or the controllable reverse diffusion process to adjust the data generation process, thereby forming a closed-loop optimization process between data generation and verification.

[0097] Through the above process, synthetic financial time series data will be used to generate enhanced financial time series data.

[0098] S5: The enhanced financial time series data is fused with the standardized financial time series data to obtain a dataset for training the financial risk detection model.

[0099] Specifically, after verification, high-quality enhanced financial time series data is fused with standardized financial time series data to construct a stable dataset for training financial risk detection models.

[0100] Please see Figure 3 , Figure 3 This is a flowchart illustrating Embodiment Two of the data generation method for the financial risk detection model in this application. The data generation method for the financial risk detection model includes:

[0101] S1: Obtain multidimensional financial time series data and standardize the multidimensional financial time series data to obtain standardized financial time series data;

[0102] S101: Perform latent factor decomposition on standardized financial time series data to transform it into low-dimensional latent factor time series data expressed by a few common factors, while preserving the covariance structure of multidimensional financial time series data;

[0103] To reduce the instability of high-dimensional standardized financial time series data to subsequent data samples, we first perform latent factor decomposition on the standardized financial time series data based on factor models. Latent factor decomposition is used to extract a few common factors in the standardized financial time series data and retain the covariance structure and risk co-movement characteristics of multidimensional financial time series data.

[0104] The mathematical form of this process is:

[0105]

[0106] in For standardized financial time series data matrices, For the latent factor matrix, For the factor loading matrix, for Residual noise in the data.

[0107] This method transforms high-dimensional standardized financial time series data into low-dimensional latent factor time series data while preserving the covariance structure and risk co-movement characteristics of multi-dimensional financial time series data, ensuring that the generated data can truly reflect the overall trend of the financial market and the interrelationships of various risks.

[0108] S2: Perform a forward diffusion process on the latent factor time series data to gradually evolve the latent factor time series data into a noise distribution based on the continuous-time stochastic differential equation framework. During the forward diffusion process, the distribution gradient information of the latent factor time series data on the diffusion path is learned by training the scoring function, thereby implicitly modeling the dynamic characteristics of the forward diffusion process.

[0109] Specifically, the input data in step S2 of this embodiment is latent factor time series data. This process ensures that the covariance and distribution structure of various samples in the latent factor space are maintained, providing a theoretical basis for generating scarce financial risk scenario samples.

[0110] The subsequent steps are similar to those in Example 1, and will not be repeated here.

[0111] Embodiment 2 of this application decomposes standardized financial time series data into latent factors, and then processes the low-dimensional latent factor time series data through forward diffusion, controlled back diffusion, verification, and fusion to form a dataset for training a financial risk detection model. Embodiment 2 constructs a complete, controllable, and repeatable financial data augmentation algorithm chain, which not only effectively alleviates the problems of sample scarcity and insufficient data in extreme scenarios in financial risk detection, but also provides a data generation method with clear financial mechanism explanations, strong controllability, and applicability to various financial application scenarios.

[0112] Please see Figure 4 , Figure 4 This is a schematic diagram of an embodiment of the data generation system used in the financial risk detection model of this application. The data generation system includes:

[0113] Data preprocessing module 21 is used to acquire multidimensional financial time series data and perform standardization processing to obtain standardized financial time series data;

[0114] SDE modeling module 22 is used to provide a framework for continuous-time stochastic differential equations;

[0115] The diffusion process construction module 23 is used to perform a forward diffusion process on standardized financial time series data based on the continuous-time stochastic differential equation framework, so that it gradually evolves into a noise distribution, while training the scoring function to learn the distribution gradient information of standardized financial time series data on the diffusion path.

[0116] The conditional sampling and generation module 24 is used to construct a conditional scoring function based on preset financial conditional variables and scoring functions, and to use the conditional scoring function to perform a controllable back diffusion process on the noise distribution to generate synthetic financial time series data that conforms to the financial conditional variables.

[0117] The data verification and reconstruction module 25 is used to verify the synthetic financial time series data and correct the synthetic financial time series data according to the verification results to obtain enhanced financial time series data.

[0118] The fusion module 26 is used to fuse enhanced financial time series data with standardized financial time series data to obtain a dataset for training a financial risk detection model.

[0119] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments 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. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the aforementioned method embodiment one, and will not be repeated here.

[0120] Please see Figure 5 , Figure 5 This is a schematic diagram of the hardware structure of an embodiment of the electronic device of this application. The device includes a memory and a processor coupled to each other. The processor is used to execute program instructions stored in the memory to implement the corresponding process as described in the foregoing method embodiment. In a specific implementation scenario, the electronic device 30 may include, but is not limited to, a microcomputer or a server. In addition, the electronic device 30 may also include mobile devices such as laptops and tablets, which are not limited here.

[0121] The electronic device may include a processor 301 and a memory 302 storing program instructions.

[0122] When processor 301 executes the program, it implements the steps in any of the above method embodiments.

[0123] For example, the program can be divided into one or more modules / units, one or more of which are stored in memory 302 and executed by processor 301 to complete this application. The one or more modules / units can be a series of program instruction segments capable of performing a specific function, which describe the execution process of the program in the device.

[0124] Specifically, the processor 301 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0125] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 302 is non-volatile solid-state memory.

[0126] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the methods according to one aspect of this disclosure.

[0127] The processor 301 implements any of the methods described above by reading and executing program instructions stored in the memory 302.

[0128] In one example, the electronic device may also include a communication interface 303 and a bus 310. The processor 301, memory 302, and communication interface 303 are connected via the bus 310 and communicate with each other.

[0129] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0130] Bus 310 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 310 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0131] In addition, the process in conjunction with the foregoing method embodiments, such as Figure 6 As shown, this application embodiment can be implemented using a computer-readable storage medium 40. The storage medium 40 stores program instructions 401; these program instructions 401 are executed by a processor to implement any step in the above-described method embodiment one.

[0132] This application also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various processes of the above method embodiments and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0133] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0134] This application provides a computer program product, which is stored in a storage medium and executed by at least one processor to implement the various processes of the above method embodiments and achieve the same technical effects. To avoid repetition, it will not be described again here.

[0135] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0136] The functional modules shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on machine-readable media or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable media" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer grids such as the Internet, intranets, etc.

[0137] The foregoing flowcharts and / or block diagrams of methods, systems, and program products according to embodiments of this disclosure have described various aspects of the present disclosure. It should be understood that each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to create a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the function / action specified in one or more blocks of the flowcharts and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified function or action, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0138] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A data generation method for a financial risk detection model, characterized in that, Includes the following steps: S1: Obtain multidimensional financial time series data and perform standardization processing to obtain standardized financial time series data; S2: Based on the continuous-time stochastic differential equation framework, the standardized financial time series data is subjected to a forward diffusion process, which gradually evolves it into a noise distribution. At the same time, the scoring function is trained to learn the distribution gradient information of the standardized financial time series data on the diffusion path. S3: Construct a conditional scoring function based on the preset financial condition variables and the scoring function, and use the conditional scoring function to perform a controllable back diffusion process on the noise distribution to generate synthetic financial time series data that conforms to the financial condition variables; S4: Verify the synthetic financial time series data and correct the synthetic financial time series data according to the verification results to obtain enhanced financial time series data; S5: The enhanced financial time series data is fused with the standardized financial time series data to obtain a dataset for constructing the financial risk detection model.

2. The data generation method according to claim 1, characterized in that, In step S3, the controllable reverse diffusion process is achieved through a controllable reverse nonlinear diffusion mechanism; The controllable reverse nonlinear diffusion mechanism is achieved by explicitly embedding the financial condition variables into the drift and fluctuation terms of the stochastic differential equation corresponding to the controllable reverse diffusion process.

3. The data generation method according to claim 2, characterized in that, The controllable reverse nonlinear diffusion mechanism includes a conditional drift function and a conditional wave function; The conditional drift function and the conditional fluctuation function both take the potential state, time and the financial condition variables in the controllable back diffusion process as inputs, and their parameters are learned through a data-driven approach.

4. The data generation method according to claim 3, characterized in that, The value of the conditional drift function is determined by subtracting the product of the square of a diffusion intensity function and the conditional score function from the value of the basic drift term defined in the forward diffusion process.

5. The data generation method according to claim 4, characterized in that, The conditional scoring function is configured to learn the log probability density gradient of the corresponding latent state in the controlled backdiffusion process of the standardized financial time series data, given the financial condition variables.

6. The data generation method according to any one of claims 2 to 5, characterized in that, The financial condition variables include at least one variable that represents market conditions, volatility levels, macroeconomic indicators, or risk level labels.

7. The data generation method according to claim 1, characterized in that, In step S2, the forward diffusion process is constructed based on a continuous-time stochastic differential equation framework, and the noise distribution is controlled by a time-dependent drift term and a noise intensity scheduling function.

8. The data generation method according to claim 7, characterized in that, The scoring function trained in step S2 is configured to take the time of the forward diffusion process as input and can adaptively capture the dynamic changes in the distribution of the standardized financial time series data, thereby implicitly modeling the time-dependent dynamics of the forward diffusion process.

9. The data generation method according to claim 1, characterized in that, In step S4, the verification of the synthetic financial time series data includes: The consistency of the statistical distribution characteristics and / or time series dynamic characteristics of the synthesized financial time series data is verified.

10. The data generation method according to claim 9, characterized in that, In step S4, correcting the synthetic financial time series data based on the verification results includes: The synthetic financial time series data is reconstructed based on the verification results; The verification result is then fed back to the forward diffusion process or the controllable reverse diffusion process to adjust the subsequent data generation process.

11. The data generation method according to claim 1, characterized in that, After step S1 and before step S2, the method further includes: The standardized financial time series data is subjected to latent factor decomposition, which transforms it into low-dimensional latent factor time series data expressed by a few common factors, while preserving the covariance structure of the multidimensional financial time series data. Step S2 executes the forward diffusion process based on the latent factor time series data.

12. The data generation method according to claim 1, characterized in that, The multidimensional financial time series data includes asset return series, trading volume, volatility indicators, interest rate factors, and / or information spreads.

13. A data generation system for a financial risk detection model, characterized in that, The data generation system includes: The data preprocessing module is used to acquire multidimensional financial time series data and perform standardization processing to obtain standardized financial time series data. The SDE modeling module is used to provide a framework for continuous-time stochastic differential equations; The diffusion process construction module is used to perform a forward diffusion process on the standardized financial time series data based on the continuous-time stochastic differential equation framework, so that it gradually evolves into a noise distribution, while training the scoring function to learn the distribution gradient information of the standardized financial time series data on the diffusion path; The conditional sampling and generation module is used to construct a conditional scoring function based on preset financial conditional variables and the scoring function, and to perform a controllable back diffusion process on the noise distribution using the conditional scoring function to generate synthetic financial time series data that conforms to the financial conditional variables. The data verification and reconstruction module is used to verify the synthetic financial time series data and correct the synthetic financial time series data according to the verification results to obtain enhanced financial time series data. The fusion module is used to fuse the enhanced financial time series data with the standardized financial time series data to obtain a dataset for constructing the financial risk detection model.

14. An electronic device, characterized in that, The device includes a memory and a processor coupled to each other, the processor being configured to execute program instructions stored in the memory to implement the data generation method for a financial risk detection model as described in any one of claims 1 to 12.

15. A computer-readable storage medium having program instructions stored thereon, characterized in that, When the program instructions are executed by the processor, they implement the data generation method for a financial risk detection model as described in any one of claims 1 to 12.

Images