Financial risk early warning analysis method and system and storage medium

Abstract

The invention provides a financial risk early warning analysis method and system, a storage medium and a device, relates to the field of enterprise risk management, and solves the technical problem that the existing financial risk early warning technology is not timely and inaccurate. The method comprises the following steps: acquiring financial data, operation data and external environment data of an enterprise to obtain a plurality of index data; classifying the plurality of index data into a plurality of risk domains, calculating the risk field intensity of each risk domain based on the index data in the risk domains, and forming a risk vector; obtaining the risk field intensity of each risk domain in a preset period of the enterprise, inputting the risk field intensity of each risk domain in the preset period into a first network based on a graph neural network, and outputting a cross-risk domain infection probability; constructing an adjacent matrix according to the cross-risk domain infection probability; and calculating the maximum eigenvalue of the adjacent matrix and the risk vector, and triggering graded early warning based on the maximum eigenvalue and the derivative of the maximum eigenvalue. The method and the device are used in the whole-process financial risk management process of enterprises.

Description

Technical Field

[0001] The present application relates to the field of enterprise risk management, and in particular to a financial risk early warning analysis method, system and storage medium. Background Art

[0002] In modern business operations, accurate financial risk forecasting is crucial for sales-oriented companies to maintain robust operations. The diversity of supply chain nodes and fluctuations in product return rates directly impact a company's cash flow, inventory turnover, and debt repayment capacity. For example, supplier delivery delays can cause production line downtime, leading to raw material inventory backlogs and overdue accounts payable. Rising product return rates can compress gross profit margins, and failure to provide timely warnings can further precipitate liquidity risks and market credibility crises.

[0003] However, since traditional risk warning methods rely solely on financial data and mostly adopt a fixed weight system, they lack real-time monitoring of abnormal fluctuations in indicators, which can easily lead to delays in risk calculations and fail to promptly reflect the actual impact of different aspects of the company on financial stability, resulting in untimely and inaccurate warnings. Summary of the Invention

[0004] The present application provides a financial risk early warning analysis method, system and storage medium, which solve the technical problems in the prior art of enterprise risk early warning being inaccurate and prone to delay.

[0005] To achieve the above objectives, this application adopts the following technical solutions:

[0006] In a first aspect, a financial risk early warning analysis method is provided, comprising:

[0007] Obtain the company's financial data, operating data, and external environment data to obtain several indicator data;

[0008] Classify several indicator data into multiple risk domains, and calculate the risk field intensity of each risk domain based on the indicator data in the risk domain to form a risk vector;

[0009] Obtain the risk field strength of each risk domain within a preset period of the enterprise, input the risk field strength of each risk domain within the preset period into a first network based on a graph neural network, and output the probability of contagion across risk domains;

[0010] Construct an adjacency matrix based on the probability of contagion across risk domains;

[0011] Calculate the maximum eigenvalue of the adjacency matrix and the risk vector, and trigger graded warnings based on the maximum eigenvalue and its derivative.

[0012] Based on the above technical solution, in a financial risk early warning analysis method provided in this application, by integrating corporate financial data, operating data and external environmental data and classifying them into multiple risk domains, structured decomposition and quantitative analysis of multi-dimensional risks are achieved; the risk field intensity is calculated based on risk domain indicators and a risk vector is composed, which provides a mathematical representation basis for the spatial distribution and dynamic evolution of risks; a graph neural network is introduced to construct a first network to analyze the probability of contagion across risk domains, effectively capturing the nonlinear correlations and complex topological relationships in risk transmission, and breaking through the limitations of traditional linear models in depicting risk linkage mechanisms; the adjacency matrix is ​​constructed using the contagion probability and combined with the risk vector to calculate the maximum eigenvalue, and the overall stability of the risk system is quantitatively evaluated based on the linear algebra eigenvalue theory. The mapping relationship between the maximum eigenvalue and the risk field intensity provides a mathematical criterion for graded early warning, enabling the early warning mechanism to dynamically respond to differences in the speed and intensity of risk evolution.

[0013] Furthermore, the financial data include: cash ratio, quick ratio, net cash flow from operating activities, debt-to-asset ratio, interest coverage ratio, debt ratio, gross profit margin, and return on equity;

[0014] The operating data include: supplier delivery delay rate, inventory turnover days, production line utilization rate, product qualification rate, order cancellation rate, and supplier concentration;

[0015] The external environment data include: GDP growth rate, benchmark interest rate, industry average debt ratio, and raw material price volatility.

[0016] Furthermore, the classification of the indicator data into multiple risk domains includes:

[0017] Liquidity risk domain, including cash ratio, quick ratio, net cash flow from operating activities, and inventory turnover days;

[0018] Debt repayment risk domain, including asset-liability ratio, interest coverage ratio, debt ratio, and industry average debt ratio;

[0019] Supply chain risk domain, including supplier delivery delay rate, production line utilization rate, product qualification rate, order cancellation rate, and supplier concentration;

[0020] Market risk domain, including gross profit margin, return on net assets, GDP growth rate, benchmark interest rate, and raw material price volatility.

[0021] Furthermore, the calculation of the risk field intensity of each risk domain based on the indicator data within the risk domain includes:

[0022] Calculate the mutation degree ΔI of indicator i in risk domain k k，i (t), the formula is: Where Δt represents the time difference, Ii (t) represents the value of index i on day t, σ i (T) represents the rolling standard deviation of the value of indicator i in period T;

[0023] Calculate the initial weight of indicator i in risk domain k using entropy weight method

[0024] Based on the initial weight superimposed on the time attenuation factor, the weight value w of indicator i in risk domain k is obtained k，i , the formula is: Where λ represents the time attenuation coefficient, and days represents the time difference between the time when the most recent mutation degree of indicator i in risk domain k was greater than the preset mutation threshold and the current time;

[0025] Determine the domain toxicity coefficient α of risk domain k based on financial data k , exogenous shock coefficient γ k , and combined with the mutation degree and weight value of several indicators in the risk area, a risk field intensity calculation function is constructed, which is used to calculate the risk field intensity of each risk domain.

[0026] Furthermore, the risk field intensity calculation function satisfies: Among them, RFI k (t) represents the risk field intensity calculation function, β represents the risk transmission sensitivity, which is calculated based on the product of the number of supply chain nodes and the supplier concentration, and Ψ(t) represents the comprehensive impact intensity. is the Sigmoid function.

[0027] Furthermore, the comprehensive impact strength is determined as follows:

[0028] Extract the value of the external impact factor E at time t from the financial data j (t) The external impact factors include GDP growth rate, benchmark interest rate, industry average debt ratio, and raw material price volatility;

[0029] Calculate the normalized offset Ψ of the external shock factor j j (t), the formula is: Among them, μ j represents the historical mean of the external shock factor j, σ j represents the historical standard deviation of external shock factor j;

[0030] The shock coupling weight of the external shock factor j is calculated according to the formula: Among them, REF represents the average value of the risk field intensity of each risk domain in the historical period, corr(E j(t), RFI(t)) represents the correlation coefficient between the external impact factor j on day t and the average value, and τ represents a preset influencing parameter, which can be determined by experimental calibration;

[0031] The comprehensive shock coupling strength is calculated based on the normalized offset and shock coupling weight, and the formula is: Where N represents the number of external shock factors.

[0032] Furthermore, the domain toxicity coefficient α of the risk domain k is determined based on the financial data. k , exogenous shock coefficient γ k ,include:

[0033] Determine the core indicators of risk domain k, where the core indicators include positive indicators and negative indicators, and each risk domain includes one core indicator;

[0034] Obtain the data of the core indicators of risk domain k in the historical period, and based on the current value I of the core indicators k (t) Calculate the historical percentile P k (t);

[0035] If the core indicator of risk domain k is a positive indicator, then the domain toxicity coefficient α k The expression is: k =1.0+0.5×(1-P k (t));

[0036] If the core indicator of risk domain k is a negative indicator, then the domain toxicity coefficient α k The expression is: k =1.0+0.5×P k (t);

[0037] Calculate the absolute value of the correlation coefficient ρ between each external impact factor and the wind direction field intensity of the risk domain k in the historical period k，j , the formula is: k，j =|corr(E j (t), RFI k )|;where corr() represents the Pearson correlation coefficient function;

[0038] The maximum absolute value of the correlation coefficient is taken as the sensitivity benchmark S of the risk domain k. k , the formula is:

[0039] Calculate the exogenous shock coefficient γ of risk domain k based on the sensitivity benchmark k , the formula is: γ k =0.1+0.8×S k .

[0040] Furthermore, the method for obtaining the infection probability across risk domains includes:

[0041] Construct an independent graph instance G = (V, E) for each historical risk event; where V represents the node of the independent graph, which is used to represent the risk domain category, and E represents the edge of the independent graph, which is used to represent the risk transmission path from risk domain i to risk domain j;

[0042] A graph convolutional network is constructed based on each independent graph instance, and the node feature vector v of the graph convolutional network is k v k =[RFI k (t), ΔRFI k / Δt, domain_type], the edge weight w of the graph convolutional network ij is the normalized value of the number of risk transmission events from risk domain i to risk domain j in historical risk events; where ΔRFI k / Δt represents the wind intensity change rate of risk domain k within the time difference Δt, and domain_type represents the one-hot encoding of the risk domain category;

[0043] Train and optimize the graph convolutional network to obtain the first network;

[0044] Determining an input node feature vector of the first network based on the risk field intensity of each risk domain within a preset period of the enterprise;

[0045] The input node feature vector is fed into the first network, and the contagion probability across risk domains is output.

[0046] Furthermore, the triggering of graded warning based on the maximum eigenvalue and the derivative of the maximum eigenvalue includes:

[0047] When the maximum eigenvalue is greater than or equal to the first preset threshold and less than the second preset threshold, and the derivative dμ / dt of the maximum eigenvalue μ is greater than 0, a first-level warning is triggered;

[0048] When the maximum eigenvalue is greater than or equal to the second preset threshold and less than the third preset threshold, and the derivative dμ / dt of the maximum eigenvalue μ is greater than the first preset derivative threshold and less than the second preset derivative threshold, a first-level warning is triggered;

[0049] When the maximum eigenvalue is greater than or equal to the third preset threshold, and the derivative dμ / dt of the maximum eigenvalue μ is greater than the second preset derivative threshold, a first-level warning is triggered.

[0050] In a second aspect, the present application provides a financial risk early warning system, comprising: a data acquisition module, a risk analysis module and an early warning trigger module; wherein,

[0051] The data acquisition module is used to obtain the financial data, operating data and external environment data of the enterprise to obtain a number of indicator data;

[0052] The risk analysis module is used to classify a number of indicator data into a plurality of risk domains, calculate the risk field intensity of each risk domain based on the indicator data in the risk domain, and form a risk vector, and

[0053] Obtain the risk field intensity of each risk domain within the preset period of the enterprise, pre-process the risk field intensity of each risk domain within the preset period, and input it into the first network based on the graph neural network to output the probability of contagion across risk domains;

[0054] The warning trigger module is used to construct an adjacency matrix based on the infection probability across risk domains, calculate the maximum eigenvalue of the adjacency matrix and the risk vector, and trigger a graded warning based on the maximum eigenvalue and the derivative of the maximum eigenvalue.

[0055] In a third aspect, the present application provides a financial risk early warning analysis device, comprising: a communication unit and a processing unit;

[0056] The communication unit is used to receive and transmit the enterprise's financial data, operating data, external environment data, and the risk field intensity of each risk domain within a preset period;

[0057] The processing unit is configured to execute:

[0058] Classify several indicator data into multiple risk domains, and calculate the risk field intensity of each risk domain based on the indicator data in the risk domain to form a risk vector;

[0059] Input the risk field intensity of each risk domain within a preset period into the first network based on the graph neural network, and output the probability of infection across risk domains;

[0060] Construct an adjacency matrix based on the probability of contagion across risk domains;

[0061] Calculate the maximum eigenvalue of the adjacency matrix and the risk vector, and trigger graded warnings based on the maximum eigenvalue and its derivative.

[0062] In a fourth aspect, the present application provides a financial risk early warning analysis device, comprising: a processor and a storage medium; the storage medium comprising instructions, the processor configured to execute the instructions to implement the method described in the first aspect and any possible implementation of the first aspect. The financial risk early warning analysis device may be an electronic device or a chip within the electronic device.

[0063] In the fifth aspect, the present application provides a computer-readable storage medium, which stores instructions. When the instructions are run on a financial risk warning analysis device, the financial risk warning analysis device executes the method described in the first aspect and any possible implementation of the first aspect.

[0064] In a sixth aspect, the present application provides a computer program product comprising instructions, which, when run on a financial risk warning analysis device, enables the financial risk warning analysis device to execute the method described in the first aspect and any possible implementation of the first aspect.

[0065] This application provides a financial risk early warning analysis method that enables systematic integration and structured analysis of multi-dimensional risk data. By integrating financial data, operating data, and external environmental data, indicators are classified into risk domains such as liquidity, debt repayment, supply chain, and market, addressing the one-sidedness of traditional single-indicator analysis. Furthermore, a risk transmission model is constructed based on a graph neural network, and the probability of contagion across risk domains is learned through historical event training, enabling dynamic modeling of complex relationships between risk domains.

[0066] At the risk assessment and early warning level, we introduce a mutation degree to measure abnormal fluctuations in indicators, combine it with a time decay factor to strengthen near-term risk weights, and calculate a domain toxicity coefficient based on the historical percentiles of core indicators, enabling the intensity of the risk field to reflect the priority of risk evolution in real time. By constructing a risk coupling matrix and solving for the maximum eigenvalue, and combining these eigenvalues ​​to trigger three levels of early warning, we achieve timely early warning and dynamic evolution monitoring of enterprise risks, providing decision-making support for enterprises from risk identification to tiered disposal.

[0067] It should be understood that the description of technical features, technical solutions, beneficial effects or similar language in this application does not imply that all features and advantages can be realized in any single embodiment. On the contrary, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution or beneficial effect is included in at least one embodiment. Therefore, the description of a technical feature, technical solution or beneficial effect in this specification does not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions and beneficial effects described in the present embodiment can also be combined in any appropriate manner. Those skilled in the art will understand that the embodiment can be implemented without one or more specific technical features, technical solutions or beneficial effects of a specific embodiment. In other embodiments, additional technical features and beneficial effects can also be identified in specific embodiments that do not embody all embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0069] Figure 1 A schematic diagram of the architecture of a financial risk early warning analysis system provided in an embodiment of the present application;

[0070] Figure 2 A flowchart of a financial risk early warning analysis method provided in an embodiment of the present application;

[0071] Figure 3 A flowchart of another financial risk early warning analysis method provided in an embodiment of the present application;

[0072] Figure 4 A schematic diagram of the structure of a financial risk early warning analysis device provided in an embodiment of the present application;

[0073] Figure 5 A schematic diagram of the hardware structure of a financial risk early warning analysis device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0074] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0075] In the description of this application, unless otherwise specified, " / " means "or", for example, A / B can mean A or B. "And / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, "at least one" means one or more, and "a plurality" means two or more. Words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not limit them to be different.

[0076] It should be noted that, in this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this application as "exemplary" or "for example" should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0077] The financial risk early warning analysis method provided in the embodiment of the present application can be applied to Figure 1 In a financial risk early warning analysis system shown in FIG. Figure 1 As shown, the communication system includes: a data acquisition module, a risk analysis module and an early warning trigger module; wherein,

[0078] The data acquisition module is used to obtain the financial data, operating data and external environment data of the enterprise to obtain a number of indicator data;

[0079] The risk analysis module is used to classify a number of indicator data into a plurality of risk domains, calculate the risk field intensity of each risk domain based on the indicator data in the risk domain, and form a risk vector, and

[0080] Obtain the risk field intensity of each risk domain within the preset period of the enterprise, pre-process the risk field intensity of each risk domain within the preset period, and input it into the first network based on the graph neural network to output the probability of contagion across risk domains;

[0081] The warning trigger module is used to construct an adjacency matrix based on the infection probability across risk domains, calculate the maximum eigenvalue of the adjacency matrix and the risk vector, and trigger a graded warning based on the maximum eigenvalue and the derivative of the maximum eigenvalue.

[0082] To address the technical issues of inaccurate and untimely risk prediction in the prior art, the present invention provides a financial risk early warning analysis method, which includes:

[0083] S1, obtain the company's financial data, operating data and external environment data to obtain several indicator data;

[0084] S2, classifying a number of indicator data into multiple risk domains, and calculating the risk field intensity of each risk domain based on the indicator data within the risk domain to form a risk vector;

[0085] S3: Obtain the risk field strength of each risk domain within a preset period of the enterprise, input the risk field strength of each risk domain within the preset period into a first network based on a graph neural network, and output the probability of contagion across risk domains;

[0086] S4, construct an adjacency matrix based on the probability of contagion across risk domains;

[0087] S5, calculate the maximum eigenvalue of the adjacency matrix and the risk vector, and trigger a graded warning based on the maximum eigenvalue and its derivative.

[0088] Based on this, in the financial risk warning analysis method provided in this application, a mathematical logic chain is constructed from data collection, risk quantification to transmission analysis and warning triggering through multi-source data fusion, risk domain structured modeling, graph neural network nonlinear analysis and eigenvalue system evaluation, which realizes the multi-dimensional and accurate characterization of corporate financial risks, cross-domain transmission dynamic tracking and graded warning response, and improves the scientificity, comprehensiveness and timeliness of risk warning.

[0089] like Figure 2 As shown, the embodiment of the present application provides a financial risk early warning analysis method, including:

[0090] S1. Obtain the company's financial data, operating data, and external environment data to obtain several indicator data;

[0091] Among them, financial data includes structured data that reflects the company's financial status and operating results, operating data includes execution efficiency and business status data during the company's production and operation process, and external environment data includes data on external factors such as macroeconomics, industry trends and market fluctuations that affect the company's operations.

[0092] In some implementations, this data can be collected through the enterprise's internal information system interface, a third-party data platform, or manual entry.

[0093] It should be pointed out that data collection must ensure timeliness and accuracy to support the reliability of subsequent risk analysis.

[0094] For example, financial data can be obtained in real time through the ERP system, operational data can be obtained through the supply chain management platform, and external environmental data can be obtained through the public economic database.

[0095] S2. Classify a number of indicator data into multiple risk domains, and calculate the risk field intensity of each risk domain based on the indicator data in the risk domain to form a risk vector.

[0096] Among them, the risk domain is used to represent risk categories of different dimensions, realize the structured decomposition and centralized analysis of multi-dimensional risks, and facilitate the quantitative assessment of risks.

[0097] In some implementations, the classification of risk domains can be set independently, or can be obtained based on principal component analysis or hierarchical clustering methods.

[0098] It should be pointed out that the risk field intensity is used to characterize the risk severity and dynamic change trend of a single risk domain, reflecting the comprehensive fluctuation level and potential threats of indicator data within the risk domain.

[0099] Exemplarily, the risk vector may be expressed as [risk field intensity of risk domain 1, risk field intensity of risk domain 2, risk field intensity of risk domain 3, risk field intensity of risk domain 4]T.

[0100] S3. Obtain the risk field intensity of each risk domain within a preset period of the enterprise, input the risk field intensity of each risk domain within the preset period into a first network based on a graph neural network, and output the probability of contagion across risk domains.

[0101] The risk field intensity of each risk domain within the preset period is also obtained based on the above step S2.

[0102] In some implementations, the underlying architecture of a graph neural network can be based on a graph convolutional network (GCN), a graph attention network (GAT), or a graph recurrent network (GRN).

[0103] It should be pointed out that the training data of the first network is the risk field intensity sequence corresponding to the company's historical financial risk events, and the labels are the actual contagion results between risk domains. It is used to learn the nonlinear correlation pattern of risk transmission and can capture the complex contagion paths and probability distributions across risk domains.

[0104] Exemplarily, after normalizing the risk field intensity of each risk domain within a preset period, the matrix is ​​spliced ​​into a node feature matrix in time series, input into the first network, and the infection probability matrix between the risk domains is obtained through multi-layer graph convolution operations.

[0105] S4. Construct an adjacency matrix based on the probability of infection across risk domains.

[0106] Among them, the adjacency matrix is ​​used to represent the transmission relationship and probability intensity between risk domains.

[0107] In some implementations, the adjacency matrix can be expressed as ij = Probability of infection i->j , where A ij It represents the probability of risk transmission from the i-th risk domain to the j-th risk domain.

[0108] It should be pointed out that the adjacency matrix must satisfy the properties of non-negativity and diagonal elements being 0, and the diagonal elements represent the conduction of the risk domain itself, which usually has no practical meaning.

[0109] For example, if the probability of transmission from risk domain 1 to risk domain 2 is 0.3, then the element A at the corresponding position in the adjacency matrix is 12 =0.3.

[0110] S5. Calculate the maximum eigenvalue of the adjacency matrix and the risk vector, and trigger a graded warning based on the maximum eigenvalue and its derivative.

[0111] Among them, the maximum eigenvalue represents the overall stability and potential diffusion capacity of the risk system. The larger the value, the stronger the risk linkage effect.

[0112] In some implementations, the calculation formula for the maximum eigenvalue may be based on an orthogonal triangular QR decomposition method, which solves the maximum eigenvalue of the characteristic equation by performing a matrix multiplication operation on the adjacency matrix and the risk vector.

[0113] It should be pointed out that the triggering conditions of graded warnings can be adjusted according to factors such as the company's risk tolerance and industry characteristics to achieve personalized warning strategies.

[0114] For example, the warning triggering condition can be set to trigger a first-level warning when the maximum characteristic value is greater than the first threshold and the intensity of a risk domain exceeds 1.5 times the baseline value, and trigger a higher-level warning when it exceeds 2 times.

[0115] Based on the above technical solution, this application realizes multi-dimensional quantitative analysis of corporate financial risks, cross-domain transmission dynamic tracking and graded early warning response, which improves the scientificity and comprehensiveness of risk warning.

[0116] In a possible implementation of the embodiment of the present application, the data in S1 may specifically include:

[0117] Financial data includes but is not limited to: cash ratio, quick ratio, net cash flow from operating activities, debt-to-asset ratio, interest coverage ratio, debt ratio, gross profit margin, and return on equity;

[0118] Operational data includes but is not limited to: supplier delivery delay rate, inventory turnover days, production line utilization rate, product qualification rate, order cancellation rate, and supplier concentration;

[0119] External environmental data include but are not limited to: GDP growth rate, benchmark interest rate, industry average debt ratio, and raw material price volatility.

[0120] Since the generation and evolution of corporate financial risks are affected by internal financial conditions (such as debt-to-asset ratio, cash flow level) and operational efficiency (such as inventory turnover, supplier delivery capabilities), and are also closely related to factors such as the external economic environment (such as GDP growth rate, industry debt level), market fluctuations (such as raw material prices, benchmark interest rates), single-dimensional data cannot fully reveal the complexity and transmission mechanism of risks; and the existing technology lacks systematic integration and analysis of multi-source data, resulting in one-sided risk assessment. Therefore, the corporate financial risk assessment of this application requires not only the company's internal financial data (reflecting debt repayment ability, profitability, etc.) and operating data (reflecting supply chain stability, production efficiency, etc.), but also external environmental data (reflecting macroeconomics, industry trends and market shocks, etc.). Only through the fusion analysis of multi-dimensional data can we more accurately identify risk sources, quantify risk intensity and track cross-domain transmission paths.

[0121] In one possible implementation, combining Figure 2 ,like Figure 3 As shown, the specific implementation steps of step S2 may include:

[0122] S201. Classify indicator data into preset risk domains.

[0123] Among them, risk domain is a multidimensional risk category divided according to the nature of the risk, which is used to achieve structured management of multidimensional risks.

[0124] In some implementations, the risk domain includes four categories: liquidity, debt repayment, supply chain, and market. The classification is based on the correlation between the indicator and the risk type (for example, liquidity risk focuses on short-term cash turnover, and market risk focuses on external economic and profit fluctuations).

[0125] It should be pointed out that risk domain classification needs to cover all dimensions of enterprise operations to avoid missing key risks.

[0126] Exemplary:

[0127] Liquidity risk domain: includes cash ratio, quick ratio, net cash flow from operating activities, and inventory turnover days;

[0128] Debt repayment risk domain: including asset-liability ratio, interest coverage ratio, debt ratio, and industry average debt ratio;

[0129] Supply chain risk domain: including supplier delivery delay rate, production line utilization rate, product qualification rate, order cancellation rate, and supplier concentration;

[0130] Market risk domain: includes gross profit margin, return on net assets, GDP growth rate, benchmark interest rate, and raw material price volatility.

[0131] S202. Calculate the mutation degree and dynamic weight of indicators within the risk domain.

[0132] Among them, the mutation degree is used to measure the abnormal fluctuation degree of indicator data, and the weight is used to characterize the contribution of the indicator to the risk domain.

[0133] Specifically, the mutation degree is first calculated:

[0134] For indicator i in risk domain k, the mutation degree formula is: Among them, Δt is the time difference in days, σ i (T) is the rolling standard deviation of indicator i in period T.

[0135] In some implementations, when |ΔI k，i (t)|When the value exceeds the preset mutation threshold (e.g., 1.5), the indicator is judged to be significantly abnormal.

[0136] Next, the dynamic weights of each indicator are calculated: the initial weights are calculated using the entropy weight method to reflect the objective variability of the indicator data, including:

[0137] First, normalize the indicator data. The formula is: Among them, min I k，i 、max I k，i Respectively represent the minimum and maximum values ​​of indicator i in the historical period T;

[0138] Then calculate the information entropy e of each indicator k,i , the formula is: Where m represents the number of samples, that is, the number of observation data points in the historical period T. For example, if the historical period T is 30 days and one observation value is generated every day, the number of samples is 30.

[0139] Calculate the coefficient of variation: d k，i =1-e k，i ;

[0140] Calculate the initial weights:

[0141] Superimpose the time attenuation factor to obtain the weight value w of indicator i in risk domain k k，i for: Among them, λ is the time attenuation coefficient, and days is the difference between the time when the indicator's mutation degree exceeded the threshold and the current time.

[0142] It should be pointed out that the time decay factor gives recent abnormal indicators a higher weight, which can improve the timeliness of risk assessment.

[0143] For example, if a certain indicator suddenly changes 3 days ago, the default value λ=0.1, then its weight is e times the initial weight. -0.3≈0.74 times.

[0144] S203: Construct a risk field intensity calculation function and output a risk vector.

[0145] Among them, the risk field intensity RFI k Used to comprehensively reflect the inherent risks, external sensitivities and dynamic impacts of the risk domain.

[0146] Specifically, first determine the domain toxicity coefficient α of each risk domain k :

[0147] (1) Identify the core indicators for each risk domain in the financial data and determine the positivity of the core indicators, including:

[0148] Liquidity risk domain: cash ratio, which is a positive indicator;

[0149] Debt repayment risk domain: debt-to-asset ratio, which is a negative indicator;

[0150] Supply chain risk domain: supplier delivery delay rate, a negative indicator;

[0151] Market risk domain: gross profit margin, a positive indicator;

[0152] (2) Obtain the data of the core indicators of risk domain k in the historical period, and based on the current value of the core indicators I k (t) Calculate the historical percentile P k (t);

[0153] (3) Perform toxicity mapping:

[0154] If the core indicator of risk domain k is a positive indicator, then the domain toxicity coefficient α k The expression is: k =1.0+0.5×(1-P k (t));

[0155] If the core indicator of risk domain k is a negative indicator, then the domain toxicity coefficient α k The expression is: k =1.0+0.5×P k (t).

[0156] For example, if the debt-to-asset ratio of the debt repayment risk domain is k (t) = 70%, its historical percentile P k (t) = 0.9 (i.e. more than 90% of historical data), then: k =1.0+0.5×0.9=1.45.

[0157] Then calculate the exogenous shock coefficient γ k :

[0158] (1) Calculate the absolute value of the Pearson correlation coefficient between each external impact factor and the wind direction field intensity of risk domain k in the historical period: ρ k，j =|corr(E j (t), RFI k )|, and take the maximum value as the sensitivity benchmark S k ;

[0159] (2) Calculate the exogenous shock coefficient according to the formula: γ k =0.1+0.8×S k .

[0160] Among them, external impact factors include GDP growth rate, benchmark interest rate, industry average debt ratio, and raw material price volatility.

[0161] In this application, the domain toxicity coefficient is used to reflect the severity of the intrinsic risk of the risk domain itself, and the exogenous impact coefficient is used to characterize the sensitivity of the risk domain to changes in the external environment.

[0162] It should be noted that the calculation of the domain toxicity coefficient and the exogenous impact coefficient are based on historical data. The historical data can be all the historical data of a certain indicator, or the historical data of the most recent period obtained according to the preset period. When the collection period of historical data has not been reached, α can be directly set. k and γ k The initial value of, for example, the initial default value is: α k =1.0,γ k =0.5. The initial value can be gradually revised and optimized as historical data accumulates to adapt to the actual risk characteristics of the enterprise.

[0163] For example, in calculating the normalized offset of the external impact factor j, μ j It represents the historical mean of the external impact factor j, which can be the mean of all historical data, the historical mean of the last two months, or the historical mean from the last time an early warning (including level one, level two, and level three warnings) was triggered to the present.

[0164] Then, the domain toxicity coefficient α of the risk domain k is determined based on the financial data k , exogenous shock coefficient γ k , and combined with the mutation degree and weight values ​​of several indicators in the risk area, the risk field intensity calculation function is constructed as follows: Among them, RFI k (t) represents the risk field intensity calculation function, β represents the risk transmission sensitivity, and β = number of supply chain nodes × supplier concentration, Ψ(t) represents the comprehensive impact intensity, is the Sigmoid function.

[0165] Finally, the RFI of each risk domain k (t) Arrange in order to form a risk vector: V risk =[RFI 流动性 , RFI 偿债 , RFI 供应链 , RFI 市场 ] T .

[0166] β is used to quantify the impact of supply chain structure on the speed and intensity of risk transmission. The more supply chain nodes there are, or the higher the supplier concentration, the more sensitive the risk transmission within the supply chain is, and the larger the β value.

[0167] In some implementations, the number of supply chain nodes can be obtained through an enterprise's supply chain management system or manually compiled and counted. This refers to the number of independent entities in the enterprise's supply chain that directly or indirectly participate in the raw material supply, production, distribution, and other aspects, such as first-tier suppliers, second-tier suppliers, and logistics providers. A greater number of nodes indicates a more complex supply chain network, with more paths for risk transmission across nodes and a higher potential transmission sensitivity.

[0168] Supplier concentration can be calculated based on a company's procurement data, typically using the proportion of procurement value from the top N suppliers (e.g., the proportion of procurement value from the top three suppliers to total procurement value) or the Herfindahl Index (HHI). For example, if the top two suppliers account for 70% of procurement, the supplier concentration is 70%. Supplier concentration reflects a company's reliance on key suppliers. The higher the concentration, the greater the supply chain's reliance on a small number of suppliers, and the greater the probability and intensity of transmission of risks from a single supplier (such as delivery delays and quality issues) to the company.

[0169] For example, if a company's supply chain contains 5 nodes (such as raw material suppliers, processors, assembly plants, distributors, and retailers), and the supplier concentration is 60% (the purchase share of the top two suppliers), then: β = 5 × 60% = 3.

[0170] It should be noted that the comprehensive impact intensity Ψ(t) reflects the real-time fluctuation degree of various impact factors in the external environment and their dynamic coupling effect with corporate risks. Through standardization and correlation weighting, it quantitatively presents the overall impact intensity of the macroeconomy, industry cycles and market fluctuations on corporate financial risks, providing real-time external driving signals for the dynamic calculation of risk field intensity.

[0171] In some implementations, the calculation process of the comprehensive impact strength includes:

[0172] Extract the value of the external impact factor E at time t from the financial data j (t) The external impact factors include GDP growth rate, benchmark interest rate, industry average debt ratio, and raw material price volatility;

[0173] Calculate the normalized offset Ψ of the external shock factor j j (t), the formula is: Among them, μ j represents the historical mean of the external shock factor j, σ j represents the historical standard deviation of external shock factor j;

[0174] The shock coupling weight of the external shock factor j is calculated according to the formula: Among them, REF represents the average value of the risk field intensity of each risk domain in the historical period, corr(E j (t), RFI(t)) represents the correlation coefficient between the external impact factor j on day t and the average value, τ represents a preset influencing parameter, which can be determined by experimental calibration and has a default value of 1;

[0175] The comprehensive shock coupling strength is calculated based on the normalized offset and shock coupling weight, and the formula is: Where N represents the number of external shock factors.

[0176] Based on the above technical solution, a quantitative correlation analysis between the external environment and corporate risks is achieved, providing key parameter support for risk field intensity calculation and cross-domain transmission model.

[0177] In a possible implementation of the embodiment of the present application, the above S3 can be implemented specifically through S301 and S302, which are specifically described below:

[0178] S301: Construction and training of the first network, specifically including:

[0179] (1) Construct independent graph instances and node features.

[0180] First, we construct an independent graph instance G = (V, E) for each historical risk event, where V represents a set of nodes, each node corresponds to a risk domain (liquidity, debt repayment, supply chain, market); E represents an edge set, edge E ij It represents the potential transmission path of risk from risk domain i to j (which can be initially preset based on business experience, such as liquidity risk may be transmitted to debt repayment risk).

[0181] Among them, independent graph instances are used to model risk domains and their transmission relationships, and node features must include risk dynamics and category information.

[0182] Then define the node feature vector of each node as v k =[RFI k (t), ΔRFI k / Δt, domain_type], where RFI k(t) is the risk field intensity of risk domain k at time t, ΔRFI k / Δt represents the risk field intensity change rate of risk domain k within the time difference Δt, reflecting the dynamic trend of risk. domain_type represents the one-hot encoding of the risk domain category.

[0183] Then proceed with data acquisition and preprocessing:

[0184] Extract the risk field intensity sequence of a preset period (e.g., the past 60 days) from the company's historical financial and operating data;

[0185] For continuous features (RFI k (t), rate of change) are standardized (such as subtracting the mean and dividing by the standard deviation), and the category feature (domain_type) is one-hot encoded.

[0186] In some implementations, the graph structure can be dynamically adjusted: if a risk domain has not been transmitted in historical events, the corresponding edge E i,j The weight is set to 0; if there is a conduction record, the edge weight is initialized according to the conduction frequency.

[0187] It should be pointed out that node features need to simultaneously include static attributes (risk domain type) and dynamic indicators (intensity, rate of change) to ensure that graph neural networks capture the spatiotemporal characteristics of risks.

[0188] For example, the node characteristics of a supply chain risk domain are: 供应链 =[0.8,0.05,[0,0,1,0]]; where “[0,0,1,0]” represents the category code of the supply chain risk domain, which means that the supply chain risk domain ranks third in the risk vector.

[0189] (2) Prepare training data and labels.

[0190] Collect the graph instance sets {G1, G2, ..., G n}, the node feature matrix of each graph instance is (N is the number of nodes, D is the feature dimension).

[0191] Then, the actual infection probability is obtained through expert annotation or historical transmission record statistics to construct label data, and the label of each graph instance is the cross-risk domain infection probability matrix where Y ij =1 means the risk spreads from domain i to domain j, Y ij =0 means no infection.

[0192] The dataset is divided into a training set (70%), a validation set (20%), and a test set (10%) to ensure that each set contains different types of risk events.

[0193] In some implementations, if there is a lack of clear infection labels, a weak supervision method can be used: if the intensity of a risk domain exceeds a threshold, and the intensity of the adjacent risk domain increases significantly in subsequent cycles, it is considered that there is infection, and the label can be set to 0.5-1; if there is no obvious change, it is considered that there is no infection, and the label can be set to 0-0.2.

[0194] It should be noted that the label data must be strictly aligned with the graph instance to ensure the temporal consistency of the input features and output labels during training.

[0195] (3) Training and optimizing graph convolutional networks (GCNs).

[0196] A graph convolutional network is used as the initial network of the first network. The training set and validation set are then input into the initial network for iterative training. The cross entropy loss function or mean square error (MSE) loss is used, and the optimization goal is to minimize the difference between the predicted infection probability and the label: in represents the infection probability predicted by the model.

[0197] During the iterative training process, the Adam optimizer can be used, with the initial learning rate set to 0.001 and a 10% decay every 50 rounds of training; and the validation set loss is monitored. If it does not decrease for 10 consecutive rounds, the training is terminated early (early stopping strategy).

[0198] In some implementations, a graph attention mechanism (GAT) can be introduced to replace GCN to enhance the learning of key transmission paths by calculating the attention weights between nodes.

[0199] It should be pointed out that the training data needs to cover multiple risk transmission modes, such as single-domain triggering and multi-domain linkage, to ensure the generalization ability of the model.

[0200] For example, after 300 rounds of training, the MSE loss of the model on the test set dropped to 0.05, and the correlation coefficient between the predicted infection probability and the actual label reached 0.82, indicating that the model converged well. The model weights at this time were saved and applied as the first network.

[0201] S302: Application and output of the first network.

[0202] The risk field intensity within a preset period is input into the trained first network to achieve real-time risk conduction analysis.

[0203] First, obtain the risk field intensity sequence {RFI} of each risk domain in the current period (e.g., the past 15 days) from the enterprise information system. k(tn), ..., RFI k (t)}, calculate the rate of change Standardize and encode the features according to the preprocessing method of S301 to generate the node feature matrix X to be input test =[RFI 流动性 , ΔRFI 流动性 / Δt,[1,0,0,0],...,RFI 市场 , ΔRFI 市场 / Δt,[0,0,0,1]].

[0204] X test Input the trained first network and calculate the cross-risk domain infection probability matrix P through forward propagation, where P ij It represents the predicted probability of risk transmission from domain i to domain j (value range [0,1]).

[0205] The output results are threshold filtered (e.g., edges with probability < 0.3 are filtered) to retain significant conduction paths.

[0206] In some implementations, a dynamic threshold may be set: when the overall risk field strength mean is greater than 0.8, the threshold is lowered to 0.2 to capture weak conduction signals in high-risk scenarios.

[0207] It should be pointed out that when applying the model, it is necessary to ensure the timeliness of the input data (such as daily updates) to avoid prediction bias caused by the use of lagged data.

[0208] Based on the above technical solution, by constructing a graph instance containing dynamic features and category information, designing a multi-dimensional labeling system and adopting graph convolutional network modeling, the first network can effectively capture the nonlinear transmission relationship between risk domains and output the infection probability distribution across risk domains, providing key parameter support for the subsequent risk coupling matrix construction and system risk assessment.

[0209] In a possible implementation, the specific implementation steps of step S4 may include:

[0210] S401. Obtain the probability of cross-risk domain transmission.

[0211] After analyzing the risk field intensity within the preset period, the first network outputs an N×N infection probability matrix P, where N is the number of risk domains (e.g., when there are 4 risk domains, N=4), and P ij It represents the probability of risk transmission from domain i to domain j, and its value range is [0,1].

[0212] It should be pointed out that the infection probability matrix must satisfy non-negativity and the diagonal elements must be 0, that is, the risk domain itself is not transmitted.

[0213] For example, if the first network outputs P流动性->偿债 =0.6, then the adjacency matrix corresponds to position A 1,2 =0.6.

[0214] S402. Construct an adjacency matrix, i.e., a risk coupling matrix.

[0215] Matrix definition: That is, the diagonal elements are 0, and the off-diagonal elements are the infection probability. The matrix dimension is consistent with the risk vector dimension.

[0216] In some implementations, the infection probability may be normalized to ensure that the sum of elements in each row is 1, which complies with the properties of the probability transfer matrix.

[0217] It should be pointed out that the adjacency matrix needs to be updated in real time and synchronized with the risk field intensity of the current cycle.

[0218] For example, the adjacency matrix of the four risk domains is as follows: The normalized result is:

[0219] In a possible implementation of the embodiment of the present application, the above S5 can be specifically implemented by the following steps:

[0220] S501. Construct a risk system equation and solve the eigenvalue.

[0221] The overall stability of the risk system can be evaluated by the linear transformation equation: RCM·V risk =μ·V risk Then, the QR decomposition method (a standard method for numerical calculation) is used to solve all eigenvalues ​​of the matrix RCM: {μ1, μ2, ..., μ N};

[0222] Extract the eigenvalue μ with the largest absolute value max =max(|μ i |), which is used to reflect the overall risk intensity of the risk system.

[0223] In some implementations, an efficient computing library (such as Python's NumPy) can be used to implement QR decomposition to improve computing efficiency.

[0224] It should be pointed out that the eigenvalue calculation needs to be performed in the real number domain. If there is a complex eigenvalue, its modulus is taken as the effective eigenvalue.

[0225] S502: Calculate the derivative of the maximum eigenvalue.

[0226] Derivative formula: Where Δt is the monitoring time window (default is 1 day). A positive derivative indicates a decrease in the stability of the risk system, while a negative derivative indicates a reduction in risk.

[0227] In some implementations, a sliding average method may be used to smooth the derivative curve to avoid interference from single-period fluctuations (eg, calculating a 3-day average derivative).

[0228] It should be noted that derivative calculation requires continuous monitoring of data for at least two cycles.

[0229] S503. Trigger a graded warning based on the characteristic value and the risk field intensity.

[0230] The warning rules are shown in Table 1:

[0231] Table 1, early warning rules table

[0232]

[0233] Among them, T2>T1, M3>M2>M1; the default values ​​are: T1=0.8, T2=1.2, T3=1.8, M1=0.3, M2=0.5.

[0234] In some implementations, customized warning rules may be allowed, such as max Absolute value trigger.

[0235] It should be pointed out that the warning level needs to be set in combination with the company's risk tolerance. For example, financial companies can set stricter thresholds.

[0236] Based on the above technical solution, step S4 constructs a risk coupling matrix, transforming the cross-domain contagion probability into a computable mathematical structure. Step S5, through eigenvalue analysis and dynamic monitoring, achieves a quantitative mapping from micro-risk domains to macro-systemic risks. This solution combines linear algebra theory with dynamic threshold rules, enabling the early warning mechanism to capture both risk intensity and the speed of risk evolution, enhancing the scientific nature and dynamic adaptability of early warnings.

[0237] The above mainly introduces the scheme of the embodiment of the present application from the perspective of device implementation. It can be understood that each device, for example, a financial risk early warning analysis device, contains at least one of the hardware structure and software modules corresponding to the execution of each function in order to realize the above functions. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0238] The embodiment of the present application can divide the functional units of a financial risk warning analysis device according to the above method example. For example, each functional unit can be divided according to each function, or two or more functions can be integrated into one processing unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. It should be noted that the division of units in the embodiment of the present application is schematic and is only a logical functional division. There may be other division methods in actual implementation.

[0239] In the case of an integrated unit, Figure 4 A possible structural diagram of a financial risk warning analysis device (referred to as a financial risk warning analysis device 40) involved in the above embodiment is shown. The financial risk warning analysis device 40 includes a processing unit 401 and a communication unit 402, and may also include a storage unit 403. Figure 4 The structural diagram shown can be used to illustrate the structure of a financial risk early warning analysis device involved in the above embodiment.

[0240] when Figure 4 The structural schematic diagram shown is used to illustrate the structure of a financial risk warning analysis device involved in the above embodiment. The processing unit 401 is used to control and manage the actions of a financial risk warning analysis device, the communication unit 402 is used for a financial risk warning analysis device to communicate with other devices, and the storage unit 403 is used to store the program code and data of a financial risk warning analysis device.

[0241] For example, the communication unit 402 is used to receive and transmit the enterprise's financial data, operating data, external environment data, and the risk field intensity of each risk domain within a preset period;

[0242] The processing unit 402 is configured to execute:

[0243] Classify several indicator data into multiple risk domains, and calculate the risk field intensity of each risk domain based on the indicator data in the risk domain to form a risk vector;

[0244] Input the risk field intensity of each risk domain within a preset period into the first network based on the graph neural network, and output the probability of infection across risk domains;

[0245] Construct an adjacency matrix based on the probability of contagion across risk domains;

[0246] Calculate the maximum eigenvalue of the adjacency matrix and the risk vector, and trigger graded warnings based on the maximum eigenvalue and its derivative.

[0247] In a possible implementation, the processing unit 402 is further configured to train the graph neural network, optimize model parameters based on historical risk event data, and obtain a first network.

[0248] In a possible implementation, the communication unit 402 is further configured to send warning information to the enterprise manager terminal, and the processing unit 401 is further configured to dynamically adjust the warning rules according to the threshold parameters input by the user.

[0249] Among them, the processing unit 401 can be a processor or a controller, and the communication unit 402 can be a communication interface, a transceiver, a transceiver, a transceiver circuit, a transceiver device, etc. Among them, the communication interface is a general term and can include one or more interfaces. The storage unit 403 can be a memory. When a financial risk warning analysis device 40 is a chip, the processing unit 401 can be a processor or a controller, and the communication unit 402 can be an input interface and / or output interface, a pin or a circuit, etc. The storage unit 403 can be a storage unit within the chip (for example, a register, a cache, etc.), or it can be a storage unit located outside the chip (for example, a read-only memory (ROM), a random access memory (RAM), etc.).

[0250] Among them, the communication unit can also be called a transceiver unit. The antenna and control circuit with transceiver functions in a financial risk warning analysis device 40 can be regarded as a communication unit 402 of a financial risk warning analysis device 40, and the processor with processing function can be regarded as a processing unit 401 of a financial risk warning analysis device 40. Optionally, the device used to implement the receiving function in the communication unit 402 can be regarded as a communication unit, and the communication unit is used to perform the receiving steps in the embodiment of the present application. The communication unit can be a receiver, a receiver, a receiving circuit, etc. The device used to implement the sending function in the communication unit 402 can be regarded as a sending unit, and the sending unit is used to perform the sending steps in the embodiment of the present application. The sending unit can be a transmitter, a transmitter, a sending circuit, etc.

[0251] Figure 4If the integrated units are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application, 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, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The storage medium for storing computer software products includes various media that can store program codes, such as USB flash drives, mobile hard drives, read-only memories, random access memories, magnetic disks or optical disks.

[0252] Figure 4 A unit in a can also be called a module, for example, a processing unit can be called a processing module.

[0253] The present application also provides a hardware structure diagram of a financial risk early warning analysis method, system, and storage medium device (referred to as a financial risk early warning analysis device 50), see Figure 5 The financial risk early warning analysis device 50 includes a processor 501 and, optionally, a memory 502 connected to the processor 501 .

[0254] In the first possible implementation, see Figure 5 A financial risk early warning analysis device 50 also includes a transceiver 503. The processor 501, the memory 502, and the transceiver 503 are connected via a bus. The transceiver 503 is used to communicate with other devices or a communication network. Optionally, the transceiver 503 may include a transmitter and a receiver. The device used to implement the receiving function in the transceiver 503 can be regarded as a receiver, and the receiver is used to perform the receiving step in the embodiment of the present application. The device used to implement the sending function in the transceiver 503 can be regarded as a transmitter, and the transmitter is used to perform the sending step in the embodiment of the present application.

[0255] Based on the first possible implementation, Figure 5 The structural diagram shown can be used to illustrate the structure of a financial risk early warning analysis device involved in the above embodiment.

[0256] in, Figure 5 It can also illustrate a system chip in a financial risk early warning analysis device. In this case, the actions performed by the above-mentioned financial risk early warning analysis device can be implemented by the system chip. The specific actions performed can be found above and will not be repeated here.

[0257] During implementation, each step of the method provided in this embodiment can be completed by hardware integrated logic circuits in a processor or by software instructions. The steps of the method disclosed in the embodiments of this application can be directly implemented as execution by a hardware processor, or as a combination of hardware and software modules in a processor.

[0258] The processor in this application may include, but is not limited to, at least one of the following: a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a microcontroller unit (MCU), or an artificial intelligence processor, and other types of computing devices that run software. Each computing device may include one or more cores for executing software instructions to perform operations or processing. The processor may be a separate semiconductor chip or integrated into a semiconductor chip with other circuits. For example, it may form an SoC (system on chip) with other circuits (such as a codec circuit, a hardware acceleration circuit, or various bus and interface circuits), or it may be integrated into the ASIC as a built-in processor of the ASIC. The ASIC with the integrated processor may be packaged separately or with other circuits. In addition to the core for executing software instructions to perform operations or processing, the processor may further include necessary hardware accelerators, such as a field programmable gate array (FPGA), a PLD (programmable logic device), or a logic circuit that implements dedicated logic operations.

[0259] The memory in the embodiments of the present application may include at least one of the following types: read-only memory (ROM) or other types of static storage devices that can store static information and instructions, random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or electrically erasable programmable read-only memory (EEPROM). In some scenarios, the memory may also be a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to this.

[0260] An embodiment of the present application also provides a computer-readable storage medium, comprising instructions, which, when executed on a computer, enables the computer to execute any of the above methods.

[0261] An embodiment of the present application also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute any of the above methods.

[0262] An embodiment of the present application also provides a chip, which includes a processor and an interface circuit, the interface circuit is coupled to the processor, the processor is used to run a computer program or instruction to implement the above method, and the interface circuit is used to communicate with other modules outside the chip.

[0263] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using a software program, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more media integrated therein. The available media may be magnetic media (eg, floppy disks, hard disks, magnetic tapes), optical media (eg, DVDs), or semiconductor media (eg, solid state disks (SSDs)).

[0264] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art may understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0265] Although the present application has been described with reference to specific features and embodiments thereof, it is apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and the drawings are merely illustrative of the present application as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, the present application is intended to encompass such modifications and variations as fall within the scope of the claims of the present application and their equivalents.

[0266] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A financial risk early warning analysis method, characterized in that: include: Obtain the company's financial data, operating data, and external environment data to obtain several indicator data; Classify several indicator data into multiple risk domains, and calculate the risk field intensity of each risk domain based on the indicator data in the risk domain to form a risk vector; Obtain the risk field strength of each risk domain within a preset period of the enterprise, input the risk field strength of each risk domain within the preset period into a first network based on a graph neural network, and output the probability of contagion across risk domains; Construct an adjacency matrix based on the probability of contagion across risk domains; Calculate the maximum eigenvalue of the adjacency matrix and the risk vector, and trigger graded warnings based on the maximum eigenvalue and its derivative.

2. A financial risk early warning analysis method according to claim 1, characterized in that: The financial data includes at least one of the following: cash ratio, quick ratio, net cash flow from operating activities, debt-to-asset ratio, interest coverage ratio, debt ratio, gross profit margin, and return on equity; The operating data includes at least one of the following: supplier delivery delay rate, inventory turnover days, production line utilization rate, product qualification rate, order cancellation rate, and supplier concentration; The external environment data includes at least one of the following: GDP growth rate, benchmark interest rate, industry average debt ratio, and raw material price volatility.

3. A financial risk early warning analysis method according to claim 2, characterized in that: The aforementioned classification of several indicator data into multiple risk domains includes: Liquidity risk domain, including cash ratio, quick ratio, net cash flow from operating activities, and inventory turnover days; Debt repayment risk domain, including asset-liability ratio, interest coverage ratio, debt ratio, and industry average debt ratio; Supply chain risk domain, including supplier delivery delay rate, production line utilization rate, product qualification rate, order cancellation rate, and supplier concentration; Market risk domain, including gross profit margin, return on net assets, GDP growth rate, benchmark interest rate, and raw material price volatility.

4. A financial risk early warning analysis method according to claim 1, characterized in that: The calculation of the risk field intensity of each risk domain based on the indicator data within the risk domain includes: Calculate the mutation degree ΔI of indicator i in risk domain k k，i (t), the formula is: Where Δt represents the time difference, I i (t) represents the value of index i on day t, σ i (T) represents the rolling standard deviation of the value of indicator i in period T; Calculate the initial weight of indicator i in risk domain k using entropy weight method Based on the initial weight superimposed on the time attenuation factor, the weight value w of indicator i in risk domain k is obtained k，i , the formula is: Where λ represents the time attenuation coefficient, and days represents the time difference between the time when the most recent mutation degree of indicator i in risk domain k was greater than the preset mutation threshold and the current time; Determine the domain toxicity coefficient α of risk domain k based on financial data k , exogenous shock coefficient γ k , and combined with the mutation degree and weight value of several indicators in the risk area, a risk field intensity calculation function is constructed, which is used to calculate the risk field intensity of each risk domain.

5. A financial risk early warning analysis method according to claim 4, characterized in that: The risk field intensity calculation function satisfies: Among them, RFI k (t) represents the risk field intensity calculation function, β represents the risk transmission sensitivity, which is calculated based on the product of the number of supply chain nodes and the supplier concentration, and Ψ(t) represents the comprehensive impact intensity. is the Sigmoid function.

6. A financial risk early warning analysis method according to claim 5, characterized in that: The comprehensive impact strength is determined as follows: Extract the value of the external impact factor E at time t from the financial data j (t) The external impact factors include GDP growth rate, benchmark interest rate, industry average debt ratio, and raw material price volatility; Calculate the normalized offset Ψ of the external shock factor j j (t), the formula is: Among them, μ j represents the historical mean of the external shock factor j, σ j represents the historical standard deviation of external shock factor j; The shock coupling weight of the external shock factor j is calculated according to the formula: Among them, REF represents the average value of the risk field intensity of each risk domain in the historical period, corr(E j (t), RFI(t)) represents the correlation coefficient between the external impact factor j on day t and the average value, and τ represents a preset influencing parameter, which can be determined by experimental calibration; The comprehensive shock coupling strength is calculated based on the normalized offset and shock coupling weight, and the formula is: Where N represents the number of external shock factors.

7. A financial risk early warning analysis method according to claim 6, characterized in that: The domain toxicity coefficient α of the risk domain k is determined based on financial data k , exogenous shock coefficient γ k ,include: Determine the core indicators of risk domain k, where the core indicators include positive indicators and negative indicators, and each risk domain includes one core indicator; Obtain the data of the core indicators of risk domain k in the historical period, and based on the current value I of the core indicators k (t) Calculate the historical percentile P k (t); If the core indicator of risk domain k is a positive indicator, then the domain toxicity coefficient α k The expression is: k =1.0+0.5×(1-P k (t)); If the core indicator of risk domain k is a negative indicator, then the domain toxicity coefficient α k The expression is: k =1.0+0.5×P k (t); Calculate the absolute value of the correlation coefficient ρ between each external impact factor and the wind direction field intensity of the risk domain k in the historical period k，j , the formula is: k,j =|corrr(E j (t),RFI k )|;where corr() represents the Pearson correlation coefficient function; The maximum absolute value of the correlation coefficient is taken as the sensitivity benchmark S of the risk domain k. k , the formula is: Calculate the exogenous shock coefficient γ of risk domain k based on the sensitivity benchmark k , the formula is: γ k =0.1+0.8×S k .

8. A financial risk early warning analysis method according to claim 3, characterized in that: The method for obtaining the infection probability across risk domains includes: Construct an independent graph instance G = (V, E) for each historical risk event; where V represents the node of the independent graph, which is used to represent the risk domain category, and E represents the edge of the independent graph, which is used to represent the risk transmission path from risk domain i to risk domain j; A graph convolutional network is constructed based on each independent graph instance, and the node feature vector v of the graph convolutional network is k v k =[RFI k (t),ΔRFI k / Δt,domain_type], the edge weight w of the graph convolutional network ij is the normalized value of the number of risk transmission events from risk domain i to risk domain j in historical risk events; where ΔRFI k / Δt represents the wind intensity change rate of risk domain k within the time difference Δt, and domain_type represents the one-hot encoding of the risk domain category; Train and optimize the graph convolutional network to obtain the first network; Determining an input node feature vector of the first network based on the risk field intensity of each risk domain within a preset period of the enterprise; The input node feature vector is fed into the first network, and the contagion probability across risk domains is output.

9. A financial risk early warning analysis system, characterized in that: include: Data collection module, risk analysis module and early warning trigger module; among them, The data acquisition module is used to obtain the financial data, operating data and external environment data of the enterprise to obtain a number of indicator data; The risk analysis module is used to classify a number of indicator data into a plurality of risk domains, calculate the risk field intensity of each risk domain based on the indicator data in the risk domain, and form a risk vector, and Obtain the risk field intensity of each risk domain within the preset period of the enterprise, pre-process the risk field intensity of each risk domain within the preset period, and input it into the first network based on the graph neural network to output the probability of contagion across risk domains; The warning trigger module is used to construct an adjacency matrix based on the infection probability across risk domains, calculate the maximum eigenvalue of the adjacency matrix and the risk vector, and trigger a graded warning based on the maximum eigenvalue and the derivative of the maximum eigenvalue.

10. A storage medium, characterized in that: The storage medium is a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the computer program is used to implement the financial risk early warning analysis method according to any one of claims 1 to 8, including: Obtain the company's financial data, operating data, and external environment data to obtain several indicator data; Classify several indicator data into multiple risk domains, and calculate the risk field intensity of each risk domain based on the indicator data in the risk domain to form a risk vector; Obtain the risk field strength of each risk domain within a preset period of the enterprise, input the risk field strength of each risk domain within the preset period into a first network based on a graph neural network, and output the probability of contagion across risk domains; Construct an adjacency matrix based on the probability of contagion across risk domains; Calculate the maximum eigenvalue of the adjacency matrix and the risk vector, and trigger graded warnings based on the maximum eigenvalue and its derivative.

Images