Large model driven time series capital chain intelligent risk transmission control system and method

Abstract

The present invention discloses a large-scale model-driven intelligent risk transmission control system and method for a time-series capital chain. The system includes a data integration module, a time-series relationship extraction module, a risk transmission probability calculation module, a graph structure dynamic update module, a risk disposal strategy generation module, and a control instruction execution module, each of which is connected in sequence and works in coordination. The data integration module collects original transaction data and related data and preliminarily integrates them; the time-series relationship extraction module uses a large model to perform time-series reasoning on the data and analyze the complex relationships in the capital chain; the risk transmission probability calculation module calculates the risk transmission probability of capital flow; the graph structure dynamic update module dynamically updates the graph structure based on the calculation results; the risk disposal strategy generation module generates a risk disposal strategy based on the updated graph structure; and the control instruction execution module executes the corresponding control instructions. Each module works in coordination to achieve accurate identification, dynamic assessment, and effective disposal of capital flow risks.

Description

Technical Field

[0001] The present invention relates to financial risk control technologies, and in particular to a large-scale model-driven time-series capital chain intelligent risk conduction control system and method. Background Art

[0002] With the increasing complexity of financial operations and the continuous expansion of transaction volumes, the risk transmission problem of capital chains has become increasingly prominent. Traditional risk analysis methods have numerous limitations when dealing with large-scale, complex capital flows and multi-dimensional risk factors. These include inefficient reliance on manual data processing, difficulty effectively incorporating unstructured data, delayed response to time-sensitive scenarios, and the lack of flexibility based on static models. Therefore, there is an urgent need for a system and method that can leverage large-scale model technology to achieve intelligent risk transmission control of time-series capital chains, in order to meet the modern financial market's requirements for efficient, accurate, and real-time risk prevention and control. Summary of the Invention

[0003] In view of this, the present invention addresses the deficiencies in the existing technology, and its main purpose is to provide a large-model-driven time-series capital chain intelligent risk transmission control system and method, which realizes the accurate identification, dynamic assessment and effective disposal of capital flow risks by making various modules work together.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A large-model-driven intelligent risk transmission control system for a time-series capital chain comprises a data integration module, a time-series relationship extraction module, a risk transmission probability calculation module, a graph structure dynamic update module, a risk disposal strategy generation module and a control instruction execution module, wherein the modules are sequentially connected and work in coordination; the data integration module is used to collect original transaction data and related data, and perform preliminary integration; the time-series relationship extraction module calls a large model to perform time-series reasoning on the data output by the data integration module, and parse the complex relationship of the capital chain; the risk transmission probability calculation module calculates the risk transmission probability of the capital flow based on the output result of the time-series relationship extraction module; the graph structure dynamic update module dynamically updates the graph structure according to the result of the risk transmission probability calculation module; the risk disposal strategy generation module generates a risk disposal strategy based on the graph structure updated by the graph structure dynamic update module; the control instruction execution module executes corresponding control instructions according to the strategy generated by the risk disposal strategy generation module.

[0006] As a preferred solution, the temporal relationship extraction module uses a specific temporal reasoning algorithm when calling the large model to analyze the complex relationship of the capital chain. The expression is:

[0007] ;

[0008] Among them, X t Indicates the state of the capital link at the current time t; X t−τ represents the state of the capital chain at the historical moment t−τ, where τ is the time step; α τ is the weight coefficient corresponding to the time step τ; β is the noise coefficient; ϵ t is the random noise at the current time t.

[0009] As a preferred solution, the risk transmission probability calculation module calculates the risk transmission probability according to the following formula:

[0010] ;

[0011] Among them, P(r i →r j ) indicates that funds are transferred from risk point ri to risk point r j The probability of S(r i ) represents the risk point r i The risk intensity of C(r i ,r j ) represents the risk point r i and r j The degree of correlation between i ,r j ) represents the time of risk transmission; λ is the time attenuation coefficient; D(r i ) represents the risk point r j δ is a smoothing factor used to prevent the denominator from being zero.

[0012] As a preferred solution, the risk handling strategy generation module combines the risk level assessment results and preset risk handling rules when generating the risk handling strategy.

[0013] As a preferred solution: the data integration module supports access to multiple data sources, including but not limited to bank transaction systems, stock exchanges, and corporate financial systems, and can perform standardized processing on data in different formats.

[0014] As a preferred solution, the graph structure dynamic update module adopts an incremental update method when updating the graph structure, and only updates the changed parts to improve the real-time performance and efficiency of the system.

[0015] As a preferred solution: when executing control instructions, the control instruction execution module can take different levels of control measures according to different risk levels, including but not limited to early warning prompts, transaction restrictions, and account freezing.

[0016] A method for applying the large model-driven time-series capital chain intelligent risk conduction control system comprises the following steps:

[0017] S1: The data integration module collects original transaction data and related data and performs preliminary integration; S2: The time series relationship extraction module calls the big model to perform time series reasoning on the integrated data and analyze

[0018] Complex relationships in the funding chain;

[0019] S3: The risk transmission probability calculation module calculates the risk transmission probability of capital flow based on the time series relationship extraction results;

[0020] S4: The graph structure dynamic update module dynamically updates the graph structure based on the risk transmission probability calculation results;

[0021] S5: The risk disposal strategy generation module generates a risk disposal strategy based on the updated graph structure; S6: The control instruction execution module executes the corresponding control instruction according to the generated risk disposal strategy.

[0022] As a preferred solution: in step S2, when calling the large model for time series reasoning, a hybrid model architecture combining LSTM network and Transformer structure is specifically adopted to improve the processing capability and accuracy of time series capital link data.

[0023] As a preferred solution: in step S5, when generating the risk disposal strategy, similar cases and corresponding disposal effects in the historical risk case library are combined, and the big model is used to optimize and recommend the strategy to ensure the scientificity and effectiveness of the disposal strategy.

[0024] Compared with the existing technology, the present invention has obvious advantages and beneficial effects. Specifically, it can be seen from the above technical solution that through the close coordination of various modules and the use of advanced algorithms, significant advantages in many aspects are achieved in the field of financial risk prevention and control. The system can not only accurately capture the complex risk transmission paths in the capital chain and improve the accuracy of risk identification, but also ensure the real-time nature of risk assessment through dynamic graph structure updates, and effectively respond to rapid market changes. Personalized disposal strategies based on risk levels balance risk control and business development and avoid excessive intervention. At the same time, the system has good scalability and adaptability, can be flexibly optimized as financial business expands, and is widely applicable to various financial scenarios. It provides financial institutions with powerful and practical risk management tools, greatly enhancing the ability and efficiency of financial risk prevention and control.

[0025] To more clearly illustrate the structural features and effects of the present invention, it is described in detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of the system architecture of the present invention;

[0027] Figure 2 Schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0028] The present invention Figure 1 and Figure 2 As shown, a large model-driven time-series capital chain intelligent risk transmission control system and method includes a data integration module, a time-series relationship extraction module, a risk transmission probability calculation module, a graph structure dynamic update module, a risk disposal strategy generation module and a control instruction execution module, wherein:

[0029] The data integration module is responsible for collecting raw transaction data and related data, including bank transaction records, securities market data, and corporate financial information, and also accesses supporting data on market conditions and policies and regulations. It performs preliminary cleaning, format conversion, and integration operations on the collected data to meet subsequent processing requirements, laying a solid data foundation for large-scale model analysis.

[0030] The data integration module is the data entry point of the entire system and interacts with other modules through data flows. After the data it collects is initially integrated, it is passed to the temporal relationship extraction module for further processing.

[0031] The data integration module is responsible for collecting raw transaction data and related data from various data sources, including but not limited to banking trading systems, stock exchanges, and corporate financial systems. It also integrates market data and supporting information such as policy and regulatory data. During the data collection process, the module performs preliminary data cleansing and format conversion to remove irrelevant noise and erroneous information, fill in missing values, and ensure data integrity and consistency. For example, it standardizes transaction time formats across different data sources to the standard YYYY-MM-DD HH:MM:SS format and converts currency units to a unified RMB unit. The integrated data is organized into a structured dataset, containing key fields such as transaction time, transaction amount, transaction parties, and transaction type, to meet the processing requirements of subsequent modules.

[0032] The temporal relationship extraction module uses a large model trained on a large amount of historical capital flow data and risk events to perform temporal reasoning analysis on the integrated data. This large model has a strong ability to capture temporal relationships, analyzing the temporal dependencies and complex interaction patterns of each node in the capital chain, identifying potential risk transmission paths and key nodes, and providing a basis for accurately calculating risk transmission probabilities.

[0033] The temporal relationship extraction module is a key step in the data processing process. It receives integrated data from the data integration module and passes the processed results to the risk transmission probability calculation module. It is closely connected with the data integration module to ensure seamless data flow.

[0034] The temporal relationship extraction module utilizes a large model based on a combination of an LSTM network and a Transformer architecture to perform temporal reasoning on the integrated data and analyze the complex relationships within the capital chain. This large model, trained on a large amount of historical capital flow data and risk events, possesses powerful capabilities for capturing temporal relationships. It can deeply mine the temporal dependencies and complex interaction patterns within the data, extracting key characteristics of capital flows and potential pathways of risk transmission. For example, by analyzing transaction time series data, it can identify patterns in capital flows between different accounts and potential hidden indicators of risk transmission. Specifically, it takes transaction time series data as input and uses an LSTM network to model the temporal dependencies within the sequence. It also employs a Transformer architecture to capture long-term dependencies between different time steps within the sequence. This approach extracts key feature vectors of capital flows. These vectors encompass the temporal, transaction, and risk characteristics of capital flows, providing a foundation for subsequent risk transmission probability calculations.

[0035] The formula used for temporal relationship extraction is as follows:

[0036] ;

[0037] Among them, X t Indicates the state of the capital chain at the current time t. In practical applications, X t Is a vector containing multiple feature variables, such as transaction amount, transaction frequency, and transaction counterparty. t−τ represents the state of the capital chain at the historical moment t−τ, where τ is the time step, indicating the historical time interval considered. For example, if the capital flow is analyzed hourly, τ can be 1 hour or 2 hours. τ is the weight coefficient corresponding to the time step τ, which indicates the degree of influence of different historical moments on the current state. These weight coefficients can be learned through model training. β is the noise coefficient, which is used to control the random noise ϵ t The degree of impact on the current state. t is the random noise at the current time t, simulating the unpredictable part of capital flow.

[0038] The risk transmission probability calculation module, after obtaining the output of the temporal relationship extraction module, uses a specific algorithm model, combining multiple factors such as capital flow, transaction amount, transaction frequency, and node association attributes, to calculate the probability of risk transmission between different nodes. This quantitative analysis clearly presents the risk distribution and transmission trends along the capital chain, providing a precise assessment for the formulation of risk management strategies.

[0039] The risk transmission probability calculation module is one of the system's core analysis modules. Based on the feature vectors output by the time series relationship extraction module, it calculates the risk transmission probability of capital flows and passes the results to the graph structure dynamic update module. It is closely connected to the time series relationship extraction module to ensure the continuity of analysis.

[0040] The risk transmission probability calculation module uses a specific algorithmic model, combining multiple factors such as capital flow direction, transaction amount, transaction frequency, and inter-node correlations, to accurately assess the likelihood of risk transmission when funds flow between different nodes. Specifically, it first constructs a capital chain graph based on capital flow direction, where nodes represent capital entities or transaction links, and edges represent the capital flow relationships between them. It then uses the feature vectors output by the temporal relationship extraction module, combined with transaction amount and frequency information, to calculate the risk transmission probability for each edge. This calculation considers the directionality of capital flow, meaning that the risk transmission probability of funds flowing from one node to another may vary depending on the direction of flow. Furthermore, it adjusts the risk transmission probability based on inter-node correlations, such as the business relationship and credit ratings of the two parties involved in the transaction. For example, if two nodes have a long-term, stable cooperative relationship and both have high credit ratings, the risk transmission probability between them is likely to be relatively low. This approach yields a comprehensive and accurate capital chain risk transmission probability matrix, which clearly illustrates the risk distribution and transmission trends along the capital chain, providing precise risk assessment results for the subsequent formulation of risk management strategies.

[0041] The formula for calculating the risk transmission probability is as follows:

[0042] ;

[0043] In this formula, P(r i →r j ) indicates that the funds are transferred from the risk point r i Transmitted to risk point r j The probability of S(r i ) represents the risk point r i The risk intensity can be determined by analyzing historical data and expert evaluation. For example, if r iis a high-risk counterparty and its risk intensity may be assessed as high. i ,r j ) represents the risk point r i and r j The degree of correlation between two risk points can be determined by analyzing the transaction relationship and business connection between the two. For example, if there is frequent capital transactions and close business cooperation between two risk points, their correlation will be high. i ,r j ) represents the time of risk transmission, that is, the risk from r i Conducted to r j The time required. This time can be calculated based on historical data statistics or estimated based on business logic. λ is the time decay coefficient, which indicates the degree to which the risk decreases over time. This coefficient can be determined based on business experience and data fitting. For example, if the risk decreases rapidly over time, the value of λ will be larger. D(r j ) represents the risk point r j The risk resistance of a company can be determined by analyzing its financial status and credit rating factors. For example, a large financial institution’s risk resistance is usually stronger than that of a small enterprise, so its D(r j ) will be higher. Finally, δ is a smoothing factor that prevents the denominator from being zero, ensuring the stability and computability of the formula.

[0044] The graph structure dynamic update module dynamically updates the capital chain graph structure in real time based on the results of the risk transmission probability calculation module. In the graph structure, nodes represent capital entities or transaction links, edges represent capital flow relationships, and edge weights represent risk transmission probabilities. After calculating the new probabilities, the module promptly adjusts edge weights and adds or removes nodes and edges to reflect the latest risk status of the capital chain, providing real-time decision-making support for the risk management strategy generation module.

[0045] The graph structure dynamic update module receives the risk transmission probability matrix from the risk transmission probability calculation module and provides the updated graph structure to the risk disposal strategy generation module. It is closely connected with the risk transmission probability calculation module to ensure that the graph structure can reflect the latest risk status in real time.

[0046] The graph structure dynamic update module dynamically updates the capital chain graph structure in real time based on the results obtained by the risk transmission probability calculation module. It uses an incremental update method, updating only the changed parts to improve the system's real-time performance and efficiency. Specifically, when a new risk transmission probability is calculated, it updates the edge weights in the graph structure. Edge weights represent the risk transmission probability of the corresponding capital flow. An increase in edge weight indicates an increase in the risk of capital flow between the two nodes; a decrease indicates a decrease in risk. Furthermore, based on changes in risk transmission probability, it determines whether to add or delete nodes and edges. For example, if a new capital entity begins trading and the frequency and amount of transactions between it and existing capital entities gradually increase, a node will be added to represent the new capital entity and corresponding edges will be added to represent the capital flow relationships between it and other capital entities. Conversely, if a capital entity gradually withdraws from trading and its transaction relationships with existing capital entities gradually disappear, the corresponding nodes and edges will be deleted. This ensures that the graph structure accurately and in real time reflects the latest risk status of the capital chain, providing real-time, dynamic decision-making basis for the risk management strategy generation module.

[0047] The risk management strategy generation module uses the updated graph structure, combined with pre-set rules and a strategy library, to analyze different risk scenarios using a large model and generate personalized risk management strategies. These strategies include enhanced monitoring of high-risk accounts, adjustments to fund transaction limits, issuance of risk warnings, and freezing of abnormal transactions, aiming to curb the spread of risk and mitigate the impact on the capital chain.

[0048] The risk management strategy generation module is the core module of the system's decision-making. Based on the latest graph structure provided by the graph structure dynamic update module, it generates risk management strategies and passes these strategies to the control instruction execution module. It is closely connected with the graph structure dynamic update module to ensure timely and accurate decision-making.

[0049] The risk management strategy generation module uses the updated graph structure, pre-set rules, and a policy library to analyze different risk scenarios using a large model and generate personalized risk management strategies. It first analyzes the graph structure to identify high-risk nodes and edges. Then, it combines the risk level assessment results with pre-set risk management rules to determine appropriate management measures. Specifically, it applies different levels of control measures based on the risk level. For example, low-risk nodes and edges might be subject to early warning alerts to alert relevant personnel; medium-risk nodes and edges might be subject to transaction restrictions, such as limits on transaction amount and frequency; and high-risk nodes and edges might be subject to account freezes to prevent further risk transmission and spread. During the strategy generation process, the large model analyzes and infers different risk scenarios. It then optimizes and recommends management strategies based on similar cases and corresponding management outcomes from a historical risk case library. For example, if there is a case in historical data similar to the current risk scenario, the module analyzes the management outcomes of that case and adjusts and optimizes the current management strategy based on the analysis results to ensure its scientific and effective nature. The resulting disposal strategy will detail the specific control measures for each high-risk node and edge, as well as the corresponding execution order and priority, providing a clear action guide for the control instruction execution module.

[0050] The control instruction execution module receives the risk management strategy generation module's strategy, converts it into specific control instructions, and sends them to financial service systems or devices for execution. For example, it can issue an account freeze instruction to a bank system or a trading restriction instruction to a securities trading system. This ensures that risk management strategies are implemented quickly and accurately, effectively controlling capital chain risks.

[0051] The control instruction execution module is the system's execution unit. It receives the risk management strategy from the risk management strategy generation module, converts it into specific control instructions, and sends them to the corresponding financial business system or device for execution. It is closely connected with the risk management strategy generation module to ensure that the management strategy can be implemented quickly and accurately.

[0052] The control instruction execution module converts risk management strategies into specific control instructions and sends them to the corresponding financial business systems or devices for execution. For example, if the management strategy requires freezing an account, it will send an account freeze instruction to the banking system; if the management strategy requires restricting a transaction, it will send a transaction restriction instruction to the securities trading system. During execution, it sorts and schedules multiple control instructions based on the execution order and priority specified in the management strategy, ensuring that high-priority instructions are executed first. It also provides feedback on the execution results to the risk management strategy generation module to adjust and optimize subsequent management strategies. For example, if a control instruction is successfully executed, it will be fed back to the risk management strategy generation module. The module may use this information to adjust subsequent management strategies, such as reducing the focus on that node. If a control instruction fails, it will be fed back to the risk management strategy generation module. The module may use this information to reassess the risk situation, regenerate a new management strategy, and send the control instruction again for execution. In this way, it ensures that risk management strategies are implemented quickly and accurately, effectively controlling capital chain risks.

[0053] The collaborative workflow of each module is as follows:

[0054] The data integration module collects original transaction data and related data from multiple data sources and performs preliminary integration, including cleaning and format conversion operations, to ensure data integrity and consistency.

[0055] The integrated data is then passed to the temporal relationship extraction module. This module uses a large model based on a combination of an LSTM network and a Transformer architecture to perform temporal reasoning on the data, analyze the complex relationships within the capital chain, and extract key characteristics of capital flows and potential paths for risk transmission.

[0056] The extracted feature vectors are passed to the risk transmission probability calculation module. This module uses a specific algorithm model, combining multiple factors such as capital flow direction, transaction amount, transaction frequency, and the correlation properties between nodes, to calculate the risk transmission probability of funds flowing between different nodes, thereby obtaining a comprehensive and accurate capital chain risk transmission probability matrix.

[0057] The risk transmission probability matrix is ​​passed to the graph structure dynamic update module. This module uses an incremental update method to dynamically update the capital chain graph structure in real time based on the probability matrix. It only updates the parts that have changed, including updating edge weights and adding or deleting nodes and edges. This ensures that the graph structure can accurately and real-timely reflect the latest risk status of the capital chain.

[0058] The updated graph structure is passed to the risk management strategy generation module. This module combines pre-set rules with a policy library, applies a large model to analyze different risk scenarios, and generates personalized risk management strategies. It applies different levels of control measures based on the risk level and optimizes and recommends strategies based on similar cases and corresponding management outcomes from a historical risk case library.

[0059] The resulting disposal strategy is passed to the control instruction execution module. This module converts the disposal strategy into specific control instructions and sends them to the corresponding financial business system or device for execution. It sorts and schedules multiple control instructions based on the execution order and priority of the disposal strategy and provides feedback on the execution results, ensuring that the risk disposal strategy can be implemented quickly and accurately, achieving effective control of capital chain risks.

[0060] The control method applied to the system comprises the following steps:

[0061] S1: The data integration module collects raw transaction data and related data and performs preliminary integration. It gathers data from various financial data sources, including but not limited to banking trading systems, stock exchanges, and corporate financial systems, while also integrating market data and supplementary information such as policy and regulatory data. The collected data is cleaned and formatted to remove irrelevant noise and erroneous information, and missing values ​​are filled in to meet the requirements of subsequent analysis.

[0062] S2: The temporal relationship extraction module uses a large model to perform temporal reasoning on the integrated data, analyzing the complex relationships within the capital chain. This large model, based on a hybrid architecture combining an LSTM network and a Transformer structure, can deeply explore temporal dependencies and complex interaction patterns within the data, extracting key characteristics of capital flows and potential paths of risk transmission.

[0063] S3: The risk transmission probability calculation module calculates the risk transmission probability of capital flows based on the time series relationship extraction results. Specifically, this module uses the above formula, combined with multiple factors such as capital flow direction, transaction amount, transaction frequency, and the correlation properties between nodes, to accurately assess the risk transmission probability of capital flows between different nodes.

[0064] S4: The graph structure dynamic update module dynamically updates the graph structure based on the risk transmission probability calculation results. This incremental update approach only updates the changed parts, improving the system's real-time performance and efficiency. This update method ensures that the graph structure accurately reflects the risk status of the capital chain while reducing computational complexity and ensuring the system can quickly respond to market changes.

[0065] S5: The risk management strategy generation module generates risk management strategies based on the updated graph structure. By integrating similar cases and corresponding management outcomes from the historical risk case library, the module applies a large-scale model to optimize and recommend strategies, ensuring the scientific and effective nature of the management strategies. Management strategies employ varying levels of control measures, including but not limited to early warning alerts, transaction restrictions, and account freezes, depending on the risk level.

[0066] S6: The control instruction execution module executes the corresponding control instructions based on the generated risk management strategy. It converts the risk management strategy into specific control instructions and sends them to the corresponding financial business system or device for execution, ensuring that the risk management strategy can be implemented quickly and accurately, achieving effective control of capital chain risks.

[0067] Taking bank fund transaction monitoring as an example, the system uses the data integration module to collect various internal bank transaction data in real time, such as customer transfer records, loan issuance and repayment records, as well as external data on market interest rate fluctuations and macroeconomic policy changes. The temporal relationship extraction module uses a large model to analyze this data, identifying patterns in fund flows between different customer accounts and business departments, as well as potential hidden indicators of risk transmission. The risk transmission probability calculation module uses this data to calculate the probability of risk transmission between each transaction link and account. For example, if a corporate account is involved in a high-risk industry, the probability of risk transmission when its funds flow out to other accounts is higher. The graph structure dynamic update module promptly updates the bank's fund transaction graph, marking high-risk nodes and edges. Based on the updated graph structure, the risk management strategy generation module formulates a risk management strategy for focusing on monitoring the corporate account and issuing risk warnings to its related accounts. The control instruction execution module then executes the corresponding instructions, such as strengthening the review of transactions in the corporate account and sending risk warnings to related account holders. This effectively prevents the transmission and spread of risks within the bank's capital chain and ensures the safe and stable operation of the bank's financial services.

[0068] In this case, assuming the current time is t, we need to calculate the current capital link status

[0069] X t According to the time series relationship extraction formula, the capital chain status X at multiple past moments (such as t-1, t-2) is collected. t−1 、X t−2 , and the corresponding time step weight coefficients α1 and α2 are determined. At the same time, random noise ϵ is considered t Substituting these data into the formula, we can calculate the current funding link status X t This state vector contains multiple characteristic variables and can fully reflect the comprehensive situation of capital flow at the current moment.

[0070] Next, you need to calculate the funds from a high-risk enterprise account r i Transfer to another linked account j The probability P(r i →r j ). According to the risk transmission probability formula, we first analyze the risk point r i The risk intensity S(r i ), considering that the company is involved in high-risk industries and has recently had unusually frequent transactions, its risk intensity is assessed as high. i and r j The degree of correlation between i ,r j ), it was found that there were frequent capital transactions and business cooperation between the two, so the correlation was high. The risk transmission time T(r i ,r j ), according to historical data, funds from r i Conducted to r j It usually takes 2 working days. Based on business experience, the time attenuation coefficient λ is determined to be 0.3, which means that the risk will be reduced to a certain extent over time. j , analyzed its financial status and credit rating, and determined its risk resilience D(r j ) is a medium level. Finally, a smoothing factor δ is added to ensure the stability of the calculation. Substituting all the above parameters into the formula, the calculation results show that the funds from r i Conducted to r j The probability P(r i →r j ), providing a quantitative basis for the formulation of risk management strategies.

[0071] The design focus of the present invention is:

[0072] First, by making full use of the large model's powerful temporal reasoning capabilities and ability to understand complex relationships, it can deeply explore the potential risk transmission paths and key nodes in the capital chain. Compared with traditional rule-based or simple statistical model methods, it can identify risks more accurately and comprehensively, effectively solving the problem that traditional methods are difficult to deal with complex capital flows and multi-dimensional risk factors, improving the accuracy and reliability of risk identification, and reducing the risk of financial losses caused by insufficient risk identification.

[0073] Second, it realizes the dynamic assessment and real-time update of capital chain risks. Through the dynamic adjustment of the graph structure, it can timely reflect the changes in risks in the capital chain, so that risk disposal strategies can always be formulated based on the latest risk status, ensuring the timeliness and effectiveness of risk control, overcoming the shortcomings of traditional static models in responding to rapid changes in the market environment, and enhancing the risk prevention and control capabilities of the financial system in a complex and changing market environment.

[0074] Third, based on precise risk assessment and intelligent analysis of large models, the generated personalized risk management strategies are more targeted and operational, and can effectively curb the transmission and spread of risks. At the same time, they avoid unnecessary interference and impact of overly conservative or overly aggressive risk management measures on normal business, achieving a balance between risk control and business development, and helping to improve the stability and sustainable development capabilities of financial business.

[0075] Fourth, the architectural design of the entire system and method has good scalability and adaptability, and can be flexibly expanded and optimized with the development of financial business and the expansion of data scale. At the same time, the continuous learning and updating capabilities of the large model also ensure that the system can continuously adapt to new risk characteristics and changing trends. It has strong practicality and foresight, and can be widely used in various financial fields such as banking, securities, and insurance, providing strong technical support and guarantee for the risk management work of financial institutions.

[0076] The above description is merely a preferred embodiment of the present invention and does not limit the technical scope of the present invention. Therefore, any slight modifications, variations and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A large-scale model-driven time-series capital chain intelligent risk transmission control system, characterized by: It includes data integration module, time series relationship extraction module, risk transmission probability calculation module, graph structure dynamic update module, risk disposal strategy generation module and control instruction execution module. Each module is connected in sequence and works together. The data integration module is used to collect original transaction data and related data and perform preliminary integration; The temporal relationship extraction module calls the large model to perform temporal reasoning on the data output by the data integration module to analyze the complex relationship of the capital chain; when calling the large model for temporal reasoning, a hybrid model architecture combining LSTM network and Transformer structure is specifically adopted; The risk transmission probability calculation module calculates the risk transmission probability of capital flow based on the output result of the time series relationship extraction module; The graph structure dynamic update module dynamically updates the graph structure according to the result of the risk transmission probability calculation module; The risk handling strategy generation module generates a risk handling strategy based on the graph structure updated by the graph structure dynamic update module; The control instruction execution module executes corresponding control instructions according to the strategy generated by the risk disposal strategy generation module; The temporal relationship extraction module uses a specific temporal reasoning algorithm when calling the large model to analyze the complex relationship of the capital chain. Its expression is: ； Among them, X t Indicates the state of the capital link at the current time t; X t−τ represents the state of the capital chain at the historical moment t−τ, where τ is the time step; α τ is the weight coefficient corresponding to the time step τ; β is the noise coefficient; ϵ t is the random noise at the current time t; The risk transmission probability calculation module calculates the risk transmission probability according to the following formula: ； Among them, P(r i →r j ) indicates that funds are transferred from risk point ri to risk point r j The probability of S(r i ) represents the risk point r i The risk intensity of C(r i ,r j ) represents the risk point r i and r j The degree of correlation between i ,r j ) represents the time of risk transmission; λ is the time attenuation coefficient; D(r i ) represents the risk point r j δ is a smoothing factor used to prevent the denominator from being zero.

2. The large model-driven time-series capital chain intelligent risk transmission control system according to claim 1 is characterized in that: The risk handling strategy generation module combines the risk level assessment results and the preset risk handling rules when generating the risk handling strategy.

3. The large model-driven time-series capital chain intelligent risk transmission control system according to claim 1 is characterized in that: The data integration module supports access to multiple data sources, including but not limited to bank transaction systems, stock exchanges, and corporate financial systems, and can perform standardized processing on data in different formats.

4. The large model-driven time-series capital chain intelligent risk transmission control system according to claim 1 is characterized in that: When updating the graph structure, the graph structure dynamic update module adopts an incremental update method and only updates the changed parts.

5. The large model-driven time-series capital chain intelligent risk transmission control system according to claim 1 is characterized in that: When executing control instructions, the control instruction execution module can take different levels of control measures according to different risk levels, including but not limited to early warning prompts, transaction restrictions, and account freezing.

6. A method for applying to the large model driven time series capital chain intelligent risk transmission control system as described in any one of claims 1 to 5, characterized in that: The following steps are involved: S1. The data integration module collects raw transaction data and related data and performs preliminary integration. S2. The time series relationship extraction module uses a large model to perform time series reasoning on the integrated data and analyze the complex relationships in the capital chain. S3, the risk transmission probability calculation module calculates the risk transmission probability of capital flow based on the time series relationship extraction results; S4, the graph structure dynamic update module dynamically updates the graph structure according to the risk transmission probability calculation results; S5. The risk disposal strategy generation module generates a risk disposal strategy based on the updated graph structure; S6. The control instruction execution module executes corresponding control instructions according to the generated risk disposal strategy.

7. The method according to claim 6, characterized in that In step S2, when calling the large model for time series reasoning, a hybrid model architecture combining the LSTM network and the Transformer structure is specifically adopted to improve the processing capability and accuracy of time series capital chain data.

8. The method according to claim 6, characterized in that In step S5, when generating risk disposal strategies, the big model is used to optimize and recommend strategies by combining similar cases and corresponding disposal effects in the historical risk case library.

Images