Enterprise risk intelligent identification system based on privacy computing and chain quantization prediction model

Abstract

The present application relates to the technical field of artificial intelligence, and more particularly to an enterprise risk intelligent identification system based on privacy calculation and chain quantization prediction model, comprising: a data processing module, which collects risk data and pre-processes to obtain standardized risk data, and determines the real-time updating capability of the risk data according to the updating frequency of the risk data; a privacy calculation module, which performs hierarchical asynchronous cache aggregation on the ciphertext update buffer pool according to the generation delay length of the privacy protection risk data; a risk feature construction module, which constructs risk features according to the privacy protection risk data; a chain quantization prediction module, which performs cross-link fusion on the time series risk data according to the number of interruptions of the multi-level risk transmission chain; a risk early warning module, which obtains risk analysis results and generates risk early warning information; and a model optimization updating module, which updates and optimizes the time series prediction model. The present application improves the enterprise risk identification capability.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an intelligent enterprise risk identification system based on privacy computing and a chain-based quantitative prediction model. Background Technology

[0002] In existing technologies, enterprise risk identification and default prediction are core components of financial risk control and supply chain management. Traditional risk control models face three major bottlenecks: First, data silos and privacy compliance conflicts: data from banks, tax authorities, judiciaries, and industrial and commercial bureaus belong to different entities, and centralized aggregation of raw data violates privacy protection requirements, resulting in incomplete training samples and limited feature dimensions. Second, limitations of static single-point prediction: existing models are mostly based on single time segments or simple time windows, lacking a chain-like characterization of the entire lifecycle of enterprise risk occurrence, transmission, and deterioration, making it difficult to capture cross-period risk migration patterns. Third, the contradiction between black-box decision-making and compliance auditing: while pure machine learning models possess non-linear fitting capabilities, they lack traceable quantitative decision-making paths, making them difficult to implement in regulatory penetration audits and cross-institutional collaborative risk control. Therefore, there is an urgent need for a systematic risk identification architecture that balances data privacy and security, risk chain evolution modeling, and quantitative interpretability.

[0003] Chinese Patent Publication No. CN120655397A discloses a method for constructing a bank anti-money laundering data model based on privacy computing. The method includes: acquiring fund flows from various financial institutions; considering transaction complexity and data fragmentation, using hierarchical sampling to integrate multi-institutional network path data to obtain preliminary flow map fragments; identifying path density and concealment levels; perturbing key paths to obtain a global fund flow map encryption framework; identifying node role diversity and topological mutation characteristics through the encryption topology framework; extracting transaction paths and multi-layered nesting relationships between nodes; analyzing the role distribution of transaction entities to obtain a dynamic path mutation set; filtering accounts with short usage cycles and accounts with frequent changes in transaction entities based on the dynamic path mutation set, and marking them as potentially suspicious transaction entities to obtain a preliminary list of suspicious transaction entities; identifying transaction records of the preliminary suspicious transaction entity list, and filtering out those with fraudulent transactions. High-risk transaction entities with fake registration information or missing key identity information are identified as a subset of high-risk transaction entities. Small, multiple transactions within this subset are statistically analyzed to identify their distribution characteristics and multi-layered transfer paths. Fund splitting patterns and hidden flow characteristics are extracted to obtain a high-risk transaction path feature map. The irregular distribution of transaction time intervals and the frequency of regional switching are analyzed using this feature map. If the time interval deviates from a preset pattern or the frequency of regional switching exceeds a preset threshold, it is marked as an abnormal transaction pattern, resulting in a set of abnormal transaction patterns. Based on this set, the differential distribution of transaction amounts at different levels and the visualization degree of inter-node relationships are calculated to determine core transaction nodes and key fund flow paths, resulting in a core transaction subgraph. Combining the high-risk transaction entity subset, the path feature map, and the set of abnormal transaction patterns, a graph neural network is used to predict anti-money laundering risks and output a money laundering risk warning. Therefore, the aforementioned privacy-based computing-based bank anti-money laundering data model construction method lacks the ability to depict the chain-like evolution trajectory of risks based on static data and lacks the ability to model the temporal evolution patterns of risks, leading to insufficient enterprise risk identification capabilities. Summary of the Invention

[0004] To address this, the present invention provides an intelligent enterprise risk identification system based on privacy computing and a chain-based quantitative prediction model, which overcomes the problem that existing technologies lack the ability to characterize the chain-based evolution trajectory of risks based on static data and lack the ability to model the temporal evolution law of risks, resulting in insufficient enterprise risk identification capabilities.

[0005] To achieve the above objectives, this invention provides an intelligent enterprise risk identification system based on privacy computing and a chain-based quantitative prediction model, comprising: The data processing module is used to preprocess the risk data collected in the enterprise's operation process to obtain standardized risk data, and to determine whether the real-time update capability of the risk data meets the requirements based on the update frequency of the risk data. The privacy computing module is used to perform secure association and encrypted computation on the standardized risk data based on privacy computing technology to obtain privacy protection risk data, calculate risk features and perform quantitative screening to output a set of effective risk features, and perform hierarchical asynchronous caching aggregation of the encrypted update buffer pool according to the generation delay of the privacy protection risk data. The risk feature construction module is used to construct risk features based on the privacy protection risk data to obtain static risk features, risk evolution features, and text risk features. The chain-based quantitative prediction module is used to construct a multi-level risk transmission chain based on the transmission relationship between enterprise risk factors. It inputs the risk evolution characteristics into the time series prediction model to obtain the chain-based risk prediction results, and performs cross-chaining and fusion of time series risk data based on the number of interruptions in the multi-level risk transmission chain. The risk warning module is used to predict the risk data based on the machine learning model to obtain risk analysis results, and generate risk warning information by combining the risk attribution path. The model optimization and update module is used to update and optimize the time series prediction model based on historical risk identification results and actual risk event results.

[0006] Furthermore, the privacy computing module is used to perform secure association and encrypted computation on the standardized risk data based on privacy computing technology to obtain privacy protection risk data; The privacy computing module is used to calculate risk features based on the privacy protection risk data, and to quantitatively filter the risk features to output a set of effective risk features.

[0007] Furthermore, the chain-based quantitative prediction module is used to construct a multi-level risk transmission chain based on the transmission relationship between enterprise risk factors; The chain-based quantitative prediction module is used to input the risk evolution characteristics into the time series prediction model to obtain the hidden features of the risk state; The chain-based quantitative prediction module is used to perform step-by-step prediction based on the risk transmission chain, the hidden features of the risk state, and the set of effective risk features to obtain the chain-based risk prediction result.

[0008] Furthermore, the risk warning module is used to train the initial model based on the risk characteristics to obtain a machine learning model; The risk warning module is used to predict risk data based on the machine learning model to obtain risk analysis results.

[0009] Furthermore, the risk warning module is used to trace risk triggering nodes according to the risk transmission chain to generate a risk attribution path; The risk warning module is used to generate risk warning information based on the risk analysis results and the risk attribution path.

[0010] Furthermore, the data processing module is used to determine whether the real-time update capability of the risk data meets the requirements based on whether the update frequency of the risk data is greater than or equal to a preset frequency; The data processing module is used to determine that the real-time update capability of the risk data does not meet the requirements if the update frequency of the risk data is less than the preset frequency.

[0011] Furthermore, the privacy computing module is used to determine that the real-time processing capability of the secret-state collaboration meets the requirements based on the generation delay of the privacy protection risk data being less than or equal to a preset first duration. The privacy computing module is used to perform hierarchical asynchronous caching aggregation on the secret state update buffer pool based on the fact that the generation delay of the privacy protection risk data is greater than the preset first duration and less than the preset second duration. The privacy computing module is used to preliminarily determine whether the temporal connection capability of the risk transmission chain does not meet the requirements based on the generation delay duration of the privacy protection risk data being greater than or equal to the preset second duration, and to determine whether the temporal connection capability of the risk transmission chain meets the requirements based on the number of interruptions in the multi-level risk transmission chain.

[0012] Furthermore, the privacy computing module is used to perform state analysis and classification management on the received encrypted update data, and store it in the corresponding cache layer; The privacy computing module is used to asynchronously perform collaborative aggregation based on the data update status of each cache layer, and to perform fusion processing on the aggregation results.

[0013] Furthermore, the chain-based quantitative prediction module is used to determine whether the temporal connection capability of the risk transmission chain meets the requirements based on whether the number of interruptions in the multi-level risk transmission chain is less than or equal to the preset number of interruptions. The chain-based quantitative prediction module is used to determine that the time-series connection capability of the risk transmission chain does not meet the requirements if the number of interruptions in the multi-level risk transmission chain is greater than the preset number of interruptions, and to perform cross-chaining and fusion of time-series risk data.

[0014] Furthermore, the chain-based quantitative prediction module is used to perform correlation analysis on risk data from different data sources at the same point in time; The chain-based quantization prediction module is used to perform cross-source cross-validation and fusion interpolation on broken data links based on existing risk status information in order to restore a complete and continuous time series.

[0015] Compared with existing technologies, the beneficial effects of the present invention are as follows: the system of the present invention, by setting up a data processing module, a privacy computing module, a risk feature construction module, a chain-based quantitative prediction module, a risk early warning module, and a model optimization and update module, determines the real-time update capability of risk data according to a preset frequency, performs hierarchical asynchronous caching and aggregation of the dense update buffer pool according to the generation delay of privacy-protected risk data, and performs cross-chaining and fusion of time-series risk data according to the number of interruptions in the multi-level risk transmission chain. This enables effective identification of enterprise risks and characterization of risk evolution trajectories through dynamic chain transmission, avoiding the problem of insufficient enterprise risk identification capability caused by static data, and improving the enterprise risk identification capability.

[0016] Furthermore, the system described in this invention determines the real-time update capability of risk data by setting a preset frequency. Due to the lag in data updates of multi-source heterogeneous data of enterprises, it is impossible to form complete and synchronous risk data input in a timely manner, resulting in insufficient data correlation analysis in the privacy computing process, which in turn leads to insufficient timeliness of risk data analysis. By determining the real-time update capability of risk data, the synchronization timeliness of multi-source heterogeneous risk data of enterprises can be evaluated, the reasons for insufficient privacy computing correlation analysis can be located, and the risk identification capability of enterprises can be further improved.

[0017] Furthermore, the system of the present invention adjusts the hierarchical asynchronous caching aggregation of the encrypted update buffer pool by setting a preset first duration and a preset second duration. Due to the communication latency of the encrypted channel and the high latency of cross-organization encrypted data transmission, privacy protection risk data cannot be aligned and aggregated in a timely manner, resulting in data synchronization lag among multiple parties involved in the calculation and insufficient fusion of encrypted features, which in turn causes delays in risk feature extraction. By performing hierarchical asynchronous caching aggregation on the encrypted update buffer pool, privacy protection risk data from different sources can be asynchronously cached and processed hierarchically according to their update status, reducing the cross-organization data synchronization waiting time, improving the efficiency of encrypted gradient collaborative aggregation, and enabling the effective risk feature set to be updated in a timely manner, thereby reducing the generation delay of privacy protection risk data and further improving the enterprise's risk identification capabilities.

[0018] Furthermore, the system described in this invention performs cross-chain fusion of time-series risk data by setting a preset number of interruptions. Due to insufficient continuity of time-series risk data and the lack of historical state information during the risk evolution process, the temporal dependencies between some risk nodes in the chain-like risk transmission process cannot be accurately extracted, resulting in chain-level truncation. This further leads to biases in the identification of risk evolution trends in the chain-like risk prediction results. By performing cross-chain fusion of time-series risk data, the temporal correlation between risk data from different sources at the same time point can be used to supplement the missing historical risk states, restore the temporal dependencies between risk nodes, reduce the probability and severity of chain-level truncation, and enable upstream-level risk information to be effectively transmitted to downstream levels. This improves the accuracy of risk evolution trend identification during the chain-like risk prediction process and further enhances the enterprise's risk identification capabilities. Attached Figure Description

[0019] Figure 1 This is a block diagram of the overall structure of the enterprise risk intelligent identification system based on privacy computing and chain-based quantitative prediction model, according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the real-time update capability of the enterprise risk intelligent identification system based on privacy computing and chained quantitative prediction model in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the hierarchical asynchronous caching aggregation process of the enterprise risk intelligent identification system based on privacy computing and chained quantitative prediction model, as described in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the process of cross-chaining and fusion of time-series risk data in the enterprise risk intelligent identification system based on privacy computing and chained quantitative prediction model, as described in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0021] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0022] Please see Figure 1 The diagram shown is an overall structural block diagram of the enterprise risk intelligent identification system based on privacy computing and chained quantitative prediction model according to an embodiment of the present invention. The system described in this embodiment of the present invention includes: The data processing module is used to preprocess the risk data collected in the enterprise's operation process to obtain standardized risk data, and to determine whether the real-time update capability of the risk data meets the requirements based on the update frequency of the risk data. The privacy computing module is connected to the data processing module. It is used to perform secure association and encrypted computation on the standardized risk data based on privacy computing technology to obtain privacy protection risk data, calculate risk features and perform quantitative screening to output a set of effective risk features, and perform hierarchical asynchronous caching aggregation of the encrypted update buffer pool according to the generation delay of the privacy protection risk data. A risk feature construction module, which is connected to the privacy computing module, is used to construct risk features based on the privacy protection risk data to obtain static risk features, risk evolution features, and text risk features. The chain-based quantitative prediction module is connected to the risk feature construction module. It is used to construct a multi-level risk transmission chain based on the transmission relationship between enterprise risk factors, input the risk evolution characteristics into the time series prediction model to obtain the chain-based risk prediction result, and perform cross-chaining and fusion of time series risk data based on the number of interruptions in the multi-level risk transmission chain. The risk warning module, which is connected to the chain-based quantitative prediction module, is used to predict the risk data based on the machine learning model to obtain risk analysis results, and generate risk warning information in combination with the risk attribution path. The model optimization and update module, which is connected to the risk warning module, is used to update and optimize the time series prediction model based on historical risk identification results and actual risk event results.

[0023] Specifically, risk data includes corporate financial data, legal risk data, and external risk data related to public opinion.

[0024] Specifically, preprocessing includes filling in missing values, removing outliers, and standardizing data for risk data.

[0025] Specifically, standardized risk data includes corporate financial data with missing values ​​filled in, legal risk data with outliers removed, and external risk data related to public opinion after data standardization.

[0026] Specifically, secure association is achieved through Privacy Set Intersection (PSI) based on Elliptic Curve Diffie-Hellman (ECDH) or Unintentional Transport (OT) protocols. Each participating party performs a salted hash mask (H(ID || salt)) on its local enterprise entity identifier. The encrypted intersection operation is completed at the scheduling node, and a globally aligned index table is output. The original ID and business data are kept in the local domain throughout the process, and only the encrypted intersection vector is exchanged, thus achieving the secure construction of the joint feature space.

[0027] Specifically, the encrypted computation employs a secure multi-party computation (MPC) framework or a trusted execution environment (TEE) enclave to perform cross-domain distribution statistics. Each party locally computes its feature frequency histogram and quantile intervals, and then uploads encrypted fragments to the aggregation node via the Shamir secret sharing protocol. The aggregation node reconstructs the joint distribution in encrypted form and calculates the information value (IV) and Kolmogorov-Smirnov statistic (KS). The system automatically removes weakly discriminative features with IV < 0.02 and uses the quantile corresponding to the maximum KS value as the privacy-preserving risk cutoff threshold for that feature, achieving objective quantification of feature selection.

[0028] Specifically, privacy-protected risk data is used as the basis for risk characterization and risk prediction.

[0029] Specifically, static characteristics include basic attribute indicators, risk warning indicators, and operational health indicators.

[0030] Specifically, the risk evolution characteristics are derived by constructing the rate of change and volatility of risk indicators over the past 1 month, 3 months, and 6 months. The trend characteristics of the risk indicator series are extracted, the first difference is calculated to characterize the rate of change of the risk indicators, and the second difference is calculated to characterize the change in the rate of change of risk, thereby identifying the trend characteristics of the risk status.

[0031] Specifically, the text risk features are derived from extracting fields such as cause of action, judgment result, amount involved, and judgment date from judicial documents. Regular expressions are used to match fixed-format data such as amounts, dates, and identification numbers to extract keywords from judicial penalty texts and construct risk word frequency vectors.

[0032] Specifically, the time-series prediction model is a deep learning model used to process risk evolution characteristics, extract time-series dependencies, and output chain-like risk prediction results.

[0033] Specifically, the chain-based risk prediction results include hierarchical risk scores, risk masks, and temporal evolution characteristics.

[0034] Specifically, the risk analysis results are a set of quantitative conclusions output by reasoning and calculating the input risk characteristics through a machine learning model.

[0035] Specifically, risk warning information includes risk level, probability of default, and warning timestamp.

[0036] Specifically, the initial model is a basic framework for enterprise risk identification that has the capability to extract basic risk features and is applicable to privacy computing and chain-based quantitative prediction.

[0037] Specifically, the machine learning model can be a long short-term memory network model, a random forest model, or a graph neural network model, with the preferred embodiment being a long short-term memory network model.

[0038] In this embodiment, the system described in this invention sets up a data processing module, a privacy computing module, a risk feature construction module, a chain-based quantitative prediction module, a risk warning module, and a model optimization and update module. It determines the real-time update capability of risk data based on a preset frequency, performs layered asynchronous caching and aggregation of the secret update buffer pool based on the generation delay of privacy-protected risk data, and performs cross-chaining and fusion of time-series risk data based on the number of interruptions in the multi-level risk transmission chain. This enables effective identification of enterprise risks and characterization of risk evolution trajectories through dynamic chain transmission, avoiding the problem of insufficient enterprise risk identification capability caused by static data and improving enterprise risk identification capabilities.

[0039] Specifically, the privacy computing module is used to perform secure association and cryptographic calculation on the standardized risk data based on privacy computing technology to obtain privacy protection risk data; The privacy computing module is used to calculate risk features based on the privacy protection risk data, and to quantitatively filter the risk features to output a set of effective risk features.

[0040] Specifically, the chain-based quantitative prediction module is used to construct a multi-level risk transmission chain based on the transmission relationship between enterprise risk factors; The chain-based quantitative prediction module is used to input the risk evolution characteristics into the time series prediction model to obtain the hidden features of the risk state; The chain-based quantitative prediction module is used to perform step-by-step prediction based on the risk transmission chain, the hidden features of the risk state, and the set of effective risk features to obtain the chain-based risk prediction result.

[0041] Specifically, the risk warning module is used to track risk triggering nodes according to the risk transmission chain to generate a risk attribution path; The risk warning module is used to generate risk warning information based on the risk analysis results and the risk attribution path.

[0042] Specifically, the risk warning module is used to train the initial model based on the risk characteristics to obtain a machine learning model; The risk warning module is used to predict risk data based on the machine learning model to obtain risk analysis results.

[0043] Please see Figure 2As shown, it is a logical flowchart of the process of determining the real-time update capability of risk data in an embodiment of the present invention. In this embodiment, the real-time update capability of risk data is quantified based on the update frequency of risk data.

[0044] Specifically, the data processing module is used to determine whether the real-time update capability of the risk data meets the requirements based on whether the update frequency of the risk data is greater than or equal to a preset frequency. The data processing module is used to determine that the real-time update capability of the risk data does not meet the requirements if the update frequency of the risk data is less than the preset frequency.

[0045] It is understood that the preset frequency is based on the statistical distribution of historical statistical samples as determined by those skilled in the art. Preferably, the update frequency of risk data is positively correlated with the real-time update capability of risk data, wherein... The preset frequency is based on the number of historical statistical samples. When the update frequency of risk data is less than this value, it indicates that there are problems such as untimely data synchronization and excessively long data refresh cycles during the risk data update process, which makes it impossible for some risk information to reflect the current risk status of the enterprise in a timely manner.

[0046] Regarding the size of the preset threshold, those skilled in the art can determine the base number based on the number of historical statistical samples, and can set it according to the actual situation, which will not be elaborated here; The number of historical statistical samples is set as a standardized reference value pre-determined by those skilled in the art based on the historical update patterns of multi-source heterogeneous risk data of enterprises, the need for real-time analysis of risk data, and the average level of risk data update frequency under stable operating conditions.

[0047] In this embodiment, the system described in this invention determines the real-time update capability of risk data by setting a preset frequency. Due to the lag in data updates of multi-source heterogeneous data of enterprises, it is impossible to form complete and synchronous risk data input in a timely manner, resulting in insufficient data correlation analysis in the privacy computing process, which in turn leads to insufficient timeliness of risk data analysis. By determining the real-time update capability of risk data, the synchronization timeliness of multi-source heterogeneous risk data of enterprises can be evaluated, the reasons for insufficient privacy computing correlation analysis can be located, and the risk identification capability of enterprises can be further improved.

[0048] Please see Figure 3 The diagram shown is a logical flowchart of the hierarchical asynchronous cache aggregation process for the dense-state update buffer pool according to an embodiment of the present invention. In this embodiment, the generation delay of privacy protection risk data is used to quantify the real-time processing capability of dense-state collaborative processing. The generation delay of privacy risk data is determined by the difference between the actual generation time of the privacy risk data and the standard generation time. The standard generation time is a standard reference value determined by the average processing time during the generation of privacy protection risk data under stable operating conditions.

[0049] Specifically, the privacy computing module is used to determine that the real-time processing capability of the secret-state collaboration meets the requirements based on the generation delay of the privacy protection risk data being less than or equal to a preset first duration. The privacy computing module is used to determine that the real-time processing capability of the secret state collaboration does not meet the requirements if the generation delay of the privacy protection risk data is greater than the preset first duration. The privacy computing module is used to perform hierarchical asynchronous caching aggregation on the secret state update buffer pool based on the fact that the generation delay of the privacy protection risk data is greater than the preset first duration and less than the preset second duration. The privacy computing module is used to preliminarily determine whether the temporal connection capability of the risk transmission chain does not meet the requirements based on the generation delay duration of the privacy protection risk data being greater than or equal to the preset second duration, and to determine whether the temporal connection capability of the risk transmission chain meets the requirements based on the number of interruptions in the multi-level risk transmission chain.

[0050] In this embodiment, the preset first duration is less than the preset second duration, wherein... If the generation delay of privacy protection risk data is less than or equal to the preset first time, the corresponding situation is that the real-time processing capability of secret-state collaboration meets the requirements. If the generation delay of privacy protection risk data is greater than the preset first time and less than the preset second time, the corresponding situation is that due to the communication delay of the encrypted channel, the cross-organization encrypted data transmission delay is high, which makes it impossible for privacy protection risk data to complete the encrypted state alignment and collaborative aggregation in a timely manner, causing the synchronization of data from multiple parties involved in the calculation to be delayed and the fusion of encrypted state features to be insufficient, thus causing the risk feature extraction delay. If the generation delay of privacy protection risk data is greater than or equal to the preset second duration, the corresponding situation is that due to insufficient continuity of time-series risk data, historical state information in the risk evolution process is missing, making it impossible to accurately extract the temporal dependencies between some risk nodes in the chain risk transmission process, causing chain hierarchical truncation, and further leading to the risk evolution trend identification deviation in the chain risk prediction results.

[0051] It is understood that the preset first duration and preset second duration are values ​​determined by those skilled in the art based on the statistical distribution of the generation delay duration of privacy protection risk data recorded in the real-time update capability of historical risk data, and in combination with the real-time processing capability of encrypted collaborative processing corresponding to different delay durations. Preferably, the generation delay duration of privacy protection risk data is negatively correlated with the real-time processing capability of encrypted collaborative processing, wherein, The preset first duration is the delay duration corresponding to the inflection point where the update frequency of risk data in the historical statistical distribution begins to decline. When the generation delay duration is less than or equal to this value, the update frequency of risk data remains at a high level, indicating that the cross-organization dense data transmission, alignment and collaborative aggregation process is relatively stable and the dense collaborative real-time processing capability meets the requirements. The inflection point is determined based on the pairing data of the historical synchronously recorded risk data update frequency and the generation delay of privacy protection risk data. It is determined by statistically analyzing the downward trend of the risk data update frequency as the generation delay changes, and identifying the node where the rate of decrease of the risk data update frequency first increases significantly. The baseline for setting the risk data update frequency is a standard reference value pre-set by those skilled in the art, taking into account the cross-institutional risk data transmission cycle, the privacy computing task processing cycle, and the average level of risk data update frequency under historical stable operating conditions.

[0052] The second preset duration is the generation delay duration corresponding to the point where the update frequency of risk data in the historical statistical distribution drops significantly. When the generation delay duration is greater than this value, it indicates that the dense collaborative processing process can no longer effectively guarantee the real-time update of risk data, resulting in a lag in the update of the risk feature set and further affecting the continuity of the risk transmission chain. The critical decline point is determined based on the pairing data of the risk data update frequency and the privacy protection risk data generation delay time recorded in historical synchronization. It is determined by statistically analyzing the downward trend of the risk data update frequency as the generation delay time changes, and identifying the critical node where the risk data update frequency drops rapidly. Specifically, the process of determining the severe decline point based on the pairing of historically synchronized risk data update frequency and privacy protection risk data generation delay duration is as follows: historical data are sorted from small to large according to privacy protection risk data generation delay duration, and the decline rate of risk data update frequency is statistically analyzed segment by segment. When the decline rate of risk data update frequency significantly increases after a certain delay duration node and exceeds the preset update frequency decline rate benchmark, then that node is determined as the severe decline point.

[0053] The size of the preset first duration and the preset second duration depends on the number of historical statistical samples and the ability of those skilled in the art to update risk data in real time. The base can be set according to the actual situation, and will not be elaborated here.

[0054] Specifically, the privacy computing module is used to perform state analysis and classification management on the received encrypted update data and store it in the corresponding cache layer; The privacy computing module is used to asynchronously perform collaborative aggregation based on the data update status of each cache layer, and to perform fusion processing on the aggregation results.

[0055] Specifically, using the dense-state update data as the object to be processed, a state analysis dimension is constructed based on the data generation time, update frequency, and data correlation of the dense-state update data. The state of the dense-state update data is identified to determine the update state and cache category corresponding to the dense-state update data. The dense-state update data is then stored in the corresponding cache layer according to different cache categories.

[0056] Specifically, the encrypted update data is risk data to be updated formed after privacy computing processing. It retains the data correlation and feature information used for risk analysis without exposing the original risk data content.

[0057] Specifically, the collaborative aggregation involves jointly calculating encrypted update data from multiple cache layers without decrypting the encrypted update data, in order to achieve secure fusion of risky data from different sources.

[0058] In this embodiment, the system described in this invention adjusts the hierarchical asynchronous caching aggregation of the encrypted update buffer pool by setting a preset first duration and a preset second duration. Due to the communication latency of the encrypted channel and the high latency of cross-organization encrypted data transmission, privacy protection risk data cannot be aligned and aggregated in a timely manner, resulting in delayed synchronization of multi-party data and insufficient fusion of encrypted features, which in turn causes delays in risk feature extraction. By performing hierarchical asynchronous caching aggregation on the encrypted update buffer pool, privacy protection risk data from different sources can be asynchronously cached and processed hierarchically according to their update status, reducing the waiting time for cross-organization data synchronization, improving the efficiency of encrypted gradient collaborative aggregation, and enabling the effective risk feature set to be updated in a timely manner. This reduces the generation latency of privacy protection risk data and further improves the enterprise's risk identification capabilities.

[0059] Please see Figure 4 As shown, it is a logical flowchart of the process of cross-chaining and fusion of time-series risk data in an embodiment of the present invention.

[0060] In this embodiment, the temporal continuity capability of the risk transmission chain is quantified based on the number of interruptions in the multi-level risk transmission chain.

[0061] Specifically, the chain-based quantitative prediction module is used to determine whether the timing connection capability of the risk transmission chain meets the requirements based on whether the number of interruptions in the multi-level risk transmission chain is less than or equal to the preset number of interruptions. The chain-based quantitative prediction module is used to determine that the time-series connection capability of the risk transmission chain does not meet the requirements if the number of interruptions in the multi-level risk transmission chain is greater than the preset number of interruptions, and to perform cross-chaining and fusion of time-series risk data.

[0062] In this embodiment, if the number of interruptions in the multi-level risk transmission chain is less than or equal to the preset number of interruptions, the corresponding situation is that the timing connection capability of the risk transmission chain is determined to meet the requirements. If the number of interruptions in the multi-level risk transmission chain exceeds the preset number of interruptions, the corresponding situation is that due to insufficient continuity of time-series risk data and missing historical state information in the risk evolution process, the time-series dependencies between some risk nodes in the chain-like risk transmission process cannot be accurately extracted, resulting in chain-level truncation, and further leading to bias in the identification of risk evolution trends in the chain-like risk prediction results.

[0063] It is understood that the preset number of interruptions is a value determined by those skilled in the art based on the statistical distribution of the number of interruptions in multi-level risk transmission chains in historical risk data, and in combination with the temporal continuity capability of risk transmission chains corresponding to the number of interruptions in different multi-level risk transmission chains. Preferably, the number of interruptions in multi-level risk transmission chains is negatively correlated with the temporal continuity capability of risk transmission chains. The preset number of interruptions is the number of interruptions that begins to increase as the generation delay of privacy protection risk data in the historical statistical distribution increases with the number of interruptions in the multi-level risk transmission chain. When the number of interruptions is less than or equal to this value, it indicates that the temporal correlation between risk nodes remains stable, the risk transmission chain can continuously transmit historical risk status information, and the risk evolution trend identification capability in the chain-like risk prediction process meets the requirements. The inflection point is determined by pairing historically synchronized data on the generation delay of privacy protection risk data with the number of interruptions in the multi-level risk transmission chain. It is determined by statistically analyzing the growth trend of the generation delay of privacy protection risk data with the number of interruptions in the risk transmission chain and identifying the node where the growth rate of the generation delay first significantly increases.

[0064] The number of preset interruptions depends on the number of historically collected samples and the benchmark set by those skilled in the art for the real-time processing capability of dense-state collaboration. It can be set according to the actual situation and will not be elaborated here.

[0065] Specifically, the chain-based quantitative prediction module is used to perform correlation analysis on risk data from different data sources at the same point in time; The chain-based quantization prediction module is used to perform cross-source cross-validation and fusion interpolation on broken data links based on existing risk status information in order to restore a complete and continuous time series.

[0066] Specifically, risk data from different sources collected within the same time window are used as the objects of correlation analysis. A multi-source risk correlation space is constructed based on the time synchronization relationship between risk data, the correlation relationship between risk factors, and the data change trend. Risk status information from different data sources is jointly matched to determine the degree of correlation between different risk data, so as to obtain multi-source fusion risk characteristics for risk transmission analysis.

[0067] Specifically, the multi-source risk association space is a joint analysis space composed of the time consistency dimension, the risk factor association dimension, and the state change trend dimension.

[0068] Specifically, the broken data link refers to a data association link where the temporal continuity of risky data is lost due to data acquisition delays, abnormal data sources, or delays in privacy computing processing.

[0069] Specifically, the cross-source cross-validation is based on the correlation between risk data from multiple different sources to make a consistency judgment on the data nodes to be repaired, so as to determine the credible basis for compensation of missing data.

[0070] Specifically, the fusion interpolation involves jointly estimating missing risk data based on historical risk status change trends, correlation weights between different data sources, and risk transmission chain constraints, in order to generate supplementary data that conforms to the laws of risk evolution.

[0071] In this embodiment, the system described in this invention performs cross-chain fusion of time-series risk data by setting a preset number of interruptions. Due to insufficient continuity of time-series risk data and missing historical state information during the risk evolution process, the temporal dependencies between some risk nodes in the chain-like risk transmission process cannot be accurately extracted, resulting in chain-level truncation. This further leads to biases in the identification of risk evolution trends in the chain-like risk prediction results. By performing cross-chain fusion of time-series risk data, the temporal correlation between risk data from different sources at the same time point can be used to supplement the missing historical risk states, restore the temporal dependencies between risk nodes, reduce the probability and severity of chain-level truncation, and enable upstream-level risk information to be effectively transmitted to downstream levels. This improves the accuracy of risk evolution trend identification during the chain-like risk prediction process and further enhances the enterprise's risk identification capabilities.

[0072] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A smart enterprise risk identification system based on privacy computing and a chain-based quantitative prediction model, characterized in that, include: The data processing module is used to preprocess the risk data collected in the enterprise's operation process to obtain standardized risk data, and to determine whether the real-time update capability of the risk data meets the requirements based on the update frequency of the risk data. The privacy computing module is used to perform secure association and encrypted computation on the standardized risk data based on privacy computing technology to obtain privacy protection risk data, calculate risk features and perform quantitative screening to output a set of effective risk features, and perform hierarchical asynchronous caching aggregation of the encrypted update buffer pool according to the generation delay of the privacy protection risk data. The risk feature construction module is used to construct risk features based on the privacy protection risk data to obtain static risk features, risk evolution features, and text risk features. The chain-based quantitative prediction module is used to construct a multi-level risk transmission chain based on the transmission relationship between enterprise risk factors. It inputs the risk evolution characteristics into the time series prediction model to obtain the chain-based risk prediction results, and performs cross-chaining and fusion of time series risk data based on the number of interruptions in the multi-level risk transmission chain. The risk warning module is used to predict the risk data based on the machine learning model to obtain risk analysis results, and generate risk warning information by combining the risk attribution path. The model optimization and update module is used to update and optimize the time series prediction model based on historical risk identification results and actual risk event results.

2. The enterprise risk intelligent identification system based on privacy computing and chained quantitative prediction model according to claim 1, characterized in that, The privacy computing module is used to perform secure association and encrypted computation on the standardized risk data based on privacy computing technology to obtain privacy protection risk data; The privacy computing module is used to calculate risk features based on the privacy protection risk data, and to quantitatively filter the risk features to output a set of effective risk features.

3. The enterprise risk intelligent identification system based on privacy computing and chain-based quantitative prediction model according to claim 1, characterized in that, The chain-based quantitative prediction module is used to construct a multi-level risk transmission chain based on the transmission relationship between enterprise risk factors; The chain-based quantitative prediction module is used to input the risk evolution characteristics into the time series prediction model to obtain the hidden features of the risk state; The chain-based quantitative prediction module is used to perform step-by-step prediction based on the risk transmission chain, the hidden features of the risk state, and the set of effective risk features to obtain the chain-based risk prediction result.

4. The enterprise risk intelligent identification system based on privacy computing and chained quantitative prediction model according to claim 1, characterized in that, The risk warning module is used to train the initial model based on the risk characteristics to obtain a machine learning model; The risk warning module is used to predict risk data based on the machine learning model to obtain risk analysis results.

5. The enterprise risk intelligent identification system based on privacy computing and chain-based quantitative prediction model according to claim 1, characterized in that, The risk warning module is used to track risk triggering nodes according to the risk transmission chain to generate risk attribution paths; The risk warning module is used to generate risk warning information based on the risk analysis results and the risk attribution path.

6. The enterprise risk intelligent identification system based on privacy computing and chained quantitative prediction model according to claim 1, characterized in that, The data processing module is used to determine whether the real-time update capability of the risk data meets the requirements based on whether the update frequency of the risk data is greater than or equal to a preset frequency. The data processing module is used to determine that the real-time update capability of the risk data does not meet the requirements if the update frequency of the risk data is less than the preset frequency.

7. The enterprise risk intelligent identification system based on privacy computing and chained quantitative prediction model according to claim 6, characterized in that, The privacy computing module is used to determine whether the real-time processing capability of the secret-state collaboration meets the requirements if the generation delay of the privacy protection risk data is less than or equal to a preset first duration. The privacy computing module is used to perform hierarchical asynchronous caching aggregation on the secret state update buffer pool based on the fact that the generation delay of the privacy protection risk data is greater than the preset first duration and less than the preset second duration. The privacy computing module is used to preliminarily determine whether the temporal connection capability of the risk transmission chain does not meet the requirements based on the generation delay duration of the privacy protection risk data being greater than or equal to the preset second duration, and to determine whether the temporal connection capability of the risk transmission chain meets the requirements based on the number of interruptions in the multi-level risk transmission chain.

8. The enterprise risk intelligent identification system based on privacy computing and chained quantitative prediction model according to claim 7, characterized in that, The privacy computing module is used to perform state analysis and classification management on the received encrypted update data and store it in the corresponding cache layer; The privacy computing module is used to asynchronously perform collaborative aggregation based on the data update status of each cache layer, and to perform fusion processing on the aggregation results.

9. The enterprise risk intelligent identification system based on privacy computing and chained quantitative prediction model according to claim 8, characterized in that, The chain-based quantitative prediction module is used to determine whether the temporal connection capability of the risk transmission chain meets the requirements based on whether the number of interruptions in the multi-level risk transmission chain is less than or equal to the preset number of interruptions. The chain-based quantitative prediction module is used to determine that the time-series connection capability of the risk transmission chain does not meet the requirements if the number of interruptions in the multi-level risk transmission chain is greater than the preset number of interruptions, and to perform cross-chaining and fusion of time-series risk data.

10. The enterprise risk intelligent identification system based on privacy computing and chain-based quantitative prediction model according to claim 9, characterized in that, The chain-based quantitative prediction module is used to perform correlation analysis on risk data from different data sources at the same point in time; The chain-based quantization prediction module is used to perform cross-source cross-validation and fusion interpolation on broken data links based on existing risk status information in order to restore a complete and continuous time series.

Images