A risk monitoring method and multi-scene adaptive system based on dynamic threshold early warning

Abstract

The application relates to the technical field of data monitoring. A risk monitoring method based on a dynamic threshold early warning comprises the following steps: acquiring risk trigger information in multiple scenes, acquiring credit type scenes, transaction type scenes and association type scenes according to the risk trigger information, judging whether the credit type scenes, the transaction type scenes and the association type scenes enter a risk monitoring state, acquiring structured monitoring data of the multiple scenes if the risk monitoring state is entered, and acquiring historical credit change data of the credit type scenes, real-time transaction flow data of the transaction type scenes and associated party risk exposure data of the association type scenes according to the structured monitoring data; and acquiring cross-scene risk characteristics according to the historical credit change data and the real-time transaction flow data. The application integrates credit type scene data, transaction type scene data and association type scene data, covers multiple-dimensional risk sources, avoids risk omissions caused by single data source analysis, and comprehensively controls systemic risks.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of data monitoring, in particular to a risk monitoring method based on dynamic threshold early warning and a multi-scene adaptive system. BACKGROUND

[0002] With the deep development of digital economy, the digital transformation of the fields of finance, business trade, supply chain management, etc. is continuously deepening, showing the significant characteristics of multi-scene, massive data and risk correlation. On the one hand, the scenes of credit business of financial institutions, transaction processes of e-commerce platforms, upstream and downstream cooperation of industrial chains, etc. are increasingly subdivided, and there are dozens of sub-scenes in the field of finance alone, such as personal credit, enterprise financing and cross-border payment, and the risk forms of various scenes are essentially different. On the other hand, the data flow between multiple scenes is significantly enhanced, and the risk is no longer limited to a single field, but shows a complex situation of “cross-scene conduction, time evolution and multi-source triggering”.

[0003] Traditional risk monitoring methods mostly use fixed risk thresholds (such as a single transaction amount upper limit and a fixed credit score standard), but the market environment, user behavior and associated network are in dynamic change, and static thresholds are prone to “false alarms” in the short-term fluctuations of risks, or “missed alarms” in the long-term evolution of risks (such as missing the intervention opportunity when credit slowly deteriorates without reaching the fixed threshold), which seriously reduces the reliability of early warning. SUMMARY

[0004] The application provides a risk monitoring method based on dynamic threshold early warning to solve the problem that single modal data cannot process multi-dimensional information correlation analysis, such as the contradictory situation of voice content and facial expressions easily leading to interactive logic fault.

[0005] The application provides a risk monitoring method based on dynamic threshold early warning, comprising:

[0006] obtaining risk trigger information in multiple scenes, and obtaining credit scenes, transaction scenes and associated scenes according to the risk trigger information,

[0007] determining whether the credit scenes, the transaction scenes and the associated scenes enter a risk monitoring state, if the risk monitoring state is entered, obtaining structured monitoring data of the multiple scenes, and obtaining historical credit change data of the credit scenes, real-time transaction flow data of the transaction scenes and associated party risk exposure data of the associated scenes according to the structured monitoring data;

[0008] obtaining cross-scene risk characteristics according to the historical credit change data and the real-time transaction flow data, and obtaining multi-scene basic risk characteristics according to the cross-scene risk characteristics;

[0009] According to the historical credit change data and the associated party risk exposure data, an associated risk feature is obtained, and a trend feature quantity is obtained according to the associated risk feature;

[0010] According to the multi-scene basic risk feature and the trend feature quantity, a comprehensive risk index is generated;

[0011] It is judged whether the comprehensive risk index is greater than a preset safety threshold;

[0012] If less, continue to monitor;

[0013] If greater, a sub-risk level is obtained according to the comprehensive risk index, and a multi-scene adaptive sub-level early warning information is obtained according to the sub-risk level.

[0014] Preferably, the step of judging whether the credit class scene, the transaction class scene and the associated class scene enter the risk monitoring state comprises:

[0015] The risk trigger information under the multi-scene is split according to the scene category to obtain credit state data corresponding to credit class scene exclusive trigger index, transaction behavior data corresponding to transaction class scene exclusive trigger index, and associated party risk transmission data corresponding to associated class scene exclusive trigger index;

[0016] According to the credit state data, the transaction behavior data and the associated party risk transmission data integration, scene-index association relationship data is obtained;

[0017] According to the scene-index association relationship data, associated risk transmission abnormal information is obtained;

[0018] According to the associated risk transmission abnormal information, the real-time trigger index data of the credit class scene, the transaction class scene and the associated class scene is respectively compared with the scene trigger condition data to obtain index-condition coincidence degree data;

[0019] According to the index-condition coincidence degree data, scene risk monitoring start judgment result data is obtained;

[0020] According to whether the scene risk monitoring start judgment result data meets the preset regulation;

[0021] If it meets, the determination result is that the coincidence degree meets the requirement, and it is determined that the corresponding scene enters the risk monitoring state;

[0022] If it does not meet, the determination result is that the coincidence degree does not meet the requirement, and it is determined that the scene does not enter the risk monitoring state.

[0023] Preferably, the step of obtaining the structured monitoring data of the multi-scene, and obtaining the historical credit change data of the credit class scene, the real-time transaction flow data of the transaction class scene and the associated party risk exposure data of the associated class scene according to the structured monitoring data comprises:

[0024] According to the structured monitoring data, the acquisition dimension data of the structured monitoring data corresponding to the credit type, the transaction type and the association type are obtained;

[0025] According to the acquisition dimension data of the structured monitoring data corresponding to the credit type, the transaction type and the association type, the credit type data storage source information, the transaction type data storage source information and the association type data storage source information are obtained;

[0026] According to the credit type data storage source information, the transaction type data storage source information and the association type data storage source information, the corresponding credit type original data, the transaction type original data and the association type original data are obtained;

[0027] The credit type original data, the transaction type original data and the association type original data are subjected to field integrity check to obtain the credit type standardized data, the transaction type standardized data and the association type standardized data;

[0028] According to the credit type standardized data, the historical credit change data are obtained;

[0029] According to the transaction type standardized data, the real-time transaction flow data are obtained;

[0030] According to the association type standardized data, the associated party risk exposure data are obtained.

[0031] Preferably, the step of obtaining the cross-scene risk features according to the historical credit change data and the real-time transaction flow data, and obtaining the multi-scene basic risk features according to the cross-scene risk features, comprises:

[0032] According to the historical credit change data, the change trajectory of the credit state in different time periods is obtained to obtain the credit change trajectory data;

[0033] According to the credit change trajectory data, the trend and the change amplitude features of the credit state are identified to obtain the credit trend feature data;

[0034] According to the real-time transaction flow data of the transaction type, the amount distribution, the transaction object party distribution and the time distribution behavior features of the transaction are extracted to obtain the transaction behavior feature data;

[0035] According to the credit trend feature data and the transaction behavior feature data, the cross-scene feature correlation data are obtained;

[0036] According to the cross-scene feature correlation data, the core correlation features capable of jointly reflecting the risk are screened out to obtain the cross-scene risk feature data;

[0037] According to the cross-scene risk characteristic data, multi-scene basic risk characteristic data is acquired.

[0038] Preferably, the step of acquiring the correlation risk characteristic according to the historical credit change data and the correlation party risk exposure data comprises:

[0039] According to the historical credit change data, credit state and correlation dimension of the correlation party are extracted to obtain credit correlation dimension data;

[0040] According to the correlation party risk exposure data, risk type, exposure scale and influence range risk dimension of the correlation party are extracted to obtain correlation risk dimension data;

[0041] According to the credit correlation dimension data and the correlation risk dimension data, correlation risk conduction analysis data is acquired;

[0042] According to the correlation risk conduction analysis data, correlation risk characteristic data is acquired;

[0043] According to the correlation risk characteristic data, correlation risk characteristics in different time periods are arranged in time sequence to form time sequence data of the correlation risk characteristic;

[0044] According to the time sequence data of the correlation risk characteristic, trend characteristic quantity data is generated.

[0045] Preferably, the step of generating the comprehensive risk index according to the multi-scene basic risk characteristic and the trend characteristic quantity comprises:

[0046] According to the multi-scene basic risk characteristic data, basic risk dimension data reflecting credit and basic risk dimension data of transaction cross-scene risk are extracted;

[0047] According to the trend characteristic quantity data, trend risk dimension data reflecting the change trend of the correlation risk is extracted;

[0048] According to the basic risk dimension data reflecting credit, the basic risk dimension data of transaction cross-scene risk and the trend risk dimension data, risk data fusion data is generated;

[0049] According to the risk data fusion data, fused risk data is acquired;

[0050] According to the fused risk data, risk quantization data is acquired;

[0051] According to the risk quantization data, comprehensive risk index data is acquired.

[0052] The application discloses a risk monitoring and multi-scene adaptation system based on dynamic threshold early warning, which comprises:

[0053] The data acquisition module is configured to acquire risk trigger information in multiple scenarios, and acquire credit scenarios, transaction scenarios, and association scenarios according to the risk trigger information,

[0054] The first judgment module is configured to judge whether the credit scenarios, the transaction scenarios, and the association scenarios enter a risk monitoring state, and if so, acquire structured monitoring data of the multiple scenarios, and acquire historical credit change data of the credit scenarios, real-time transaction flow data of the transaction scenarios, and association party risk exposure data of the association scenarios according to the structured monitoring data.

[0055] The multi-scenario basic risk feature acquisition module is configured to acquire cross-scenario risk features according to the historical credit change data and the real-time transaction flow data, and acquire multi-scenario basic risk features according to the cross-scenario risk features.

[0056] The trend feature quantity acquisition module is configured to acquire association risk features according to the historical credit change data and the association party risk exposure data, and acquire trend feature quantities according to the association risk features.

[0057] The comprehensive risk index generation module is configured to generate a comprehensive risk index according to the multi-scenario basic risk features and the trend feature quantities.

[0058] The second judgment module is configured to judge whether the comprehensive risk index is greater than a preset safety threshold.

[0059] If not, the monitoring is continued.

[0060] If so, a sub-risk level is acquired according to the comprehensive risk index, and sub-classified early warning information matched with multi-scenario adaptation is acquired according to the sub-risk level.

[0061] Preferably, the first judgment module comprises:

[0062] The data acquisition unit is configured to split risk trigger information in multiple scenarios according to scene categories to obtain credit state data corresponding to credit scenario exclusive trigger indicators, transaction behavior data corresponding to transaction scenario exclusive trigger indicators, and association party risk transmission data corresponding to association scenario exclusive trigger indicators.

[0063] The association relationship data acquisition unit is configured to integrate the credit state data, the transaction behavior data, and the association party risk transmission data to obtain scene-indicator association relationship data.

[0064] The association risk transmission abnormal information acquisition unit is configured to acquire association risk transmission abnormal information according to the scene-indicator association relationship data.

[0065] The fitting degree data acquisition unit is configured to acquire index-condition fitting degree data by comparing the fitting degrees of real-time trigger index data and scene trigger condition data of credit type scenes, transaction type scenes and correlation type scenes according to the associated risk conduction abnormal information.

[0066] The determination result data acquisition unit is configured to acquire scene risk monitoring start determination result data according to the index-condition fitting degree data.

[0067] The judgment unit is configured to determine whether the scene risk monitoring start determination result data meets a preset regulation.

[0068] If the determination result is that the fitting degree meets the requirement, it is determined that the corresponding scene enters a risk monitoring state.

[0069] If the determination result is that the fitting degree does not meet the requirement, it is determined that the scene does not enter the risk monitoring state.

[0070] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the risk monitoring method based on dynamic threshold early warning when executing the computer program.

[0071] The application further provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the risk monitoring method based on dynamic threshold early warning when executed by a processor.

[0072] Compared with the related art, the risk monitoring method based on dynamic threshold early warning provided by the application has at least the following technical effects:

[0073] The application integrates credit type, transaction type and correlation type scene data, covers multi-dimensional risk sources, avoids risk omissions caused by single data source analysis, and comprehensively controls systemic risks; through cross-scene risk feature correlation analysis and trend feature quantity extraction, the risk conduction law is captured, compared with traditional static evaluation, the evolution direction of the risk can be dynamically predicted, the early warning foresight is greatly improved; through data standardization processing, multi-dimensional fusion and comprehensive risk indicator quantization, the risk evaluation accuracy is improved, the hierarchical early warning mechanism can accurately adapt to the needs of multiple scenes, and excessive response or insufficient response is avoided.

[0074] The details of one or more embodiments of the application are presented in the following drawings and description to make other features, objects and advantages of the application more apparent. BRIEF DESCRIPTION OF DRAWINGS

[0075] The accompanying drawings, which are included to provide a further understanding of the application, constitute a part of the application and illustrate the illustrative embodiments of the application and its description serve to explain the application, and do not constitute an improper limitation on the application. In the drawings:

[0076] Figure 1 A flow chart of a method according to an embodiment of the application.

[0077] Figure 2 A schematic diagram of a system according to an embodiment of the application.

[0078] Figure 3 A schematic diagram of a computer device according to an embodiment of the application. DETAILED DESCRIPTION

[0079] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application is described and explained below in connection with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort belong to the scope of the present application.

[0080] Obviously, the accompanying drawings in the following description are only some examples or embodiments of the present application, and for those of ordinary skill in the art, the present application can be applied to other similar scenarios without creative effort based on these drawings. In addition, it can be understood that although the efforts made in this development process can be complex and lengthy, for those of ordinary skill in the art related to the content disclosed in the present application, some design, manufacturing or production changes based on the technical content disclosed in the present application are only routine technical means, and should not be understood as insufficient disclosure of the content disclosed in the present application.

[0081] In the present application, "embodiment" means that the specific features, structures or characteristics described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor is it an independent or alternative embodiment that is not mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in the present application can be combined with other embodiments without conflict.

[0082] Unless otherwise defined, technical terms and scientific terms used in the present application shall have the meanings that are commonly understood by one of ordinary skill in the art to which this application belongs. The terms "a", "an", "one", "this", and the like, do not denote a limitation of quantity, but rather denote the presence of at least one of the referenced item. The terms "include", "comprise", "have", and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, product, or device that comprises a list of steps or units (elements) not only comprises those steps or units but can also comprise other steps or units not expressly listed or inherent to such process, method, product, or device. The terms "connected", "coupled", and "linked", and the like, do not necessarily denote a direct connection or linkage, but also include an indirect connection or linkage. The term "multiple" means two or more. The term "and / or" describes associated objects in association relationship, which means that there can be three relationships, for example, "A and / or B" can mean that A exists alone, A and B exist together, and B exists alone. The character " / " generally means that the associated objects are in an "or" relationship. The terms "first", "second", "third", and the like, merely distinguish similar objects, and do not represent a specific order.

[0083] Embodiment 1

[0084] The embodiment of the present application provides a risk monitoring method based on dynamic threshold early warning. Figure 1 is a flowchart according to an exemplary embodiment. As shown in Figure 1 , it comprises:

[0085] S1, acquiring risk trigger information in multiple scenarios, and acquiring credit type scenarios, transaction type scenarios, and association type scenarios according to the risk trigger information,

[0086] S2, judging whether the credit type scenarios, the transaction type scenarios, and the association type scenarios enter a risk monitoring state, if entering the risk monitoring state, acquiring structured monitoring data of the multiple scenarios, and acquiring historical credit change data of the credit type scenarios, real-time transaction flow data of the transaction type scenarios, and association party risk exposure data of the association type scenarios according to the structured monitoring data;

[0087] S3, acquiring cross-scenario risk features according to the historical credit change data and the real-time transaction flow data, and acquiring multiple-scenario basic risk features according to the cross-scenario risk features;

[0088] S4, acquiring association risk features according to the historical credit change data and the association party risk exposure data, and acquiring trend feature quantities according to the association risk features;

[0089] S5, generating a comprehensive risk index according to the multi-scene basic risk features and trend feature quantities;

[0090] S6, judging whether the comprehensive risk index is greater than a preset safety threshold;

[0091] If not, continuous monitoring is performed;

[0092] If yes, a sub-risk level is obtained according to the comprehensive risk index, and a multi-scene adaptive sub-level early warning information is obtained according to the sub-risk level.

[0093] As described in S1-S6 above, the present application integrates the risk trigger information of credit scenarios, transaction scenarios and associated scenarios, extracts cross-scene risk features and trend feature quantities, generates a comprehensive risk index, and then determines whether to start risk monitoring, hierarchical early warning or response measures according to the index, solves the problem that single-scene risk assessment cannot comprehensively cope with systemic risks, and realizes accurate identification and timely disposal of multi-dimensional risks. From the physical meaning, risk management is a dynamic and complex process. With the continuous changes of credit environment, market transaction mode and associated behaviors, single-scene assessment cannot cover potential systemic risks, especially in the case of high correlation of multi-scene data, traditional methods cannot accurately identify the risk transmission effect, therefore, a forward-looking dynamic monitoring system integrating multi-scene data needs to be built to timely capture risk signs and prevent spread. The deficiency of the prior art is that traditional risk monitoring focuses on a single scene, lacks deep integration of cross-scene risk correlation effects, can only judge risks in a specific scene, and cannot comprehensively and real-time assess multi-dimensional systemic risks. The present application integrates multi-scene risk trigger information, and each step is based on cross-scene data integration ideas, ensuring the comprehensiveness and accuracy of risk monitoring.

[0094] Multi-scene risk trigger information is collected from API interfaces, internal databases, data grabbing tools and other channels, including historical transaction data, credit data, associated party behavior data and the like, and then classified by clustering algorithms, rule engines and the like into credit, transaction and association scenarios; risk trigger information is the basis of risk assessment, and the information of each scene is a preliminary signal of potential risk, and classification and extraction can lay a data foundation for subsequent monitoring, for example, in the financial scenario, large high-frequency transactions of an account can trigger a transaction class alarm, and credit score fluctuations trigger credit class monitoring.

[0095] Thresholds are set according to the risk trigger information of each scene by decision trees, random forest models or rule logic to judge whether to enter a risk monitoring state, and if the information exceeds the threshold, monitoring is started; the preliminary screening of the risk state prevents high-risk scenarios from spreading to other fields, for example, when a user transfers a large amount of money to an unknown account, the transaction scenario will automatically enter the monitoring state.

[0096] Based on historical credit change data and real-time transaction flow data, principal component analysis (PCA), neural network and other feature extraction algorithms are used to fuse data and extract cross-scenario risk features. The physical meaning of this step is to mine potential correlations in multiple scenarios. Single-scenario risk may not be significant, but joint analysis can reveal deep risk patterns. For example, a user's transactions and credit are abnormal in multiple scenarios, and the cross-scenario risk features are higher than normal values.

[0097] Extract historical credit change data and associated risk exposure data, and then fuse the data through time series analysis and clustering to extract associated risk features and finally calculate trend feature quantities. Capture the dynamic changes of credit risk and the risk transmission effect of associated parties. For example, a company's credit score fluctuates greatly within three months, and the associated supplier's financial situation deteriorates, which can generate a "credit risk transmission index" to predict risk evolution.

[0098] Quantify the basic risk features of each scenario, and then synthesize the comprehensive risk index through weighted averaging or machine learning algorithms according to the contribution of each scenario to the overall risk. This reflects the overall risk level of the subject and avoids the one-sidedness of single-scenario evaluation. For example, the market, credit, and operational risk indices of a company are 0.6, 0.7, and 0.5 respectively, and the weighted comprehensive index is 0.6, providing a comprehensive basis for decision-making.

[0099] Based on historical data and industry standards, set a safety threshold, and then compare the comprehensive risk index with the threshold. If the index exceeds the threshold, intervention is required, otherwise continuous monitoring is performed. The physical meaning is that the threshold is the safety boundary of risk tolerance, which can avoid unnecessary intervention. Traditional methods set rough and static thresholds, while dynamic comparison achieves intelligent control, ensuring that response is initiated only when risk is truly exceeded. For example, if the index exceeds the threshold, a graded early warning is triggered, otherwise regular monitoring is maintained. Through the integration and dynamic analysis of multi-scenario data, the risk assessment is significantly improved in terms of forward-looking and accuracy, providing scientific support for risk management.

[0100] In one embodiment, the step of determining whether the credit scenario, transaction scenario, and associated scenario enter the risk monitoring state includes:

[0101] S201, splitting the risk trigger information in multiple scenarios according to the scene category to obtain credit state data corresponding to credit scenario-specific trigger indicators, transaction behavior data corresponding to transaction scenario-specific trigger indicators, and associated party risk transmission data corresponding to associated scenario-specific trigger indicators;

[0102] S202, integrating the credit state data, transaction behavior data, and associated party risk transmission data to obtain scene-index association relationship data;

[0103] S203, obtaining associated risk transmission abnormal information according to the scene-index association relationship data;

[0104] S204, respectively compare the real-time trigger index data of the credit type scene, the transaction type scene and the correlation type scene with the scene trigger condition data according to the associated risk conduction abnormal information, and obtain index-condition matching degree data;

[0105] S205, obtaining scene risk monitoring start determination result data according to the index-condition matching degree data;

[0106] S206, whether the scene risk monitoring start determination result data meets the preset regulation;

[0107] If it meets, the determination result is that the matching degree meets the requirement, and it is determined that the corresponding scene enters the risk monitoring state;

[0108] If it does not meet, the determination result is that the matching degree does not meet the requirement, and it is determined that the scene does not enter the risk monitoring state.

[0109] In summary, the present application can accurately determine the risk monitoring start condition of each scene by fine splitting, integrating and multi-dimensional comparison of multiple types of risk data, identify and respond to potential risks in time, and avoid the spread of single scene risk or cross-scene risk conduction causing systemic problems. From step S201, the risk characteristics of each scene need to be embodied by exclusive trigger indicators and data types. Only by extracting risk trigger information according to scene classification can a foundation be laid for subsequent targeted analysis. The deficiency of traditional technology is that it adopts a global unified data processing method, ignores the differences in risk performance of different scenes, and leads to disordered information and fuzzy analysis, which cannot accurately locate the source of risk. The solution is to first collect multiple scene risk trigger information from API interfaces, internal databases and other channels, and then split the information according to scene categories through clustering algorithms, rule engines and other technologies to form credit status data exclusive to credit type scenes, transaction behavior data exclusive to transaction type scenes, and associated party risk conduction data exclusive to correlation type scenes. For example, in the financial field, credit score fluctuations of an account can be classified as credit status data, and large high-frequency transfers can be classified as transaction behavior data, which clearly defines the data boundary for subsequent analysis.

[0110] By integrating the split data of different scenarios, the correlation between scenarios and risk indicators is established, forming a unified analysis perspective to explore the potential connection of cross-scenario risks. The defect of traditional technology is that it processes different types of data sources separately, lacks cross-scenario and cross-indicator data fusion and correlation analysis capability, and cannot identify the transmission path between risks, reducing the accuracy of risk monitoring. Through data integration algorithm, trigger data of credit, transaction and association scenarios are included in a unified correlation model. This model can reflect the internal relationship between risk indicators in different scenarios, such as using historical credit data of credit scenarios to predict the risk occurrence probability of transaction scenarios, and identifying cross-border risk transmission risks through association scenarios' fund flow information, providing integrated data support for subsequent anomaly detection.

[0111] From the integrated scenario-indicator correlation data, abnormal correlation patterns are mined. These abnormal patterns often represent potential risk transmission paths and are important signals of risk hazards. Traditional technology relies on a single data source or simple statistical models for anomaly detection, which cannot cover multi-dimensional and cross-scenario risk transmission anomalies, and is prone to miss deep risks. Through complex anomaly detection algorithms, the abnormal rules in scenario-indicator correlation data are analyzed. For example, in association scenarios, by analyzing the fund flow frequency and amount changes between different enterprises, it can be found that the credit deterioration of an enterprise may be transmitted to its partners through the fund chain, thus identifying the abnormal transmission chain of "credit problem-fund flow-risk diffusion" and providing key evidence for risk prediction.

[0112] By comparing real-time trigger indicator data with preset scenario trigger conditions, the degree of fit is calculated to dynamically determine the probability of scenario risk occurrence. The higher the degree of fit, the greater the risk possibility, providing a dynamic reference for risk monitoring start. Traditional technology relies on fixed preset risk thresholds, which are difficult to adapt to real-time changes in scenario conditions and data fluctuations, and are prone to false positives or false negatives. Combined with a dynamic threshold adjustment mechanism, real-time trigger indicator data is compared with scenario trigger condition data. The degree of fit is calculated by methods such as cosine similarity and Euclidean distance. If the degree of fit reaches the set standard (e.g., the real-time frequency of a transaction scenario and the abnormal trigger condition fit degree exceeds 80%), it is preliminarily determined that the scenario risk possibility is high, providing a quantitative basis for subsequent start decision.

[0113] Based on the index-condition matching degree data, multiple dimensions need to be considered to ensure accuracy; traditional technology often uses a single index threshold to determine the monitoring start, which is easy to ignore the complexity of risk factors in multiple scenarios and multiple data environments, resulting in errors; by using a weighted average algorithm, multiple dimension risk trigger indicators (such as credit scenario score fluctuation range, transaction scenario amount abnormality degree, and correlation scenario conduction speed) are combined to optimize the matching degree data, for example, higher weight is given to high-impact indicators (such as sudden credit score drop), and finally the scenario risk monitoring start determination result data is generated, avoiding decision bias caused by a single indicator.

[0114] The scenario risk monitoring start determination result is finally compared with the preset monitoring start condition to determine whether the scenario enters the risk monitoring state, which is the final decision-making link of the risk monitoring process; traditional risk monitoring systems rely on static rules and lack flexibility and dynamic adjustment capability, and are easy to fail to respond to new risks due to rule lag; by comparing the determination result with the preset risk monitoring standard and adjusting the strategy flexibly according to the real-time data change trend, if the determination result meets the preset regulation (such as the matching degree exceeds the threshold for 5 minutes and the risk trend is rising), it is determined that the corresponding scenario enters the risk monitoring state; if it does not meet, it is determined that the scenario does not enter the monitoring state, avoiding unnecessary resource waste and misjudgment.

[0115] In one embodiment, the step of acquiring structured monitoring data of multiple scenarios and acquiring historical credit change data of credit scenarios, real-time transaction flow data of transaction scenarios, and associated party risk exposure data of associated scenarios according to the structured monitoring data comprises:

[0116] S207, acquiring structured monitoring data collection dimension data corresponding to credit scenarios, transaction scenarios, and associated scenarios according to the structured monitoring data;

[0117] S208, acquiring credit scenario data storage source information, transaction scenario data storage source information, and associated scenario data storage source information according to the collection dimension data of the structured monitoring data corresponding to the credit scenarios, transaction scenarios, and associated scenarios;

[0118] S209, acquiring corresponding credit scenario original data, transaction scenario original data, and associated scenario original data according to the credit scenario data storage source information, transaction scenario data storage source information, and associated scenario data storage source information;

[0119] S210, performing field integrity check on the credit scenario original data, transaction scenario original data, and associated scenario original data to obtain credit scenario standardized data, transaction scenario standardized data, and associated scenario standardized data;

[0120] S211, acquire historical credit change data according to the credit scenario standardized data;

[0121] S212, acquire real-time transaction flow data according to the transaction scenario standardized data;

[0122] S213, acquire associated party risk exposure data according to the associated scenario standardized data.

[0123] In summary, the present application identifies the structured monitoring data collection dimensions of credit, transaction and association scenarios according to the system preset rules. These dimensions clearly define the boundaries and core content of data collection for each scenario. The credit scenario collection dimensions include historical credit score, overdue records, loan repayment, etc. The transaction scenario collection dimensions include transaction time, transaction amount, transaction party identity information, transaction purpose, etc. The associated scenario collection dimensions involve associated party fund flow record, cooperation period, guarantee relationship, associated party credit status, etc. This step clearly defines the collection dimensions to ensure that the subsequent collected data is complete and has analysis value, avoiding data loss or redundancy due to ambiguous dimensions. For example, in the financial field, the credit scenario needs to collect credit score change dimensions for nearly 3 years, which can ensure accurate analysis of credit trends.

[0124] Based on the determined collection dimensions of each scenario, the corresponding data storage source information is further acquired. These information includes data storage location, data format, access permission and data update frequency. Since multi-scenario data is often scattered in different database systems or even across data centers, the mapping and matching of collection dimensions and storage sources is realized through automated means.

[0125] According to the acquired data storage source information of each scenario, the original data is extracted from the corresponding storage source. The credit scenario original data is extracted from the credit database or third-party credit platform, the transaction scenario original data is extracted from the bank transaction system and e-commerce platform transaction log, and the associated scenario original data is extracted from the associated business management system and enterprise cooperation archives. The technical core of this step is to design an efficient and secure data retrieval mechanism. For example, distributed data synchronization technology is used to extract data from multiple storage sources in parallel, encrypted transmission protocols are used to ensure data security, and data integrity check points are set to avoid data omission or incorrect acquisition, ensuring that the original data can fully reflect the actual situation of each scenario.

[0126] The original data of the three types of scenarios are subjected to field integrity check and standardization processing: the field integrity check checks whether the data has missing key fields, format errors, and numerical abnormalities through preset rules, fills in the missing fields by interpolation with the same scene and dimension data, converts the format error data in batches, and marks and reextracts the numerical abnormal data; the standardization processing unifies the data of different formats and units into the system standard format, and finally outputs the standardized data of credit, transaction, and association scenarios.

[0127] The historical credit change data is extracted from the standardized data of the credit scenario, the fields related to the credit state change in the standardized data are filtered, and the historical credit change data set is arranged in chronological order, for example, the credit score, overdue times, and unpaid loan amount of a subject in the past 24 months are summarized monthly, which can intuitively present the evolution trend of the credit status of the subject, and provide time series data support for subsequent analysis of the credit risk change rule and prediction of future credit behavior.

[0128] For the standardized data of the transaction scenario, real-time transaction stream data is extracted, the transaction timestamp, transaction amount, transaction party, and transaction type fields in the standardized data are analyzed in real time, and a real-time transaction stream list is generated according to the transaction occurrence order, which includes the unique identifier, occurrence time, core information, and risk label of each transaction. The core of this step is to ensure real-time performance, and the real-time analysis and stream generation of standardized data are realized through stream processing technology, and a data error correction mechanism is set to ensure that the real-time transaction stream data accurately reflects the current transaction behavior of the subject, and provides data support for real-time monitoring of abnormal transactions and timely blocking of risks.

[0129] The associated party risk exposure data is extracted from the standardized data of the association scenario, the risk related fields of the associated party in the standardized data are identified, and the associated party risk exposure data set is filtered and integrated according to the risk exposure definition, for example, the credit rating of an associated supplier of a subject is reduced from A to C, and the supplier still owes the subject 1 million yuan of overdue goods, which will be included in the associated party risk exposure data. The technical difficulty of this step lies in the comprehensive identification of associated parties and accurate definition of risk exposure, which is achieved by constructing an associated party relationship graph to avoid missing associated parties and by setting a risk exposure judgment rule to ensure accuracy. The associated party risk exposure data obtained finally can help identify potential indirect risks and provide a basis for preventing cross-subject risk transmission.

[0130] In one embodiment, the step of obtaining cross-scenario risk features from the historical credit change data and real-time transaction stream data, and obtaining multi-scenario basic risk features according to the cross-scenario risk features, comprises:

[0131] S301, obtain the change trajectory of the credit state in different time periods according to historical credit change data, and obtain credit change trajectory data;

[0132] S302, identify the trend and change amplitude characteristics of the credit state according to the credit change trajectory data, and obtain credit trend characteristic data;

[0133] S303, extract the amount distribution, transaction object party distribution and time distribution behavior characteristics of the transaction according to the real-time transaction flow data of the transaction type scene, and obtain transaction behavior characteristic data;

[0134] S304, obtain cross-scene feature association data according to the credit trend characteristic data and the transaction behavior characteristic data;

[0135] S305, filter out core associated features that can jointly reflect risks according to the cross-scene feature association data, and obtain cross-scene risk feature data;

[0136] S306, obtain multi-scene basic risk feature data according to the cross-scene risk feature data.

[0137] In summary, the application extracts credit trend characteristics and transaction behavior characteristics through multi-dimensional analysis of credit scene historical data and transaction scene real-time data, filters out core risk features through cross-scene association analysis, and finally generates multi-scene basic risk feature data, which provides core data support for subsequent comprehensive and accurate risk assessment and early warning, and solves the problem of incomplete and inaccurate risk identification caused by traditional technology relying on single scene or single dimension data. From the physical meaning, the change of credit state has time sequence regularity, and the transaction behavior directly reflects the real-time activity of the subject, and the two respectively carry the risk information of different scenes. Through the time sequence tracing, trend extraction of credit data and the multi-dimensional feature extraction of transaction data, the two types of scene characteristics are associated and integrated, potential risk patterns (such as the linkage of credit deterioration and abnormal transactions) that cannot be found in a single scene can be mined. The selection of core risk features and the generation of basic risk features can further focus on key risk signals, reduce redundant data interference, and lay a precise data foundation for risk assessment.

[0138] Obtain credit change trajectory data of different time periods through historical credit change data: technically, use autoregressive model (AR), moving average method (MA), exponential smoothing method and other time series analysis techniques to decompose historical credit data by period, fit the credit fluctuation trend of each period, mark the credit rising, falling or stable mode, and form a continuous trajectory; for example, a user's credit score dropped from 820 to 650 in 12 months due to multiple credit card overdue, the time series model can capture this continuous decline trajectory and the sharp drop node after each overdue. This step traces back the time sequence evolution of credit status, provides original time sequence data for subsequent trend analysis, and makes up for the limitations of traditional single time window data.

[0139] Identify credit trend direction and change amplitude characteristics based on credit change trajectory data to generate credit trend characteristic data: technically, use linear regression model to judge long-term trend, use moving average method and Holt-Winters exponential smoothing method to filter short-term noise, and calculate standard deviation, variance and other indicators to evaluate fluctuation amplitude; distinguish between long-term trend (such as continuous deterioration) and short-term fluctuation of credit change, and solve the problem that traditional static data cannot distinguish between risk nature.

[0140] Extract transaction amount distribution, transaction object distribution, and time distribution characteristics from real-time transaction flow data in transaction scenarios to generate transaction behavior characteristic data: technically, first, count the interval distribution of transaction amount to identify abnormal consumption, analyze the type of transaction object to judge the risk of transaction object, and extract the time distribution of transaction to find time anomalies; describe the subject's consumption habits and potential risks through multi-dimensional transaction characteristics, for example, a user transfers a large amount of money to a strange network merchant at night within a week, and the amount, object, and time distribution characteristics will jointly point to the risk of abnormal transactions, making up for the one-sidedness of traditional single-dimensional transaction analysis.

[0141] Obtain cross-scene feature association data through correlation analysis of credit trend characteristic data and transaction behavior characteristic data: technically, use correlation analysis, collaborative filtering and other methods to obtain the correlation of the two types of characteristics, for example, find that "continuous decline in credit score" and "increase in large transaction frequency" and "increase in proportion of transactions with strangers" have strong correlation; reveal the potential linkage between credit status and transaction behavior, break the traditional credit and transaction analysis, for example, a user's credit deterioration is accompanied by an increase in abnormal transactions, this association data can be used as an important risk signal to provide direction for subsequent core feature screening.

[0142] Filtering core associated features that can jointly reflect risks from cross-scenario associated feature data to generate cross-scenario risk feature data: technically, first standardize the data to ensure comparability, then calculate the association between features using chi-square test, mutual information, Pearson correlation coefficient, etc., filter features that show strong association in multiple scenarios, and eliminate redundant features; focus on multi-scenario common risk features to reduce irrelevant data interference, for example, the filtered "credit score monthly drop over 50 points and stranger transaction proportion over 30%" can accurately point to the situation where high credit risk and transaction risk coexist, solving the problem of narrow coverage of traditional single-scenario features.

[0143] Based on cross-scenario risk feature data, obtain multi-scenario basic risk feature data: technically, first integrate core feature data into a unified data set, then reduce redundancy by principal component analysis, aggregate similar features by clustering analysis, and finally standardize to generate basic risk feature data that adapts to multiple scenarios; further integrate and optimize cross-scenario core risk features to form a simple and unified risk data set, for example, "credit-transaction linkage risk factor 0.85" can directly represent high risk level, providing standardized input for subsequent comprehensive risk index calculation, solving the problem of traditional multi-scenario data integration difficulty and poor comparability.

[0144] In one embodiment, the step of obtaining associated risk features according to the historical credit change data and associated party risk exposure data, and obtaining trend feature quantities according to the associated risk features, comprises:

[0145] S401, according to the historical credit change data, extracting the association dimension of credit status and associated parties to obtain credit association dimension data;

[0146] S402, according to the associated party risk exposure data, extracting the risk type, exposure scale, and impact range risk dimension of the associated party to obtain associated risk dimension data;

[0147] S403, obtaining associated risk transmission analysis data according to the credit association dimension data and the associated risk dimension data;

[0148] S404, obtaining associated risk feature data according to the associated risk transmission analysis data;

[0149] S405, according to the associated risk feature data, arranging the associated risk features in different time periods in time sequence to form time series data of the associated risk features;

[0150] S406, generating trend feature quantity data according to the associated risk feature time series data.

[0151] In summary, the present application extracts the associated risk characteristics that can reflect the linkage between the two by deeply analyzing the historical credit change data and the associated party risk exposure data, and then generates trend characteristic quantity data of quantifiable risk change rules through the time series processing of these characteristics, finally provides accurate and comprehensive data support for subsequent risk analysis, early warning and decision making. The core is to solve the problems of single data source dependence, weak time series modeling capability and simplified risk dimension extraction in traditional risk analysis, and to realize the dynamic and deep control of the risk transmission effect of associated parties. From the physical meaning, with the increasing complexity of the financial market and the business environment, the risk transmission between associated parties has become a key factor affecting the stable operation of enterprises. The historical credit change data is the "time series archive" of the credit status of enterprises, and the associated party risk exposure data is the "mapping carrier" of the risk transmission path. Combining the two types of data can not only discover potential associated risk signals (such as the linkage between the decline of enterprise credit score and the default of associated party debt) that cannot be found by single data source, but also can quantify the future evolution direction of risk through trend characteristics, thereby optimizing the forward-looking and targetedness of risk management and avoiding the risk of risk missing caused by incomplete coverage in traditional single data dimension analysis.

[0152] The credit status and the associated dimension of the associated party are extracted from the historical credit change data to generate credit associated dimension data. The credit status of an enterprise does not exist in isolation, and the financial health and cooperation stability of associated parties will directly affect the credit change of the enterprise. Conversely, the credit fluctuation of the enterprise may also be transmitted to the associated party. Extracting the associated dimension of the two can clearly define the basic associated path of risk transmission. The credit status of an enterprise is analyzed by using static credit data, and the dynamic interaction between the credit and the associated party risk is completely ignored, which leads to the inability to capture the potential impact of credit change on the risk exposure of associated parties. The solution of the present application is to extract the time series credit information of the credit score, default record, debt repayment progress, etc. of the enterprise at multiple time nodes from the historical credit change data, and then match the financial health score, cooperation business default situation, and abnormal fund flow record, etc. information of the associated party at the corresponding time node from the associated party risk exposure data. Through the establishment of a mapping model of "enterprise credit indicators and associated party risk indicators", the two types of time series data are aligned according to the time dimension, and finally the credit associated dimension data that can reflect the dynamic association between the two is formed, laying a foundation for subsequent analysis of how credit change affects associated risk transmission.

[0153] According to the associated party risk exposure data, the risk type, exposure size and influence range risk dimensions of the associated party are extracted, and the associated risk dimension data is obtained. The associated party risk is not a single form. Different risk types, different exposure sizes and different influence ranges have great differences in the risk impact on enterprises. Comprehensive extraction of these dimensions can completely depict the real characteristics of the associated party risk. Often, only a single risk indicator is used to analyze the risk, which not only ignores the comprehensive effect of different types of risks such as market risk and credit risk, but also fails to consider the transmission chain of risks among different associated parties. The solution of the present application is to realize multi-dimensional extraction through multi-channel data analysis: for the risk type, analyze the profit fluctuation reason, contract default clause, regulatory penalty record and the like in the financial statements of the associated party, and define the risk category; for the exposure size, the cumulative debt amount, the unfulfilled contract amount, the abnormal fund occupation amount and the like of the associated party in the cooperation period are counted, and the specific risk exposure value is calculated; for the influence range, the cooperation business range, the number of upstream and downstream enterprises and the industry field that the associated party risk may affect are evaluated in combination with the industry database, market research report and enterprise cooperation network graph, and finally the information is integrated to generate the associated risk dimension data, and the overall risk characteristics of the associated party are completely presented.

[0154] The two types of dimension data extracted in the first two steps provide basic information from the perspectives of “credit association” and “associated risk”. Only by analyzing the two can the specific risk transmission mechanism between the enterprise and the associated party be revealed, such as whether the decrease of the enterprise credit score will lead to the tightening of the cooperation credit of the associated party, or whether the credit risk of the associated party will affect the debt repayment ability of the enterprise. Credit analysis and risk exposure analysis are generally carried out separately, either only assessing the credit risk of the enterprise or only analyzing the risk exposure of the associated party, lacking systematic research on the interaction between the two, and unable to deeply explore the deep influence of credit change on the transmission of associated risk. The solution of the present application is to build a multi-dimensional associated analysis model, and to correlate the indicators such as “enterprise credit change amplitude, credit deterioration duration” in the credit association dimension data with the indicators such as “associated party risk type, exposure size growth rate” in the associated risk dimension data: Pearson correlation coefficient is used to analyze the linear correlation between the two, Granger causality test is used to determine the causal direction of credit change and associated risk exposure, regression analysis is used to quantify the influence coefficient of credit change on associated risk exposure, and finally the associated risk transmission analysis data containing risk transmission path, transmission strength and transmission cycle is generated, providing basis for accurately predicting the risk transmission trend.

[0155] According to the correlation risk conduction analysis data, the correlation risk characteristic data is acquired, the correlation risk conduction analysis data discloses a conduction mechanism, but still belongs to multi-dimensional original analysis result, and the core characteristics with clear risk indication significance need to be extracted from the multi-dimensional original analysis result, and the core characteristics are key basis for judging the severity of the correlation risk and formulating a coping strategy, are limited to identifying a single risk source, lack comprehensive and deep analysis of correlation risk characteristics, and cannot extract effective early warning signals for future risks from the conduction data, the solution of the application is that the correlation risk conduction analysis data is extracted by a data mining technology: for risk fluctuation characteristics, the standard deviation and the coefficient of variation of the exposure scale of the correlation risk in different time periods are calculated, and the risk fluctuation amplitude is obtained, for risk conduction efficiency characteristics, the average length from the occurrence of credit change to the appearance of the correlation risk exposure is counted, and the risk conduction time delay is obtained, for risk periodicity characteristics, the Fourier transform is used to analyze the periodic law of the change of the risk exposure scale with time, and the risk fluctuation period is obtained, and the extracted characteristics are integrated into structured correlation risk characteristic data, and clear characteristic dimensions are provided for subsequent time series analysis.

[0156] According to the correlation risk characteristic data, the correlation risk characteristics in different time periods are arranged in time sequence, and the time sequence data of the correlation risk characteristics is formed, the correlation risk characteristics are not fixed and unchangeable, but dynamically evolve with time, the characteristics are arranged into time sequence data according to time periods, the evolution track of the risk characteristics can be directly and intuitively displayed, and support is provided for discovering long-term trends and short-term abnormalities, the processing of the time sequence data is extremely rough, and the analysis is mostly performed on quarterly or annual summary data, the change details of the correlation risk characteristics in a finer time dimension cannot be captured, and the turning points of the risk trend cannot be accurately identified, the solution of the application is that the reasonable time period is determined first, then the characteristic indexes such as the fluctuation amplitude, the conduction time delay and the periodicity in the correlation risk characteristic data are arranged in time sequence according to the time period, the time sequence corresponding to each characteristic index is formed, then the time sequence is optimized by using a time sequence preprocessing technology, and the time dependence relationship and the potential change mode in the sequence are mined by using an ARIMA model, an LSTM network and other algorithms, and finally the correlation risk characteristic time sequence data capable of clearly reflecting the evolution law of the correlation risk characteristics with time is generated.

[0157] The trend characteristic quantity data finally generated according to the correlation risk characteristic time sequence data is quantitative presentation of the correlation risk change trend, can convert abstract time sequence data into intuitive and quantified information for decision-making, provides timely and accurate trend signals for the risk early warning system, and helps decision-makers to make response measures in advance; the risk trend prediction depends on artificial experience or simple rules, lacks systematic and scientific prediction based on historical time sequence data, and has low prediction accuracy and strong subjectivity; the solution of the present application is that a plurality of trend analysis algorithms are used to process the correlation risk characteristic time sequence data; for linear trend characteristics, a least square method is used to fit a trend line, and a trend slope is calculated; for nonlinear trend characteristics, an exponential smoothing method is used to process seasonal and periodic fluctuations in the sequence, and future trend changes are predicted; the trend quantitative results output by these algorithms are integrated to generate trend characteristic quantity data, which can not only accurately predict the change direction and amplitude of the correlation risk in a future period of time, but also provide reliability evaluation of the trend prediction, and provide scientific and reliable basis for subsequent risk monitoring and decision-making.

[0158] In one embodiment, the step of generating a comprehensive risk index according to the multi-scene basic risk characteristics and trend characteristics comprises:

[0159] S501, extracting basic risk dimension data reflecting credit and basic risk dimension data of transaction cross-scene risk according to multi-scene basic risk characteristic data;

[0160] S502, extracting trend risk dimension data reflecting the correlation risk change trend according to the trend characteristic quantity data;

[0161] S503, generating risk data fusion data according to the basic risk dimension data reflecting credit, the basic risk dimension data of transaction cross-scene risk and the trend risk dimension data;

[0162] S504, obtaining fused risk data according to the risk data fusion data;

[0163] S505, obtaining risk quantization data according to the fused risk data;

[0164] S506, obtaining comprehensive risk index data according to the risk quantization data.

[0165] In summary, the present application generates comprehensive risk index data according to multi-scene basic risk characteristics and trend characteristic quantities, and through multi-step data fusion and quantitative processing, finally obtains comprehensive risk index data that can reflect the overall risk situation, provides quantitative basis for risk management and assists decision support; in view of the complex and changeable risk environment, a single risk dimension cannot fully present the overall risk situation, especially in the fields of credit and transaction cross-scene, it is necessary to fuse data from multiple dimensions to obtain a more representative risk index, so as to accurately assess risks and respond to changing market conditions and complex business scenarios. In the prior art, risk index generation often relies on single-dimensional evaluation or only focuses on specific scenes, ignores the interaction of multi-scene and trend characteristics, and can only give static evaluation, which is difficult to cope with dynamic and complex situations; therefore, the present application innovatively proposes a comprehensive risk assessment method based on multi-scene basic risk characteristics and trend characteristic data, by extracting multi-dimensional risk data, using fusion algorithm to integrate multi-source data and quantifying it into comprehensive risk index, not only solves the problem of insufficient integration of multi-scene and multi-dimensional risk information in traditional technology, but also improves the accuracy and dynamic adaptability of risk assessment.

[0166] According to the multi-scene basic risk characteristic data, the basic risk dimension data reflecting the credit and the basic risk dimension data of the transaction cross-scene risk are extracted: the basic risk dimension data is a preliminary measurement of risk, which can reflect the core factors affecting risk, and the credit and transaction cross-scene dimension data are extracted specifically, which can ensure that the two key risk sources are covered; the extraction of multi-scene basic risk characteristics is relatively extensive, and the core risk dimensions of different scenes are not accurately distinguished, which is easy to miss the key risk information of credit or transaction cross-scene; the technical implementation needs to collect multi-scene original data first, clean and denoise the data to ensure accuracy, then use feature extraction techniques such as principal component analysis and clustering algorithm, and adjust the extraction strategy flexibly according to the needs of different scenes, analyze user credit history, payment ability and other information for credit basic risk dimension data; for transaction cross-scene risk basic risk dimension data, focus on mining risk interaction and transmission characteristics between different scenes, and finally obtain two types of subdivided basic risk dimension data.

[0167] According to the trend characteristic quantity data, trend risk dimension data reflecting the change trend of the associated risk is extracted: the risk change trend can reveal the evolution law of the risk at different times and scenes, and is the key basis for predicting the future risk trend; although the traditional technology can make trend judgment through time series analysis, it can only capture short-term data fluctuations and cannot comprehensively cover long-term change trends, especially in a multi-scene and multi-dimensional data environment, it is difficult to identify potential associated risk trends; the technical implementation adopts a method combining time series analysis and machine learning, first processes the historical trend characteristic quantity data, extracts long-term and short-term risk trend information through moving average, exponential smoothing and other methods, then introduces decision tree, support vector machine and other machine learning algorithms to analyze the relevance of these trend information and historical risk events, and finally forms trend risk dimension data that can reflect the evolution law of the associated risk over time, providing reliable support for subsequent risk prediction.

[0168] According to the basic risk dimension data reflecting credit, the basic risk dimension data of transaction cross-scene risk and the trend risk dimension data, risk data fusion data is generated: through multi-dimensional data fusion, the local limitations of single data source can be eliminated, and different dimensional risk information can be integrated, so that the comprehensive state of the risk can be more comprehensively and accurately reflected, and the comprehensiveness and prediction accuracy of risk assessment can be improved; the data fusion method of the traditional technology is relatively simple, and is mostly focused on a single dimension or scene, and the correlation between dimensions is not fully utilized, resulting in insufficient precision of the fusion result; the technical implementation adopts a weighted fusion algorithm, first determines the weighted coefficients of the dimensional data through regression analysis according to the historical risk data and expert experience, for example, the credit basic risk dimension weight is 0.4, the transaction cross-scene risk dimension weight is 0.3, and the trend risk dimension weight is 0.3; then the three types of dimensional data are weighted and summed according to the weighted coefficients, the correlation between the dimensions is fully utilized, and finally the risk data fusion data covering the multi-scene and trend change characteristics is obtained.

[0169] According to the risk data fusion data, the fused risk data is obtained: the fused risk data is the basic input for subsequent risk quantitative analysis, and needs to comprehensively reflect the comprehensive influence of different dimensional risks to ensure that it can provide a true and consistent data source for quantitative processing; the technical implementation needs to further process the risk data fusion data, first carries out data normalization, standardization and other preprocessing to eliminate data format and magnitude differences; then adopts a fusion method with strong numerical stability to check and correct abnormal values that may occur in the fusion process, ensures the accuracy and consistency of the data in the calculation process, and finally outputs the fused risk data that can truly reflect the risk status.

[0170] Risk quantification data is obtained from the integrated risk data: Risk quantification data transforms abstract risk information into actionable numerical indicators, which can clearly present the severity of the risk and provide a scientific basis for the actual operation of risk management; The technical implementation adopts professional quantitative algorithms, such as converting the integrated risk data into standardized risk values ​​through risk coefficient calculation, so that managers can intuitively understand the level of risk. For example, if a certain entity's risk quantification score is 75 points, it indicates that it is at a high risk level.

[0171] Comprehensive risk index data is obtained based on risk quantification data: The comprehensive risk index data is the final quantitative result of all risk data and is a core tool for risk management decision-making. It needs to accurately quantify the comprehensive risk level in different scenarios to provide a direct basis for enterprise risk decisions. The technical implementation adopts weighted average or multi-dimensional fusion methods. Based on risk quantification data and combined with the risk preferences of different business scenarios, the comprehensive risk index is calculated. For example, when using the weighted average method, the quantitative sub-items related to credit, transactions, and correlation trends in the risk quantification data are weighted according to preset weights to finally generate comprehensive risk index data. This index can uniformly quantify the comprehensive risk level in multiple scenarios, helping enterprises to clearly identify risk levels and thus make accurate risk response decisions.

[0172] Example 2

[0173] This application also provides a risk monitoring and multi-scenario adaptation system based on dynamic threshold early warning, such as... Figure 2 As shown, it includes:

[0174] Data acquisition module 1 is used to acquire risk trigger information under multiple scenarios, and to identify credit-related scenarios, transaction-related scenarios, and association-related scenarios based on the risk trigger information.

[0175] The first judgment module 2 is used to determine whether the credit scenario, transaction scenario, and related scenario have entered the risk monitoring state. If the risk monitoring state has entered, it obtains the structured monitoring data of multiple scenarios, and obtains the historical credit change data of the credit scenario, the real-time transaction flow data of the transaction scenario, and the related party risk exposure data of the related scenario based on the structured monitoring data.

[0176] The multi-scenario basic risk feature acquisition module 3 is used to acquire cross-scenario risk features based on historical credit change data and real-time transaction flow data, and to acquire multi-scenario basic risk features based on the cross-scenario risk features.

[0177] Trend feature quantity acquisition module 4 is used to acquire related risk features based on the historical credit change data and related party risk exposure data, and to acquire trend feature quantities based on the related risk features.

[0178] The comprehensive risk index generation module 5 is configured to generate a comprehensive risk index according to the multi-scenario basic risk features and trend feature quantities;

[0179] The second judgment module 6 is configured to judge whether the comprehensive risk index is greater than a preset safety threshold value;

[0180] If not, the monitoring is continued;

[0181] If yes, a sub-risk level is obtained according to the comprehensive risk index, and a multi-scenario adaptive sub-level early warning information is obtained according to the sub-risk level.

[0182] In one embodiment, the first judgment module 2 comprises:

[0183] The data acquisition unit is configured to split the risk trigger information under the multi-scenarios according to the scene categories to obtain credit status data corresponding to credit-scene-specific trigger indicators, transaction behavior data corresponding to transaction-scene-specific trigger indicators, and associated party risk transmission data corresponding to associated-scene-specific trigger indicators;

[0184] The correlation relationship data acquisition unit is configured to integrate the credit status data, the transaction behavior data and the associated party risk transmission data to obtain scene-indicator correlation relationship data;

[0185] The associated risk transmission abnormal information acquisition unit is configured to acquire associated risk transmission abnormal information according to the scene-indicator correlation relationship data;

[0186] The fitting degree data acquisition unit is configured to compare the real-time trigger indicator data of the credit-scene, the transaction-scene and the associated-scene with the scene trigger condition data according to the associated risk transmission abnormal information to obtain indicator-condition fitting degree data;

[0187] The determination result data acquisition unit is configured to acquire scene risk monitoring start determination result data according to the indicator-condition fitting degree data;

[0188] The judgment unit is configured to judge whether the scene risk monitoring start determination result data meets a preset regulation;

[0189] If yes, the determination result is that the fitting degree meets the requirement, and it is determined that the corresponding scene enters a risk monitoring state;

[0190] If not, the determination result is that the fitting degree does not meet the requirement, and it is determined that the scene does not enter the risk monitoring state.

[0191] Embodiment 3

[0192] As Figure 3As shown, the embodiment 3 of the present application provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above-mentioned risk monitoring method for early warning based on a dynamic threshold when executing the computer program.

[0193] The present application also provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the above-mentioned risk monitoring method for early warning based on a dynamic threshold when executed by a processor.

[0194] The above-mentioned embodiments only express several embodiments of the present application, which are described in detail and specifically, but should not be understood as a limitation to the patent scope of the present application. It should be pointed out that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, which all belong to the protection scope of the present application. Therefore, the patent protection scope of the present application should be subject to the appended claims.

Claims

1. A risk monitoring method based on dynamic threshold early warning, characterized in that, include: Obtain risk trigger information in multiple scenarios, and based on the risk trigger information, identify credit-related scenarios, transaction-related scenarios, and association-related scenarios. Determine whether credit-related scenarios, transaction-related scenarios, and association-related scenarios have entered a risk monitoring state. If they have entered a risk monitoring state, obtain structured monitoring data for multiple scenarios, and obtain historical credit change data for credit-related scenarios, real-time transaction flow data for transaction-related scenarios, and related party risk exposure data for association-related scenarios based on the structured monitoring data. Credit change trajectory data is obtained by acquiring the trajectory of credit status changes over different time periods based on historical credit change data. Based on credit change trajectory data, identify the trend and magnitude of changes in credit status to obtain credit trend feature data; Based on real-time transaction data in transaction scenarios, we extract the transaction amount distribution, transaction counterparty distribution, and time distribution behavioral characteristics to obtain transaction behavior characteristic data. Obtain cross-scenario feature correlation data based on credit trend feature data and transaction behavior feature data; Based on cross-scenario feature correlation data, core correlation features that can jointly reflect risks are selected to obtain cross-scenario risk feature data; Obtain basic risk characteristic data for multiple scenarios based on cross-scenario risk characteristic data; Based on the historical credit change data, the correlation dimension between credit status and related parties is extracted to obtain credit correlation dimension data; Based on the related party risk exposure data, the risk types, exposure scales, and impact scopes of the related parties are extracted to obtain related risk dimension data. Based on the aforementioned credit-related dimension data and related risk dimension data, obtain related risk transmission analysis data; Based on the associated risk transmission analysis data, obtain associated risk characteristic data; Based on the associated risk characteristics data, the associated risk characteristics in different time periods are arranged in time sequence to form time series data of associated risk characteristics; Generate trend characteristic data based on time series data of associated risk characteristics; A comprehensive risk index is generated based on the basic risk characteristics and trend characteristics of the multiple scenarios. Determine whether the comprehensive risk index is greater than the preset safety threshold; If it is less than, continue monitoring; If the risk level is greater than the comprehensive risk index, then the risk level is obtained based on the comprehensive risk index, and the graded early warning information matching multiple scenarios is obtained based on the risk level.

2. The risk monitoring method based on dynamic threshold early warning according to claim 1, characterized in that, The steps to determine whether credit-related scenarios, transaction-related scenarios, and association-related scenarios have entered a risk monitoring state include: Risk trigger information in multiple scenarios is broken down by scenario category to obtain credit status data corresponding to credit-related scenario-specific trigger indicators, transaction behavior data corresponding to transaction-related scenario-specific trigger indicators, and related party risk transmission data corresponding to related scenario-specific trigger indicators. By integrating credit status data, transaction behavior data, and risk transmission data from related parties, scenario-indicator correlation data is obtained. Based on the scenario-indicator correlation data, obtain abnormal information on the transmission of associated risks; Based on the abnormal information of related risk transmission, the matching degree between the real-time trigger indicator data and the scenario trigger condition data of credit scenario, transaction scenario and related scenario is compared respectively to obtain indicator-condition matching degree data; Obtain the scenario risk monitoring activation judgment result data based on indicator-condition fit data; Based on whether the scenario risk monitoring activation judgment data meets the preset regulations; If the conditions are met, the result is that the fit meets the requirements, and the corresponding scenario is determined to enter the risk monitoring state. If it does not meet the requirements, the result is that the fit is not met, and the scenario will not be put into risk monitoring for the time being.

3. The risk monitoring method based on dynamic threshold early warning according to claim 1, characterized in that, The steps of acquiring structured monitoring data from multiple scenarios, and obtaining historical credit change data for credit-related scenarios, real-time transaction flow data for transaction-related scenarios, and related party risk exposure data for related scenarios based on the structured monitoring data, include: Based on the structured monitoring data, obtain the structured monitoring data collection dimensions corresponding to credit, transaction, and association scenarios respectively; Based on the collection dimensions of structured monitoring data corresponding to credit-related scenarios, transaction-related scenarios, and association-related scenarios, obtain the data storage source information for credit-related scenarios, transaction-related scenarios, and association-related scenarios. Based on the data storage source information for credit-related scenarios, transaction-related scenarios, and related scenarios, obtain the corresponding original data for credit-related scenarios, transaction-related scenarios, and related scenarios. Field integrity checks are performed on the raw data from credit scenarios, transaction scenarios, and related scenarios to obtain standardized data for credit scenarios, transaction scenarios, and related scenarios. Historical credit change data are obtained based on the standardized data of the aforementioned credit scenarios; Real-time transaction flow data is obtained based on the standardized data of the aforementioned transaction scenarios; Risk exposure data of related parties is obtained based on the standardized data of the aforementioned related scenarios.

4. The risk monitoring method based on dynamic threshold early warning according to claim 1, characterized in that, The step of generating a comprehensive risk index based on the basic risk characteristics and trend characteristics of the multiple scenarios includes: Based on multi-scenario basic risk characteristic data, extract basic risk dimension data reflecting credit and basic risk dimension data reflecting cross-scenario transaction risks; Based on the trend characteristic data, extract the trend risk dimension data that reflects the changing trend of associated risks; Risk data fusion data is generated based on basic risk dimension data reflecting credit, basic risk dimension data reflecting cross-scenario risk of transactions, and trend risk dimension data. Obtain the merged risk data based on the risk data fusion data; Obtain risk quantification data based on the merged risk data; Comprehensive risk index data is obtained based on risk quantification data.

5. A risk monitoring and multi-scenario adaptation system based on dynamic threshold early warning, characterized in that, include: The data acquisition module is used to acquire risk trigger information in multiple scenarios, and to identify credit-related scenarios, transaction-related scenarios, and association-related scenarios based on the risk trigger information. The first judgment module is used to determine whether credit-related scenarios, transaction-related scenarios, and related scenarios have entered the risk monitoring state. If they have entered the risk monitoring state, the module obtains structured monitoring data for multiple scenarios and obtains historical credit change data for credit-related scenarios, real-time transaction flow data for transaction-related scenarios, and related party risk exposure data for related scenarios based on the structured monitoring data. The multi-scenario basic risk feature acquisition module is used to obtain the change trajectory of credit status in different time periods based on historical credit change data, and obtain credit change trajectory data. Based on credit change trajectory data, identify the trend and magnitude of changes in credit status to obtain credit trend feature data; Based on real-time transaction data in transaction scenarios, we extract the transaction amount distribution, transaction counterparty distribution, and time distribution behavioral characteristics to obtain transaction behavior characteristic data. Obtain cross-scenario feature correlation data based on credit trend feature data and transaction behavior feature data; Based on cross-scenario feature correlation data, core correlation features that can jointly reflect risks are selected to obtain cross-scenario risk feature data; Obtain basic risk characteristic data for multiple scenarios based on cross-scenario risk characteristic data; The trend feature acquisition module is used to extract the correlation dimension between credit status and related parties based on the historical credit change data, and obtain credit correlation dimension data. Based on the related party risk exposure data, the risk types, exposure scales, and impact scopes of the related parties are extracted to obtain related risk dimension data. Based on the aforementioned credit-related dimension data and related risk dimension data, obtain related risk transmission analysis data; Based on the associated risk transmission analysis data, obtain associated risk characteristic data; Based on the associated risk characteristics data, the associated risk characteristics in different time periods are arranged in time sequence to form time series data of associated risk characteristics; Generate trend characteristic data based on time series data of associated risk characteristics; The comprehensive risk index generation module is used to generate a comprehensive risk index based on the basic risk characteristics and trend characteristics of the multiple scenarios. The second judgment module is used to determine whether the comprehensive risk index is greater than the preset safety threshold. If it is less than, continue monitoring; If the risk level is greater than the comprehensive risk index, then the risk level is obtained based on the comprehensive risk index, and the graded early warning information matching multiple scenarios is obtained based on the risk level.

6. A risk monitoring and multi-scenario adaptation system based on dynamic threshold early warning according to claim 5, characterized in that, The first judgment module includes: The data acquisition unit is used to break down risk trigger information in multiple scenarios into credit status data corresponding to credit-related scenario-specific trigger indicators, transaction behavior data corresponding to transaction-related scenario-specific trigger indicators, and related party risk transmission data corresponding to related scenario-specific trigger indicators. The correlation data acquisition unit is used to integrate credit status data, transaction behavior data, and risk transmission data of related parties to obtain scenario-indicator correlation data. The associated risk transmission anomaly information acquisition unit is used to acquire associated risk transmission anomaly information based on scenario-indicator correlation data; The matching data acquisition unit is used to compare the matching degree between real-time trigger indicator data and scenario trigger condition data of credit scenario, transaction scenario and related scenario based on the abnormal information of related risk transmission, and to obtain indicator-condition matching degree data. The judgment result data acquisition unit is used to acquire the scenario risk monitoring start judgment result data based on the indicator-condition fit data. The judgment unit is used to determine whether the data from the scenario risk monitoring startup meets the preset requirements. If the conditions are met, the result is that the fit meets the requirements, and the corresponding scenario is determined to enter the risk monitoring state. If it does not meet the requirements, the result is that the fit is not met, and the scenario will not be put into risk monitoring for the time being.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images