A risk data processing method based on artificial intelligence

Abstract

The present invention provides a risk data processing method based on artificial intelligence, including collecting multi-source risk data and constructing a basic data framework of a risk data processing model based on artificial intelligence; identifying the three domains of historical risk data, real-time risk data and risk feature data in the above basic data framework; dividing the data processing module of the historical risk data domain according to the preset data processing frequency; determining the optimization scheme of risk feature extraction according to the type and association logic of risk data, dividing the data processing modules of the real-time risk data domain and the risk feature data domain; integrating the data processing modules of the above three domains to obtain the construction result of the risk data processing model based on artificial intelligence. The present invention can realize the efficient processing and feature extraction of risk data, and improve the accuracy and response speed of risk management.

Description

Technical Field

[0001] The present invention relates to the technical field of risk data processing, and more specifically, to a risk data processing method based on artificial intelligence. Background Art

[0002] Existing risk data processing technologies mainly rely on traditional data collection and analysis methods, which are often inefficient in processing multi-source, large-scale risk data and difficult to adapt to the rapidly changing market environment. In the fields of finance, insurance, and healthcare, the real-time and accuracy of risk data are crucial for decision-making, but existing technologies often fail to effectively integrate internal and external risk data, resulting in inaccurate risk assessment and prediction. In addition, traditional data processing models lack flexibility and intelligence in feature extraction and risk prediction, making it difficult to cope with complex and changing risk scenarios.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: insufficient efficiency of data fusion and processing, insufficient accuracy of risk feature extraction, and the scalability and real-time performance of the data processing model need to be improved. Summary of the invention

[0004] The present invention provides a risk data processing method based on artificial intelligence, comprising:

[0005] Collect multi-source risk data and build a basic data framework for risk data processing models based on artificial intelligence;

[0006] Identify the three domains of historical risk data, real-time risk data and risk characteristic data in the above basic data framework;

[0007] Divide the data processing modules of the historical risk data domain according to the preset data processing frequency;

[0008] According to the type and association logic of risk data, determine the optimization scheme for risk feature extraction and divide the data processing modules into real-time risk data domain and risk feature data domain;

[0009] Integrate the data processing modules of the above three domains to obtain the construction result of the risk data processing model based on artificial intelligence.

[0010] Furthermore, the basic data framework for collecting multi-source risk data and constructing an artificial intelligence-based risk data processing model further includes:

[0011] Collect risk data stored in the internal system and build an internal risk data sub-framework, including a data storage layer, a data transmission layer, and a data transfer channel composed of a data conversion layer;

[0012] The data storage layer and the data transmission layer share a data identification system;

[0013] Collecting externally accessed risk data and constructing an external risk data subframe, wherein the external risk data subframe includes a data collection interface and a plurality of risk data formats transmitted on the data collection interface using a specific protocol;

[0014] Within the above-mentioned data flow channel, the above-mentioned external risk data sub-framework is integrated according to preset rules to complete the construction of the basic data framework of the risk data processing model based on artificial intelligence.

[0015] Furthermore, in the basic data framework of the risk data processing model based on artificial intelligence, the identification conversion rules between risk data from different sources are unified standards, and the fusion method between different types of risk data is based on risk weight allocation;

[0016] Each of the risk data subframes is an extensible structure, with an updateable data module inside and the same number of risk data interfaces with the same specifications connected to the outside;

[0017] Moreover, the number of risk data interfaces shall be ≥2.

[0018] Furthermore, the data storage layer and the data transmission layer have the same architecture but different functions;

[0019] The fusion method of risk data subframes is as follows: each type of risk data subframe is orderly integrated along the data flow channel based on the data identification system, and risk data of different types are staggered and integrated according to risk weights, with a total of n types;

[0020] The n is obtained by the following formula ];

[0021] In the formula, m is the processing capacity of the data flow channel, k is the data volume of a single risk data subframe, and [] is the rounding operation;

[0022] The fusion weight w between risk data from different sources is determined by the following inequality:

[0023]

[0024]

[0025] Where W is the data processing accuracy of a single risk data subframe, and w is the risk data fusion weight.

[0026] Furthermore, the step of identifying the three domains of historical risk data domain, real-time risk data domain and risk characteristic data domain in the above basic data framework further includes:

[0027] Obtain any data slice of the basic data framework of the AI-based risk data processing model;

[0028] Identify the scope of each historical risk data subframe in the data slice and the area where the corresponding data processing flow is located, use the area within the scope as the processing domain of the historical risk data, and arrange the processing domains of all historical risk data to form the historical risk data domain of the basic data framework;

[0029] Identify the data source of each real-time risk data subframe in the data slice, construct an auxiliary boundary 1 according to the historical risk data domain boundary, connect adjacent risk data interface edge points outside the auxiliary boundary 1 with a radius of k+d to construct an auxiliary boundary 2, and use the area between the auxiliary boundaries 1 and 2 as the processing domain of the real-time risk data. All the processing domains of the real-time risk data are arranged to form the real-time risk data domain of the basic data framework;

[0030] All historical risk data fields and real-time risk data fields in the data slice are removed to obtain the risk characteristic data fields of the basic data framework.

[0031] Furthermore, the step of dividing the data processing modules of the historical risk data domain according to the preset data processing frequency further includes:

[0032] In the historical risk data domain, a data screening area is made with the data identifier of each historical risk data subframe as the center and W as the processing accuracy range to obtain a circumscribed regular polygon of the screening area;

[0033] Among them, for the historical risk data subframe within the preset distance from the core of the data flow channel, an circumscribed regular pentagon is obtained, and for the historical risk data subframe outside the preset distance from the core of the data flow channel, an circumscribed regular hexagon is obtained;

[0034] The data identifier of each historical risk data subframe is used as the division center, and the corresponding circumscribed regular polygon is used as the boundary. The hash algorithm is used to evenly divide 6p data subsets outward in the data slice, p ≥ 2, until the historical risk data domain in each circumscribed regular polygon is divided into the same number of submodules;

[0035] According to the rule that the size of the largest data block is smaller than the smallest risk factor size in the risk assessment process and the number of data points at the borders of adjacent submodules is consistent, hierarchical data processing is performed on each submodule;

[0036] The historical risk data domain after layered data processing is expanded along the data flow direction to the bottom plane of the data flow channel, and then the expanded structure is divided into equal intervals along the data dimension with an interval length of W to complete the division of the data processing modules of the historical risk data domain.

[0037] Furthermore, the optimization scheme for risk feature extraction is as follows: the maximum size of the first layer of data processing close to the historical risk data domain in the real-time risk data domain does not exceed 0.01MB; and, on the data processing boundary of each risk data subframe, the number of data points N wide on the width of the risk data interface and the number of data points N thick on the thickness of the risk data interface satisfy In the formula, q represents the width of the risk data interface;

[0038] The number of data points Narc on the arc length between adjacent risk data interfaces and the number of data points Nthick on the thickness of the risk data interface satisfy .

[0039] Further, the data processing module of the real-time risk data domain is divided by the following steps: identifying all endpoints on the inner contour of each real-time risk data domain close to the contour of the historical risk data domain;

[0040] The data source of each real-time risk data subframe is used as the division center, and the outer contour of the real-time risk data domain is used as the boundary. A clustering algorithm is used to evenly divide the line segments in the data slice outward by twice the number of risk data interfaces, so that the intersection set of the division line segments and the outer contour contains all the above endpoints;

[0041] Remove all the endpoints in the above intersection set, construct two parallel lines connecting the remaining intersections and the division center as auxiliary lines, so that one end of each auxiliary line coincides with the position of the nearest endpoint, and the other end is on the outer contour of the real-time risk data domain, and divide each real-time risk data domain into 4 times the number of risk data interfaces. Processing submodules;

[0042] According to the rule that the size of the largest data block is smaller than the smallest risk factor size in the risk assessment process, the number of data points at the boundaries of adjacent processing submodules is consistent, and the number of data points at the boundaries of the real-time risk data domain and the historical risk data domain is consistent, hierarchical data processing is performed on each processing submodule;

[0043] According to the optimization scheme of risk feature extraction, the real-time risk data domain after the layered data processing is completed is further optimized;

[0044] The further optimized real-time risk data domain is extended along the data flow direction to the bottom plane of the data flow channel, and then the extended structure is divided into equal intervals along the data dimension with an interval length of W to complete the division of the data processing modules of the real-time risk data domain.

[0045] Furthermore, the data processing module of the risk feature data domain is divided through the following steps:

[0046] After the data processing module division of the real-time risk data domain and the historical risk data domain is completed, the positions of all data points on the boundary of the real-time risk data domain and the historical risk data domain that overlap with the risk feature data domain are obtained;

[0047] The data source of each risk feature data subframe is used as the division center, and the outer contour of the risk feature data domain is used as the boundary. A decision tree algorithm is used to divide line segments in the data slice to the intersection of each risk data interface and the risk feature data domain, and the risk feature data domain of each risk feature data subframe is divided into processing submodules twice the number of risk data interfaces;

[0048] According to the rule that the number of data points on the overlapping boundaries of the risk feature data domain, the real-time risk data domain, and the historical risk data domain is consistent, hierarchical data processing is performed on each processing submodule;

[0049] The risk feature data domain that has completed the hierarchical data processing is expanded along the data flow direction to the bottom plane of the data flow channel, and then the expanded structure is divided into equal intervals along the data dimension with an interval length of W to complete the division of the data processing modules of the risk feature data domain.

[0050] Furthermore, through data backup and recovery technology, the processing of key data modules in the risk feature data domain, the real-time risk data domain, and the historical risk data domain is completed;

[0051] The interval length W is less than or equal to 2MB and no more than one tenth of the total amount of risk data;

[0052] Furthermore, the delay in data transmission between risk data subframes shall not exceed 10 milliseconds;

[0053] In the risk feature data domain, the data processing of each type of risk feature data axis includes at least 5 layers;

[0054] The maximum scale of data processing shall not be greater than half of the smallest risk factor scale in the risk assessment process.

[0055] The above-mentioned embodiments of the present invention have at least the following beneficial effects: by constructing a risk data processing model based on artificial intelligence, the present invention can collect and integrate multi-source risk data, including risk data stored in the internal system and risk data accessed externally, thereby improving the comprehensiveness and accuracy of data processing. The model can achieve seamless connection and efficient integration between risk data of different sources and types through a unified data identification system and fusion rules, thereby enhancing the manageability and usability of risk data.

[0056] In addition, the method of the present invention can dynamically divide the data processing modules of the historical risk data domain, the real-time risk data domain and the risk feature data domain according to the preset data processing frequency and the type and association logic of the risk data, and optimize the risk feature extraction scheme. This dynamic division and optimization mechanism can improve the efficiency and effect of risk data processing, making risk assessment and prediction more accurate, and can also enhance the adaptability and flexibility of the model to cope with the ever-changing risk environment and business needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, in which:

[0058] Figure 1 A flowchart of a risk data processing method based on artificial intelligence provided in accordance with an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0060] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, device, apparatus, method or computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0061] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0062] Reference below Figure 1 , Figure 1 The following is a flow chart of a risk data processing method based on artificial intelligence provided by an embodiment of the present invention. Figure 1 As shown, a risk data processing method 100 based on artificial intelligence includes:

[0063] Step 101, collecting multi-source risk data and building a basic data framework of a risk data processing model based on artificial intelligence;

[0064] Step 102, identifying the three domains of historical risk data domain, real-time risk data domain and risk characteristic data domain in the above basic data framework;

[0065] Step 103, dividing the data processing module of the historical risk data domain according to the preset data processing frequency;

[0066] Step 104, determining an optimization scheme for risk feature extraction according to the type and association logic of risk data, and dividing the data processing modules of the real-time risk data domain and the risk feature data domain;

[0067] Step 105, integrating the data processing modules of the above three domains to obtain the construction result of the risk data processing model based on artificial intelligence.

[0068] It should be noted that this method involves collecting multi-source risk data and building a basic data framework for an AI-based risk data processing model. In this process, multi-source risk data refers to risk data from different sources, such as internal system data of financial institutions and external market data. The basic data framework refers to the data structure that constitutes the core of the risk data processing model, which includes the data storage layer, the data transmission layer, and the data conversion layer.

[0069] Specifically, the process of collecting multi-source risk data includes building an internal risk data subframe from risk data stored in the internal system, and building an external risk data subframe from risk data accessed from the outside. The internal risk data subframe involves a data flow channel consisting of a data storage layer, a data transmission layer, and a data conversion layer. These layers share a data identification system to ensure data consistency and traceability. The external risk data subframe includes a data collection interface and transmission protocols for multiple risk data formats.

[0070] Preferably, the construction of the basic data framework can integrate the external risk data subframe into the internal risk data subframe through preset rules. This integration method ensures the effective integration of different types of risk data based on risk weight allocation. Each risk data subframe is an extensible structure with an updateable data module inside and the same number of risk data interfaces with the same specifications connected to the outside. The number of risk data interfaces is at least 2 to meet the access requirements of different data sources.

[0071] In some embodiments, the method of collecting multi-source risk data and constructing a basic data framework for an artificial intelligence-based risk data processing model further includes:

[0072] Collect risk data stored in the internal system and build an internal risk data sub-framework, including a data storage layer, a data transmission layer, and a data transfer channel composed of a data conversion layer;

[0073] The data storage layer and the data transmission layer share a data identification system;

[0074] Collecting externally accessed risk data and constructing an external risk data subframe, wherein the external risk data subframe includes a data collection interface and a plurality of risk data formats transmitted on the data collection interface using a specific protocol;

[0075] Within the above-mentioned data flow channel, the above-mentioned external risk data sub-framework is integrated according to preset rules to complete the construction of the basic data framework of the risk data processing model based on artificial intelligence.

[0076] It should be noted that this implementation describes in detail the process of collecting multi-source risk data and building a basic data framework for an artificial intelligence-based risk data processing model. In this process, the risk data stored in the internal system refers to the data generated and maintained within the financial institution, such as transaction records, customer information, etc. Externally accessed risk data refers to data from outside the financial institution, such as market conditions, economic indicators, etc. The basic data framework is the data structure that constitutes the core of the risk data processing model, which includes a data storage layer, a data transmission layer, and a data conversion layer.

[0077] Specifically, collecting risk data stored in the internal system involves building an internal risk data subframework, which consists of a data flow channel consisting of a data storage layer, a data transmission layer, and a data conversion layer. The data storage layer is responsible for storing data, the data transmission layer is responsible for moving data, and the data conversion layer is responsible for converting data formats to adapt to different usage scenarios. These layers share a data identification system to ensure the consistency and traceability of data throughout the entire flow process. Collecting risk data from external access involves building an external risk data subframework, which includes a data collection interface and a transmission protocol that supports multiple risk data formats.

[0078] Preferably, the construction of the basic data framework can integrate the external risk data subframe into the internal risk data subframe through preset rules. This integration method ensures the effective integration of different types of risk data based on risk weight allocation. Each risk data subframe is an extensible structure with an updateable data module inside and the same number of risk data interfaces with the same specifications connected to the outside. The number of risk data interfaces is at least 2 to meet the access requirements of different data sources.

[0079] Furthermore, in practical applications, the rules of the data identification system, as well as the processing capacity of the data flow channel and the data volume of a single risk data subframe can be adjusted according to specific business needs and data volume to optimize data integration and processing efficiency.

[0080] In some embodiments, in the basic data framework of the risk data processing model based on artificial intelligence, the identification conversion rules between risk data from different sources are unified standards, and the fusion method between different types of risk data is based on risk weight allocation;

[0081] Each of the risk data subframes is an extensible structure, with an updateable data module inside and the same number of risk data interfaces with the same specifications connected to the outside;

[0082] Moreover, the number of risk data interfaces shall be ≥2.

[0083] It should be noted that this implementation further elaborates on the identification conversion rules and fusion methods between risk data from different sources in the basic data framework of the risk data processing model based on artificial intelligence. Here, the unified standard refers to the common rules followed by all risk data from different sources when performing identification conversion to ensure data consistency and compatibility. Risk weight refers to the weight assigned according to the importance and impact of risk data, which is used to determine the priority of data fusion.

[0084] Specifically, the identification conversion rules between risk data from different sources in the basic data framework are unified standards, which means that whether it is internal or external risk data, it needs to be identified and converted according to the same rules when entering the basic data framework to ensure the consistency and comparability of the data.

[0085] More specifically, the fusion method between different types of risk data is based on risk weight allocation, that is, the order and method of data fusion are determined according to the importance of the data and the degree of impact on risk assessment. Each risk data subframe is an extensible structure, with an updateable data module inside and the same number of risk data interfaces with the same specifications outside, which allows the model to be flexibly expanded as business needs change.

[0086] Preferably, the number of risk data interfaces is ≥ 2, which ensures that the model can process multiple data sources simultaneously and enhances the flexibility and robustness of data processing. In practical applications, the number and specifications of risk data interfaces can be adjusted according to the diversity of data sources and business needs.

[0087] Furthermore, the width and thickness of the risk data interface can be adjusted according to data flow and processing requirements to optimize data transmission efficiency. In the allocation of risk weights, historical data analysis and expert experience can be used to determine the rationality and effectiveness of weight allocation.

[0088] In some embodiments, the data storage layer and the data transmission layer have the same architecture but different functions;

[0089] The fusion method of risk data subframes is as follows: each type of risk data subframe is orderly integrated along the data flow channel based on the data identification system, and risk data of different types are staggered and integrated according to risk weights, with a total of n types;

[0090] The n is obtained by the following formula ];

[0091] In the formula, m is the processing capacity of the data flow channel, k is the data volume of a single risk data subframe, and [] is the rounding operation;

[0092] The fusion weight w between risk data from different sources is determined by the following inequality:

[0093]

[0094]

[0095] Where W is the data processing accuracy of a single risk data subframe, and w is the risk data fusion weight.

[0096] It should be noted that this implementation describes in detail the architecture of the data storage layer and the data transmission layer and the integration of the risk data subframework. Here, the data storage layer refers to the system layer used to store risk data, and the data transmission layer refers to the layer responsible for transmitting data within the system or between systems. The risk data subframework refers to the independent but interrelated data modules that constitute the basic data framework.

[0097] Specifically, although the data storage layer and the data transmission layer have the same architecture, their functions are different. The data storage layer is responsible for persistently storing risk data to ensure the security and accessibility of the data, while the data transmission layer is responsible for moving data from one location to another for further processing or analysis.

[0098] More specifically, the integration of risk data subframes is orderly integrated along the data flow channel through the data identification system, which means that each type of risk data subframe will find the corresponding position in the data flow channel according to its data identification and integrate with other subframes. Different types of risk data are interlaced and integrated according to risk weights, with a total of n types, where n is calculated by a specific formula, which represents the ratio of the processing capacity of the data flow channel to the data volume of a single risk data subframe.

[0099] Preferably, the fusion weight between risk data from different sources is determined by a specific inequality, which involves the data processing accuracy W of a single risk data subframe and the risk data fusion weight w. In practical applications, the value of n can be adjusted according to the actual data flow channel processing capacity and the data volume of a single risk data subframe to optimize the efficiency of data fusion. At the same time, the determination of the fusion weight w can also be adjusted according to the actual risk assessment requirements and the data processing accuracy W to ensure the accuracy and effectiveness of data fusion.

[0100] Furthermore, different data fusion algorithms or technologies can be used to achieve orderly fusion of risk data sub-frameworks, such as machine learning algorithms or optimization algorithms, to further improve the level of intelligence in data processing.

[0101] In some embodiments, the step of identifying the three domains of historical risk data domain, real-time risk data domain and risk characteristic data domain in the above basic data framework further includes:

[0102] Obtain any data slice of the basic data framework of the AI-based risk data processing model;

[0103] Identify the scope of each historical risk data subframe in the data slice and the area where the corresponding data processing flow is located, use the area within the scope as the processing domain of the historical risk data, and arrange the processing domains of all historical risk data to form the historical risk data domain of the basic data framework;

[0104] Identify the data source of each real-time risk data subframe in the data slice, construct an auxiliary boundary one according to the historical risk data domain boundary, connect adjacent risk data interface edge points outside the auxiliary boundary one with the k+d radius to construct an auxiliary boundary two, and use the area between the auxiliary boundaries one and two as the processing domain of the real-time risk data. All the processing domains of the real-time risk data are arranged to form the real-time risk data domain of the basic data framework;

[0105] All historical risk data fields and real-time risk data fields in the data slice are removed to obtain the risk characteristic data fields of the basic data framework.

[0106] It should be noted that this implementation method describes in detail how to identify the historical risk data domain, real-time risk data domain and risk feature data domain in the basic data framework. Here, data slices refer to data sets containing specific information extracted from the basic data framework for further analysis and processing. The historical risk data domain refers to a specific area containing historical risk data and its processing flow.

[0107] Specifically, the implementation method first obtains any data slice of the basic data framework, and then identifies the range formed by each historical risk data subframe in the data slice and the area where the corresponding data processing flow is located. The area within this range is regarded as the processing domain of the historical risk data, and all the processing domains of the historical risk data are arranged together to form the historical risk data domain of the basic data framework.

[0108] More specifically, for the identification of the real-time risk data domain, it is necessary to construct an auxiliary boundary one based on the boundary of the historical risk data domain, and to connect the adjacent risk data interface edge points with a specific radius to construct an auxiliary boundary two. The area between auxiliary boundaries one and two is the processing domain of the real-time risk data, and all the processing domains of the real-time risk data are arranged to form the real-time risk data domain.

[0109] Preferably, for the identification of risk characteristic data domains, it is necessary to remove all historical risk data domains and real-time risk data domains in the data slice, and the remaining part is the risk characteristic data domain. In actual operation, automated data processing tools can be used to assist in identifying these data domains, such as using artificial intelligence algorithms to analyze data slices and automatically identify and mark historical risk data domains, real-time risk data domains, and risk characteristic data domains.

[0110] Furthermore, parameters such as the size of data slices, identification criteria for historical risk data subframes, etc. can be set to accommodate different data sizes and complexities. Different data identification techniques, such as pattern recognition or machine learning, can also be considered to improve the accuracy and efficiency of identification.

[0111] In some embodiments, the step of dividing the data processing module of the historical risk data domain according to the preset data processing frequency further includes:

[0112] In the historical risk data domain, a data screening area is made with the data identifier of each historical risk data subframe as the center and W as the processing accuracy range to obtain a circumscribed regular polygon of the screening area;

[0113] Among them, for the historical risk data subframe within the preset distance from the core of the data flow channel, an circumscribed regular pentagon is obtained, and for the historical risk data subframe outside the preset distance from the core of the data flow channel, an circumscribed regular hexagon is obtained;

[0114] The data identifier of each historical risk data subframe is used as the division center, and the corresponding circumscribed regular polygon is used as the boundary. The hash algorithm is used to evenly divide 6p data subsets outward in the data slice, p ≥ 2, until the historical risk data domain in each circumscribed regular polygon is divided into the same number of submodules;

[0115] According to the rule that the size of the largest data block is smaller than the smallest risk factor size in the risk assessment process and the number of data points at the borders of adjacent submodules is consistent, hierarchical data processing is performed on each submodule;

[0116] The historical risk data domain after layered data processing is expanded along the data flow direction to the bottom plane of the data flow channel, and then the expanded structure is divided into equal intervals along the data dimension with an interval length of W to complete the division of the data processing modules of the historical risk data domain.

[0117] It should be noted that this implementation describes how to divide the data processing module of the historical risk data domain according to the preset data processing frequency. Here, the data processing module refers to the manageable data units into which the historical risk data domain is subdivided for subsequent analysis and processing. Data identification refers to a set of information used to uniquely identify data, which can help the system quickly locate and process specific data.

[0118] Specifically, the implementation method mentions that within the historical risk data domain, the data identifier of each historical risk data subframe is used as the center, and the data processing accuracy W is used as the processing accuracy range to make a data screening area, and obtain the circumscribed regular polygon of the screening area. This means that the data identifier of each historical risk data subframe is used as a reference point, and around this point, an area is determined according to the processing accuracy W, and the boundary of the area forms a regular polygon. For the historical risk data subframe within the preset distance from the core of the data flow channel, the circumscribed regular pentagon is obtained; for the historical risk data subframe outside the preset distance from the core of the data flow channel, the circumscribed regular hexagon is obtained.

[0119] Preferably, the data identifier of each historical risk data subframe is used as the division center, and the corresponding circumscribed regular polygon is used as the boundary. The hash algorithm is used to evenly divide 6p data subsets outward in the data slice, where p ≥ 2, until the historical risk data domain in each circumscribed regular polygon is divided into the same number of submodules. According to the rule that the size of the largest data block is smaller than the smallest risk factor size in the risk assessment process, and the number of data points at the boundaries of adjacent submodules is consistent, hierarchical data processing is performed on each submodule.

[0120] Furthermore, the division of data processing modules not only considers the geometric distribution of data, but also the actual scale and boundary consistency of data to ensure the uniformity and effectiveness of data processing. In actual operation, the shape and size of the circumscribed regular polygon and the number of data subset divisions 6p can be adjusted according to the core distance of the data flow channel and the specific value of the data processing accuracy W to adapt to different data processing requirements and optimize performance.

[0121] In some embodiments, the optimization scheme for risk feature extraction is: the maximum scale of the first layer of data processing close to the historical risk data domain in the real-time risk data domain does not exceed 0.01MB; and on the data processing boundary of each risk data subframe, the number of data points N wide on the width of the risk data interface and the number of data points N thick on the thickness of the risk data interface meet In the formula, q represents the width of the risk data interface;

[0122] The number of data points Narc on the arc length between adjacent risk data interfaces and the number of data points Nthick on the thickness of the risk data interface satisfy .

[0123] It should be noted that this implementation describes in detail the optimization scheme for risk feature extraction, especially the scale limit of data processing in the real-time risk data domain and the proportional relationship between the number of data points on the data processing boundary of the risk data subframe. Here, the real-time risk data domain refers to the area containing the risk data updated in real time, and the risk data subframe refers to the basic unit constituting the risk data domain, which is responsible for processing and transmitting the risk data.

[0124] Specifically, the implementation method mentions that the maximum size of the first layer of data processing in the real-time risk data domain close to the historical risk data domain does not exceed 0.01MB. This means that in the real-time risk data domain, the amount of data in the data processing layer closest to the historical risk data domain is limited to a very small range to ensure the real-time data processing capability and the response speed of the system.

[0125] More specifically, on the data processing boundary of each risk data subframe, the number of data points Nwidth on the width of the risk data interface and the number of data points Nthickness on the thickness of the risk data interface satisfy the proportional relationship of Nwidth / Nthickness, and the number of data points Narc on the arc length between adjacent risk data interfaces and the number of data points Nthickness on the thickness of the risk data interface satisfy the proportional relationship of Narc / Nthickness. These proportional relationships ensure the uniformity and consistency of data processing.

[0126] Preferably, the optimization scheme for risk feature extraction can be further refined into specific parameter settings for data processing in the real-time risk data domain. For example, the maximum scale limit of the first-layer data processing can be adjusted according to the actual data processing requirements and system performance to adapt it to different data flows and processing capabilities. At the same time, the design of the risk data interface can be optimized, and the specific values ​​of N width, N thickness, and N arc can be adjusted to meet different data processing accuracy and efficiency requirements.

[0127] Furthermore, we can consider using different data compression techniques and algorithms to further reduce the scale of data processing, or adopt parallel processing technology to increase the speed of data processing. These alternatives can be customized and optimized according to specific application scenarios and business needs.

[0128] In some embodiments, the data processing module of the real-time risk data domain is divided into the following steps:

[0129] Identify all endpoints on the inner contour of each real-time risk data domain that are close to the contour of the historical risk data domain;

[0130] The data source of each real-time risk data subframe is used as the division center, and the outer contour of the real-time risk data domain is used as the boundary. A clustering algorithm is used to evenly divide the line segments in the data slice outward by twice the number of risk data interfaces, so that the intersection set of the division line segments and the outer contour contains all the above endpoints;

[0131] Remove all the endpoints in the above intersection set, construct two parallel lines connecting the remaining intersections and the division center as auxiliary lines, so that one end of each auxiliary line coincides with the position of the nearest endpoint, and the other end is on the outer contour of the real-time risk data domain, and divide each real-time risk data domain into 4 times the number of risk data interfaces. Processing submodules;

[0132] According to the rule that the size of the largest data block is smaller than the smallest risk factor size in the risk assessment process, the number of data points at the boundaries of adjacent processing submodules is consistent, and the number of data points at the boundaries of the real-time risk data domain and the historical risk data domain is consistent, hierarchical data processing is performed on each processing submodule;

[0133] According to the optimization scheme of risk feature extraction, the real-time risk data domain after the layered data processing is completed is further optimized;

[0134] The further optimized real-time risk data domain is extended along the data flow direction to the bottom plane of the data flow channel, and then the extended structure is divided into equal intervals along the data dimension with an interval length of W to complete the division of the data processing modules of the real-time risk data domain.

[0135] It should be noted that this embodiment describes a method for dividing the data processing modules of the real-time risk data domain. Here, the real-time risk data domain refers to the area containing the risk data updated in real time, and the data processing module refers to the manageable data units into which the real-time risk data domain is subdivided for subsequent analysis and processing. The inner contour refers to the inner boundary of the real-time risk data domain, and the outer contour refers to the outer boundary of the real-time risk data domain.

[0136] Specifically, the implementation method mentions identifying all endpoints on the inner contour of each real-time risk data domain close to the contour of the historical risk data domain, and using the data source of each real-time risk data subframe as the division center, and the outer contour of the real-time risk data domain as the boundary, using a clustering algorithm to evenly divide the line segments in the data slice outward by twice the number of risk data interfaces, so that the intersection set of the dividing line segments and the outer contour contains all the above endpoints. This means that through the clustering algorithm, the real-time risk data domain can be divided into multiple processing submodules, each of which is associated with an endpoint, and the boundaries of these submodules intersect with the outer contour.

[0137] Preferably, for the data processing module division of the real-time risk data domain, the operation steps can be further refined. For example, after removing all endpoints in the intersection set, two parallel lines connecting the remaining intersections and the division center can be constructed as auxiliary lines, so that one end of each auxiliary line coincides with the position of the nearest endpoint and the other end is on the outer contour of the real-time risk data domain. In this way, each real-time risk data domain can be divided into 4 processing submodules as many as the number of risk data interfaces.

[0138] Furthermore, hierarchical data processing can be performed on each processing submodule according to the rule that the size of the largest data block is smaller than the smallest risk factor size in the risk assessment process, the number of data points at the boundaries of adjacent processing submodules is consistent, and the number of data points at the boundaries of the real-time risk data domain and the historical risk data domain are consistent. According to the optimization scheme for risk feature extraction, the real-time risk data domain after the hierarchical data processing is completed is further optimized. Finally, the optimized real-time risk data domain is extended along the data flow direction to the bottom plane of the data flow channel, and then the extended structure is divided into equal intervals along the data dimension, with an interval length of W, to complete the division of the data processing modules of the real-time risk data domain.

[0139] Furthermore, in actual operation, the parameters of the clustering algorithm can be adjusted according to the characteristics of the data and processing requirements, or other data processing algorithms, such as decision trees, neural networks, etc., can be selected to improve the accuracy and efficiency of data processing.

[0140] In some embodiments, the data processing module of the risk feature data domain is divided into the following steps:

[0141] After the data processing module division of the real-time risk data domain and the historical risk data domain is completed, the positions of all data points on the boundary of the real-time risk data domain and the historical risk data domain that overlap with the risk feature data domain are obtained;

[0142] The data source of each risk feature data subframe is used as the division center, and the outer contour of the risk feature data domain is used as the boundary. A decision tree algorithm is used to divide line segments in the data slice to the intersection of each risk data interface and the risk feature data domain, and the risk feature data domain of each risk feature data subframe is divided into processing submodules twice the number of risk data interfaces;

[0143] According to the rule that the number of data points on the overlapping boundaries of the risk feature data domain, the real-time risk data domain, and the historical risk data domain is consistent, hierarchical data processing is performed on each processing submodule;

[0144] The risk feature data domain that has completed the hierarchical data processing is expanded along the data flow direction to the bottom plane of the data flow channel, and then the expanded structure is divided into equal intervals along the data dimension with an interval length of W to complete the division of the data processing modules of the risk feature data domain.

[0145] It should be noted that this embodiment describes a method for dividing the data processing modules of the risk characteristic data domain. Here, the risk characteristic data domain refers to a specific area containing risk characteristic data, which is essential for risk assessment and decision-making. The data processing module refers to the manageable data units into which the risk characteristic data domain is subdivided for subsequent analysis and processing.

[0146] Specifically, the implementation method mentions that after the division of the data processing modules of the real-time risk data domain and the historical risk data domain is completed, the positions of all data points on the boundary of the two domains that overlap with the risk feature data domain are obtained. This means that it is necessary to determine the boundary points where the real-time risk data domain and the historical risk data domain intersect with the risk feature data domain, and these points will serve as the basis for the subsequent division of processing submodules. Next, taking the data source of each risk feature data subframe as the division center and the outer contour of the risk feature data domain as the boundary, a decision tree algorithm is used to divide line segments in the data slice to each intersection of the risk data interface and the risk feature data domain, and the risk feature data domain of each risk feature data subframe is divided into processing submodules that are twice the number of risk data interfaces.

[0147] Preferably, the operation steps can be further refined for the division of data processing modules of the risk characteristic data domain. For example, according to the rule that the number of data points on the overlapping boundaries of the risk characteristic data domain, the real-time risk data domain, and the historical risk data domain are consistent, hierarchical data processing can be performed on each processing submodule separately. This means that the number of data points in each processing submodule should be kept consistent to ensure the uniformity and consistency of data processing. The risk characteristic data domain that has completed the hierarchical data processing is expanded along the data flow direction to the bottom plane of the data flow channel, and then the expanded structure is divided into equal intervals along the data dimension, with an interval length of W, to complete the division of the data processing modules of the risk characteristic data domain.

[0148] Furthermore, in actual operation, the parameters of the decision tree algorithm can be adjusted according to the characteristics of the data and processing requirements, or other data processing algorithms, such as support vector machines, random forests, etc., can be selected to improve the accuracy and efficiency of data processing.

[0149] In some embodiments, the processing of key data modules in the risk feature data domain, the real-time risk data domain, and the historical risk data domain is completed through data backup and recovery technology;

[0150] The interval length W is less than or equal to 2MB and no more than one tenth of the total amount of risk data;

[0151] Furthermore, the delay in data transmission between risk data subframes shall not exceed 10 milliseconds;

[0152] In the risk feature data domain, the data processing of each type of risk feature data axis includes at least 5 layers;

[0153] The maximum scale of data processing shall not be greater than half of the smallest risk factor scale in the risk assessment process.

[0154] Formula description: Assume that the risk data processing model based on artificial intelligence is ; The historical risk data domain is , the real-time risk data domain is , the risk characteristic data domain is ; The data storage layer is , the data transmission layer is , the data conversion layer is ; The data acquisition interface is ; The data processing module is ; The data processing frequency is ; Risk data type is ;The number of risk data interfaces is ; The data processing accuracy is ; Risk data fusion weight is ; The width of the risk data interface is , the thickness of the risk data interface is .

[0155] It should be noted that this implementation describes how to complete the processing of key data modules in the risk feature data domain, the real-time risk data domain, and the historical risk data domain through data backup and recovery technology. Here, data backup and recovery technology refers to technical means for protecting data from accidental loss or damage, including data replication, archiving, and recovery when needed.

[0156] Specifically, the implementation method mentions that the size of the interval length W is less than or equal to 2MB, and not more than one tenth of the total amount of risk data. This means that when dividing the data processing modules, the size of each module is controlled within a smaller range for easy management and processing. At the same time, the delay time of data transmission between risk data subframes does not exceed 10 milliseconds, which ensures the real-time nature of data processing.

[0157] More specifically, in the risk feature data domain, the data processing of each risk feature data axis includes at least 5 layers, which ensures the depth and meticulousness of data processing. The maximum scale of data processing is no more than half of the minimum risk factor scale in the risk assessment process, which helps to maintain the accuracy and effectiveness of data processing.

[0158] Preferably, the operation steps can be further refined for the implementation of data backup and recovery technology. For example, an automated backup strategy can be set to regularly copy key data modules to a secure storage location to prevent data loss. At the same time, a fast recovery process can be implemented to quickly restore data when data is damaged or lost. In addition, advanced data compression technology can be used to reduce the size of backup data, thereby saving storage space and improving backup efficiency.

[0159] Furthermore, in terms of data transmission, network infrastructure can be optimized and high-speed transmission protocols can be used to ensure low latency in data transmission. For the processing of risk feature data, multi-level analysis methods can be adopted, such as using machine learning technology to conduct in-depth analysis of data to extract more refined risk features. These alternatives can be adjusted and optimized according to specific business needs and technical environment.

[0160] The above-mentioned embodiments of the present invention have the following beneficial effects: The risk data processing method based on artificial intelligence described in the present invention can improve the integration and processing efficiency of risk data. By constructing a basic data framework including historical risk data domain, real-time risk data domain and risk feature data domain, the method can accurately identify and divide different risk data domains to achieve refined management of risk data. In addition, by presetting the data processing frequency and the risk data type association logic, the method can optimize the risk feature extraction scheme and improve the accuracy of risk prediction.

[0161] Furthermore, this technology can ensure the scalability and flexibility of the risk data processing model. Each risk data subframe is an extensible structure, allowing the update of internal data modules and the connection of external risk data interfaces, which can enhance the model's ability to adapt to different data sources and formats. At the same time, this method can ensure the security of key data modules, reduce data transmission delays, and ensure the real-time and reliability of data processing through data backup and recovery technology. The application of these technologies can effectively improve the overall effectiveness of risk management and reduce losses caused by potential risks.

[0162] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0163] The above descriptions are only some preferred embodiments of the present invention and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A risk data processing method based on artificial intelligence, characterized in that: The steps include: Collect multi-source risk data and build a basic data framework for risk data processing models based on artificial intelligence; Identify the three domains of historical risk data, real-time risk data and risk characteristic data in the above basic data framework; Divide the data processing modules of the historical risk data domain according to the preset data processing frequency; According to the type and association logic of risk data, determine the optimization scheme for risk feature extraction and divide the data processing modules into real-time risk data domain and risk feature data domain; Integrate the data processing modules of the above three domains to obtain the construction results of the risk data processing model based on artificial intelligence; The basic data framework for collecting multi-source risk data and constructing an artificial intelligence-based risk data processing model further includes: Collect risk data stored in the internal system and build an internal risk data sub-framework, including a data storage layer, a data transmission layer, and a data transfer channel composed of a data conversion layer; The data storage layer and the data transmission layer share a data identification system; Collecting risk data from external access and constructing an external risk data subframe, wherein the external risk data subframe includes a data collection interface and a plurality of risk data formats transmitted on the data collection interface using a specific protocol; In the above data flow channel, the above external risk data sub-framework is integrated according to preset rules to complete the construction of the basic data framework of the risk data processing model based on artificial intelligence; The data storage layer and the data transmission layer have the same architecture but different functions; The fusion method of risk data subframes is as follows: each type of risk data subframe is orderly integrated along the data flow channel based on the data identification system, and risk data of different types are staggered and integrated according to risk weights, with a total of n types; The n is obtained by the following formula ; In the formula, m is the processing capacity of the data flow channel, k is the data volume of a single risk data subframe, and [] is the rounding operation; The fusion weight w between risk data from different sources is determined by the following inequality: Where W is the data processing accuracy of a single risk data subframe, and w is the risk data fusion weight.

2. The risk data processing method based on artificial intelligence according to claim 1 is characterized in that: In the basic data framework of the risk data processing model based on artificial intelligence, the identification conversion rules between risk data from different sources are unified standards, and the integration method between different types of risk data is based on risk weight allocation; Each of the risk data subframes is an extensible structure, with an updateable data module inside and the same number of risk data interfaces with the same specifications connected to the outside; Moreover, the number of risk data interfaces is ≥2.

3. The risk data processing method based on artificial intelligence according to claim 1 is characterized in that: The step of identifying the three domains of historical risk data domain, real-time risk data domain and risk characteristic data domain in the above basic data framework further includes: Obtain any data slice of the basic data framework of the AI-based risk data processing model; Identify the scope of each historical risk data subframe in the data slice and the area where the corresponding data processing flow is located, use the area within the scope as the processing domain of the historical risk data, and arrange the processing domains of all historical risk data to form the historical risk data domain of the basic data framework; Identify the data source of each real-time risk data subframe in the data slice, construct an auxiliary boundary one according to the historical risk data domain boundary, connect adjacent risk data interface edge points outside the auxiliary boundary one with the k+d radius to construct an auxiliary boundary two, and use the area between the auxiliary boundaries one and two as the processing domain of the real-time risk data. All the processing domains of the real-time risk data are arranged to form the real-time risk data domain of the basic data framework; d is the risk data interface thickness; All historical risk data fields and real-time risk data fields in the data slice are removed to obtain the risk characteristic data fields of the basic data framework.

4. The risk data processing method based on artificial intelligence according to claim 3 is characterized in that: The step of dividing the data processing modules of the historical risk data domain according to the preset data processing frequency further includes: In the historical risk data domain, a data screening area is made with the data identifier of each historical risk data subframe as the center and W as the processing accuracy range to obtain a circumscribed regular polygon of the screening area; Among them, for the historical risk data subframe within the preset distance from the core of the data flow channel, an circumscribed regular pentagon is obtained, and for the historical risk data subframe outside the preset distance from the core of the data flow channel, an circumscribed regular hexagon is obtained; The data identifier of each historical risk data subframe is used as the division center, and the corresponding circumscribed regular polygon is used as the boundary. The hash algorithm is used to evenly divide 6p data subsets outward in the data slice, p ≥ 2, until the historical risk data domain in each circumscribed regular polygon is divided into the same number of submodules; According to the rule that the size of the largest data block is smaller than the smallest risk factor size in the risk assessment process and the number of data points at the borders of adjacent submodules is consistent, hierarchical data processing is performed on each submodule; The historical risk data domain after layered data processing is expanded along the data flow direction to the bottom plane of the data flow channel, and then the expanded structure is divided into equal intervals along the data dimension with an interval length of L to complete the division of the data processing modules of the historical risk data domain.

5. The risk data processing method based on artificial intelligence according to claim 3 or 4, characterized in that: The optimization scheme for extracting risk features is as follows: the maximum size of the first layer of data processing close to the historical risk data domain in the real-time risk data domain does not exceed 0.01MB; Moreover, on the data processing boundary of each risk data subframe, the number of data points N wide on the width of the risk data interface and the number of data points N thick on the thickness of the risk data interface satisfy In the formula, q represents the width of the risk data interface; The number of data points Narc on the arc length between adjacent risk data interfaces and the number of data points Nthick on the thickness of the risk data interface satisfy .

6. The risk data processing method based on artificial intelligence according to claim 5 is characterized in that: The data processing module for dividing the real-time risk data domain is performed by the following steps: identifying all endpoints on the inner contour of each real-time risk data domain that are close to the contour of the historical risk data domain; The data source of each real-time risk data subframe is used as the division center, and the outer contour of the real-time risk data domain is used as the boundary. A clustering algorithm is used to evenly divide the line segments in the data slice outward by twice the number of risk data interfaces, so that the intersection set of the division line segments and the outer contour contains all the above endpoints; Remove all the endpoints in the above intersection set, construct two parallel lines connecting the remaining intersections and the division center as auxiliary lines, so that one end of each auxiliary line coincides with the position of the nearest endpoint, and the other end is on the outer contour of the real-time risk data domain, and divide each real-time risk data domain into 4 times the number of risk data interfaces. Processing submodules; According to the rule that the size of the largest data block is smaller than the smallest risk factor size in the risk assessment process, the number of data points at the boundaries of adjacent processing submodules is consistent, and the number of data points at the boundaries of the real-time risk data domain and the historical risk data domain is consistent, hierarchical data processing is performed on each processing submodule; According to the optimization scheme of risk feature extraction, the real-time risk data domain after the layered data processing is completed is further optimized; The further optimized real-time risk data domain is extended along the data flow direction to the bottom plane of the data flow channel, and then the extended structure is divided into equal intervals along the data dimension with an interval length of L to complete the division of the data processing modules of the real-time risk data domain.

7. The risk data processing method based on artificial intelligence according to claim 6 is characterized in that: The data processing modules of the risk feature data domain are divided into the following steps: After the data processing module division of the real-time risk data domain and the historical risk data domain is completed, the positions of all data points on the boundary of the real-time risk data domain and the historical risk data domain that overlap with the risk feature data domain are obtained; The data source of each risk feature data subframe is used as the division center, and the outer contour of the risk feature data domain is used as the boundary. A decision tree algorithm is used to divide line segments in the data slice to the intersection of each risk data interface and the risk feature data domain, and the risk feature data domain of each risk feature data subframe is divided into processing submodules twice the number of risk data interfaces; According to the rule that the number of data points on the overlapping boundaries of the risk feature data domain, the real-time risk data domain, and the historical risk data domain is consistent, hierarchical data processing is performed on each processing submodule; The risk feature data domain that has completed the hierarchical data processing is expanded along the data flow direction to the bottom plane of the data flow channel, and then the expanded structure is divided into equal intervals along the data dimension with an interval length of L to complete the division of the data processing modules of the risk feature data domain.

8. The risk data processing method based on artificial intelligence according to claim 7 is characterized in that: The data processing modules of the risk feature data domain are divided into the following steps: After the data processing module division of the real-time risk data domain and the historical risk data domain is completed, the positions of all data points on the boundary of the real-time risk data domain and the historical risk data domain that overlap with the risk feature data domain are obtained; The data source of each risk feature data subframe is used as the division center, and the outer contour of the risk feature data domain is used as the boundary. A decision tree algorithm is used to divide line segments in the data slice to the intersection of each risk data interface and the risk feature data domain, and the risk feature data domain of each risk feature data subframe is divided into processing submodules twice the number of risk data interfaces; According to the rule that the number of data points on the overlapping boundaries of the risk feature data domain, the real-time risk data domain, and the historical risk data domain is consistent, hierarchical data processing is performed on each processing submodule; The risk feature data domain that has completed the hierarchical data processing is expanded along the data flow direction to the bottom plane of the data flow channel, and then the expanded structure is divided into equal intervals along the data dimension with an interval length of L to complete the division of the data processing modules of the risk feature data domain.

Images