Data security protection method and system for insurance business risk control platform

Abstract

The invention discloses a data security protection method and system for an insurance business risk control platform, and relates to the technical field of insurance information security, and the method comprises the steps: collecting multi-source data, carrying out the preprocessing of the collected multi-source data, employing a zero filling subset matrix, generating a low-rank feature matrix, employing a matrix value to fill a correlation coefficient into a mapping matrix, and carrying out the protection of the correlation coefficient. Fusing by using a weighted aggregation method to generate a sample matrix; the method comprises the following steps: generating a pseudo-random stream by using an SHA-256 hash function, generating a disturbance direction vector, constructing a quantum projection matrix by using a projection disturbance coding method, and disturbing each row of a sample matrix to obtain a disturbed sample matrix; a low-rank feature matrix is generated through tensor reconstruction and an information bottleneck method, calculation complexity is reduced, feature extraction efficiency is improved, a sample matrix is subjected to safety disturbance through an SHA-256 hash function and a projection disturbance coding method, and explanatory performance of feature embedding and anti-attack capacity of a model are improved.

Description

Technical Field

[0001] The present invention relates to the field of insurance information security technology, and in particular to a data security protection method and system for an insurance business risk control platform. Background Art

[0002] With the in-depth application of big data and artificial intelligence technologies in the financial and insurance industries, insurance business risk control platforms have gradually developed into comprehensive systems integrating data collection, risk identification, and intelligent decision-making. Traditional risk control mechanisms mainly rely on rule engines and statistical models. Although they can identify and prevent risk events to a certain extent, in the face of the reality of diverse data dimensions, complex structures, and strong correlations in time series, existing technologies are unable to meet the requirements in terms of modeling accuracy and data processing capabilities. Methods such as deep learning, graph neural networks, and federated learning have been attempted to be introduced into insurance risk control modeling to improve the intelligence level of the system. However, most of these methods rely on large-scale centralized processing and exposure of raw data, making it difficult to take into account data privacy protection needs.

[0003] Existing insurance risk control data modeling technologies generally have the following key deficiencies: the lack of the ability to uniformly structure multi-source heterogeneous data, especially the weak alignment and dynamic modeling mechanisms in the time dimension, resulting in insufficient expression of the temporal evolution of risk characteristics. Current methods generally ignore the in-depth analysis and feature optimization of the topological structure in the data feature space, which makes it easy for key risk factors to be concealed, reducing the accuracy of risk control strategy judgment. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a data security protection method and system for an insurance business risk control platform to solve the problem of lack of ability to perform unified structured processing of multi-source heterogeneous data, especially weak alignment and dynamic modeling mechanisms in the time dimension, resulting in insufficient expression of the temporal evolution of risk characteristics. Current methods generally ignore the in-depth analysis and feature optimization of the topological structure in the data feature space, which makes key risk factors easily concealed and reduces the judgment accuracy of risk control strategies.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a data security protection method for an insurance business risk control platform, which comprises: Collect multi-source data, pre-process the collected multi-source data, generate time slice data, use tensor reconstruction to fill the time slice data into the tensor to obtain a third-order tensor, extract the time slice data in the third-order tensor, define it as a sample point, splice the correlation vectors of all samples by row, generate a topological feature matrix, extract the corresponding topological feature dimension, generate a subset matrix, use zero to fill the subset matrix, generate a low-rank feature matrix, use matrix assignment to fill the correlation coefficient into the mapping matrix, use the weighted aggregation method to fuse, and generate a sample matrix; Use the SHA-256 hash function to generate a pseudo-random stream, generate a perturbation direction vector, use the projection perturbation coding method to construct a quantum projection matrix, perturb each row of the sample matrix to obtain the perturbed sample matrix, construct an approximate reconstruction matrix, calculate the error, and use the PID control algorithm to monitor and optimize the error; Build a visual interface to display monitoring results and store multi-source data generated by collection and analysis.

[0007] As a preferred solution of the data security protection method of the insurance business risk control platform described in the present invention, the method includes: collecting multi-source data, generating time slice data, using tensor reconstruction to fill the time slice data into the tensor to obtain a third-order tensor, extracting the time slice data in the third-order tensor, defining it as a sample point, splicing the correlation vectors of all samples by row, generating a topological feature matrix, extracting the corresponding topological feature dimension, generating a subset matrix, using zero to fill the subset matrix, generating a low-rank feature matrix, using matrix assignment to fill the correlation coefficient into the mapping matrix, using a weighted aggregation method for fusion, and generating a sample matrix, including: Use the APL interface to collect multi-source data from the insurance business risk control platform, including insurance type, insured amount, login frequency, number of claims, and timestamp data; The preprocessed multi-source data is aligned according to the timestamps and spliced ​​using the splicing method to generate a two-dimensional matrix. The timestamp column of the two-dimensional matrix is ​​extracted, and the minimum value is defined as the reference time. The time interval is set using a sliding time window, and the index of the time dimension is calculated. The two-dimensional matrix is ​​grouped according to the index of the time dimension to generate time slice data. The time slice data is filled into the tensor using tensor reconstruction to obtain a third-order tensor. Extract time slice data from the third-order tensor, define it as sample points, use the Euclidean distance formula to calculate the Euclidean distance between sample points, generate a distance matrix, use the maximum distance ratio to set the filter radius, use the Vietoris-Rips complex to connect sample points with a distance less than the filter radius, calculate persistent homology, identify sample points of topological features, use the geometric mean method to calculate the center point of each topological feature, use topological data analysis to calculate the persistence score and sort in descending order, use the fixed threshold method to select the topological features with the highest scores, where k is the number of topological features, use the Gaussian kernel function to calculate the association between each sample and the k topological features, and generate an association vector; The correlation vectors of all samples are spliced ​​row by row to generate a topological feature matrix. The j-th column element of the topological feature matrix is ​​extracted to obtain the topological feature dimension. The mutual information formula is used to calculate the mutual information value between each topological feature dimension and the time slice data, and the data are sorted in descending order. The fixed threshold method is used to select the top The feature dimensions with the highest mutual information are is the number of topological feature dimensions, extract the corresponding topological feature dimensions, and generate a subset matrix. , initialize the low-rank feature matrix, set all elements to 0, and fill the feature dimension of the subset matrix into the front of the initialized low-rank feature matrix Column, if , directly initialize the low-rank feature matrix as a subset matrix, generate a low-rank feature matrix, and perform normalization; Based on the low-rank feature matrix, the correlation coefficient is calculated using the Pearson correlation coefficient formula, the correlation coefficient is filled into the mapping matrix using matrix assignment, and normalized; The mean of the normalized low-rank feature matrix and the mapping matrix are calculated respectively using mean aggregation, and the mean of the normalized low-rank feature matrix and the mapping matrix are fused using the weighted aggregation method to generate a sample matrix.

[0008] As a preferred solution of the data security protection method of the insurance business risk control platform described in the present invention, wherein: the SHA-256 hash function is used to generate a pseudo-random stream, a perturbation direction vector is generated, a projection perturbation coding method is used to construct a quantum projection matrix, and each row of the sample matrix is ​​perturbed to obtain the perturbed sample matrix, including: Use the SHA-256 hash function to generate a pseudo-random stream and use the Quickselect algorithm to extract the floating point numbers, where is the number of floating-point numbers, and is normalized to generate a perturbation direction vector. Based on the perturbation direction vector, the quantum projection matrix is ​​constructed using the projection perturbation coding method. Perturb each row of the sample matrix to obtain the perturbed sample matrix.

[0009] As a preferred solution of the data security protection method of the insurance business risk control platform of the present invention, the steps of constructing an approximate reconstruction matrix, calculating the error, and using a PID control algorithm to monitor and optimize the error include: Multiply the perturbed sample matrix by the transpose of the normalized mapping matrix to construct an approximate reconstruction matrix; Calculate the mean of the time slice data in the third-order tensor as a reference value, extract the elements of the approximate reconstruction matrix, and use the absolute error method to calculate the error between the elements of the approximate reconstruction matrix and the reference value; Use statistical analysis to set the judgment threshold, compare the error with the judgment threshold, and use the PID control algorithm to adjust the error rate that is greater than or equal to the judgment threshold until the difference is less than the judgment threshold. Stop adjustment and continue monitoring the error.

[0010] As a preferred solution of the data security protection method of the insurance business risk control platform of the present invention, the pre-processing of the collected multi-source data includes: One-hot encoding was used to encode the insurance category data to obtain a binary vector, which was then normalized. The interquartile range method was used to identify and delete outliers in the insured amount, login frequency, number of claims, and timestamp data. The mean imputation method was used to fill in missing data and normalize the data.

[0011] As a preferred solution of the data security protection method of the insurance business risk control platform of the present invention, wherein: the construction of a visual interface to display monitoring results includes: Use the front-end framework React.js to build a visual interface to display errors and monitoring results; Users who have passed real-name verification are allowed to view the information.

[0012] As a preferred solution of the data security protection method of the insurance business risk control platform of the present invention, the multi-source data generated by storage, collection and analysis includes: The collected multi-source data and the errors generated by the analysis are stored in the central database, and security access measures are set up. The central database will back up the stored data to the cloud and regularly perform integrity checks on the stored data and backup data. After the test is completed, the integrity test record will be generated and stored synchronously in the central database.

[0013] In a second aspect, the present invention provides a data security protection system for an insurance business risk control platform, comprising: The collection matrix module is used to collect multi-source data, pre-process the collected multi-source data, generate time slice data, use tensor reconstruction to fill the time slice data into the tensor to obtain a third-order tensor, extract the time slice data in the third-order tensor, define it as a sample point, splice the correlation vectors of all samples by row, generate a topological feature matrix, extract the corresponding topological feature dimension, generate a subset matrix, use zero to fill the subset matrix, generate a low-rank feature matrix, use matrix assignment to fill the correlation coefficient into the mapping matrix, use the weighted aggregation method to fuse, and generate a sample matrix; The perturbation monitoring module is used to generate a pseudo-random stream using the SHA-256 hash function, generate a perturbation direction vector, construct a quantum projection matrix using the projection perturbation coding method, perturb each row of the sample matrix to obtain the perturbed sample matrix, construct an approximate reconstruction matrix, calculate the error, and use the PID control algorithm to monitor and optimize the error; The visualization storage module is used to build a visualization interface to display monitoring results and store multi-source data generated by collection and analysis.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the data security protection method for the insurance business risk control platform as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the data security protection method for the insurance business risk control platform as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: the present invention collects multi-source data to generate time slice data, uses tensor reconstruction to fill the time slice data into the tensor to obtain a third-order tensor, extracts the time slice data in the third-order tensor, defines it as a sample point, splices the correlation vectors of all samples by row, generates a topological feature matrix, extracts the corresponding topological feature dimension, generates a subset matrix, uses zero to fill the subset matrix, generates a low-rank feature matrix, uses matrix assignment to fill the correlation coefficient into the mapping matrix, uses the weighted aggregation method to fuse, and generates a sample matrix; uses the SHA-256 hash function to generate a pseudo-random stream, generates a perturbation direction vector, uses the projection perturbation coding method to construct a quantum projection matrix, perturbs each row of the sample matrix to obtain the perturbed sample matrix, constructs an approximate reconstruction matrix, calculates the error, and uses the PID control algorithm to monitor and optimize the error; enhances the discrimination accuracy of the risk control strategy, reduces the computational complexity, and improves the efficiency of feature extraction, the interpretability of feature embedding, and the anti-attack capability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flow chart of the data security protection method for the insurance business risk control platform in Example 1.

[0019] Figure 2 This is a schematic diagram of the data security protection system of the insurance business risk control platform in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0023] Example 1, with reference to Figure 1 This is the first embodiment of the present invention, which provides a data security protection method for an insurance business risk control platform, comprising the following steps: S1. Collect multi-source data, pre-process the collected multi-source data, generate time slice data, use tensor reconstruction to fill the time slice data into the tensor to obtain a third-order tensor, extract the time slice data in the third-order tensor, define it as a sample point, splice the correlation vectors of all samples by row, generate a topological feature matrix, extract the corresponding topological feature dimension, generate a subset matrix, use zero to fill the subset matrix, generate a low-rank feature matrix, use matrix assignment to fill the correlation coefficient into the mapping matrix, use the weighted aggregation method to fuse, and generate a sample matrix; Specifically, multi-source data is collected and pre-processed, including: Use the APL interface to collect multi-source data from the insurance business risk control platform, including insurance type, insured amount, login frequency, number of claims, and timestamp data; One-hot encoding was used to encode the insurance category data to obtain a binary vector, which was then normalized. The interquartile range method was used to identify and delete outliers in the insured amount, login frequency, number of claims, and timestamp data. The mean imputation method was used to fill in missing data and normalize the data.

[0024] Multi-source data is collected through API interfaces to capture the full picture of customer behavior and provide a rich source of information for risk control models. One-hot encoding converts non-numerical insurance categories into standardized binary vectors, adapting to numerical calculations. The interquartile range method effectively removes outliers and reduces noise interference. Mean padding and normalization ensure data integrity and scale consistency, improving the stability of subsequent feature extraction.

[0025] Furthermore, the preprocessed multi-source data are aligned according to the timestamps and spliced ​​using the splicing method to generate a two-dimensional matrix. The timestamp column of the two-dimensional matrix is ​​extracted, the minimum value is defined as the reference time, the time interval is set using a sliding time window, the index of the time dimension is calculated, the two-dimensional matrix is ​​grouped according to the index of the time dimension, and time slice data is generated. The time slice data is filled into the tensor using tensor reconstruction to obtain a third-order tensor. The formula is: , in is the index of the time dimension, is the timestamp of the i-th multi-source data, is the reference time, is the time interval; Extract the time slice data from the third-order tensor, define it as sample points, use the Euclidean distance formula to calculate the Euclidean distance between the sample points, generate a distance matrix, and use the maximum distance ratio to set the filter radius. The formula is: , in is the filter radius, is a fixed proportional coefficient (set using the empirical fixed coefficient method), is a two-dimensional matrix at time slice t, representing the sample data at time t; Using the Vietoris-Rips complex, connect sample points whose distance is less than the filter radius, calculate persistent homology, identify sample points of topological features, calculate the center point of each topological feature using the geometric mean method, calculate the persistence score using topological data analysis, and sort them in descending order. Use the fixed threshold method to select the topological features with the highest scores, where k is the number of topological features. Use the Gaussian kernel function to calculate the association between each sample and the k topological features, and generate the association vector. The formula is: , in is the correlation between the i-th sample and the j-th topological feature, is the i-th row of the time slice matrix, representing the eigenvector of the i-th sample, is the center point of the j-th topological feature, is the Euclidean distance between the feature vector of the i-th sample and the center point of the j-th topological feature, σ is the bandwidth parameter of the Gaussian kernel (set using the Silverman rule); The correlation vectors of all samples are spliced ​​row by row to generate a topological feature matrix. The j-th column element of the topological feature matrix is ​​extracted to obtain the topological feature dimension, which represents the correlation value of n samples on the j-th topological feature. The mutual information formula is used to calculate the mutual information value between each topological feature dimension and the time slice data, and the data are sorted in descending order. The fixed threshold method is used to select the top The feature dimensions with the highest mutual information are is the number of topological feature dimensions, extract the corresponding topological feature dimensions, and generate a subset matrix. , initialize the low-rank feature matrix, set all elements to 0, and fill the feature dimension of the subset matrix into the front of the initialized low-rank feature matrix Column, if , directly let the initialized low-rank feature matrix be the subset matrix, generate the low-rank feature matrix (k is the number of topological features selected after persistent homology calculation, which directly determines the number of columns of the topological feature matrix, The high-information dimensions are further screened out from the k dimensions of the topological feature matrix. They are all in the same feature space (topological embedding space), representing the initial topological feature dimensions and the screened high-information dimensions, which are comparable. They are then normalized. Based on the low-rank feature matrix, the Pearson correlation coefficient formula is used to calculate the correlation coefficient, and the correlation coefficient is filled into the mapping matrix using matrix assignment and normalized. The formula is: , in is the correlation coefficient between the value of the p-th multi-source data of the time slice data (such as the insurance amount of the i-th customer after preprocessing) and the j-th low-rank feature, n is the number of time slice data, is the value of the p-th multi-source data of the i-th sample in the two-dimensional matrix of time slice t, is the mean of the p-th multi-source data in the two-dimensional matrix of time slice t, is the embedding value of the i-th sample in the j-th topological feature dimension in the low-rank feature matrix, is the mean value of the jth topological feature dimension in the low-rank feature matrix; The mean of the normalized low-rank feature matrix and the mapping matrix are calculated respectively using mean aggregation, and the mean of the normalized low-rank feature matrix and the mapping matrix are fused using the weighted aggregation method to generate a sample matrix.

[0026] The sliding time window converts timestamps into discrete indexes, retaining the dynamic changes of insurance data. The third-order tensor facilitates subsequent topological analysis and feature extraction, enhancing the model's ability to recognize complex risk patterns. TDA captures the topological structure in the data, overcomes the limitations of traditional linear methods (such as principal component analysis), and improves the risk control model's ability to recognize complex risk patterns. It persistently and homologically adapts to the high-dimensional sparse characteristics of insurance data to ensure the stability of feature extraction. By selecting high-scoring topological features, high-dimensional data is compressed into low-dimensional representations, reducing computational complexity and overfitting risks. Mutual information screening retains high-information feature dimensions. Weighted aggregation integrates the topological information of the low-rank feature matrix and the field correlation of the mapping matrix to generate a more comprehensive sample representation, improving the generalization ability of the risk control model. Indirect feature representation is introduced through the reconstruction terms of the mapping matrix to reduce the risk of reversibility of the original data. Through topological feature compression and indirect representation of the mapping matrix, the risk of original data exposure is reduced while retaining core risk information, which is superior to traditional differential privacy methods.

[0027] S2. Use the SHA-256 hash function to generate a pseudo-random stream, generate a perturbation direction vector, use the projection perturbation coding method to construct a quantum projection matrix, perturb each row of the sample matrix to obtain the perturbed sample matrix, construct an approximate reconstruction matrix, calculate the error, and use the PID control algorithm to monitor and optimize the error; Specifically, the SHA-256 hash function is used to generate a pseudo-random stream, a perturbation direction vector is generated, and the projection perturbation coding method is used to construct a quantum projection matrix. Each row of the sample matrix is ​​perturbed to obtain the perturbed sample matrix, including: Use the SHA-256 hash function to generate a pseudo-random stream and use the Quickselect algorithm to extract the floating point numbers, where is the number of floating-point numbers, and is normalized to generate a perturbation direction vector. Based on the perturbation direction vector, the quantum projection matrix is ​​constructed using the projection perturbation coding method. The formula is: , , in is the disturbance direction vector, For the A pseudo-random stream of floating point numbers, R is the transpose, is the identity matrix (diagonal elements are 1 and the rest are 0), is the perturbation intensity factor (set using grid search optimization); Perturb each row of the sample matrix to obtain the perturbed sample matrix. The formula is: , in is the sample vector after perturbation, is the i-th row of the sample matrix, representing the embedding vector of the i-th sample.

[0028] SHA-256's anti-collision and high entropy properties ensure the unpredictability of pseudo-random streams, preventing attackers from deducing perturbation patterns through reverse engineering. It is superior to traditional linear congruential generators. The Quickselect algorithm extracts floating-point numbers with linear time complexity, adapting to the real-time perturbation requirements of large-scale insurance data sets. Normalization ensures that the perturbation direction vector is located on the unit sphere, maintaining the stability of the perturbation amplitude and avoiding feature distortion caused by randomness. By fixing the seed input, SHA-256 can generate deterministic pseudo-random streams, which is convenient for debugging and verifying the perturbation effect, while retaining randomness to protect privacy. The quantum projection matrix introduces irreversible noise through low-rank perturbations. It effectively prevents the reverse recovery of original features (such as customer behavior characteristics), and is superior to the traditional Gaussian noise addition method. The perturbation intensity factor ensures that the perturbation amplitude is controllable and retains the core structure of the topological features (such as the clustering pattern of high-risk customer groups). Row-by-row perturbation ensures that each customer sample (such as insurance behavior characteristics) is subject to consistent privacy protection, preventing the deduction of the overall pattern through partial data. The low-rank characteristics of the quantum projection matrix concentrate the perturbation in a specific direction, retaining the geometric structure of the topological features (such as the clustering characteristics of the customer group) to the greatest extent, which is superior to random noise perturbation. Normalization processing eliminates the scale changes caused by the perturbation, ensuring that the matrix after perturbation is compatible with downstream risk control models (such as fraud detection models).

[0029] Furthermore, an approximate reconstruction matrix is ​​constructed, the error is calculated, and the PID control algorithm is used to monitor and optimize the error, including: Multiply the perturbed sample matrix by the transpose of the normalized mapping matrix to construct an approximate reconstruction matrix. The formula is: , in is the approximate reconstruction matrix, G is the sample matrix after perturbation, is the transpose of the normalized mapping matrix; Calculate the mean of the time slice data in the third-order tensor as a reference value, extract the elements of the approximate reconstruction matrix, and use the absolute error method to calculate the error between the elements of the approximate reconstruction matrix and the reference value; Use statistical analysis to set the judgment threshold, compare the error with the judgment threshold, and use the PID control algorithm to adjust the error rate that is greater than or equal to the judgment threshold until the difference is less than the judgment threshold. Stop adjustment and continue monitoring the error.

[0030] The deviation between the approximate reconstructed matrix and the original data reflects the disturbance effect. The larger the deviation, the stronger the privacy protection. The more controllable the deviation, the higher the data utility. The error rate provides a statistical indicator of the global disturbance effect, reflecting whether the disturbance achieves the expected privacy protection goal (for example, a high error rate indicates strong privacy protection). The differential term of PID predicts the error change trend, suppresses the sharp fluctuation of the disturbance intensity, enhances the system stability, and adapts to the dynamic changes of insurance data. Through the feedback mechanism, the PID algorithm automatically adapts to the characteristics of different data sets and improves the versatility of the disturbance process. Quantum perturbation and PID control dynamically optimize the disturbance intensity to ensure that customer sensitive information (such as the insured amount) cannot be reversed while retaining the core structure of the risk control feature.

[0031] S3, build a visual interface to display monitoring results, store and analyze multi-source data; Specifically, a visual interface is built to display monitoring results, including: Use the front-end framework React.js to build a visual interface to display errors and monitoring results; Users who have passed real-name verification are allowed to view the information.

[0032] React.js' component-based design and powerful rendering capabilities of the visualization library present complex error analysis results (such as the change in error rate with PID control iterations) in intuitive charts. The responsive layout ensures consistency of the interface on PC and mobile devices, making it convenient for risk control personnel to access monitoring results anytime and anywhere, improving the platform's practicality and user experience. Real-name verification and multi-factor authentication effectively prevent unauthorized access, protect sensitive data (such as customer insurance amounts and claims records), and comply with privacy regulations.

[0033] Furthermore, the multi-source data generated by storage, collection and analysis includes: The collected multi-source data and the errors generated by the analysis are stored in the central database, and security access measures are set up. The central database will back up the stored data to the cloud and regularly perform integrity checks on the stored data and backup data. After the test is completed, the integrity test record will be generated and stored synchronously in the central database.

[0034] The central database uniformly stores multi-source data and analysis results, ensuring data consistency and traceability, facilitating subsequent analysis and model reproduction. Cloud-based backup ensures that data can be quickly restored in the event of hardware failure or network attacks (such as ransomware), enhancing the business continuity of the risk control platform. The incremental backup strategy significantly reduces storage and transmission costs, adapting to the backup needs of large-scale data on the insurance platform, and outperforming the resource consumption of full backup. Regular backup and detection are automatically executed through scripts, reducing manual intervention and improving operation and maintenance efficiency.

[0035] Example 2, reference Figure 2 In a second embodiment of the present invention, a data security protection system for an insurance business risk control platform includes: The collection matrix module is used to collect multi-source data, pre-process the collected multi-source data, generate time slice data, use tensor reconstruction to fill the time slice data into the tensor to obtain a third-order tensor, extract the time slice data in the third-order tensor, define it as a sample point, splice the correlation vectors of all samples by row, generate a topological feature matrix, extract the corresponding topological feature dimension, generate a subset matrix, use zero to fill the subset matrix, generate a low-rank feature matrix, use matrix assignment to fill the correlation coefficient into the mapping matrix, use the weighted aggregation method to fuse, and generate a sample matrix; The perturbation monitoring module is used to generate a pseudo-random stream using the SHA-256 hash function, generate a perturbation direction vector, construct a quantum projection matrix using the projection perturbation coding method, perturb each row of the sample matrix to obtain the perturbed sample matrix, construct an approximate reconstruction matrix, calculate the error, and use the PID control algorithm to monitor and optimize the error; The visualization storage module is used to build a visualization interface to display monitoring results and store multi-source data generated by collection and analysis.

[0036] This embodiment also provides a computer device, which is suitable for the data security protection method of the insurance business risk control platform, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the data security protection method of the insurance business risk control platform proposed in the above embodiment.

[0037] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0038] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the data security protection method for implementing the insurance business risk control platform as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0039] In summary, the present invention collects multi-source data to generate time slice data, uses tensor reconstruction to fill the time slice data into the tensor to obtain a third-order tensor, extracts the time slice data in the third-order tensor, defines it as a sample point, splices the correlation vectors of all samples by row, generates a topological feature matrix, extracts the corresponding topological feature dimension, generates a subset matrix, uses zero to fill the subset matrix, generates a low-rank feature matrix, uses matrix assignment to fill the correlation coefficient into the mapping matrix, uses the weighted aggregation method to fuse, and generates a sample matrix; uses the SHA-256 hash function to generate a pseudo-random stream and a perturbation direction vector, uses the projection perturbation coding method to construct a quantum projection matrix, perturbs each row of the sample matrix to obtain the perturbed sample matrix, constructs an approximate reconstruction matrix, calculates the error, and uses the PID control algorithm to monitor and optimize the error; enhances the discrimination accuracy of the risk control strategy, not only reduces the computational complexity, but also improves the interpretability of feature embedding and the anti-attack capability of the model.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A data security protection method for an insurance business risk control platform, characterized by: The steps include: Collect multi-source data, pre-process the collected multi-source data, generate time slice data, use tensor reconstruction to fill the time slice data into the tensor to obtain a third-order tensor, extract the time slice data in the third-order tensor, define it as a sample point, splice the correlation vectors of all samples by row, generate a topological feature matrix, extract the corresponding topological feature dimension, generate a subset matrix, use zero to fill the subset matrix, generate a low-rank feature matrix, use matrix assignment to fill the correlation coefficient into the mapping matrix, use the weighted aggregation method to fuse, and generate a sample matrix; Use the SHA-256 hash function to generate a pseudo-random stream, generate a perturbation direction vector, use the projection perturbation coding method to construct a quantum projection matrix, perturb each row of the sample matrix to obtain the perturbed sample matrix, construct an approximate reconstruction matrix, calculate the error, and use the PID control algorithm to monitor and optimize the error; Build a visual interface to display monitoring results and store multi-source data generated by collection and analysis.

2. The data security protection method for the insurance business risk control platform according to claim 1, characterized in that: The method collects multi-source data, generates time slice data, uses tensor reconstruction to fill the time slice data into the tensor to obtain a third-order tensor, extracts the time slice data in the third-order tensor, defines it as a sample point, concatenates the correlation vectors of all samples by row, generates a topological feature matrix, extracts the corresponding topological feature dimension, generates a subset matrix, fills the subset matrix with zeros, generates a low-rank feature matrix, uses matrix assignment to fill the correlation coefficient into the mapping matrix, and uses a weighted aggregation method for fusion to generate a sample matrix, including: Use the APL interface to collect multi-source data from the insurance business risk control platform, including insurance type, insured amount, login frequency, number of claims, and timestamp data; The preprocessed multi-source data is aligned according to the timestamps and spliced ​​using the splicing method to generate a two-dimensional matrix. The timestamp column of the two-dimensional matrix is ​​extracted, and the minimum value is defined as the reference time. The time interval is set using a sliding time window, and the index of the time dimension is calculated. The two-dimensional matrix is ​​grouped according to the index of the time dimension to generate time slice data. The time slice data is filled into the tensor using tensor reconstruction to obtain a third-order tensor. Extract time slice data from the third-order tensor, define it as sample points, use the Euclidean distance formula to calculate the Euclidean distance between sample points, generate a distance matrix, use the maximum distance ratio to set the filter radius, use the Vietoris-Rips complex to connect sample points with a distance less than the filter radius, calculate persistent homology, identify sample points of topological features, use the geometric mean method to calculate the center point of each topological feature, use topological data analysis to calculate the persistence score and sort in descending order, use the fixed threshold method to select the topological features with the highest scores, where k is the number of topological features, use the Gaussian kernel function to calculate the association between each sample and the k topological features, and generate an association vector; The correlation vectors of all samples are spliced ​​row by row to generate a topological feature matrix. The j-th column element of the topological feature matrix is ​​extracted to obtain the topological feature dimension. The mutual information formula is used to calculate the mutual information value between each topological feature dimension and the time slice data, and the data are sorted in descending order. The fixed threshold method is used to select the top The feature dimensions with the highest mutual information are is the number of topological feature dimensions, extract the corresponding topological feature dimensions, and generate a subset matrix. , initialize the low-rank feature matrix, set all elements to 0, and fill the feature dimension of the subset matrix into the front of the initialized low-rank feature matrix Column, if , directly initialize the low-rank feature matrix as a subset matrix, generate a low-rank feature matrix, and perform normalization; Based on the low-rank feature matrix, the correlation coefficient is calculated using the Pearson correlation coefficient formula, the correlation coefficient is filled into the mapping matrix using matrix assignment, and normalized; The mean of the normalized low-rank feature matrix and the mapping matrix are calculated respectively using mean aggregation, and the mean of the normalized low-rank feature matrix and the mapping matrix are fused using the weighted aggregation method to generate a sample matrix.

3. The data security protection method for an insurance business risk control platform according to claim 2, characterized in that: The method uses the SHA-256 hash function to generate a pseudo-random stream, generates a perturbation direction vector, constructs a quantum projection matrix using a projection perturbation coding method, and perturbs each row of the sample matrix to obtain a perturbed sample matrix, including: Use the SHA-256 hash function to generate a pseudo-random stream and use the Quickselect algorithm to extract the floating point numbers, where is the number of floating-point numbers, and is normalized to generate a perturbation direction vector. Based on the perturbation direction vector, the quantum projection matrix is ​​constructed using the projection perturbation coding method. Perturb each row of the sample matrix to obtain the perturbed sample matrix.

4. The data security protection method for an insurance business risk control platform according to claim 3, characterized in that: The method of constructing an approximate reconstruction matrix, calculating an error, and using a PID control algorithm to monitor and optimize the error includes: Multiply the perturbed sample matrix by the transpose of the normalized mapping matrix to construct an approximate reconstruction matrix; Calculate the mean of the time slice data in the third-order tensor as a reference value, extract the elements of the approximate reconstruction matrix, and use the absolute error method to calculate the error between the elements of the approximate reconstruction matrix and the reference value; Use statistical analysis to set the judgment threshold, compare the error with the judgment threshold, and use the PID control algorithm to adjust the error rate that is greater than or equal to the judgment threshold until the difference is less than the judgment threshold. Stop adjustment and continue monitoring the error.

5. The data security protection method for the insurance business risk control platform according to claim 1, characterized in that: The preprocessing of the collected multi-source data includes: One-hot encoding was used to encode the insurance category data to obtain a binary vector, which was then normalized. The interquartile range method was used to identify and delete outliers in the insured amount, login frequency, number of claims, and timestamp data. The mean imputation method was used to fill in missing data and normalize the data.

6. The data security protection method for the insurance business risk control platform according to claim 4, characterized in that: The construction of a visual interface to display monitoring results includes: Use the front-end framework React.js to build a visual interface to display errors and monitoring results; Users who have passed real-name verification are allowed to view the information.

7. The data security protection method for an insurance business risk control platform according to claim 4, characterized in that: The multi-source data generated by the storage, collection and analysis includes: The collected multi-source data and the errors generated by the analysis are stored in the central database, and security access measures are set up. The central database will back up the stored data to the cloud and regularly perform integrity checks on the stored data and backup data. After the test is completed, the integrity test record will be generated and stored synchronously in the central database.

8. A data security protection system for an insurance business risk control platform, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The collection matrix module is used to collect multi-source data, pre-process the collected multi-source data, generate time slice data, use tensor reconstruction to fill the time slice data into the tensor to obtain a third-order tensor, extract the time slice data in the third-order tensor, define it as a sample point, splice the correlation vectors of all samples by row, generate a topological feature matrix, extract the corresponding topological feature dimension, generate a subset matrix, use zero to fill the subset matrix, generate a low-rank feature matrix, use matrix assignment to fill the correlation coefficient into the mapping matrix, use the weighted aggregation method to fuse, and generate a sample matrix; The perturbation monitoring module is used to generate a pseudo-random stream using the SHA-256 hash function, generate a perturbation direction vector, construct a quantum projection matrix using the projection perturbation coding method, perturb each row of the sample matrix to obtain the perturbed sample matrix, construct an approximate reconstruction matrix, calculate the error, and use the PID control algorithm to monitor and optimize the error; The visualization storage module is used to build a visualization interface to display monitoring results and store multi-source data generated by collection and analysis.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the data security protection method for the insurance business risk control platform described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the data security protection method for the insurance business risk control platform described in any one of claims 1 to 7 are implemented.

Images