A sensitivity automatic modeling method and system for the whole life cycle of a device

Abstract

The present application relates to the technical field of data privacy protection, and a sensitivity automatic modeling method and system for the whole life cycle of equipment, comprising: sensitivity scoring on a benchmark data category set to obtain a benchmark sensitivity set; classifying a to-be-layered data set using the benchmark data category set to obtain multiple homogeneous data sets; intra-set data correlation degree evaluation based on the homogeneous data sets to obtain a data correlation group set; density clustering of the homogeneous data sets using the data correlation group set to obtain multiple current clustering data sets; adjustment of original data frequency using the multiple current clustering data sets to obtain target data frequency; correction of preliminary sensitivity based on the target data frequency to obtain target sensitivity; and aggregation of the target sensitivity to obtain a target sensitivity set, and obtaining multiple allocation data sets using the target sensitivity set. The present application can improve the accuracy and intelligence of sensitivity evaluation and improve the accuracy of privacy protection.

Description

Technical Field

[0001] This invention relates to the field of data privacy protection technology, and in particular to a method and system for automatic sensitivity modeling throughout the entire lifecycle of a device. Background Technology

[0002] In industrial equipment data management scenarios, equipment operation data, fault records, maintenance information, etc., often contain a large amount of sensitive information, such as trade secrets and personal privacy. To ensure the security of data during the release or sharing process, it is necessary to carry out differentiated privacy protection processing on this data. Sensitivity modeling, as a prerequisite for differential privacy protection, directly affects the balance between the rationality of noise addition and data availability. Therefore, achieving automated and accurate sensitivity modeling has important practical significance.

[0003] Traditional sensitivity modeling methods are usually based solely on static scoring of data types. While this method can initially distinguish the sensitivity of different data, it cannot address the differences in actual risk between rare and common data, leading to over-protection or under-protection. Therefore, the level of intelligence is not high. Furthermore, this method does not consider link attacks and background knowledge attacks that may be caused by external databases, thus underestimating the actual risk of privacy leakage. Summary of the Invention

[0004] This invention provides a method and system for automatic sensitivity modeling throughout the entire lifecycle of a device. Its main purpose is to improve the accuracy and intelligence of sensitivity assessment and enhance the accuracy of privacy protection.

[0005] To achieve the above objectives, the present invention provides an automatic sensitivity modeling method for the entire lifecycle of equipment, comprising:

[0006] Receive sensitivity modeling instructions, determine the dataset to be stratified based on the sensitivity modeling instructions, the dataset to be stratified includes data generated during the equipment procurement, acceptance, use, maintenance, verification, decommissioning and scrapping stages;

[0007] Construct a set of benchmark data types, and assign sensitivity scores to each benchmark data type in the set to obtain a benchmark sensitivity set;

[0008] The dataset to be stratified is classified using a benchmark data category set to obtain multiple datasets of the same type. Each dataset of the same type corresponds one-to-one with a benchmark data category in the benchmark data category set.

[0009] Extract similar datasets sequentially from multiple similar datasets, evaluate the data correlation within the set based on the similar datasets, and obtain a data correlation set. The data correlation set includes multiple data correlation groups, and the data correlation group includes: data correlation, shared data, and similar data.

[0010] Density clustering of similar datasets is performed using data association sets to obtain multiple current clustered datasets;

[0011] Extract the data to be modeled from the same dataset, count the original data frequency of the data to be modeled in the dataset to be stratified, and adjust the original data frequency using multiple current clustering datasets to obtain the target data frequency.

[0012] The initial sensitivity corresponding to the data to be modeled is identified in the benchmark sensitivity set. The initial sensitivity is then corrected based on the target data frequency and a preset public external database to obtain the target sensitivity.

[0013] Summarize the target sensitivities corresponding to the data to be modeled to obtain a target sensitivity set. Use the target sensitivity set to assign sensitivity to the same type of dataset to obtain an assigned dataset.

[0014] By aggregating the corresponding allocation datasets of similar datasets, multiple allocation datasets are obtained, enabling automatic sensitivity modeling for the entire lifecycle of devices.

[0015] Optionally, the step of evaluating the intra-set data correlation based on similar datasets to obtain a data correlation set includes:

[0016] Extract similar data sequentially from similar datasets, and remove similar data from the same dataset to obtain a comparison dataset;

[0017] Calculate the data distance between each comparative data in the same data set and the comparison data set to obtain a data distance set. Identify the nearest neighbor set in the data distance set according to the preset nearest neighbor parameter. The number of nearest neighbors in the nearest neighbor set is equal to the number of nearest neighbor parameters.

[0018] The current nearest neighbor dataset is identified in the comparison dataset based on the nearest neighbor distance set, where the current nearest neighbor data in the current nearest neighbor dataset corresponds one-to-one with the nearest neighbor distance in the nearest neighbor distance set;

[0019] The density of similar data is obtained by evaluating the density of similar data based on the nearest neighbor distance set.

[0020] The comparison dataset is designated as the shared dataset. Shared data is extracted sequentially from the shared dataset, and the data correlation between the shared data and the data of the same type is calculated based on the density of similar data.

[0021] Data association, shared data, and similar data are merged to obtain data association groups;

[0022] By aggregating the data association groups corresponding to the shared data, multiple data association groups are obtained;

[0023] Treat the comparison dataset as a similar dataset, and return to the step of sequentially extracting similar data from the similar dataset until the comparison dataset is empty. Summarize the multiple data association groups corresponding to the similar data to obtain the data association group set.

[0024] Optionally, the step of calculating the data correlation degree between shared data and similar data based on the density of similar data includes:

[0025] Obtain the shared comparison dataset and shared data density of the shared data;

[0026] Based on the shared comparison dataset and the intersection statistics of the comparison dataset, the intersection of shared data is obtained, and the number of shares in the intersection of shared data is counted.

[0027] Determine the distance between shared data and similar data;

[0028] Data correlation is calculated based on the number of shares, shared data density, density of similar data, and distance between shared data. The data correlation is expressed as:

[0029]

[0030] Where ε represents the data correlation degree, S represents the number of shares, and ρ t ρ represents the density of similar data. g D represents the shared data density. g Indicates the distance of shared data.

[0031] Optionally, the step of using data association sets to perform density clustering on similar datasets to obtain multiple current clustered datasets includes:

[0032] Calculate the decision value for each data point in the same dataset to obtain the decision value set;

[0033] Based on a preset number of cluster centers, target decision value groups are identified in the decision value set, and cluster center data groups are confirmed in similar datasets based on the target decision value groups.

[0034] The cluster center data groups are removed from the same type of dataset to obtain the original dataset;

[0035] Cluster center data are extracted sequentially from the cluster center data group to construct a preliminary cluster dataset containing only cluster center data.

[0036] The clustering range is calculated from the cluster center data to obtain the cluster radius, and candidate datasets are identified in the original dataset based on the cluster radius;

[0037] Candidate data are extracted sequentially from the candidate dataset, the central correlation between the candidate data and the cluster center data is calculated, and it is determined whether the central correlation is greater than the preset average correlation.

[0038] If the central correlation is greater than the average correlation, the candidate data is added to the preliminary clustering dataset to obtain the updated clustering dataset.

[0039] The updated clustering dataset is used as the initial clustering dataset, and the step of sequentially extracting candidate data from the candidate dataset is returned until the candidate dataset is an empty set;

[0040] The updated clustering dataset when the candidate dataset is empty is denoted as the target clustering dataset;

[0041] By aggregating the target cluster datasets corresponding to the data of each cluster center, multiple target cluster datasets are obtained;

[0042] Identify the unassigned datasets in the original dataset, and assign them to multiple target cluster datasets to obtain multiple current cluster datasets.

[0043] Optionally, the step of assigning the unassigned dataset to multiple target clustering datasets to obtain multiple current clustering datasets includes:

[0044] Perform the following operations on all unassigned data in the unassigned dataset:

[0045] Obtain the unassigned nearest neighbor dataset of unassigned data, and sequentially extract the target cluster dataset from multiple target cluster datasets;

[0046] Based on the unassigned nearest neighbor dataset, cluster association analysis is performed on the target clustering dataset and the unassigned data to obtain cluster association values;

[0047] Summarize the cluster association values ​​corresponding to the target clustering dataset to obtain a cluster association value set. Identify the maximum association value in the cluster association value set and add the unassigned data to the target clustering dataset corresponding to the maximum association value to complete the assignment of the unassigned data.

[0048] When each unassigned data in the unassigned dataset has been assigned, the multiple target clustering datasets are recorded as multiple current clustering datasets.

[0049] Optionally, the step of performing cluster association analysis on the target clustering dataset and the unassigned data based on the unassigned nearest neighbor dataset to obtain cluster association values ​​includes:

[0050] The unassigned shared dataset between the unassigned nearest neighbor dataset and the target clustering dataset is statistically analyzed.

[0051] Calculate the unallocated distance and the number of unallocated shares between each unallocated shared data in the unallocated data and the unallocated shared data set respectively, to obtain the unallocated distance set and the unallocated share set;

[0052] Cluster association values ​​are calculated based on the unassigned distance set and the unassigned shared data set, where the cluster association values ​​are represented as follows:

[0053]

[0054] Where Y represents the cluster association value, N represents the number of unassigned distances in the unassigned distance set or the number of unassigned shared numbers in the unassigned shared number set, and exp(*) represents an exponential function with the natural constant as the base. This represents the i-th unassigned distance in the unassigned distance set. Let represent the i-th unallocated shared number in the unallocated shared number set, and let a represent the nearest neighbor parameter.

[0055] Optionally, adjusting the original data frequency using multiple current clustering datasets to obtain the target data frequency includes:

[0056] Obtain the total number of similar data in the same dataset and the total number of data to be stratified in the dataset to be stratified. Calculate the category frequency based on the total number of similar data and the total number of data to be stratified, where the category frequency is the ratio of the total number of similar data to the total number of data to be stratified.

[0057] Identify the cluster dataset to be modeled from among multiple current cluster datasets, and count the number of clusters to be modeled in the cluster dataset to be modeled.

[0058] The clustering frequency is calculated based on the number of clusters to be modeled and the total number of data in the same category. The clustering frequency is the ratio of the number of clusters to be modeled to the total number of data in the same category.

[0059] Calculate the intra-cluster center distance of the data to be modeled based on the clustering dataset to be modeled;

[0060] The target data frequency is obtained by adjusting the original data frequency based on the category frequency, cluster frequency, and intra-cluster center distance.

[0061] Optionally, the step of refining the initial sensitivity based on the target data frequency and a preset public external database to obtain the target sensitivity includes:

[0062] The initial sensitivity is adjusted using the target data frequency to obtain the adjusted sensitivity;

[0063] The frequency of occurrence of the data to be modeled is counted in a publicly available external database, and the probability of a link attack is calculated based on the frequency of occurrence.

[0064] Build a background knowledge base, and filter the public external database based on the background knowledge base to obtain a public external sub-database;

[0065] Calculate the probability of background knowledge attacks based on publicly available external sub-libraries and the data to be modeled;

[0066] The maximum attack probability of background knowledge attack and link attack is identified, and the sensitivity is adjusted for external attacks based on the maximum attack probability to obtain the target sensitivity.

[0067] Optionally, the step of adjusting the sensitivity based on the maximum attack probability to obtain the target sensitivity includes:

[0068] Target sensitivity is calculated using the following formula:

[0069]

[0070] Among them, F m F represents target sensitivity. t P represents the sensitivity adjustment, θ represents the preset attack sensitivity value, and P represents the sensitivity adjustment. max This indicates the maximum probability of an attack. This indicates the preset compression factor.

[0071] To achieve the above objectives, the present invention also provides an automatic sensitivity modeling system for the entire lifecycle of equipment, comprising:

[0072] The modeling instruction receiving module is used to receive sensitivity modeling instructions, determine the dataset to be layered based on the sensitivity modeling instructions, the dataset to be layered includes data generated during the equipment procurement, acceptance, use, maintenance, verification, shutdown and scrapping stages, construct a benchmark data category set, and assign sensitivity scores to each benchmark data category in the benchmark data category set to obtain a benchmark sensitivity set.

[0073] The data correlation assessment module is used to classify the hierarchical dataset to be classified using a benchmark data category set to obtain multiple similar datasets. Among these multiple similar datasets, each similar dataset corresponds one-to-one with the benchmark data category in the benchmark data category set. Similar datasets are extracted sequentially from these multiple similar datasets, and the intra-set data correlation is assessed based on these similar datasets to obtain a data correlation set. The data correlation set includes multiple data correlation groups, and each data correlation group includes: data correlation degree, shared data, and similar data.

[0074] The data frequency adjustment module is used to perform density clustering on similar datasets using data association sets to obtain multiple current cluster datasets, extract the data to be modeled from the similar datasets, count the original data frequency of the data to be modeled in the dataset to be stratified, and adjust the original data frequency using multiple current cluster datasets to obtain the target data frequency.

[0075] The sensitivity allocation module is used to identify the initial sensitivity corresponding to the data to be modeled in the baseline sensitivity set, correct the initial sensitivity based on the target data frequency and a preset public external database to obtain the target sensitivity, summarize the target sensitivity corresponding to the data to be modeled to obtain the target sensitivity set, use the target sensitivity set to allocate sensitivity to the same type of dataset to obtain the allocated dataset, and summarize the allocated datasets corresponding to the same type of dataset to obtain multiple allocated datasets.

[0076] To address the above problems, the present invention also provides an electronic device, the electronic device comprising:

[0077] Memory, storing at least one instruction; and

[0078] The processor executes the instructions stored in the memory to implement the above-described method for automatic sensitivity modeling throughout the device's lifecycle.

[0079] To address the aforementioned issues, the present invention also provides a computer-readable storage medium storing at least one instruction, which is executed by a processor in an electronic device to implement the aforementioned automatic sensitivity modeling method for the entire lifecycle of a device.

[0080] To address the problems described in the background, this invention first assigns sensitivity scores to each benchmark data type in the benchmark data set, resulting in a benchmark sensitivity set. This step transforms subjective data sensitivity into a calculable value in the 0-1 range, avoiding noise intensity mismatch caused by differing evaluation scales among different personnel. Next, it evaluates the intra-set data correlation based on similar datasets, obtaining data association sets. This step integrates local density and shared information through data correlation, reflecting the true correlation strength within similar data sets. This overcomes the shortcomings of traditional distance metrics that ignore distribution density, making the subsequent selection of cluster centers closer to the actual data topology and reducing sensitivity estimation bias caused by misclassification. Furthermore, this invention uses data association sets to perform density clustering on similar datasets, obtaining multiple current cluster datasets. This step embeds a correlation-driven radius calculation and unassigned point reassignment strategy into DPC clustering, preserving the suitability of the density peak algorithm for clusters of arbitrary shapes. Furthermore, this invention utilizes correlation to filter out noise points, ensuring high cohesion of each cluster at the sensitivity level, providing a clean cluster structure for subsequent frequency adjustment. In addition, the invention adjusts the original data frequency using multiple current clustering datasets to obtain the target data frequency. This step adjusts the original data frequency through category frequency, cluster frequency, and intra-cluster center distance, more realistically reflecting the representativeness and privacy risks of the data in the clustering, providing a more reasonable frequency basis for subsequent sensitivity correction. Finally, based on the target data frequency and publicly available external databases, the initial sensitivity is corrected to obtain the target sensitivity. This step explicitly incorporates the link probability of the publicly available external database and the inference probability of the background knowledge base into the sensitivity correction formula, forming a quantitative response mechanism to external re-identification attacks. This avoids the over-protection or under-protection caused by the uniform noise addition to all data in traditional differential privacy. Therefore, this invention can improve the accuracy and intelligence of sensitivity assessment and enhance the accuracy of privacy protection. Attached Figure Description

[0081] Figure 1 This is a flowchart illustrating an automatic sensitivity modeling method for the entire lifecycle of a device, provided in an embodiment of the present invention.

[0082] Figure 2 A functional block diagram of an automatic sensitivity modeling system for the entire lifecycle of equipment provided in an embodiment of the present invention;

[0083] Figure 3 This is a schematic diagram of the structure of an electronic device that implements the automatic sensitivity modeling method for the entire life cycle of a device, according to an embodiment of the present invention.

[0084] Explanation of reference numerals in the attached figures:

[0085] 10. Electronic device; 11. Processor; 12. Memory; 13. Bus.

[0086] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0087] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0088] This application provides an automatic sensitivity modeling method for the entire lifecycle of a device. The executing entity of this automatic sensitivity modeling method for the entire lifecycle of a device includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the automatic sensitivity modeling method for the entire lifecycle of a device can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0089] Reference Figure 1 The diagram shown is a flowchart illustrating an automatic sensitivity modeling method for the entire lifecycle of a device, provided by an embodiment of the present invention. In this embodiment, the automatic sensitivity modeling method for the entire lifecycle of a device includes:

[0090] S1. Receive sensitivity modeling instructions, and determine the dataset to be layered based on the sensitivity modeling instructions. The dataset to be layered includes data generated during the equipment procurement, acceptance, use, maintenance, verification, decommissioning and scrapping stages.

[0091] Understandably, the sensitivity modeling instruction refers to a manually initiated instruction to calculate the sensitivity of specific data. The dataset to be stratified refers to the specific data specified in the sensitivity modeling instruction. This dataset is industrial equipment management data, such as equipment serial numbers, equipment models, and production dates. In addition, it may include equipment fault records, maintenance services, and procurement records. This data is extremely valuable for manufacturing enterprises, industry platforms, and equipment procurement evaluation scenarios. However, it often contains sensitive information about the enterprise, such as trade secrets related to production processes and equipment operating status, as well as potentially personal privacy information of equipment operators. Therefore, differential data stratification is required. During this process, different levels of operational protection need to be applied based on the privacy risk level of different data. Thus, it is necessary to calculate the sensitivity of each piece of data to be stratified, i.e., to perform sensitivity modeling on each piece of data. It should be noted that the aforementioned dataset to be stratified will be subsequently made public.

[0092] S2. Construct a set of benchmark data types, and assign sensitivity scores to each benchmark data type in the set of benchmark data types to obtain a benchmark sensitivity set.

[0093] It is understood that the benchmark data category set includes multiple benchmark data categories, and these benchmark data categories are artificially defined categories of industrial equipment management data. For example, the cumulative operating time of equipment over a period of time can be considered a benchmark data category, as can the number of equipment failures over a period of time. The benchmark sensitivity set refers to a collection of multiple benchmark sensitivities. A benchmark sensitivity refers to the sensitivity of a specific benchmark data category within the benchmark data category set. Each benchmark sensitivity in the benchmark sensitivity set corresponds one-to-one with a benchmark data category within the benchmark data category set. Sensitivity indicates the degree of privacy risk contained in that benchmark data category; the higher the sensitivity, the higher the privacy risk, and the stronger the privacy protection required.

[0094] Furthermore, the aforementioned sensitivity assignment for each benchmark data type in the benchmark data type set refers to the relevant personnel assigning different sensitivities to data types at different levels, with sensitivity ranging from 0 to 1. For example, a certain level in the data hierarchy is the raw data layer, which contains the finest-grained, unprocessed raw equipment management and maintenance data. This data layer includes benchmark data types such as sensor readings of a single device, fault logs, maintenance work order details, and purchase contract numbers. This layer has the highest data sensitivity and is usually not directly released externally; therefore, the initial sensitivity of the benchmark data types in this layer can be set to 0.9. In the enterprise aggregation layer, macro-level statistical information at the enterprise level is mainly included, such as the total number of devices, equipment utilization rate, and total annual maintenance expenditure of a manufacturing enterprise. The initial sensitivity of the benchmark data types in this layer is lower, mainly used for horizontal comparisons between enterprises; the initial sensitivity of these benchmark data types can be set to 0.2.

[0095] S3. Classify the dataset to be stratified using the benchmark data category set to obtain multiple datasets of the same type. Among these multiple datasets of the same type, each dataset corresponds one-to-one with the benchmark data category in the benchmark data category set.

[0096] It is clear that the aforementioned similar datasets refer to a collection of multiple data sets to be stratified corresponding to a certain benchmark data category after classification. For example, a dataset to be stratified may include: multiple industrial equipment models, multiple industrial equipment downtime, and multiple industrial equipment failure counts. Then, according to different benchmark data categories (equipment model, downtime, failure count), the dataset to be stratified is classified into several similar datasets: similar dataset A contains multiple industrial equipment models, similar dataset B contains multiple industrial equipment downtime, and similar dataset C contains multiple industrial equipment failure counts.

[0097] S4. Extract similar datasets sequentially from multiple similar datasets, evaluate the data correlation within the set based on the similar datasets, and obtain a data correlation set. The data correlation set includes multiple data correlation groups, and the data correlation group includes: data correlation, shared data, and similar data.

[0098] Understandably, the data association set refers to a collection of multiple data association groups. A data association group refers to a combination of data association degree, shared data, and similar data. The data association degree represents the degree of association between shared data and similar data. The higher the data association degree, the higher the degree of association between shared data and similar data.

[0099] In detail, the evaluation of intra-set data correlation based on similar datasets to obtain data correlation sets includes:

[0100] Extract similar data sequentially from similar datasets, and remove similar data from the same dataset to obtain a comparison dataset;

[0101] Calculate the data distance between each comparative data in the same data set and the comparison data set to obtain a data distance set. Identify the nearest neighbor set in the data distance set according to the preset nearest neighbor parameter. The number of nearest neighbors in the nearest neighbor set is equal to the number of nearest neighbor parameters.

[0102] The current nearest neighbor dataset is identified in the comparison dataset based on the nearest neighbor distance set, where the current nearest neighbor data in the current nearest neighbor dataset corresponds one-to-one with the nearest neighbor distance in the nearest neighbor distance set;

[0103] The density of similar data is obtained by evaluating the density of similar data based on the nearest neighbor distance set.

[0104] The comparison dataset is designated as the shared dataset. Shared data is extracted sequentially from the shared dataset, and the data correlation between the shared data and the data of the same type is calculated based on the density of similar data.

[0105] Data association, shared data, and similar data are merged to obtain data association groups;

[0106] By aggregating the data association groups corresponding to the shared data, multiple data association groups are obtained;

[0107] Treat the comparison dataset as a similar dataset, and return to the step of sequentially extracting similar data from the similar dataset until the comparison dataset is empty. Summarize the multiple data association groups corresponding to the similar data to obtain the data association group set.

[0108] It should be explained that the comparison dataset refers to the similar dataset after the elimination of data. It should be noted that since numerical calculations need to be performed on the similar data later, if the similar data is non-numerical data, such as equipment model, equipment location, etc., it needs to be converted into calculable numerical values. The data conversion methods are: one-hot encoding, label encoding, or hash mapping, etc.

[0109] It is clear that the data distance refers to the Euclidean distance between similar data and comparison data, and the nearest neighbor parameter is a human-defined constant representing the number of nearest neighbors in the nearest neighbor distance set. Identifying the nearest neighbor distance set in the data distance set according to the preset nearest neighbor parameter means: identifying the 'a' closest data distances in the data distance set and recording these 'a' data distances as the nearest neighbor distance set, where 'a' is the nearest neighbor parameter. The current nearest neighbor dataset refers to the set of multiple comparison data corresponding to the nearest neighbors in the nearest neighbor distance set.

[0110] Furthermore, the aforementioned similar data density refers to the data density of similar data within the same dataset. Data density indicates the degree of clustering of similar data within its dataset; a higher data density indicates more similar data surrounding the data of that type, and a denser distribution. The calculation method for the aforementioned similar data density is as follows: Where, ρ t This represents the density of similar data, where 'a' represents the nearest neighbor parameter, and 'D' represents the density of similar data. i This represents the distance of the i-th nearest neighbor in the nearest neighbor distance set. The phrase "summarizing multiple data association groups corresponding to the same type of data to obtain a data association group set" means: placing multiple data groups corresponding to each type of data into the same set as a data association group; the resulting set is the data association group set.

[0111] Specifically, the calculation of the data correlation degree between shared data and similar data based on the density of similar data includes:

[0112] Obtain the shared comparison dataset and shared data density of the shared data;

[0113] Based on the shared comparison dataset and the intersection statistics of the comparison dataset, the intersection of shared data is obtained, and the number of shares in the intersection of shared data is counted.

[0114] Determine the distance between shared data and similar data;

[0115] Data correlation is calculated based on the number of shares, shared data density, density of similar data, and distance between shared data. The data correlation is expressed as:

[0116]

[0117] Where ε represents the data correlation degree, S represents the number of shares, and ρ t ρ represents the density of similar data. g D represents the shared data density. g Indicates the distance of shared data.

[0118] It is clear that the shared comparison dataset refers to the comparison dataset corresponding to the shared data, and the method for obtaining this shared comparison dataset is the same as the method for obtaining the comparison dataset corresponding to similar data. The shared data density refers to the data density of the shared data, and the calculation method for this shared data density is the same as the calculation method for the density of similar data. The shared data intersection refers to the intersection between the shared comparison dataset and the comparison dataset. The shared number refers to the number of data points in the shared data intersection. The shared data distance refers to the Euclidean distance between the shared data and similar data.

[0119] It should be explained that the above data correlation calculation formula comprehensively evaluates the correlation strength between shared data and similar data by combining shared data density, similar data density, shared quantity, and shared data distance. This calculation method considers both the local density of data distribution and introduces the distance factor, which can more accurately reflect the intrinsic relationship between data and avoid the bias caused by relying on a single indicator such as distance or density.

[0120] S5. Utilize data association sets to perform density clustering on similar datasets to obtain multiple current cluster datasets.

[0121] It is clear that the multiple current clustered datasets refer to multiple sets obtained after clustering similar datasets.

[0122] In detail, the method of using data association sets to perform density clustering on similar datasets yields multiple current clustered datasets, including:

[0123] Calculate the decision value for each data point in the same dataset to obtain the decision value set;

[0124] Based on a preset number of cluster centers, target decision value groups are identified in the decision value set, and cluster center data groups are confirmed in similar datasets based on the target decision value groups.

[0125] The cluster center data groups are removed from the same type of dataset to obtain the original dataset;

[0126] Cluster center data are extracted sequentially from the cluster center data group to construct a preliminary cluster dataset containing only cluster center data.

[0127] The clustering range is calculated from the cluster center data to obtain the cluster radius, and candidate datasets are identified in the original dataset based on the cluster radius;

[0128] Candidate data are extracted sequentially from the candidate dataset, the central correlation between the candidate data and the cluster center data is calculated, and it is determined whether the central correlation is greater than the preset average correlation.

[0129] If the central correlation is greater than the average correlation, the candidate data is added to the preliminary clustering dataset to obtain the updated clustering dataset.

[0130] The updated clustering dataset is used as the initial clustering dataset, and the step of sequentially extracting candidate data from the candidate dataset is returned until the candidate dataset is an empty set;

[0131] The updated clustering dataset when the candidate dataset is empty is denoted as the target clustering dataset;

[0132] By aggregating the target cluster datasets corresponding to the data of each cluster center, multiple target cluster datasets are obtained;

[0133] Identify the unassigned datasets in the original dataset, and assign them to multiple target cluster datasets to obtain multiple current cluster datasets.

[0134] It should be explained that the decision value refers to the decision value in the DPC algorithm. This decision value can be calculated from the data density and relative distance of the data, which is existing technology and will not be elaborated here. Unlike the traditional DPC algorithm, the density calculation method in this scheme is the same as the above-mentioned calculation method for similar data density. The number of cluster centers refers to a human-set constant, which represents the number of cluster centers when performing subsequent clustering. The target decision value group refers to the combination of the top d largest decision values ​​in the decision value set, where d is the number of cluster centers. The preliminary clustering dataset refers to the set of cluster distributions centered on the cluster center data. This preliminary clustering dataset initially contains only cluster center data, and other data will be continuously added later. It should be noted that when extracting cluster center data sequentially from the cluster center data group, the cluster center data with the largest target decision value should be extracted first.

[0135] Furthermore, the clustering radius refers to the distance threshold used to determine the data range that each cluster center can cover during the clustering process. The clustering radius is obtained by: obtaining the data density of the cluster center data in the same dataset; identifying the average density of all data densities corresponding to all data of the same type in the same dataset; identifying the nearest neighbor datasets of the cluster center data; identifying the nearest neighbor data in the nearest neighbor dataset that is farthest from the cluster center data; identifying the farthest data distance between the farthest nearest neighbor data and the cluster center data; and calculating the clustering radius using the following formula: Clustering radius = (Data density of the cluster center data in the same dataset / Average density) × Farthest data distance. The candidate dataset refers to a set of multiple original datasets whose data distance from the cluster center in the original dataset is less than the clustering radius.

[0136] It is clear that the centrality correlation refers to a numerical value that quantifies the degree of correlation between candidate data and cluster center data, and this centrality correlation is calculated in the same way as the data correlation mentioned above. The average correlation refers to the average centrality correlation between all candidate data and cluster center data in the candidate dataset. The updated cluster dataset refers to the preliminary cluster dataset after adding candidate data. The unassigned dataset refers to the set of original data in the original dataset that has not been added to any preliminary cluster dataset.

[0137] Understandably, through the clustering process described above, data with similar sensitivity characteristics can be grouped together to form multiple clusters, providing a foundation for subsequent sensitivity adjustment based on cluster structure.

[0138] Specifically, the process of assigning the unassigned dataset to multiple target clustering datasets to obtain multiple current clustering datasets includes:

[0139] Perform the following operations on all unassigned data in the unassigned dataset:

[0140] Obtain the unassigned nearest neighbor dataset of unassigned data, and sequentially extract the target cluster dataset from multiple target cluster datasets;

[0141] Based on the unassigned nearest neighbor dataset, cluster association analysis is performed on the target clustering dataset and the unassigned data to obtain cluster association values;

[0142] Summarize the cluster association values ​​corresponding to the target clustering dataset to obtain a cluster association value set. Identify the maximum association value in the cluster association value set and add the unassigned data to the target clustering dataset corresponding to the maximum association value to complete the assignment of the unassigned data.

[0143] When each unassigned data in the unassigned dataset has been assigned, the multiple target clustering datasets are recorded as multiple current clustering datasets.

[0144] It should be explained that the "unassigned nearest neighbor dataset" refers to the nearest neighbor dataset of the unassigned data within the same type of dataset. The steps for obtaining this unassigned nearest neighbor dataset are the same as those for obtaining the current nearest neighbor dataset, and will not be repeated here. The "cluster association value" refers to a numerical value that quantifies the degree of association between the unassigned data and the target cluster dataset. The larger the cluster association value, the greater the degree of association between the unassigned data and the target cluster dataset, that is, the greater the probability that the unassigned data will be assigned to the target cluster dataset. The "maximum association value" refers to the cluster association value with the largest value in the cluster association value set.

[0145] In detail, the step of performing cluster association analysis on the target clustering dataset and the unassigned data based on the unassigned nearest neighbor dataset to obtain cluster association values ​​includes:

[0146] The unassigned shared dataset between the unassigned nearest neighbor dataset and the target clustering dataset is statistically analyzed.

[0147] Calculate the unallocated distance and the number of unallocated shares between each unallocated shared data in the unallocated data and the unallocated shared data set respectively, to obtain the unallocated distance set and the unallocated share set;

[0148] Cluster association values ​​are calculated based on the unassigned distance set and the unassigned shared data set, where the cluster association values ​​are represented as follows:

[0149]

[0150] Where Y represents the cluster association value, N represents the number of unassigned distances in the unassigned distance set or the number of unassigned shared numbers in the unassigned shared number set, and exp(*) represents an exponential function with the natural constant as the base. This represents the i-th unassigned distance in the unassigned distance set. Let represent the i-th unallocated shared number in the unallocated shared number set, and let a represent the nearest neighbor parameter.

[0151] It should be explained that the unassigned shared dataset refers to the intersection between the unassigned nearest neighbor dataset and the target clustering dataset. The unassigned distance refers to the Euclidean distance between unassigned data and unassigned shared data. The unassigned shared number refers to the number of shares between unassigned data and unassigned shared data. This unassigned shared number is obtained in the same way as the step described above, which involves statistically analyzing the intersection of the shared comparison dataset and the comparison dataset to obtain the shared data intersection and then calculating the number of shares in the shared data intersection. Therefore, it will not be repeated here.

[0152] Furthermore, the formula for calculating the cluster association value above uses an exponential decay function to combine the distance and shared quantity between unassigned data and its nearest neighbors in the target cluster dataset, calculating a value reflecting the strength of the association between unassigned data and the target cluster dataset. The effect of this calculation method is that data that is closer and shares more information is more likely to be assigned to the current cluster, thereby improving the accuracy and rationality of cluster assignment.

[0153] S6. Extract the data to be modeled from the same dataset, count the original data frequency of the data to be modeled in the dataset to be stratified, and adjust the original data frequency using multiple current clustering datasets to obtain the target data frequency.

[0154] It is clear that the data to be modeled refers to similar data within the same dataset. This data is used for subsequent target sensitivity calculations. This naming convention is used to differentiate this data from other similar data in terms of its intended use. The original data frequency refers to the frequency of the data to be modeled appearing in the dataset to be stratified. The original data frequency is calculated by: obtaining the number of times the data to be modeled appears in the dataset to be stratified, counting the total number of all data to be stratified in the dataset, and dividing the number of times the data to be modeled appears by the total number to obtain the original data frequency. The target data frequency refers to the adjusted original data frequency.

[0155] Furthermore, the reason for adjusting the original data frequency is that the original data frequency only reflects the frequency of the data in the overall dataset (the dataset to be stratified), without considering its distribution characteristics in the clusters. By combining the category frequency, cluster frequency, and distance between cluster centers, the representativeness of the data in its cluster and its deviation from the cluster center can be more accurately reflected, thereby more reasonably assessing its privacy risks after the dataset to be stratified is released.

[0156] In detail, the step of adjusting the original data frequency using multiple current clustering datasets to obtain the target data frequency includes:

[0157] Obtain the total number of similar data in the same dataset and the total number of data to be stratified in the dataset to be stratified. Calculate the category frequency based on the total number of similar data and the total number of data to be stratified, where the category frequency is the ratio of the total number of similar data to the total number of data to be stratified.

[0158] Identify the cluster dataset to be modeled from among multiple current cluster datasets, and count the number of clusters to be modeled in the cluster dataset to be modeled.

[0159] The clustering frequency is calculated based on the number of clusters to be modeled and the total number of data in the same category. The clustering frequency is the ratio of the number of clusters to be modeled to the total number of data in the same category.

[0160] Calculate the intra-cluster center distance of the data to be modeled based on the clustering dataset to be modeled;

[0161] The target data frequency is obtained by adjusting the original data frequency based on the category frequency, cluster frequency, and intra-cluster center distance.

[0162] It should be explained that the total number of similar data refers to the number of similar data within the same dataset. The total number of data to be stratified refers to the number of data to be stratified within the dataset to be stratified. The category frequency represents the proportion of that category of data in the total (i.e., the dataset to be stratified). The higher the category frequency, the more common that category of data is, and the lower the relative privacy risk. The cluster dataset to be modeled refers to the current cluster dataset where the data to be modeled is located. The number of clusters to be modeled refers to the number of data in the cluster dataset to be modeled. The higher the clustering frequency, the higher the proportion of the current cluster dataset where the data to be modeled is located within the same dataset. The intra-cluster centroid distance refers to the data distance between the data to be modeled and the cluster center data in the cluster dataset to be modeled. The larger the intra-cluster centroid distance, the more the data to be modeled deviates from the cluster center, and the worse its representativeness. The calculation formula for adjusting the original data frequency based on the category frequency, clustering frequency, and intra-cluster centroid distance is: f m =f0×(f clu / f cla )×[1 / (1+D clu )], where f m f represents the target data frequency, f0 represents the original data frequency, f clu f represents the clustering frequency. cla D represents the frequency of species. clu The formula above, representing the distance between cluster centers, means that the ratio of cluster frequency to class frequency is used to amplify the frequency of the original data to highlight the representativeness of the clusters. Then, [1 / (1+D] is used to further amplify the frequency of the clusters. clu The attenuation term of the model suppresses the weights of data that deviate from the center, thereby comprehensively measuring the true probability and importance of the data to be modeled in the population and cluster, and thus more accurately reflecting its privacy risks.

[0163] S7. Identify the preliminary sensitivity corresponding to the data to be modeled in the baseline sensitivity set, and correct the preliminary sensitivity based on the target data frequency and a preset public external database to obtain the target sensitivity.

[0164] It is clear that the preliminary sensitivity refers to the baseline sensitivity corresponding to the data to be modeled. This preliminary sensitivity is obtained by: determining the type of baseline data corresponding to the data to be modeled, identifying the baseline sensitivity corresponding to that type of baseline data in the baseline sensitivity set, and recording this baseline sensitivity as the preliminary sensitivity. The publicly available external database refers to publicly available data sources that may be obtained by attackers and used for re-identification attacks, such as: equipment business registration information, publicly available equipment model databases, lists of company websites, etc. The target sensitivity refers to the preliminary sensitivity after correction.

[0165] Furthermore, the lower the frequency of the target data, the rarer the data to be modeled. This means the data appears sparser in both the overall dataset (the dataset to be stratified) and within its clusters (the current clustering dataset), making it easier for attackers to exploit and re-identify the data. Therefore, the baseline sensitivity (i.e., the subsequent preliminary sensitivity) needs to be increased to apply stronger differential privacy noise protection. When revising the preliminary sensitivity, in addition to considering the frequency of the target data, external attacks also need to be considered, such as link attacks and background knowledge attacks. Link attacks refer to attackers using records in publicly available external databases that have the same or similar attributes as the dataset to be stratified, and then re-identifying individuals through attribute matching. Background knowledge attacks refer to attackers using their prior knowledge related to the data to be stratified (such as industry common sense, publicly available company information, etc.) to infer from the published data, thereby reconstructing or speculating sensitive information.

[0166] In detail, the process of refining the initial sensitivity based on the target data frequency and a preset public external database to obtain the target sensitivity includes:

[0167] The initial sensitivity is adjusted using the target data frequency to obtain the adjusted sensitivity;

[0168] The frequency of occurrence of the data to be modeled is counted in a publicly available external database, and the probability of a link attack is calculated based on the frequency of occurrence.

[0169] Build a background knowledge base, and filter the public external database based on the background knowledge base to obtain a public external sub-database;

[0170] Calculate the probability of background knowledge attacks based on publicly available external sub-libraries and the data to be modeled;

[0171] The maximum attack probability of background knowledge attack and link attack is identified, and the sensitivity is adjusted for external attacks based on the maximum attack probability to obtain the target sensitivity.

[0172] Understandably, the frequency of occurrence refers to the number of times the data to be modeled appears in a publicly available external database. The link attack probability refers to the probability that an attacker might successfully link to and identify individual information based on the publicly available external database. This link attack probability is obtained as follows: when the frequency of occurrence is 0, it means that the data to be stratified does not exist in the publicly available external database, that is, the attacker cannot query relevant information about the data to be stratified from the publicly available external database, and the link attack probability is recorded as 0. When the frequency of occurrence is not 0, the link attack probability is recorded as the reciprocal of the frequency of occurrence. When the frequency of occurrence is not 0, it means that the attacker may be able to query relevant information about the data to be stratified from the publicly available external database. For example, when the frequency of occurrence is 1, it means that there is data in the publicly available external database that is completely identical and unique to the data to be stratified. This means that the attacker is very likely to uniquely identify an individual through this data, and therefore the link attack probability is 1.

[0173] Furthermore, the background knowledge base refers to the set of prior knowledge that an attacker might possess related to the data to be layered. This background knowledge base is constructed by extracting typical attack patterns and data association rules from historical attack cases, public reports, industry common sense, and other sources. The public external sub-database refers to a filtered public external database. Filtering the public external database based on the background knowledge base means using the attack patterns and association rules in the background knowledge base to filter the public external database, retaining only records related to the attack patterns and association rules in the background knowledge base. The background knowledge attack probability refers to the probability that an attacker can infer sensitive information from the published data using their background knowledge. The step of calculating the background knowledge attack probability based on the public external sub-database and the data to be modeled is the same as the step of statistically analyzing the frequency of occurrence of the data to be modeled in the public external database and calculating the link attack probability based on the frequency of occurrence. The maximum attack probability refers to the larger of the background knowledge attack probability and the link attack probability.

[0174] Importantly, the above-mentioned adjustment sensitivity is expressed as:

[0175] F t =F c ×[1+α×tanh(γ×(0.5-f m ))]

[0176] Among them, F t Indicates the degree of adjustment, F c The initial sensitivity is indicated by α, which represents a manually set adjustment coefficient, denoted as 0.3. tanh(*) represents the hyperbolic tangent function, which limits the output value to between -1 and 1. γ represents a manually set adjustment sensitivity value; the larger this value, the more clearly the distinction between high and low frequencies of the target data. Optionally, this adjustment sensitivity value is set to 10. mIndicates the target data frequency.

[0177] In detail, the step of adjusting the sensitivity based on the maximum attack probability to obtain the target sensitivity includes:

[0178] Target sensitivity is calculated using the following formula:

[0179]

[0180] Among them, F m F represents target sensitivity. t P represents the sensitivity adjustment, θ represents the preset attack sensitivity value, and P represents the sensitivity adjustment. max This indicates the maximum probability of an attack. This indicates the preset compression factor.

[0181] It should be explained that the attack sensitivity value refers to a manually set parameter used to control the intensity of sensitivity correction based on external attack risks. A larger attack sensitivity value indicates a more sensitive response to external attack risks. For example, a value of 0 indicates that the impact of external attacks on sensitivity adjustment is completely ignored. The compression factor is a manually set parameter used to suppress the excessive influence of extreme probability values ​​on sensitivity correction. This compression factor is set to prevent P... max The compression factor can be set to 2 to prevent excessive amplification of the output when it approaches 1.

[0182] S8. Summarize the target sensitivities corresponding to the data to be modeled to obtain a target sensitivity set. Use the target sensitivity set to allocate sensitivity to the same type of dataset to obtain an allocated dataset.

[0183] It is clear that the allocated dataset includes multiple allocated data, wherein the allocated data refers to the same type of data in the same type of dataset after sensitivity allocation, wherein sensitivity allocation refers to pairing the target sensitivity in the target sensitivity set with the corresponding same type of data in the same type of dataset to obtain the allocated data, that is, the allocated data is a combination of the target sensitivity and the corresponding same type of data.

[0184] S9. Summarize the corresponding allocation datasets of the same type of dataset to obtain multiple allocation datasets, and complete the automatic sensitivity modeling for the entire life cycle of the device.

[0185] It is clear that obtaining multiple allocation datasets completes the sensitivity modeling for each data point in the dataset to be stratified.

[0186] For example, taking the full lifecycle management of laboratory equipment (such as gas chromatographs, mass spectrometers, electronic balances, etc.) in a provincial-level testing and inspection institution as an example, the equipment data sources for the stratified dataset include:

[0187] Procurement stage: Equipment model, manufacturer, procurement contract number, supplier information;

[0188] Acceptance phase: Equipment serial number, acceptance report number, person in charge of acceptance;

[0189] Usage phase: Detection item name, sample number, operator employee number, experimental data log, and equipment operation sensor time sequence data;

[0190] Maintenance phase: maintenance time, maintenance records, repair vendor;

[0191] Verification / calibration stage: calibration organization, calibration report number, traceability certificate information;

[0192] Discontinuation phase: Reason for discontinuation, approver of discontinuation, date of discontinuation;

[0193] Scrapping stage: Scrapping decision number, scrapping method, and scrapping execution unit.

[0194] In this embodiment, the testing and inspection agency inputs the dataset to be stratified into the modeling system of the present invention and executes the following process:

[0195] Baseline sensitivity assignment: The data administrator assigns initial sensitivity values ​​to the data to be stratified for different types of baseline data. For example, the initial sensitivity value for experimental data logs is 0.9, and the initial sensitivity value for equipment models is 0.3.

[0196] Classification and correlation assessment: The dataset to be stratified is divided into multiple similar datasets (such as "identity-related class", "detection result class", "device information class"), and the density and sharing degree between similar data are calculated to obtain the data correlation set.

[0197] Clustering and Reassignment: Density peak clustering groups highly correlated data into the same cluster, that is, clustering is performed on each of the same datasets in multiple datasets of the same type.

[0198] Frequency adjustment: It was found that the "experimental data log" appeared with low frequency in the dataset to be stratified, but its centrality in clustering was high. Therefore, its adjustment sensitivity was increased.

[0199] External attack mitigation: Querying external laboratory qualification databases and equipment registration information (i.e., these are publicly available external databases) revealed that the combination of "equipment model A + supplier B" frequently appears in publicly available external databases, suggesting a high probability of link attacks. Therefore, the target sensitivity values ​​for equipment model A and supplier B were increased.

[0200] Combining sensitivity output with differential privacy: When multiple allocation datasets are formed, noise injection of different degrees is required for each allocation dataset: large noise injection (such as Laplace mechanism ε=0.3) is used for allocation data with target sensitivity ≥0.8 (such as experimental data logs, operator employee numbers), and light noise (such as ε=1.5) is used for allocation data with target sensitivity <0.5 (such as equipment model, scrap number), thereby balancing the usability of data analysis and privacy protection.

[0201] This embodiment demonstrates that the present invention can perform refined sensitivity modeling of the entire lifecycle data of testing and inspection institutions' equipment. It can identify high-risk data (such as experimental data logs and operator information) while avoiding overprotection of low-sensitivity data, thereby ensuring the data privacy and security of the institution and enhancing the value of data sharing and utilization.

[0202] To address the problems described in the background, this invention first assigns sensitivity scores to each benchmark data type in the benchmark data set, resulting in a benchmark sensitivity set. This step transforms subjective data sensitivity into a calculable value in the 0-1 range, avoiding noise intensity mismatch caused by differing evaluation scales among different personnel. Next, it evaluates the intra-set data correlation based on similar datasets, obtaining data association sets. This step integrates local density and shared information through data correlation, reflecting the true correlation strength within similar data sets. This overcomes the shortcomings of traditional distance metrics that ignore distribution density, making the subsequent selection of cluster centers closer to the actual data topology and reducing sensitivity estimation bias caused by misclassification. Furthermore, this invention uses data association sets to perform density clustering on similar datasets, obtaining multiple current cluster datasets. This step embeds a correlation-driven radius calculation and unassigned point reassignment strategy into DPC clustering, preserving the suitability of the density peak algorithm for clusters of arbitrary shapes. Furthermore, this invention utilizes correlation to filter out noise points, ensuring high cohesion of each cluster at the sensitivity level, providing a clean cluster structure for subsequent frequency adjustment. In addition, the invention adjusts the original data frequency using multiple current clustering datasets to obtain the target data frequency. This step adjusts the original data frequency through category frequency, cluster frequency, and intra-cluster center distance, more realistically reflecting the representativeness and privacy risks of the data in the clustering, providing a more reasonable frequency basis for subsequent sensitivity correction. Finally, based on the target data frequency and publicly available external databases, the initial sensitivity is corrected to obtain the target sensitivity. This step explicitly incorporates the link probability of the publicly available external database and the inference probability of the background knowledge base into the sensitivity correction formula, forming a quantitative response mechanism to external re-identification attacks. This avoids the over-protection or under-protection caused by the uniform noise addition to all data in traditional differential privacy. Therefore, this invention can improve the accuracy and intelligence of sensitivity assessment and enhance the accuracy of privacy protection.

[0203] like Figure 2 The diagram shown is a functional block diagram of an automatic sensitivity modeling system for the entire life cycle of equipment provided in an embodiment of the present invention.

[0204] The automatic sensitivity modeling system 100 for the entire lifecycle of equipment described in this invention can be installed in electronic devices. Depending on the functions implemented, the automatic sensitivity modeling system 100 for the entire lifecycle of equipment may include a modeling instruction receiving module 101, a data correlation evaluation module 102, a data frequency adjustment module 103, and a sensitivity allocation module 104. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and which are stored in the memory of the electronic device.

[0205] The modeling instruction receiving module 101 is used to receive sensitivity modeling instructions, determine the dataset to be layered based on the sensitivity modeling instructions, the dataset to be layered includes data generated during the equipment procurement, acceptance, use, maintenance, verification, shutdown and scrapping stages, construct a benchmark data category set, and assign sensitivity scores to each benchmark data category in the benchmark data category set to obtain a benchmark sensitivity set.

[0206] The data correlation evaluation module 102 is used to classify the hierarchical dataset to be classified using a benchmark data category set to obtain multiple similar datasets. Among them, the similar datasets in the multiple similar datasets correspond one-to-one with the benchmark data categories in the benchmark data category set. The similar datasets are extracted sequentially from the multiple similar datasets, and the data correlation degree within the set is evaluated based on the similar datasets to obtain a data correlation group set. The data correlation group set includes multiple data correlation groups, and the data correlation group includes: data correlation degree, shared data, and similar data.

[0207] The data frequency adjustment module 103 is used to perform density clustering on similar datasets using data association sets to obtain multiple current cluster datasets, extract the data to be modeled in the similar datasets, count the original data frequency of the data to be modeled in the dataset to be stratified, and adjust the original data frequency using multiple current cluster datasets to obtain the target data frequency.

[0208] The sensitivity allocation module 104 is used to identify the preliminary sensitivity corresponding to the data to be modeled in the baseline sensitivity set, correct the preliminary sensitivity based on the target data frequency and a preset public external database to obtain the target sensitivity, summarize the target sensitivity corresponding to the data to be modeled to obtain the target sensitivity set, use the target sensitivity set to allocate sensitivity to the same type of dataset to obtain the allocated dataset, and summarize the allocated datasets corresponding to the same type of dataset to obtain multiple allocated datasets.

[0209] In detail, the modules in the automatic sensitivity modeling system 100 for the entire lifecycle of equipment described in this embodiment of the invention employ the same methods as described above. Figure 1 The method used is the same as the automatic sensitivity modeling method for the entire life cycle of equipment described in the article, and can produce the same technical effect, so it will not be repeated here.

[0210] like Figure 3 The diagram shown is a structural schematic of an electronic device that implements an automatic sensitivity modeling method for the entire lifecycle of a device, according to an embodiment of the present invention.

[0211] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a sensitivity automatic modeling method program for the entire life cycle of the device.

[0212] The memory 11 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device 1. Furthermore, the memory 11 includes both internal storage units and external storage devices of the electronic device 1. The memory 11 can be used not only to store application software and various types of data installed on the electronic device 1, such as code for an automatic sensitivity modeling method program for the entire lifecycle of the device, but also to temporarily store data that has been output or will be output.

[0213] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., a sensitivity automatic modeling method program for the entire device lifecycle) and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0214] The bus 12 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 can be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0215] Figure 3 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0216] For example, although not shown, the electronic device 1 may also include a power supply (such as a battery) to power various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management system, thereby enabling functions such as charging management, discharging management, and power consumption management through the power management system. The power supply may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0217] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the electronic device 1 and other electronic devices.

[0218] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired or wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device 1 and to display a visual user interface.

[0219] The sensitivity automatic modeling method program for the entire lifecycle of the device, stored in the memory 11 of the electronic device 1, is a combination of multiple instructions. When run in the processor 10, it can achieve the following:

[0220] Receive sensitivity modeling instructions, determine the dataset to be stratified based on the sensitivity modeling instructions, the dataset to be stratified includes data generated during the equipment procurement, acceptance, use, maintenance, verification, decommissioning and scrapping stages;

[0221] Construct a set of benchmark data types, and assign sensitivity scores to each benchmark data type in the set to obtain a benchmark sensitivity set;

[0222] The dataset to be stratified is classified using a benchmark data category set to obtain multiple datasets of the same type. Each dataset of the same type corresponds one-to-one with a benchmark data category in the benchmark data category set.

[0223] Extract similar datasets sequentially from multiple similar datasets, evaluate the data correlation within the set based on the similar datasets, and obtain a data correlation set. The data correlation set includes multiple data correlation groups, and the data correlation group includes: data correlation, shared data, and similar data.

[0224] Density clustering of similar datasets is performed using data association sets to obtain multiple current clustered datasets;

[0225] Extract the data to be modeled from the same dataset, count the original data frequency of the data to be modeled in the dataset to be stratified, and adjust the original data frequency using multiple current clustering datasets to obtain the target data frequency.

[0226] The initial sensitivity corresponding to the data to be modeled is identified in the benchmark sensitivity set. The initial sensitivity is then corrected based on the target data frequency and a preset public external database to obtain the target sensitivity.

[0227] Summarize the target sensitivities corresponding to the data to be modeled to obtain a target sensitivity set. Use the target sensitivity set to assign sensitivity to the same type of dataset to obtain an assigned dataset.

[0228] By aggregating the corresponding allocation datasets of similar datasets, multiple allocation datasets are obtained, enabling automatic sensitivity modeling for the entire lifecycle of devices.

[0229] Specifically, the processor 10's implementation method for the above instructions can be found in [reference needed]. Figures 1 to 3 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0230] Furthermore, if the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or system capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0231] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following:

[0232] Receive sensitivity modeling instructions, determine the dataset to be stratified based on the sensitivity modeling instructions, the dataset to be stratified includes data generated during the equipment procurement, acceptance, use, maintenance, verification, decommissioning and scrapping stages;

[0233] Construct a set of benchmark data types, and assign sensitivity scores to each benchmark data type in the set to obtain a benchmark sensitivity set;

[0234] The dataset to be stratified is classified using a benchmark data category set to obtain multiple datasets of the same type. Each dataset of the same type corresponds one-to-one with a benchmark data category in the benchmark data category set.

[0235] Extract similar datasets sequentially from multiple similar datasets, evaluate the data correlation within the set based on the similar datasets, and obtain a data correlation set. The data correlation set includes multiple data correlation groups, and the data correlation group includes: data correlation, shared data, and similar data.

[0236] Density clustering of similar datasets is performed using data association sets to obtain multiple current clustered datasets;

[0237] Extract the data to be modeled from the same dataset, count the original data frequency of the data to be modeled in the dataset to be stratified, and adjust the original data frequency using multiple current clustering datasets to obtain the target data frequency.

[0238] The initial sensitivity corresponding to the data to be modeled is identified in the benchmark sensitivity set. The initial sensitivity is then corrected based on the target data frequency and a preset public external database to obtain the target sensitivity.

[0239] Summarize the target sensitivities corresponding to the data to be modeled to obtain a target sensitivity set. Use the target sensitivity set to assign sensitivity to the same type of dataset to obtain an assigned dataset.

[0240] By aggregating the corresponding allocation datasets of similar datasets, multiple allocation datasets are obtained, enabling automatic sensitivity modeling for the entire lifecycle of devices.

[0241] In the embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and actual implementations may have other classification methods.

[0242] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0243] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0244] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0245] Finally, 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 it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for sensitivity automatic modeling for the full life cycle of a device, characterized in that, The method comprises: receiving sensitivity modeling instructions, determining a to-be-layered data set based on the sensitivity modeling instructions, the to-be-layered data set comprising data generated in the stages of device procurement, acceptance, use, maintenance, calibration, decommissioning and scrapping; constructing a benchmark data category set, and assigning a sensitivity score to each benchmark data category in the benchmark data category set to obtain a benchmark sensitivity set; classifying the to-be-layered data set using the benchmark data category set to obtain a plurality of homogeneous data sets, wherein each homogeneous data set in the plurality of homogeneous data sets corresponds to a benchmark data category in the benchmark data category set; extracting homogeneous data sets from the plurality of homogeneous data sets in sequence, and performing intra-set data correlation degree evaluation based on the homogeneous data sets to obtain a data correlation group set, wherein the data correlation group set comprises a plurality of data correlation groups, and each data correlation group comprises a data correlation degree, shared data and homogeneous data; performing density clustering on the homogeneous data sets using the data correlation group set to obtain a plurality of current clustering data sets; extracting to-be-modeled data from the homogeneous data sets, and statistically analyzing the original data frequency of the to-be-modeled data in the to-be-layered data set, adjusting the original data frequency using the plurality of current clustering data sets to obtain a target data frequency; confirming a preliminary sensitivity corresponding to the to-be-modeled data in the benchmark sensitivity set, and correcting the preliminary sensitivity based on the target data frequency and a preset public external database to obtain a target sensitivity; summarizing the target sensitivities corresponding to the to-be-modeled data to obtain a target sensitivity set, and performing sensitivity allocation on the homogeneous data sets using the target sensitivity set to obtain an allocated data set; summarizing the allocated data sets corresponding to the homogeneous data sets to obtain a plurality of allocated data sets, and completing the automatic modeling of the sensitivity for the entire life cycle of the device.

2. The method for automatic modeling of sensitivity throughout the life cycle of equipment as claimed in claim 1, wherein, The intra-set data correlation degree evaluation based on the homogeneous data sets to obtain the data correlation group set comprises: extracting homogeneous data from the homogeneous data sets in sequence, and removing the homogeneous data from the homogeneous data sets to obtain a comparison data set; calculating the data distance between the homogeneous data and each comparison data in the comparison data set to obtain a data distance set, and identifying a near neighbor distance set in the data distance set according to a preset near neighbor parameter, wherein the number of near neighbor distances in the near neighbor distance set is equal to the near neighbor parameter; confirming a current near neighbor data set in the comparison data set based on the near neighbor distance set, wherein each current near neighbor data in the current near neighbor data set corresponds to a near neighbor distance in the near neighbor distance set; performing density evaluation on the homogeneous data according to the near neighbor distance set to obtain a homogeneous data density; regarding the comparison data set as a shared data set, extracting shared data from the shared data set in sequence, and calculating the data correlation degree between the shared data and the homogeneous data according to the homogeneous data density; merging the data correlation degree, the shared data and the homogeneous data to obtain a data correlation group; summarizing the data correlation groups corresponding to the shared data to obtain a plurality of data correlation groups; regarding the comparison data set as a homogeneous data set, and returning to the step of extracting homogeneous data from the homogeneous data set in sequence until the comparison data set is empty, and summarizing the plurality of data correlation groups corresponding to the homogeneous data to obtain the data correlation group set.

3. The method for automatic modeling of sensitivity throughout the life cycle of equipment as claimed in claim 2, wherein, The data correlation degree between the shared data and the same type data is calculated according to the density of the same type data, and the data correlation degree is calculated according to the shared number, the shared data density, the same type data density and the shared data distance. A shared comparison data set and a shared data density of the shared data are obtained; An intersection of the shared comparison data set and the comparison data set is calculated to obtain a shared data intersection, and a shared number in the shared data intersection is counted; The shared data distance between the shared data and the same type data is confirmed; The data correlation degree is calculated according to the shared number, the shared data density, the same type data density and the shared data distance, and the data correlation degree is represented as: wherein ε represents the data correlation degree, S represents the sharing number, ρ t represents the same kind data density, ρ g represents the shared data density, D g represents the shared data distance.

4. The method for automatic modeling of sensitivity throughout the life cycle of equipment as claimed in claim 3, wherein, The same type data set is density clustered by using the data correlation group set to obtain a plurality of current clustering data sets, and the method comprises the steps of: A decision value of each same type data in the same type data set is calculated to obtain a decision value set; A target decision value group is identified in the decision value set according to a preset clustering center number, and a clustering center data group is confirmed in the same type data set according to the target decision value group; The clustering center data group is removed from the same type data set to obtain an original data set; A clustering center data is extracted from the clustering center data group in sequence to construct a preliminary clustering data set of the clustering center data, and the preliminary clustering data set only contains the clustering center data; A clustering radius is calculated by clustering the clustering center data to identify a candidate data set in the original data set according to the clustering radius; A candidate data is extracted from the candidate data set in sequence, a center correlation degree between the candidate data and the clustering center data is calculated, and it is judged whether the center correlation degree is greater than a preset average correlation degree; If the center correlation degree is greater than the average correlation degree, the candidate data is added to the preliminary clustering data set to obtain an updated clustering data set; The updated clustering data set is taken as the preliminary clustering data set, and the step of extracting the candidate data from the candidate data set in sequence is returned until the candidate data set is empty; The updated clustering data set when the candidate data set is empty is recorded as a target clustering data set; The target clustering data sets corresponding to each clustering center data are summarized to obtain a plurality of target clustering data sets; Unallocated data sets in the original data set are confirmed, and the unallocated data sets are allocated to the plurality of target clustering data sets to obtain a plurality of current clustering data sets.

5. The method for automatic modeling of sensitivity throughout the life cycle of equipment as claimed in claim 4, wherein, The unallocated data sets are allocated to the plurality of target clustering data sets to obtain a plurality of current clustering data sets, and the method comprises the steps of: The following operations are performed on each unallocated data in the unallocated data set: An unallocated neighbor data set of the unallocated data is obtained, and a target clustering data set is extracted from the plurality of target clustering data sets in sequence; A clustering correlation analysis is performed on the target clustering data set and the unallocated data based on the unallocated neighbor data set to obtain a clustering correlation value; The clustering correlation values corresponding to the target clustering data sets are summarized to obtain a clustering correlation value set, a maximum correlation value is identified in the clustering correlation value set, and the unallocated data is added to the target clustering data set corresponding to the maximum correlation value to complete the allocation of the unallocated data; When each unallocated data in the unallocated data set is allocated, the plurality of target clustering data sets are recorded as a plurality of current clustering data sets.

6. The method for automatic modeling of sensitivity throughout the life cycle of equipment as claimed in claim 5, wherein, The clustering correlation analysis is performed on the target clustering data set and the unallocated data based on the unallocated neighbor data set to obtain a clustering correlation value, and the method comprises the steps of: counting an unassigned shared data set between the unassigned neighbor data set and the target clustering data set; calculating an unassigned distance and an unassigned shared number between each unassigned shared data in the unassigned data and the unassigned shared data set respectively, to obtain an unassigned distance set and an unassigned shared number set; calculating a clustering correlation value according to the unassigned distance set and the unassigned shared number set, wherein the clustering correlation value is expressed as: where Y denotes a cluster correlation value, N denotes a number of unassigned distances in an unassigned distance set or a number of unassigned sharing numbers in an unassigned sharing number set, and exp(*) denotes an exponential function with a natural constant as a base, denotes an i-th unassigned distance in the unassigned distance set, denotes an i-th unassigned sharing number in the unassigned sharing number set, and a denotes a near neighbor parameter.

7. The method for automatic modeling of sensitivity throughout the life cycle of equipment as claimed in claim 6, wherein, the target data frequency is obtained by adjusting the original data frequency by using the plurality of current clustering data sets, including: obtaining a same data total number of the same data set and a to-be-layered data total number of the to-be-layered data set respectively, and calculating a category frequency according to the same data total number and the to-be-layered data total number, wherein the category frequency is a ratio of the same data total number to the to-be-layered data total number; confirming a to-be-modeled clustering data set in which the to-be-modeled data is located in the plurality of current clustering data sets, and counting a to-be-modeled clustering number of the to-be-modeled clustering data set; calculating a clustering frequency according to the to-be-modeled clustering number and the same data total number, wherein the clustering frequency is a ratio of the to-be-modeled clustering number to the same data total number; calculating an intra-cluster center distance of the to-be-modeled data based on the to-be-modeled clustering data set; adjusting the original data frequency according to the category frequency, the clustering frequency and the intra-cluster center distance, to obtain the target data frequency.

8. The method for automatic modeling of sensitivity throughout the life cycle of equipment as claimed in claim 7, wherein, the target sensitivity is obtained by correcting the preliminary sensitivity based on the target data frequency and a preset public external database, including: frequency adjusting the preliminary sensitivity by using the target data frequency to obtain an adjusted sensitivity; counting the occurrence frequency of the to-be-modeled data in the public external database, and calculating a link attack probability based on the occurrence frequency; constructing a background knowledge base, and filtering in the public external database based on the background knowledge base to obtain a public external sub-base; calculating a background knowledge attack probability based on the public external sub-base and the to-be-modeled data; confirming a maximum attack probability of the background knowledge attack probability and the link attack probability, and externally attacking and correcting the adjusted sensitivity according to the maximum attack probability to obtain the target sensitivity.

9. The method for automatic modeling of sensitivity throughout the life cycle of equipment as claimed in claim 8, wherein, the target sensitivity is obtained by externally attacking and correcting the adjusted sensitivity according to the maximum attack probability, including: the target sensitivity is calculated by using the following formula: Wherein, F m represents the target sensitivity, F t represents the adjustment sensitivity, θ represents the preset attack sensitivity value, P max represents the maximum attack probability, represents the preset compression factor.

10. A sensitivity auto-modeling system for the full lifecycle of a device, characterized in that, the system includes: a modeling instruction receiving module, configured to receive a sensitivity modeling instruction, determine a to-be-layered data set based on the sensitivity modeling instruction, the to-be-layered data set including data generated in the stages of device procurement, acceptance, use, maintenance, calibration, stop using and scrap, construct a benchmark data category set, and give a sensitivity score to each benchmark data category in the benchmark data category set to obtain a benchmark sensitivity set; a data correlation evaluation module, configured to classify the to-be-layered data set by using the benchmark data category set to obtain a plurality of same data sets, wherein each same data set in the plurality of same data sets corresponds to a benchmark data category in the benchmark data category set, sequentially extract same data sets in the plurality of same data sets, and evaluate the data correlation degree in the set based on the same data sets to obtain a data correlation group set, wherein the data correlation group set includes a plurality of data correlation groups, and each data correlation group includes data correlation degree, shared data and same data. The data frequency adjustment module is configured to perform density clustering on the same type of data sets by using the data association group set, to obtain a plurality of current clustering data sets, to extract the data to be modeled from the same type of data sets, to count the original data frequency of the data to be modeled in the data to be layered, to adjust the original data frequency by using the plurality of current clustering data sets, and to obtain a target data frequency. The sensitivity allocation module is configured to confirm the preliminary sensitivity corresponding to the data to be modeled in the benchmark sensitivity set, to correct the preliminary sensitivity based on the target data frequency and a preset public external database, to obtain a target sensitivity, to collect the target sensitivity corresponding to the data to be modeled to obtain a target sensitivity set, to allocate the sensitivity to the same type of data sets by using the target sensitivity set, to obtain an allocated data set, and to collect the allocated data set corresponding to the same type of data sets to obtain a plurality of allocated data sets.

Images