A safety data analysis method and system for a chemical industrial park

Abstract

The application discloses a kind of safety data analysis method and system for chemical industry park, the sensitive sample database of chemical industry park is obtained in the application, to construct sensitive data correlation diagram, and simultaneously construct the data query diagram of data to be analyzed;Then, using graph matching technique, similar sensitive sequence of data to be analyzed is matched in sensitive data correlation diagram;Then, based on similar sensitive sequence, the sensitivity of data is accurately calculated;Finally, in combination with access information, desensitization parameter of data to be analyzed is dynamically calculated, and based on this, data to be analyzed is desensitized;In this way, the application realizes the differential desensitization processing, avoids the problem of "one size fits all" caused by traditional uniform desensitization strategy, ensures the strict protection of high-risk data in specific scenario, and can maximize the analysis value of low-risk data under the premise of safety, significantly improves data utilization efficiency.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology applied to chemical industrial parks, specifically relating to a method and system for safety data analysis in chemical industrial parks. Background Technology

[0002] As a concentrated area for the production, storage, and use of a large number of hazardous chemicals, the safety data analysis and risk management of chemical industrial parks are a key focus of the industry. At the same time, with the development of IoT, big data, and artificial intelligence technologies, chemical industrial parks are now widely equipped with various sensors and monitoring systems that can collect environmental data, equipment status data, and operational process data in real time. This massive amount of data provides important basic support for risk warning, hazard investigation, and emergency decision-making in the parks.

[0003] In the safety management practice of chemical industrial parks, the accumulated historical accident data, hazard records, and abnormal operating condition data constitute a highly sensitive sample database. These data not only reflect the risk characteristics of the park, but also serve as an important basis for data comparison and risk identification. By conducting correlation analysis and similarity matching between newly added monitoring data and historical sensitive data, potential safety risks can be discovered in a timely manner, which can help the park achieve a shift from passive response to proactive early warning.

[0004] However, there is an inherent contradiction between security data sharing and privacy protection in chemical industrial parks. On the one hand, sufficient data sharing helps improve the accuracy of risk identification; on the other hand, the leakage of sensitive data may bring serious security risks. Existing data processing methods usually adopt a uniform de-identification strategy, which makes it difficult to differentiate processing according to the risk level of the data itself and the specific circumstances of the visitors. This either leads to over-de-identification, which affects the value of data analysis, or insufficient de-identification, which poses a risk of data leakage. Therefore, based on the aforementioned shortcomings, how to provide a secure data analysis method that can dynamically adjust the de-identification strategy according to the sensitivity of the data and the access scenario has become an urgent technical problem to be solved in the field of security data management in chemical industrial parks. Summary of the Invention

[0005] The purpose of this invention is to provide a safety data analysis method and system for chemical industrial parks, in order to solve the problem that existing technologies are unable to differentiate data based on the risk level of the data itself and the specific circumstances of the visitors, resulting in over- or under-sensitization.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a safety data analysis method for chemical industrial parks is provided, including: Obtain a database of sensitive samples from chemical industrial parks and the data to be analyzed from the chemical industrial parks; Based on the sensitive sample database, a sensitive data association graph is constructed. The sensitive data association graph contains multiple nodes, and nodes with association relationships are connected by edges. Each node corresponds to a sensitive sample data. Based on the data to be analyzed, a data query graph is constructed; Using a data query graph, data matching is performed in the sensitive data association graph to obtain the sensitive data sequence with the highest similarity to the data to be analyzed; Based on the sensitive data sequence, the sensitivity of the data to be analyzed is calculated; When accessing the data to be analyzed on the data analysis terminal, obtain the access information of the data analysis terminal; Based on the access information and the sensitivity, the desensitization parameters corresponding to the data to be analyzed are calculated, and the data to be analyzed is desensitized according to the desensitization parameters to obtain the desensitized data to be analyzed. The anonymized data to be analyzed is sent to the data analysis terminal so that the data analysis terminal can perform data analysis of the chemical industrial park based on the anonymized data.

[0007] Based on the aforementioned disclosure, this invention constructs a sensitive data association graph by acquiring a sensitive sample database of chemical industrial parks, and simultaneously constructs a data query graph for the data to be analyzed. Then, using graph matching technology, similar sensitive sequences of the data to be analyzed are matched in the sensitive data association graph. Subsequently, based on these similar sensitive sequences, the sensitivity of the data is accurately calculated. Finally, combined with access information from the data analysis terminal, the desensitization parameters of the data to be analyzed are dynamically calculated, and based on these parameters, the data to be analyzed is desensitized, resulting in desensitized data for subsequent data analysis. Thus, this invention achieves differentiated desensitization processing, avoiding the "one-size-fits-all" problem caused by traditional uniform desensitization strategies. It ensures strict protection of high-risk data in specific scenarios while maximizing the preservation of the analytical value of low-risk data under secure conditions, significantly improving data utilization efficiency. Therefore, this invention effectively resolves the inherent contradiction between secure data sharing and privacy protection in chemical industrial parks, providing reliable technical support for risk management in these parks.

[0008] In one possible design, a sensitive data association graph is constructed based on a sensitive sample database, including: Data cleaning is performed on each sensitive sample in the sensitive sample database to obtain several cleaned data sets. Each cleaned data point is vectorized to obtain the initial feature vector corresponding to each sensitive sample data point. Construct a sensitive sample matrix using all initial feature vectors; The covariance matrix of the sensitive sample matrix is ​​calculated, and based on the covariance matrix, the sensitive sample matrix is ​​subjected to feature dimensionality reduction processing to obtain the dimensionality-reduced sensitive sample matrix. Based on the dimension-reduced sensitive sample matrix, the optimal sensitive sample matrix is ​​constructed, where each row of the optimal sensitive sample matrix represents the feature vector corresponding to a sensitive sample data. The sensitive data association graph is constructed using the optimal sensitive sample matrix.

[0009] In one possible design, each row in the dimensionality-reduced sensitive sample matrix is ​​used to characterize the initial dimensionality-reduced feature vector of a sensitive sample data; Among them, the optimal sensitive sample matrix is ​​constructed based on the dimensionality-reduced sensitive sample matrix, including: For any row in the reduced-dimensionality sensitive sample matrix, calculate the information entropy of the initial reduced-dimensionality feature vector; Based on the reduced-dimensionality sensitive sample matrix, the standard deviation of each feature dimension in all initial reduced-dimensionality feature vectors is calculated, and the adaptive adjustment factor is determined based on all the standard deviations. Based on the information entropy, the adaptive adjustment factor, and the initial dimensionality reduction feature vector, the feature vector corresponding to the target data is calculated. After polling all rows of the dimensionality-reduced sensitive sample matrix, the feature vector of each sensitive sample data is obtained. In order to use all feature vectors, the optimal sensitive sample matrix is ​​constructed, wherein the target data is the sensitive sample data corresponding to the initial dimensionality reduction feature vector.

[0010] In one possible design, based on the information entropy, the adaptive adjustment factor, and the initial dimensionality reduction feature vector, the feature vector corresponding to the target data is calculated, including: The feature vector corresponding to the target data is calculated according to the following formula; ; In the formula, This represents the feature vector corresponding to the target data. This represents the initial dimensionality-reduced feature vector. This represents the information entropy. Indicates the weighting coefficient. This represents the adaptive adjustment factor. This represents the total number of initial dimensionality-reduced eigenvectors in the dimensionality-reduced sensitive sample matrix.

[0011] In one possible design, each node in the sensitive data association graph also corresponds to an attribute vector. The attribute vector of each node is the feature vector of the sensitive sample data corresponding to that node, and the data query graph and the sensitive data association graph have the same graph structure. Specifically, data matching is performed in the sensitive data association graph using a data query graph to obtain the sensitive data sequence with the highest similarity to the data to be analyzed, including: For the first query node in the data query graph, based on the feature vector of the first query node and the attribute vectors corresponding to each node in the sensitive data association graph, determine the set of starting points for the graph search from the sensitive data association graph; Node constraints are constructed, including node consistency constraints and node duplication constraints. The node consistency constraints are used to constrain the consistency of labels between nodes, and the node duplication constraints are used to constrain that the same node can only match one query node. Based on node constraints, and taking each search starting point in the graph search starting point set as the query path starting point, in the sensitive data association graph, node matching is performed on each query node in the data query graph to obtain the matching node of the data query graph under each query path. Based on the matching nodes under each query path, multiple initial sensitive data sequences are generated, and based on the multiple initial sensitive data sequences, the sensitive data sequence with the highest similarity to the data to be analyzed is determined.

[0012] In one possible design, based on node constraints and using each search starting point in the graph search starting point set as the query path starting point, node matching is performed on each query node within the data query graph in the sensitive data association graph, including: For any search starting point in the set of graph search starting points, take that search starting point as the starting point of the query path of the i-th query node in the data query graph, where the initial value of i is 2; In the sensitive data association graph, find the neighbor node of the starting point of the query path, and in the sensitive data association graph, find the similar node of the i-th query node; Find the intersection of similar nodes and their neighboring nodes to form the candidate node set for the i-th query node; Candidate nodes that satisfy the node constraints are selected from the candidate node set and used as initial matching nodes; Based on the feature vector of the i-th query node and the attribute vectors of each initial matching node, the initial matching node most similar to the i-th query node is selected from all initial matching nodes and used as the matching node of the i-th query node. Update any search starting point to the matching node of the i-th query node, increment i by 1, and re-use any search starting point as the query path starting point of the i-th query node in the data query graph until i equals m, thus obtaining the matching nodes of each query node, where m is the total number of query nodes.

[0013] In one possible design, any node in the sensitive data association graph also corresponds to an attribute vector, and the attribute vector of any node is the feature vector of the sensitive sample data corresponding to that node. The sensitivity of the data to be analyzed is calculated based on the sensitive data sequence, including: Calculate the similarity between the sensitive data sequence and the data to be analyzed; From the sensitive data association diagram, determine the attribute vector corresponding to each sensitive sample data in the sensitive data sequence; Calculate the feature entropy of the sensitive data sequence based on each attribute vector; Based on the feature entropy and the similarity, the sensitivity of the data to be analyzed is calculated.

[0014] In one possible design, the access information of the data analysis terminal includes: the access permissions of the data analysis terminal and the access query type of the data analysis terminal for the data to be analyzed; Specifically, based on the access information and the sensitivity, the de-identification parameters corresponding to the data to be analyzed are calculated, including: Based on the access query type and access permissions, determine the query type adjustment factor and permission factor; Based on the query type adjustment factor, permission factor, and sensitivity, the noise parameters corresponding to the data to be analyzed are calculated; Based on the noise parameters, random noise conforming to a Gaussian distribution is generated; The desensitization parameters are determined based on the random noise.

[0015] Secondly, a safety data analysis system for chemical industrial parks is provided, including: The acquisition unit is used to acquire the sensitive sample database of the chemical industrial park and the data to be analyzed in the chemical industrial park; The graph construction unit is used to construct a sensitive data association graph based on the sensitive sample database. The sensitive data association graph contains multiple nodes, and nodes with association relationships are connected by edges. Each node corresponds to a sensitive sample data. The graph construction unit is also used to construct a data query graph based on the data to be analyzed; The matching unit is used to perform data matching in the sensitive data association graph using the data query graph, so as to obtain the sensitive data sequence with the highest similarity to the data to be analyzed; A sensitivity identification unit is used to calculate the sensitivity of the data to be analyzed based on a sensitive data sequence; The access information acquisition unit is used to acquire access information of the data analysis terminal when accessing the data to be analyzed. The desensitization unit is used to calculate the desensitization parameters corresponding to the data to be analyzed based on the access information and the sensitivity, and to perform desensitization processing on the data to be analyzed based on the desensitization parameters to obtain the desensitized data to be analyzed. The desensitization unit is also used to send the desensitized data to be analyzed to the data analysis terminal, so that the data analysis terminal can perform data analysis of the chemical industrial park based on the desensitized data.

[0016] Thirdly, a safety data analysis device for chemical industrial parks is provided. Taking the device as an electronic device as an example, it includes a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the safety data analysis method for chemical industrial parks as described in the first aspect or any possible design of the first aspect.

[0017] Fourthly, a storage medium is provided, on which instructions are stored, which, when executed on a computer, perform the safety data analysis method for chemical industrial parks as described in the first aspect or any possible design of the first aspect.

[0018] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, causes the computer to perform the safety data analysis method for chemical industrial parks as described in the first aspect or any possible design of the first aspect.

[0019] Beneficial effects: (1) This invention constructs a sensitive data association graph by acquiring a sensitive sample database of a chemical industrial park, and simultaneously constructs a data query graph of the data to be analyzed; then, using graph matching technology, it matches similar sensitive sequences of the data to be analyzed in the sensitive data association graph; then, based on the similar sensitive sequences, it accurately calculates the sensitivity of the data; finally, it combines the access information of the data analysis terminal to dynamically calculate the desensitization parameters of the data to be analyzed, and based on this, it desensitizes the data to be analyzed, and obtains the desensitized data to be analyzed for subsequent data analysis; thus, this invention achieves differentiated desensitization processing, avoiding the "one-size-fits-all" problem caused by the traditional unified desensitization strategy, ensuring strict protection of high-risk data in specific scenarios, and maximizing the preservation of the analytical value of low-risk data under the premise of security, significantly improving data utilization efficiency; thus, this invention effectively solves the inherent contradiction between secure data sharing and privacy protection in chemical industrial parks, and can provide reliable technical support for risk management in chemical industrial parks. Attached Figure Description

[0020] Figure 1 A flowchart illustrating the steps of a safety data analysis method for chemical industrial parks provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a safety data analysis system for chemical industrial parks provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0022] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0023] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0024] Example: See Figure 1 As shown, the safety data analysis method for chemical industrial parks provided in this embodiment can be executed, but is not limited to, by computer equipment with certain computing resources, such as servers, edge computers, personal computers (PCs, which are multi-purpose computers of a size, price and performance suitable for personal use; desktop computers, laptops to mini-laptops and tablets and ultrabooks are all personal computers), smartphones or personal digital assistants (PDAs). It is understood that the aforementioned execution entities do not constitute a limitation on the embodiments of this application. Accordingly, the operation steps of this method can be, but are not limited to, the steps S1 to S8 below.

[0025] S1. Obtain the sensitive sample database and the data to be analyzed from the chemical industrial park. In specific implementation, the sensitive data types that need to be protected in the chemical industrial park can be defined first (prioritizing continuous data), such as production operation data (real-time monitoring values ​​of reactor temperature, pressure, flow rate, liquid level, etc.), safety monitoring data (concentration of toxic gases (such as chlorine, ammonia, etc.), personnel information (employee ID numbers, mobile phone numbers), logistics data (trajectory of hazardous materials transport vehicles, inventory ledger of hazardous chemicals), and environmental data (surrounding air quality (PM2.5, SO2)). After obtaining the sensitive data types that need to be collected, the raw data of the aforementioned sensitive data can be obtained through industrial IoT, database interfaces, log files, etc., thus forming a sensitive sample database. Similarly, the data to be analyzed is the data that needs to be accessed and queried from all the data collected in the chemical industrial park, so that data analysis can be carried out later using the accessed and queried data, such as accessing real-time production operation data for production operation risk monitoring, and accessing environmental data for environmental monitoring of the chemical industrial park.

[0026] After obtaining the aforementioned sensitive sample database, this embodiment constructs a sensitive data association graph based on it, so as to use the graph structure to organize the multi-source and heterogeneous sensitive data in the chemical industrial park into a structured and computable mathematical representation, thereby providing data support for feature matching and sensitivity calculation of the data to be analyzed later; wherein, the construction process of the sensitive data association graph may be, but is not limited to, as shown in step S2 below.

[0027] S2. Based on the sensitive sample database, a sensitive data association graph is constructed. The sensitive data association graph contains multiple nodes, and nodes with association relationships are connected by edges. Each node corresponds to a sensitive sample data. In specific applications, this embodiment first constructs a sensitive sample matrix to characterize the initial features of the sensitive sample data, and then performs feature dimensionality reduction. Next, a multi-source information fusion and adaptive adjustment mechanism is used to generate the optimal sensitive sample matrix. Finally, the sensitive data association graph can be constructed based on the optimal sensitive sample matrix.

[0028] Optionally, the detailed construction process of the association graph described above may be, but is not limited to, as shown in steps S21 to S26 below.

[0029] S21. Perform data cleaning on each sensitive sample data in the sensitive sample database to obtain several cleaned data. In specific implementation, data cleaning may include, but is not limited to: removing duplicate records and outliers, filling missing values ​​(using the mean or difference method), unifying timestamp format and unit unification, such as unifying pressure to MPa. At the same time, if the sensitive sample data is text data, such as logs, reports, etc., its data cleaning may include: using regular expressions to remove HTML tags, and then performing word segmentation and stop word removal. Of course, data cleaning is a common preprocessing method for data processing, and its principle will not be elaborated here.

[0030] After the data cleaning of the sensitive sample database is completed, vectorization processing can be performed, as shown in step S22 below.

[0031] S22. Vectorize each cleaned data point to obtain an initial feature vector corresponding to each sensitive sample data point. In this embodiment, if the sensitive sample data is a numerical type (such as temperature or pressure), it can be used as a feature value to form the corresponding initial feature vector. If the sensitive sample data is a categorical field (such as sensor type or personnel position), it can be encoded using one-hot encoding to form the initial feature vector. If the sensitive sample data is a text field, a trained word vector model (such as FastText or Word2Vec) is used to convert it into an initial feature vector. In this way, based on the aforementioned method, all data values ​​in each sensitive sample data point can be converted into feature values. Then, all feature values ​​of each sensitive sample data point are concatenated to obtain the corresponding initial feature vector.

[0032] After obtaining several initial feature vectors, a sensitive sample matrix can be constructed based on them, as shown in step S23 below.

[0033] S23. Construct a sensitive sample matrix using all initial feature vectors. In specific implementation, each row of the sensitive sample matrix represents the initial feature vector of a sensitive sample data, while the same column represents the feature values ​​of each sensitive sample data in the same dimension. Thus, after constructing the sensitive sample matrix, in order to reduce redundancy and noise and improve the sample discrimination ability, this embodiment also performs feature dimensionality reduction, the process of which is shown in step S24 below.

[0034] S24. Calculate the covariance matrix of the sensitive sample matrix, and perform feature dimensionality reduction on the sensitive sample matrix based on the covariance matrix to obtain the dimensionality-reduced sensitive sample matrix. In this embodiment, each row of the sensitive sample matrix represents a sample, and each column represents a feature. Therefore, the calculated covariance matrix can reflect the linear correlation between features. Based on this, eigenvalue decomposition is performed on the covariance matrix to obtain multiple matrix eigenvalues ​​and their corresponding eigenvectors. Then, the multiple matrix eigenvalues ​​are sorted in descending order, and the eigenvectors corresponding to the first few matrix eigenvalues ​​are selected according to the preset dimension or variance contribution rate to form a projection matrix (i.e., principal component analysis). Finally, based on the projection matrix, the original sensitive sample matrix is ​​projected onto the principal component space to obtain the dimensionality-reduced sensitive sample matrix, that is, the sensitive sample matrix is ​​multiplied by the projection matrix to obtain the dimensionality-reduced sensitive sample matrix. Based on this, each row of the dimensionality-reduced sensitive sample matrix is ​​used to represent the initial dimensionality-reduced feature vector of a sensitive sample data (that is, the initial feature vector after dimensionality reduction).

[0035] In this way, the principal components are orthogonal, eliminating the linear correlation between the original features, while dimensionality reduction discards components with small variance, making the features more compact. That is, the first few principal components retain the most important features in the original data. Based on this, after feature dimensionality reduction, the "curse of dimensionality" can be alleviated, the algorithm efficiency can be improved, and redundant features can be avoided from interfering with similarity calculation.

[0036] After obtaining the dimensionality-reduced sensitive sample matrix, this embodiment uses it and employs a multi-source information fusion and adaptive adjustment mechanism to generate the optimal sensitive sample matrix, as shown in step S25 below.

[0037] S25. Based on the dimension-reduced sensitive sample matrix, construct the optimal sensitive sample matrix, where each row of the optimal sensitive sample matrix represents the feature vector corresponding to a sensitive sample data. In specific implementation, for example, but not limited to, the following steps S25a to S25c can be used to construct the aforementioned optimal sensitive sample matrix.

[0038] S25a. For any row of the reduced-dimensionality sensitive sample matrix, calculate the information entropy of the initial reduced-dimensionality feature vector. In this embodiment, for any row of the initial reduced-dimensionality feature vector, count the number of occurrences of each feature value in the initial reduced-dimensionality feature vector. Then, divide the number of occurrences of each feature value by the total feature length to obtain the probability of each feature value. Then, use the probability of each feature value to calculate the information entropy of the initial reduced-dimensionality feature vector for that row.

[0039] The calculation formula is as follows: ; In the formula, This represents the information entropy. Let represent the probability of the j-th dimension feature in the initial dimensionality-reduced feature vector corresponding to any row. This represents the length of the initial dimensionality-reduced feature vector.

[0040] Thus, after calculating the information entropy of the initial dimensionality-reduced feature vector corresponding to any row, the adaptive adjustment factor can be calculated, as shown in step S25b below.

[0041] S25b. Based on the reduced-dimensionality sensitive sample matrix, calculate the standard deviation of each feature dimension in all initial reduced-dimensionality feature vectors, and determine the adaptive adjustment factor based on all standard deviations. In this embodiment, assuming there are N initial reduced-dimensionality feature vectors, then the j-th feature has N values. Therefore, the standard deviation of the N j-th feature can be calculated, and the standard deviation measures the dispersion (information content) of the feature dimension. Based on this, the adaptive adjustment factor obtained by summing the standard deviations of all feature dimensions can reflect the total variability of the entire sensitive sample database.

[0042] After calculating the adaptive adjustment factor, the information entropy and the initial dimensionality reduction feature vector can be combined to calculate the feature vector corresponding to each sensitive sample data. Based on all feature vectors, the optimal sensitive sample matrix is ​​formed, as shown in step S25c below.

[0043] S25c. Based on the information entropy, the adaptive adjustment factor, and the initial dimensionality reduction feature vector, calculate the feature vector corresponding to the target data, and after polling all rows of the dimensionality-reduced sensitive sample matrix, obtain the feature vector of each sensitive sample data, so as to construct the optimal sensitive sample matrix using all feature vectors, wherein the target data is the sensitive sample data corresponding to the initial dimensionality reduction feature vector.

[0044] In specific implementation, for example, but not limited to, the feature vector corresponding to the target data can be calculated according to the following formula; ; In the formula, This represents the feature vector corresponding to the target data. This represents the initial dimensionality-reduced feature vector. This represents the information entropy. This represents the weighting coefficient (which can be preset according to the type of data). This represents the adaptive adjustment factor. This represents the total number of initial dimensionality-reduced eigenvectors in the dimensionality-reduced sensitive sample matrix.

[0045] As can be seen from the aforementioned formula, the formula organically combines information from three different sources, so that the final feature vector not only contains the original data features, but also incorporates statistical characteristics and business knowledge; in practical applications, Essentially, it's used to eliminate the impact of differences in sample size, making sample databases of different sizes comparable. This ensures that in chemical industrial parks, high-frequency normal operating condition data won't mask low-frequency abnormal events due to numerical superiority. This is a weighted adjustment term; the addition of information entropy introduces a measure of the data's inherent "uncertainty," meaning that high-entropy regions (such as hazardous process sections with complex data distribution) are given higher weights, improving the detection sensitivity of sensitive events; simultaneously, For weighted correlation, specifically, It allows for dynamic adjustment of weights based on the importance of the data source (e.g., core production facilities > auxiliary facilities) or historical sensitivity level (e.g., sensors that have experienced leaks). Principal component analysis captures the correlation structure between features, allowing similar data to naturally cluster in the feature space and improving the accuracy of pattern matching.

[0046] at last, This plays a crucial adaptive adjustment role; that is, when the overall dataset fluctuates greatly, it automatically increases to suppress excessive amplification of noise, and when the dataset is stable, it... The smaller size allows weak but important signals (such as minute leaks) to be highlighted; therefore, this adaptive property avoids the failure problem of fixed threshold methods under different data distributions.

[0047] Furthermore, The introduction of this allows the calculation formula for the eigenvector to indirectly retain the graph structure information of the data, that is, because It is obtained through principal component analysis, and its essence represents the coordinates of the sample in the principal component space (that is, projecting the original sensitive sample matrix onto the principal component space), which reflects the similarity relationship between samples; therefore, the preservation of this structured information facilitates subsequent graph matching calculation, thereby making the subgraph similarity calculation more accurate.

[0048] Furthermore, the structure of this formula can also achieve parallel computation. That is, once the total number of information entropy, adaptive adjustment factor, projection matrix and initial dimensionality reduction feature vectors is determined, the calculation of the feature vector corresponding to each sensitive sample data is completely independent. Based on this, the sample data can be divided into blocks, each block is processed by different computing nodes, and finally the calculation results are concatenated row by row to obtain the complete optimal sensitive sample matrix; thus, the computational efficiency can be greatly improved.

[0049] Furthermore, most of the global quantities in the aforementioned formulas possess additivity or recursive update properties. For example, N can be directly accumulated after it increases. The dimensionality-reduced sensitive sample matrix is ​​obtained by multiplying the projection matrix by the original sensitive sample matrix, and the projection matrix is ​​calculated from the covariance matrix. Therefore, the rank-1 update formula can be used recursively without recalculating the product of deviations for all samples. At the same time, the adaptive adjustment factor is calculated from the standard deviation, which is the square root of the variance. Therefore, after adding new data, the variance recursion formula can be used to perform variance recursion, thereby back-deriving the standard deviation without rescanning all the old data.

[0050] Therefore, this embodiment can efficiently process large-scale, dynamically growing sensitive big data, meeting the real-time and scalability requirements of chemical industry scenarios.

[0051] Therefore, this embodiment integrates data frequency, information entropy, business weight, and feature correlation into a unified framework, and achieves scale independence through adaptive normalization, thereby generating a compact and information-rich sample matrix. This can reduce computation time while improving the accuracy of subsequent sensitivity calculations.

[0052] After constructing the optimal sensitive sample matrix, a sensitive data association graph can be constructed based on it, as shown in step S26 below.

[0053] S26. Using the optimal sensitive sample matrix, construct the sensitive data association graph; In specific implementation, the optimal sensitive sample matrix provides the feature vector of each sample. Therefore, to form a graph, it is also necessary to have edges (associations) between nodes. The source of the edges can be explicit relationships (such as connections in network topology, upstream and downstream in business processes, and production records of the same batch) and implicit relationships (such as adjacency in time series, proximity in spatial location, and feature similarity higher than the threshold). Therefore, this embodiment can use the following rules to determine whether there is an edge between any two nodes in time; (1) If two sensitive sample data come from the same production device, there is an edge (i.e., same device edge); (2) If the timestamp difference between two sensitive sample data is less than 1 minute, there is an edge (time edge); (3) If there is a network communication record between two data, there is an edge (communication edge); In this way, the sensitive data association graph contains a set of nodes (corresponding to each sensitive sample data), a set of edges (containing the edge connection relationship between each node) and a combination of attributes (i.e., each node corresponds to an attribute vector, which is the feature vector of the sensitive sample data corresponding to the node in the optimal sensitive sample matrix).

[0054] Therefore, after constructing the sensitive data association graph through the aforementioned steps S21 to S26, the query graph of the data to be analyzed can be constructed in the same way, as shown in step S3 below.

[0055] S3. Based on the data to be analyzed, a data query graph is constructed. In specific implementation, for example, but not limited to, using the collection time of the data to be analyzed as the center, several data entries that are temporally adjacent to the data to be analyzed in the original data stream of the chemical industrial park are selected as neighboring data; for example, five consecutive data entries before the data to be analyzed and five consecutive data entries after the data to be analyzed are selected. Then, the data to be analyzed and each neighboring data are used as query nodes in the data query graph, and the temporal order edges of each node are retained. At the same time, the aforementioned explicit and implicit relationships are used to complete the edges between each query node. Next, the feature vector of each query node is calculated using the aforementioned method for calculating the feature vector of sensitive sample data to ensure that the features of the data to be analyzed are aligned with the features of the sample. Finally, the data query graph can be constructed based on the aforementioned query nodes, the edges between each query node, and the feature vector of each query node.

[0056] After constructing the data query graph, graph matching can be performed on the sensitive data association graph to obtain the sensitive data sequence with the highest similarity to the data to be analyzed; the graph matching process is as shown in step S4 below.

[0057] S4. Using the data query graph, data matching is performed in the sensitive data association graph to obtain the sensitive data sequence with the highest similarity to the data to be analyzed. In specific implementation, the graph structure of the data query graph and the sensitive data association graph is the same. Therefore, this embodiment proposes a multi-path graph matching algorithm based on recursion mechanism to perform data matching. The process can be, but is not limited to, the steps S41 to S44 below.

[0058] S41. For the first query node in the data query graph, based on the feature vector of the first query node and the attribute vectors corresponding to each node in the sensitive data association graph, a graph search starting point set is determined from the sensitive data association graph. In specific implementation, for the first query node, the similarity between the feature vector of the first query node and the attribute vectors of each node in the sensitive data association graph is calculated. Then, nodes with similarity greater than the similarity threshold are selected so that the selected nodes can be used to form a graph search starting point set. Thus, each search starting point in the graph search starting point set can be considered as a matching node of the first query node. Therefore, this embodiment proposes a multi-path graph matching algorithm with multiple starting points, the process of which is described in detail in the following steps.

[0059] After determining the set of starting points for the graph search, constraints are constructed so that node matching can be performed based on these constraints. The process is shown in step S42 below.

[0060] S42. Construct node constraints, which include node consistency constraints and node repetition constraints. Node consistency constraints are used to constrain label consistency between nodes, and node repetition constraints are used to constrain that the same node can only match one query node. In specific implementation, label consistency between nodes indicates data source type matching. For example, if the data corresponding to the query node is pressure sensor data, then the node matching it in the sensitive data association graph should also be pressure sensor data. The node repetition constraint constrains a node in the sensitive data association graph to match only one query node. For example, if node A is a matching node of query node a, and in the next matching, if node A is assigned to the candidate node of query node b, then according to the aforementioned node repetition constraint, node A can be directly removed from the candidate nodes. Thus, after constructing the node constraints, multi-path graph structure matching can be performed based on this and using each search starting point in the graph search starting point set as the query path starting point, as shown in step S43 below.

[0061] S43. Based on node constraints, and using each search starting point in the graph search starting point set as the query path starting point, perform node matching on each query node in the sensitive data association graph to obtain the matching node of the data query graph under each query path. In specific implementation, each query path starting point corresponds to a query path. For example, assuming there are 10 query path starting points, then there are 10 query paths when performing graph matching (e.g., query path starting point 1 corresponds to path 1, query path starting point 2 corresponds to path 2, etc. Of course, the query path here is not pre-planned, but refers to the path formed after graph matching with different starting points). Therefore, this embodiment will use each search starting point as the query path starting point to perform multi-path graph matching. Optionally, taking any search starting point as an example to illustrate the recursive graph matching mechanism, it can be, but is not limited to, the steps S43a to S43f below.

[0062] S43a. For any search starting point in the graph search starting point set, take that search starting point as the starting point of the query path for the i-th query node in the data query graph, where the initial value of i is 2. In this embodiment, the search starting point is actually the matching node of the first query node. Therefore, it is equivalent to using the matching node of the previous query node of the current query node as the starting point of the query path of the current query node. Then, with the starting point of the current query path as the starting point, find the matching node corresponding to the current query node in the sensitive data association graph. The process is as shown in steps S43b to S43f below.

[0063] S43b. In the sensitive data association graph, the neighbor nodes of the starting point of the query path are found, and the similar nodes of the i-th query node are also found. In specific implementation, nodes in the sensitive data association graph that have edge connections with the starting point of the query path are considered as neighbor nodes of the starting point of the query path. Similarly, the similarity (e.g., cosine similarity) between the feature vector of the i-th query node and the attribute vectors of each node in the sensitive data association graph is calculated. Then, nodes in the sensitive data association graph with similarity greater than a similarity threshold are considered as similar nodes of the i-th query node. Thus, the aforementioned node matching method is performed in two dimensions: one is a graph structure-based filtering method, and the other is an attribute similarity-based filtering method. For the i-th query node, if it has an edge with a query node that has already been matched in the query graph, then its matching node in the sensitive data association graph should be located among the neighbor nodes of the matching node corresponding to the previous query node. This ensures the isomorphism of the subgraph structure. Similarly, it is also necessary to satisfy that the similarity between the feature vectors of all nodes found in the sensitive data association graph and the feature vector of the i-th query node is greater than or equal to the similarity threshold.

[0064] Based on the aforementioned method, after finding neighboring and similar nodes, a candidate node set can be generated, as shown in step S43c below.

[0065] S43c. Find the intersection of similar nodes and neighboring nodes to form the candidate node set for the i-th query node. In practice, if there is no edge connection between the i-th query node and the (i-1)-th query node in the data query graph, then the candidate node set is formed directly using similar nodes.

[0066] After obtaining the set of candidate nodes, the candidate nodes can be screened based on the node constraints, as shown in step S43d below.

[0067] S43d. Select candidate nodes that satisfy the node constraints from the candidate node set to serve as initial matching nodes. In specific implementation, the i-th query node and any candidate node must have the same label, and any candidate node is not a matching node of any query node before the i-th query node. Thus, after filtering the candidate node set based on the aforementioned node constraints, the initial matching nodes can be obtained. Then, the matching node of the i-th query node can be determined from the initial matching nodes, as shown in step S43e below.

[0068] S43e. Based on the feature vector of the i-th query node and the attribute vectors of each initial matching node, select the initial matching node that is most similar to the i-th query node from all initial matching nodes, and use it as the matching node of the i-th query node; in this embodiment, the initial matching node with the highest similarity between the feature vector of the i-th query node and the attribute vectors of each initial matching node is used as the matching node of the i-th query node.

[0069] Thus, after determining the matching node of the i-th query node, it can be used as the starting point of the query path of the (i+1)-th query node. Then, the aforementioned search process is repeated to obtain the matching node of the (i+1)-th query node, and this process is repeated until all query nodes in the data query graph have been traversed. At this point, the matching nodes of each query node under the path corresponding to any search starting point are obtained. The recursive traversal process is shown in step S43f below.

[0070] S43f. Update any search starting point to the matching node of the i-th query node, increment i by 1, and re-use any search starting point as the query path starting point of the i-th query node in the data query graph until i equals m, thus obtaining the matching nodes of each query node, where m is the total number of query nodes.

[0071] In specific implementation, if the candidate node set in the aforementioned step S43c is empty, or if there is no candidate node that satisfies the node constraint conditions in step S43d, then the graph search process corresponding to any search starting point will end, and a prompt will be output that no search results were found for that search starting point.

[0072] After completing the graph matching for any search starting point through the aforementioned steps S43a to S43f, the graph matching for the other search starting points can be performed in the same way to obtain all matching nodes of the data query graph under each query path. Of course, the graph matching process with the aforementioned search starting points as the starting points of the query paths can be carried out in parallel, which can improve the graph matching efficiency.

[0073] After obtaining all matching nodes of the data query graph under each query path, multiple initial sensitive data sequences can be generated. Then, among the multiple initial sensitive data sequences, the sensitive data sequence with the highest similarity to the data to be analyzed is determined. The process is shown in step S44 below.

[0074] S44. Based on the matching nodes under each query path, generate multiple initial sensitive data sequences, and based on the multiple initial sensitive data sequences, determine the sensitive data sequence with the highest similarity to the data to be analyzed; in this embodiment, the matching nodes under each query path are used to form the initial sensitive data sequence under each query path; then, through the similarity between each matching node in the initial sensitive data sequence and the corresponding query node, calculate the total similarity between each initial sensitive data sequence and the data query graph, and use each total similarity as the similarity between each initial sensitive data sequence and the data to be analyzed; finally, the initial sensitive data sequence with the highest similarity is taken as the sensitive data sequence with the highest similarity to the data to be analyzed.

[0075] Therefore, through the aforementioned steps S41 to S44, the advantages of using a recursive multi-path graph matching algorithm in this embodiment to match similar sensitive data corresponding to the data to be analyzed are as follows: (1) Enhanced robustness to differences in graph structure. Multiple starting points provide path redundancy. Even if a search starting from one node fails, a search starting from another node may still successfully match the entire subgraph. This makes the algorithm less sensitive to the failure of matching a single node and enhances the robustness of finding the target pattern in real, noisy data environments.

[0076] (2) Optimize the search path and improve efficiency; The search space for graph matching is huge. Starting from a poor starting point may lead to a large number of invalid recursions and backtrackings, resulting in high computational costs. In this embodiment, by starting from multiple points, different search spaces can be explored in parallel, thereby entering the most promising path for successful matching earlier.

[0077] (3) Improve the accuracy and reliability of matching results; Among them, the match found by a single path may only be the local optimum, not the global optimum. It may just be similar to the query graph in a certain local structure, but it is not the actual pattern to be found. Starting from multiple starting points, multiple candidate matching subgraphs will be obtained. By comparing the overall similarity scores of these candidate subgraphs, the global optimum matching result can be selected. This "selecting the best from multiple" mechanism is much more reliable than relying on the result of a single path and can effectively avoid the trap of local optima.

[0078] After matching the sensitive data sequence that is most similar to the data to be analyzed from the sensitive data association graph, the sensitivity of the data to be analyzed can be calculated based on this, as shown in step S5 below.

[0079] S5. Calculate the sensitivity of the data to be analyzed based on the sensitive data sequence; in specific implementation, examples such as, but not limited to, the following steps S51 to S54 can be used.

[0080] S51. Calculate the similarity between the sensitive data sequence and the data to be analyzed; In this embodiment, the method of calculating the similarity between the sensitive data sequence and the data to be analyzed has been described in the aforementioned step S44, and will not be repeated in this embodiment; After calculating the similarity between the two, the feature entropy can be calculated, and the process is shown in the following steps S52 and S53.

[0081] S52. From the sensitive data association graph, determine the attribute vector corresponding to each sensitive sample data in the sensitive data sequence; after determining the attribute vector of each sensitive sample data in the sensitive data sequence, the feature entropy can be calculated based on this, as shown in step S53 below.

[0082] S53. Calculate the feature entropy of the sensitive data sequence based on each attribute vector. In practice, first calculate the Shannon entropy of each attribute vector, which is the feature vector corresponding to each sensitive sample data in the sensitive data sequence. Then, the sum of all Shannon entropies is used as the feature entropy of the sensitive data sequence. After calculating the feature entropy of the sensitive data sequence, the sensitivity of the data to be analyzed can be calculated by combining the aforementioned similarity. The process is shown in step S54 below.

[0083] S54. Based on the feature entropy and the similarity, calculate the sensitivity of the data to be analyzed; in specific implementation, for example, but not limited to, the following formula can be used to calculate the sensitivity of the data to be analyzed.

[0084] ; Indicates the similarity, Represents feature entropy, Indicates a sensitive regulatory factor. The sensitivity is expressed as described above. In this embodiment, the sensitivity adjustment factor is greater than 0. Thus, the introduction of this sensitivity adjustment factor ensures that the greater the product of feature entropy and similarity, the closer the sensitivity approaches 1, highlighting the sensitivity of data with high similarity. In practical applications, the sensitivity adjustment factor can be dynamically adjusted according to the safety requirements of chemical industrial parks (e.g., high-risk areas require extremely high sensitivity) to achieve adaptive grading. Simultaneously, the input of the exponential function in the sensitivity calculation formula adopts... and The product of similarity and feature entropy reflects the synergistic effect of similarity and feature entropy. That is, the product is large only when both are large, and the sensitivity increases significantly. If the similarity is high but the entropy is low, the product is not large and the sensitivity will not be too high. If the entropy is high but the similarity is low, there will be no misjudgment. Therefore, the formula reflects the dual conditions of "both similarity to a sensitive pattern and rich information", avoiding the dominance of a single factor, thereby improving the accuracy of sensitivity calculation.

[0085] After calculating the sensitivity of the data to be analyzed, the access information when the data analysis terminal accesses the data to be analyzed (i.e., combining the sensitivity and the specific circumstances of the visitor) can be combined to calculate the differentiated desensitization parameters, thereby achieving differentiated desensitization; the calculation process of the desensitization parameters is shown in steps S6 and S7 below.

[0086] S6. When accessing the data to be analyzed on the data analysis terminal, obtain the access information of the data analysis terminal; in specific implementation, the access information of the data analysis terminal may include, but is not limited to: the access permissions of the data analysis terminal (i.e., the access permissions of the currently logged-in user of the data analysis terminal) and the access query type of the data to be analyzed by the data analysis terminal (such as aggregation query, single point query, etc.); in this way, the desensitization parameters of the data to be analyzed can be calculated based on the current access permissions, access query type and sensitivity, and the process is shown in step S7 below.

[0087] S7. Based on the access information and the sensitivity, calculate the desensitization parameters corresponding to the data to be analyzed, and perform desensitization processing on the data to be analyzed according to the desensitization parameters to obtain the desensitized data to be analyzed; in specific implementation, for example, but not limited to, the following steps S71 to S74 can be used to calculate the aforementioned desensitization parameters.

[0088] S71. Determine the query type adjustment factor and permission factor based on the access query type and access permissions. In this embodiment, query type adjustment factors corresponding to different access query types are preset (e.g., single-point query is set to 1 (the exact value needs maximum protection, aggregate query is set to 0.3, etc.)), and permission factors corresponding to different access permissions are preset. Therefore, in actual use, matching can be performed according to the current access permissions and query type.

[0089] After obtaining the query type adjustment factor and permission factor, the noise parameter can be calculated by combining the sensitivity, as shown in step S72 below.

[0090] S72. Based on the query type adjustment factor, permission factor, and sensitivity, calculate the noise parameter corresponding to the data to be analyzed; in specific implementation, the following formula can be used, but is not limited to, to calculate the noise parameter.

[0091] ; In the formula, This represents the noise parameter (the larger the value, the stronger the noise). This represents the global scaling factor (which controls the overall noise level and is usually set according to the data range and availability requirements, such as 10% to 30% of the data standard deviation). Indicates the sensitivity, Indicates permission factor. Indicates the query type adjustment factor. It represents a constant (a very small positive number, such as 10 to the power of negative 5, used to avoid the denominator being 0). It is a non-linear adjustment index (greater than 0, used to adjust the intensity of the effect of the sensitivity / authority ratio on noise).

[0092] Thus, after calculating the noise parameters based on the aforementioned formula, random noise can be generated based on these parameters, as shown in step S73 below.

[0093] S73. Based on the noise parameters, generate random noise conforming to a Gaussian distribution; in specific implementation, a Gaussian distribution is generated based on the noise parameters. Noise, and then, from A random noise is sampled as a desensitization parameter, and the process is shown in step S74 below.

[0094] S74. Determine the desensitization parameters based on the random noise.

[0095] Based on the aforementioned steps S71 to S74, after calculating the desensitization parameters of the data to be analyzed, the data to be analyzed can be desensitized by adding random noise to the data to be analyzed, thereby obtaining the desensitized data to be analyzed (of course, if the data obtained after adding random noise exceeds the maximum data range of the data to be analyzed, then truncation is performed, that is, if it is greater than the maximum value, the maximum value is taken, and if it is less than the minimum value, the minimum value is taken).

[0096] After obtaining the anonymized data to be analyzed, it can be sent to the data analysis terminal, so that the data analysis terminal can perform data analysis based on it. The process is shown in step S8 below.

[0097] S8. Send the anonymized data to be analyzed to the data analysis terminal so that the data analysis terminal can perform data analysis of the chemical industrial park based on the anonymized data.

[0098] Therefore, through the safety data analysis method for chemical industrial parks described in detail in steps S1 to S8 above, this invention constructs a sensitive data association graph by acquiring a sensitive sample database of the chemical industrial park, and simultaneously constructs a data query graph for the data to be analyzed. Then, using graph matching technology, similar sensitive sequences of the data to be analyzed are matched in the sensitive data association graph. Subsequently, based on the similar sensitive sequences, the sensitivity of the data is accurately calculated. Finally, combined with the access information of the data analysis terminal, the desensitization parameters of the data to be analyzed are dynamically calculated, and based on this, the data to be analyzed is desensitized to obtain the desensitized data to be analyzed for subsequent data analysis. In this way, this invention achieves differentiated desensitization processing, avoiding the "one-size-fits-all" problem caused by the traditional uniform desensitization strategy. It ensures strict protection of high-risk data in specific scenarios, and maximizes the preservation of the analytical value of low-risk data under the premise of security, significantly improving data utilization efficiency. Thus, this invention effectively solves the inherent contradiction between safety data sharing and privacy protection in chemical industrial parks, and can provide reliable technical support for risk management in chemical industrial parks.

[0099] like Figure 2 As shown, the second aspect of this embodiment provides a hardware system for implementing the safety data analysis method for chemical industrial parks described in the first aspect of the embodiment, comprising: The acquisition unit is used to acquire the sensitive sample database of the chemical industrial park and the data to be analyzed in the chemical industrial park.

[0100] The graph construction unit is used to construct a sensitive data association graph based on the sensitive sample database. The sensitive data association graph contains multiple nodes, and nodes with association relationships are connected by edges. Each node corresponds to a sensitive sample data.

[0101] The graph construction unit is also used to construct a data query graph based on the data to be analyzed.

[0102] The matching unit is used to perform data matching in the sensitive data association graph using the data query graph to obtain the sensitive data sequence with the highest similarity to the data to be analyzed.

[0103] A sensitivity identification unit is used to calculate the sensitivity of the data to be analyzed based on a sensitive data sequence.

[0104] The access information acquisition unit is used to acquire access information of the data analysis terminal when accessing the data to be analyzed.

[0105] The desensitization unit is used to calculate the desensitization parameters corresponding to the data to be analyzed based on the access information and the sensitivity, and to perform desensitization processing on the data to be analyzed based on the desensitization parameters to obtain the desensitized data to be analyzed.

[0106] The desensitization unit is also used to send the desensitized data to be analyzed to the data analysis terminal, so that the data analysis terminal can perform data analysis of the chemical industrial park based on the desensitized data.

[0107] The working process, working details and technical effects of the system provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0108] like Figure 3 As shown, the third aspect of this embodiment provides a safety data analysis device for chemical industrial parks. Taking the device as an electronic device as an example, it includes: a memory, a processor, and a transceiver connected in sequence. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the safety data analysis method for chemical industrial parks as described in the first aspect of the embodiment.

[0109] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0110] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard) transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0111] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0112] The fourth aspect of this embodiment provides a storage medium that stores instructions containing the safety data analysis method for chemical industrial parks as described in the first aspect of the embodiment. That is, the storage medium stores instructions that, when executed on a computer, perform the safety data analysis method for chemical industrial parks as described in the first aspect of the embodiment.

[0113] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0114] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0115] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the safety data analysis method for chemical industrial parks as described in the first aspect of this embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0116] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A safety data analysis method for chemical industrial parks, characterized in that, include: Obtain a database of sensitive samples from chemical industrial parks and the data to be analyzed from the chemical industrial parks; Based on the sensitive sample database, a sensitive data association graph is constructed. The sensitive data association graph contains multiple nodes, and nodes with association relationships are connected by edges. Each node corresponds to a sensitive sample data. Based on the data to be analyzed, a data query graph is constructed; Using a data query graph, data matching is performed in the sensitive data association graph to obtain the sensitive data sequence with the highest similarity to the data to be analyzed; Based on the sensitive data sequence, the sensitivity of the data to be analyzed is calculated; When accessing the data to be analyzed on the data analysis terminal, obtain the access information of the data analysis terminal; Based on the access information and the sensitivity, the desensitization parameters corresponding to the data to be analyzed are calculated, and the data to be analyzed is desensitized according to the desensitization parameters to obtain the desensitized data to be analyzed. The anonymized data to be analyzed is sent to the data analysis terminal so that the data analysis terminal can perform data analysis of the chemical industrial park based on the anonymized data.

2. The method according to claim 1, characterized in that, Based on the sensitive sample database, a sensitive data association graph is constructed, including: Data cleaning is performed on each sensitive sample in the sensitive sample database to obtain several cleaned data sets. Each cleaned data point is vectorized to obtain the initial feature vector corresponding to each sensitive sample data point. Construct a sensitive sample matrix using all initial feature vectors; The covariance matrix of the sensitive sample matrix is ​​calculated, and based on the covariance matrix, the sensitive sample matrix is ​​subjected to feature dimensionality reduction processing to obtain the dimensionality-reduced sensitive sample matrix. Based on the dimension-reduced sensitive sample matrix, the optimal sensitive sample matrix is ​​constructed, where each row of the optimal sensitive sample matrix represents the feature vector corresponding to a sensitive sample data. The sensitive data association graph is constructed using the optimal sensitive sample matrix.

3. The method according to claim 2, characterized in that, Each row in the reduced-dimensionality sensitive sample matrix is ​​used to represent the initial reduced-dimensionality feature vector of a sensitive sample data; Among them, the optimal sensitive sample matrix is ​​constructed based on the dimensionality-reduced sensitive sample matrix, including: For any row in the reduced-dimensionality sensitive sample matrix, calculate the information entropy of the initial reduced-dimensionality feature vector; Based on the reduced-dimensionality sensitive sample matrix, the standard deviation of each feature dimension in all initial reduced-dimensionality feature vectors is calculated, and the adaptive adjustment factor is determined based on all the standard deviations. Based on the information entropy, the adaptive adjustment factor, and the initial dimensionality reduction feature vector, the feature vector corresponding to the target data is calculated. After polling all rows of the dimensionality-reduced sensitive sample matrix, the feature vector of each sensitive sample data is obtained. In order to use all feature vectors, the optimal sensitive sample matrix is ​​constructed, wherein the target data is the sensitive sample data corresponding to the initial dimensionality reduction feature vector.

4. The method according to claim 3, characterized in that, Based on the information entropy, the adaptive adjustment factor, and the initial dimensionality reduction feature vector, the feature vector corresponding to the target data is calculated, including: The feature vector corresponding to the target data is calculated according to the following formula; ； In the formula, This represents the feature vector corresponding to the target data. This represents the initial dimensionality-reduced feature vector. This represents the information entropy. Indicates the weighting coefficient. This represents the adaptive adjustment factor. This represents the total number of initial dimensionality-reduced eigenvectors in the dimensionality-reduced sensitive sample matrix.

5. The method according to claim 1, characterized in that, Each node in the sensitive data association graph also corresponds to an attribute vector. The attribute vector of each node is the feature vector of the sensitive sample data corresponding to that node, and the graph structure of the data query graph and the sensitive data association graph is the same. Specifically, data matching is performed in the sensitive data association graph using a data query graph to obtain the sensitive data sequence with the highest similarity to the data to be analyzed, including: For the first query node in the data query graph, based on the feature vector of the first query node and the attribute vectors corresponding to each node in the sensitive data association graph, determine the set of starting points for the graph search from the sensitive data association graph; Node constraints are constructed, including node consistency constraints and node duplication constraints. The node consistency constraints are used to constrain the consistency of labels between nodes, and the node duplication constraints are used to constrain that the same node can only match one query node. Based on node constraints, and taking each search starting point in the graph search starting point set as the query path starting point, in the sensitive data association graph, node matching is performed on each query node in the data query graph to obtain the matching node of the data query graph under each query path. Based on the matching nodes under each query path, multiple initial sensitive data sequences are generated, and based on the multiple initial sensitive data sequences, the sensitive data sequence with the highest similarity to the data to be analyzed is determined.

6. The method according to claim 5, characterized in that, Based on node constraints, and using each search starting point in the graph search starting point set as the query path starting point, node matching is performed on each query node within the data query graph in the sensitive data association graph, including: For any search starting point in the set of graph search starting points, take that search starting point as the starting point of the query path of the i-th query node in the data query graph, where the initial value of i is 2; In the sensitive data association graph, find the neighbor node of the starting point of the query path, and in the sensitive data association graph, find the similar node of the i-th query node; Find the intersection of similar nodes and their neighboring nodes to form the candidate node set for the i-th query node; Candidate nodes that satisfy the node constraints are selected from the candidate node set and used as initial matching nodes; Based on the feature vector of the i-th query node and the attribute vectors of each initial matching node, the initial matching node most similar to the i-th query node is selected from all initial matching nodes and used as the matching node of the i-th query node. Update any search starting point to the matching node of the i-th query node, increment i by 1, and re-use any search starting point as the query path starting point of the i-th query node in the data query graph until i equals m, thus obtaining the matching nodes of each query node, where m is the total number of query nodes.

7. The method according to claim 1, characterized in that, Each node in the sensitive data association graph also corresponds to an attribute vector, and the attribute vector of each node is the feature vector of the sensitive sample data corresponding to that node. The sensitivity of the data to be analyzed is calculated based on the sensitive data sequence, including: Calculate the similarity between the sensitive data sequence and the data to be analyzed; From the sensitive data association diagram, determine the attribute vector corresponding to each sensitive sample data in the sensitive data sequence; Calculate the feature entropy of the sensitive data sequence based on each attribute vector; Based on the feature entropy and the similarity, the sensitivity of the data to be analyzed is calculated.

8. The method according to claim 1, characterized in that, The access information of the data analysis terminal includes: the access permissions of the data analysis terminal and the access query type of the data analysis terminal for the data to be analyzed; Specifically, based on the access information and the sensitivity, the de-identification parameters corresponding to the data to be analyzed are calculated, including: Based on the access query type and access permissions, determine the query type adjustment factor and permission factor; Based on the query type adjustment factor, permission factor, and sensitivity, the noise parameters corresponding to the data to be analyzed are calculated; Based on the noise parameters, random noise conforming to a Gaussian distribution is generated; The desensitization parameters are determined based on the random noise.

9. A safety data analysis system for chemical industrial parks, characterized in that, include: The acquisition unit is used to acquire the sensitive sample database of the chemical industrial park and the data to be analyzed in the chemical industrial park; The graph construction unit is used to construct a sensitive data association graph based on the sensitive sample database. The sensitive data association graph contains multiple nodes, and nodes with association relationships are connected by edges. Each node corresponds to a sensitive sample data. The graph construction unit is also used to construct a data query graph based on the data to be analyzed; The matching unit is used to perform data matching in the sensitive data association graph using the data query graph, so as to obtain the sensitive data sequence with the highest similarity to the data to be analyzed; A sensitivity identification unit is used to calculate the sensitivity of the data to be analyzed based on a sensitive data sequence; The access information acquisition unit is used to acquire access information of the data analysis terminal when accessing the data to be analyzed. The desensitization unit is used to calculate the desensitization parameters corresponding to the data to be analyzed based on the access information and the sensitivity, and to perform desensitization processing on the data to be analyzed based on the desensitization parameters to obtain the desensitized data to be analyzed. The desensitization unit is also used to send the desensitized data to be analyzed to the data analysis terminal, so that the data analysis terminal can perform data analysis of the chemical industrial park based on the desensitized data.

10. A computer program product containing instructions, characterized in that, When the instructions are executed on the computer, the computer performs the safety data analysis method for chemical industrial parks as described in any one of claims 1 to 8.

Images