Data desensitization system for multi-physics model simulation based on distributed computing

Abstract

The present invention relates to the field of data desensitization technology, specifically a data desensitization system based on multi-physics model simulation of distributed computing, the system includes a data feature analysis module, a desensitization strategy customization module, a data transmission path optimization module and a multidimensional data anonymization module. In the present invention, by utilizing the comprehensive processing of distributed computing and multi-physics model data, the security and privacy of data are significantly improved, the regularization coefficient is finely adjusted and a differentiated desensitization strategy is implemented, allowing data to be accurately desensitized, and according to different attributes and sensitivities, the protection effect of data is optimized, not only the security barrier of data is enhanced, but also the analysis availability is maintained, and the data transmission path is optimized through the graphical model theory, which significantly reduces the risk of data leakage during transmission, improves the efficiency and security of transmission, ensures the high confidentiality of data in non-production environments, and enhances the practicality and scientific research value of model data.

Description

Technical Field

[0001] The present invention relates to the technical field of data desensitization, and in particular to a data desensitization system based on distributed computing and multi-physics field model simulation. Background Art

[0002] The field of data desensitization technology mainly involves the processing of sensitive data to prevent data from being leaked during processing, transmission or storage. The core purpose is to remove or transform personal or confidential information while maintaining data availability to ensure data security and privacy. It is widely used in industries such as finance, medical care, education and business analysis. Common data desensitization technologies include data masking, data segmentation, data encryption, etc., which can effectively protect the privacy of data in non-production environments while maintaining the practicality of data and supporting various data analysis and testing activities.

[0003] Among them, the data desensitization system for multi-physics model simulation of distributed computing is designed to process and desensitize model simulation data involving multiple physical phenomena through distributed computing technology, with the aim of providing accurate and efficient model processing capabilities while ensuring the security and privacy of simulation data. Such systems are often used in scientific research or engineering applications that require simultaneous processing of multiple physical parameters and data sets, such as in weather forecasting, nuclear reaction simulation, or bioengineering. Through desensitization processing, sensitive simulation data can be safely used for further research and analysis without exposing sensitive information in the original data.

[0004] Traditional data desensitization techniques, such as data masking or data segmentation, are often insufficient when processing multi-physics model data, especially in terms of flexibility and accuracy. This may cause the data to lose its accurate representation of the original physical phenomena, reducing the scientific research value of the data. In addition, these technologies fail to fully utilize the potential of distributed computing, such as dynamic processing and efficient resource utilization, especially in multi-node distributed computing environments. This deficiency leads to inefficient data processing. In terms of data transmission, existing methods do not provide effective path optimization, resulting in a high security risk during data transmission between nodes. These defects limit the effectiveness and application scope of data desensitization technology in demanding application scenarios. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and propose a data desensitization system for multi-physics field model simulation based on distributed computing.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: A data desensitization system based on multi-physics field model simulation based on distributed computing includes:

[0007] The data feature analysis module performs data sensitivity analysis based on multi-physics field model data, adjusts the regularization coefficient according to the physical properties and sensitivity assessment of the data, performs initial data desensitization operations based on the adjustment results, analyzes the data structure, extracts key information, and generates preliminary desensitized data packets;

[0008] The desensitization strategy customization module is based on the preliminary desensitized data packet. The distributed nodes select the desensitization algorithm according to the local data characteristics and processing requirements, perform differential desensitization according to the selection results, perform data anonymization processing, adjust and optimize the data format, ensure data consistency between nodes, and generate node customized desensitized data;

[0009] The data transmission path optimization module combines the node customized desensitized data to optimize the path redundancy, uses the graphical model theory to analyze the network structure between nodes, optimizes the data transmission path according to the analysis results, adjusts the connection weights between nodes, performs data packet forwarding tests, and generates an optimized data transmission path;

[0010] The multidimensional data anonymization module performs data processing based on the optimized data transmission path, reduces the dimension of the data using singular value decomposition, adjusts the data structure, performs data reconstruction, and generates the final anonymous multi-physics field model simulation data.

[0011] As a further solution of the present invention, the step of adjusting the regularization coefficient is specifically:

[0012] Perform attribute analysis on multi-physics model data and extract the physical parameters p1, p2, ..., p of each data point n , calculate the sensitivity score S, using the formula,

[0013]

[0014] Generate a sensitivity score list, where n represents the number of types of physical properties, w k represents the weight of the kth physical attribute, p ik Represents the value of data point i on attribute k;

[0015] Using the sensitivity score list, the regularization coefficient is set, using the formula,

[0016]

[0017] Get the initial regularization coefficient R, where m represents the total number of data points and S i Represents the sensitivity score of a single data point;

[0018] Fine-tune the initial regularization coefficient using the formula,

[0019]

[0020] Generate the adjusted regularization coefficient R adj , where δ represents the tuning parameter used to fine-tune the regularization coefficient so that it matches the actual sensitivity distribution of the data.

[0021] As a further solution of the present invention, the steps of obtaining the preliminary desensitized data packet are specifically as follows:

[0022] The adjusted regularization coefficient is used to desensitize the data set, and the desensitized data points are calculated using the formula:

[0023]

[0024] Among them, D i Represents the data value of the i-th data after desensitization, d i represents the value of the original data point, R adj After adjusting the regularization coefficient, the desensitized data list is obtained;

[0025] Perform structural analysis on the desensitized data list, apply clustering algorithm, and adopt formula,

[0026]

[0027] Among them, M represents the key metric of the data structure, θ i represents the clustering sensitivity parameter, N represents the total number of data points, which is used to adjust the accuracy and response of clustering and identify the key structure of the data;

[0028] Based on the data key structure, the key information of the data is extracted using the formula:

[0029]

[0030] Among them, K represents a comprehensive measure of key information extracted from the data structure, t represents the total number of data points, and α i Represents the weight of the information category, M i represents the data point value in M;

[0031] For data that needs to keep its original value, atomization technology is used to replace the value based on physical and structural parameter functions before sending the data to the distributed computing server to desensitize the data. Through the model:

[0032]

[0033] Among them, Φ, Ψ i (i∈[0,M-1]) and C are arbitrary functionals representing the model and physics, p j(j∈[0,N-1]) is a parameter, x k (k∈[0,L-1]) is the value that needs to be calculated remotely, based on the regularization coefficient evaluation, replacing the function equation with a numerical value:

[0034]

[0035] The physical and structural parameters p j The value of will be hidden in the value f i (i∈[0,M-1]) and c, generate a preliminary desensitized data packet.

[0036] As a further solution of the present invention, the selection step of the desensitization algorithm is specifically as follows:

[0037] Based on the preliminary desensitized data packet, the mean and standard deviation of the data set are calculated using the formula,

[0038]

[0039]

[0040] Generates a summary of statistical properties where x i represents a single data point, μ represents the mean, σ represents the standard deviation, and n represents the number of samples;

[0041] Using the statistical characteristic summary, the information gain evaluation method is applied to assess the relevance of data characteristics and processing requirements, using the formula,

[0042]

[0043] Determine a list of candidate desensitization algorithms, where G represents information gain and λ is a balancing factor used to adjust the sensitivity of the data;

[0044] According to the candidate desensitization algorithm list, combined with the preset information gain threshold, the desensitization algorithm is selected using the formula:

[0045]

[0046] Among them, T is the preset information gain threshold, and the desensitization algorithm selection result is obtained.

[0047] As a further solution of the present invention, the step of obtaining the node customized desensitized data is specifically as follows:

[0048] According to the selection results of the desensitization algorithm, the data is processed differently, and a dynamically adjusted standardization method is adopted, using the formula,

[0049]

[0050] Among them, yi represents the data points after differential processing, μ and σ represent the mean and standard deviation, and ∈ is a fine-tuning parameter used to enhance the flexibility of the model and obtain differentially processed data;

[0051] The differentially processed data is anonymized, and the hash function and salt value technology are used to enhance security. The formula is used.

[0052] z i = hash(y i s+salt)

[0053] Among them, hash represents the hash function, z i represents the anonymized data point, s is the data scaling factor, and salt is the randomly generated salt value to generate an anonymized data set;

[0054] Integrate the anonymized data set, unify the data format among all distributed nodes, and generate node-customized desensitized data.

[0055] As a further solution of the present invention, the steps of analyzing the network structure between the nodes are specifically:

[0056] Combined with the node customized desensitized data, the total data processing volume of each node is calculated using the formula:

[0057]

[0058] Among them, T i represents the total data processing capacity of node i, d ij is the amount of data received or sent by node i to node j, α ij is the data importance weight, adjusted based on the priority or sensitivity of the data, M is the number of nodes directly connected to node i, and the comprehensive data processing volume is obtained;

[0059] According to the comprehensive data processing volume, the graph theory is applied to construct a network graph model, using the formula,

[0060]

[0061] Among them, P opt represents the optimized network path, E represents the edge set in the network, T i and T j is the data processing amount connecting nodes i and j, β is the smoothing factor to prevent the denominator from being zero, and the network topology is obtained;

[0062] To evaluate the data transmission efficiency of the network topology, the formula is used:

[0063]

[0064] Where τ is the average data transmission time, P opt,k is the length of the kth optimized path, γ is the adjustment factor used to adjust the impact of path length, and F is the total number of paths. The network topology structure evaluation result is obtained.

[0065] As a further solution of the present invention, the step of obtaining the optimized data transmission path is specifically as follows:

[0066] According to the network topology evaluation results, the key transmission path is optimized using the formula:

[0067]

[0068] Among them, L ij is the new weight from node i to node j, H ij is the path length from node i to node j, g is the weight factor used to adjust the sensitivity of the weight, h is a constant used to prevent division by zero errors, and the new weight parameter is obtained;

[0069] Apply the new weight parameters to perform data forwarding test, using the formula,

[0070]

[0071] Among them, I ij is the estimated transmission time from node i to node j, ξ is the data transmission rate factor, and the packet forwarding test result is obtained;

[0072] According to the data packet forwarding test results, the data transmission path is adjusted and optimized, using the formula,

[0073] Q=∑ (i,j)∈Y I ij ·η ij

[0074] Where Q is the total transmission time, Y is the total number of critical paths, and η ij is the usage frequency of path (i, j), and outputs the optimized data transmission path.

[0075] As a further solution of the present invention, the steps for obtaining the final anonymous multi-physics field model simulation data are specifically as follows:

[0076] Collect node data from the optimized data transmission path, recalculate the total amount of data for each node, and use the formula,

[0077]

[0078] Among them, A i represents the total amount of data at node i, d ijis the amount of data received by node i or sent to node j, c ij is the transmission stability coefficient, ∈ is a constant to prevent division by zero errors, k j is the connection strength coefficient between nodes, log(1+k j ) represents nonlinear scaling of the connection strength, q represents the total number of nodes, and outputs summary data;

[0079] The singular value decomposition method is applied to reduce the dimension of the summary data, extract key features and reduce the data dimension. The formula is used.

[0080]

[0081] in, Represents the data matrix A i Power law transformation is applied to enhance the sensitivity of data processing. a is the adjustment coefficient, S represents the singular value, U and V T Represents the orthogonal matrix obtained by singular value decomposition, and obtains the singular value decomposition result;

[0082] According to the singular value decomposition result, the data structure is adjusted, and the formula is adopted.

[0083]

[0084] Among them, S reduced represents the key singular value, b is the index used to adjust the influence of the singular value, B is a small value used to ensure numerical stability, and the adjusted data structure A is generated adj ;

[0085] Using the adjusted data structure A adj and the singular value decomposition of V T Perform data reconstruction using the formula,

[0086]

[0087] Where c is an exponent used to enhance anonymity and generate the final anonymized multi-physics model simulation data A final , ensuring the privacy of multi-physics model simulation data and preserving key physical properties.

[0088] Compared with the prior art, the advantages and positive effects of the present invention are:

[0089] In the present invention, by utilizing distributed computing and comprehensive processing of multi-physics field model data, the security and privacy of the data are significantly improved, the regularization coefficient is finely adjusted and a differentiated desensitization strategy is implemented, allowing accurate desensitization of the data, and optimizing the data protection effect according to different attributes and sensitivities, which not only enhances the data security barrier but also maintains the analysis availability. The data transmission path is optimized through graphical model theory, which significantly reduces the risk of data leakage during the transmission process, improves the efficiency and security of the transmission, ensures the high confidentiality of the data in a non-production environment, and enhances the practicality and scientific research value of the model data. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 is a system flow chart of the present invention;

[0091] Figure 2 is a flow chart of the adjustment steps of the regularization coefficient of the present invention;

[0092] Figure 3 This is a flow chart of the steps for obtaining a preliminary desensitized data packet of the present invention;

[0093] Figure 4 A flow chart of the selection steps of the desensitization algorithm of the present invention;

[0094] Figure 5 A flowchart of the steps for obtaining desensitized data customized for the node of the present invention;

[0095] Figure 6 A flowchart of the steps for analyzing the network structure between nodes of the present invention;

[0096] Figure 7 A flow chart of the steps for obtaining the optimized data transmission path of the present invention;

[0097] Figure 8 This is a flow chart of the steps for obtaining the final anonymous multi-physics field model simulation data of the present invention. DETAILED DESCRIPTION

[0098] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0099] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0100] Embodiment 1

[0101] See also Figure 1 The data desensitization system for multi-physics model simulation based on distributed computing includes:

[0102] The data feature analysis module performs data sensitivity analysis based on multi-physics field model data, adjusts the regularization coefficient according to the physical properties and sensitivity assessment of the data, performs initial data desensitization operations based on the adjustment results, analyzes the data structure, extracts key information, and generates preliminary desensitized data packets;

[0103] The desensitization strategy customization module is based on the preliminary desensitization data packet. The distributed nodes select the desensitization algorithm according to the local data characteristics and processing requirements, perform differential desensitization according to the selection results, perform data anonymization, adjust and optimize the data format, ensure data consistency between nodes, and generate node customized desensitized data;

[0104] The data transmission path optimization module combines node customized desensitized data to optimize path redundancy, uses graphical model theory to analyze the network structure between nodes, optimizes the data transmission path based on the analysis results, adjusts the connection weights between nodes, performs data packet forwarding tests, and generates optimized data transmission paths;

[0105] The multidimensional data anonymization module performs data processing based on the optimized data transmission path, reduces the data dimension using singular value decomposition, adjusts the data structure, performs data reconstruction, and generates the final anonymous multi-physics model simulation data.

[0106] The preliminary desensitized data packet includes data sensitivity rating, data attributes and preliminary desensitization effect records. The node customized desensitized data includes node desensitization algorithm type, desensitization depth and data consistency index. The optimized data transmission path includes path efficiency records, security protocols and network topology structure analysis results. The final anonymous multi-physics field model simulation data specifically includes data anonymity level, reconstruction results and structural adjustment details.

[0107] See also Figure 2, the adjustment steps of the regularization coefficient are as follows:

[0108] Perform attribute analysis on multi-physics model data and extract the physical parameters p1, p2, ..., p of each data point n , calculate the sensitivity score S, using the formula,

[0109]

[0110] Generate a sensitivity score list, where n represents the number of types of physical properties, w k represents the weight of the kth physical attribute, p ik Represents the value of data point i on attribute k;

[0111] Using the sensitivity score list, set the regularization coefficient and use the formula,

[0112]

[0113] Get the initial regularization coefficient R, where m represents the total number of data points and S i Represents the sensitivity score of a single data point;

[0114] Fine-tune according to the initial regularization coefficient, using the formula,

[0115]

[0116] Generate the adjusted regularization coefficient R adj , where δ represents the tuning parameter used to fine-tune the regularization coefficient so that it matches the actual sensitivity distribution of the data.

[0117] Assume that the multi-physics model data has three physical attributes (n=3). For a specific data point, the attribute values ​​are p i1 =2, p i2 =-1, p i3 =0.5, assuming weights w1=0.5, w2=1.5, w3=1.

[0118] Substituting into the formula, we get:

[0119] For the first property:

[0120]

[0121] but,

[0122]

[0123] For the second property:

[0124]

[0125] but,

[0126]

[0127] For the third attribute:

[0128]

[0129] but,

[0130]

[0131] Calculate the overall sensitivity score:

[0132]

[0133] The result of 1.76 represents a comprehensive assessment of the sensitivity of the physical properties to the data point taking into account the weight. The higher the value, the greater the sensitivity of the data point in the physical model and the more significant the impact on the model.

[0134] Assume that there are 5 data points (m=5), and their sensitivity scores are S1=1.76, S2=2.05, S3=1.45, S4=2.31 and S5=1.89, respectively. Substitute the values ​​into the formula to calculate the regularization coefficient.

[0135] Calculate the square of each sensitivity score:

[0136]

[0137] Calculate the sum of the squares of all sensitivity scores:

[0138]

[0139] Calculate the regularization coefficient:

[0140]

[0141] The regularization coefficient R = 0.8558 represents the average scale of the sensitivity score of the data set, which is used to adjust the regularization strength during data processing. The value of 0.8558 indicates the standardized ratio of the average sensitivity score of each data point relative to the maximum value, and a lower value may indicate that the data points are not extremely sensitive as a whole.

[0142] Continue to use the obtained regularization coefficient R = 0.8558. Assume that the adjustment parameter δ is set to 0.5 (this is a commonly used medium-strength adjustment parameter that can be adjusted according to specific circumstances to adapt to different data sensitivities).

[0143] Calculate δ·R:

[0144] δ·R=0.5·0.8558=0.4279

[0145] Adjust using exponential function (using exponential calculation to get the ratio) and square root (taking square root provides smooth adjustment):

[0146]

[0147] Calculate the adjusted regularization coefficient R adj :

[0148] R adj =0.8558·0.6196=0.5303

[0149] Adjusted regularization coefficient R adj =0.5303 reflects the adjustment of the original regularization coefficient after considering the sensitivity response strength. R is lower than the initial R = 0.8558 adj This indicates that the data has been adjusted more carefully, probably because the dataset was considered more sensitive to specific operations in the original evaluation. Therefore, the new regularization coefficient is more suitable for handling these sensitive data to avoid possible errors caused by over-adjustment.

[0150] See also Figure 3 ,The specific steps for obtaining the initial desensitized data packet are:

[0151] Use the adjusted regularization coefficient to desensitize the data set and calculate the desensitized data points using the formula,

[0152]

[0153] Among them, D i Represents the data value of the i-th data after desensitization, d i represents the value of the original data point, R adj After adjusting the regularization coefficient, the desensitized data list is obtained;

[0154] Perform structural analysis on the anonymized data list, apply clustering algorithm, and use formula.

[0155]

[0156] Among them, M represents the key metric of the data structure, θ i represents the clustering sensitivity parameter, N represents the total number of data points, which is used to adjust the accuracy and response of clustering and identify the key structure of the data;

[0157] Based on the key structure of data, the key information of data is extracted using the formula:

[0158]

[0159] Among them, K represents a comprehensive measure of key information extracted from the data structure, t represents the total number of data points, and α i Represents the weight of the information category, M i Represents the data point value in M ​​and generates a preliminary desensitized data packet.

[0160] Assume there are 5 data points, whose original values ​​are d1=100, d2=150, d3=200, d4=250 and d5=300. Use the adjusted regularization coefficient R calculated above adj =0.5303.

[0161] Substituting into the formula, we get:

[0162]

[0163] Apply the masking formula to each data point:

[0164]

[0165] The desensitized data values ​​D1, D2, D3, D4, and D5 are:

[0166] 57.87, 86.81, 115.74, 144.68, 173.61

[0167] It shows that the original data has been normalized and scaled by considering the sensitivity adjustment coefficient to reduce data sensitivity and protect information security.

[0168] Continue to use the desensitized data D1, D2, D3, D4, D5 obtained by the above calculation, assuming that the clustering sensitivity parameter θ i Both are set to 1.

[0169] Apply the structural analysis formula to each data point:

[0170]

[0171] At large negative values, the exponential function approaches 0.

[0172] Calculate D2, D3, D4, and D5 in the same way.

[0173] Calculate the key structure M:

[0174]

[0175] The result 35.12 represents a key measure of the data structure. This value reflects the clustering characteristics of the desensitized multi-physics model simulation data and the level of abstraction of the structural information, which can be further used for data classification, grouping or other forms of analysis.

[0176] Continue to use the key structure M = 23.4 obtained in the previous calculation. Assume that all α i The weight is set to 1 to simplify the calculation, and t = 5, then:

[0177] calculate

[0178]

[0179] Apply the formula to extract key information:

[0180]

[0181] The final K = 24.2 represents a comprehensive measure of key information extracted from the data structure, which can be understood as a data aggregation indicator that reflects the core characteristics or trends of the overall data and is suitable for subsequent data analysis or decision support.

[0182] For data that strictly retains the original value, apply the value replacement function. For example:

[0183]

[0184] h=2,n=16,after replacing the values ​​we get:

[0185]

[0186] The specific function characteristics and parameter values ​​are not visible, achieving the effect of data desensitization.

[0187] See also Figure 4 , the selection steps of the desensitization algorithm are as follows:

[0188] Based on the preliminary desensitized data packets, the mean and standard deviation of the data set are calculated using the formula,

[0189]

[0190] Generates a summary of statistical properties where x i represents a single data point, μ represents the mean, σ represents the standard deviation, and n represents the number of samples;

[0191] Using statistical characteristic summaries, the information gain evaluation method is applied to assess the relevance of data characteristics and processing requirements, using the formula,

[0192]

[0193] Determine a list of candidate desensitization algorithms, where G represents information gain and λ is a balancing factor used to adjust the sensitivity of the data;

[0194] According to the list of candidate desensitization algorithms, combined with the preset information gain threshold, the desensitization algorithm is selected using the formula:

[0195]

[0196] Among them, T is the preset information gain threshold, and the desensitization algorithm selection result is obtained.

[0197] Suppose there is a data set: x = [10, 20, 30, 40, 50].

[0198] Calculate the mean μ:

[0199]

[0200] Calculate the standard deviation σ:

[0201]

[0202] The obtained mean μ = 30 and standard deviation σ = 14.14 represent the center position of the data and the breadth of the data distribution. The two statistics are helpful to evaluate the distribution characteristics of the multiphysics model simulation data to select the most suitable desensitization algorithm.

[0203] Continue to use the above calculation results, assuming that the balance factor λ is 0.5, and calculate the information gain G.

[0204] Calculate the deviation and normalize each term:

[0205]

[0206] Calculate G:

[0207]

[0208] Information gain G = -20.88 reflects the matching degree between data characteristics and desensitization algorithm. A lower G value indicates that a desensitization algorithm that can better handle data characteristics (such as asymmetric distribution) needs to be selected.

[0209] Assuming that the preset information gain threshold T is -40, then, compare T with G

[0210] -20.88>-40

[0211] Therefore, the algorithm 1 selected based on the characteristics of the data is an algorithm that is more suitable for processing multi-physics field model simulation data.

[0212] See also Figure 5 , the specific steps for obtaining node customized desensitized data are:

[0213] According to the results of the desensitization algorithm selection, the data is differentiated and a dynamically adjusted standardization method is adopted. The formula is used.

[0214]

[0215] Among them, y i represents the data points after differential processing, μ and σ represent the mean and standard deviation, and ∈ is a fine-tuning parameter used to enhance the flexibility of the model and obtain differentially processed data;

[0216] Anonymize the differentially processed data, use hash functions and salt value technology to enhance security, and use the formula,

[0217] z i = hash(y i s+salt)

[0218] Among them, hash represents the hash function, z i represents the anonymized data point, s is the data scaling factor, and salt is the randomly generated salt value to generate an anonymized data set;

[0219] Integrate anonymized data sets, unify the data format among all distributed nodes, and generate node-customized desensitized data.

[0220] Continue to use the above data set x = [10, 20, 30, 40, 50] and μ = 30, σ = 14.14, assuming the fine-tuning parameter is 0.1, then,

[0221] For each data point x i Apply differential desensitization:

[0222]

[0223] For x1=10:

[0224]

[0225] Similarly, calculate other y2, y3, y4, and y5.

[0226] The obtained y value sequence such as [-1.41, -0.70, 0, 0.70, 1.41] represents the result of standardization through the adjusted standard deviation, reflecting the standardized deviation of each data point relative to the mean, which is conducive to further anonymization processing.

[0227] Assume that the data scaling factor s is 10 and the randomly generated salt value salt is "abc".

[0228] Using the above results and the given parameters, for each y i Anonymization:

[0229] For y1 = -1.41:

[0230] z1=hash(-1.41·10+"abc")==hash(-14.1+"abc")

[0231] Similarly, calculate z2, z3, z4, and z5.

[0232] The generated z-value sequence represents the anonymized data after hashing. The hash function ensures the irreversibility and privacy of the data, which is suitable for data sharing and analysis without leaking the original information.

[0233] Finally, the processed data is integrated and the format is unified (for example, the date format is unified) to ensure the consistency of data formats among all nodes, which helps to integrate and analyze data across systems or departments and avoid data processing errors caused by inconsistent formats.

[0234] See also Figure 6 ,The specific steps for analyzing the network structure between nodes are:

[0235] Combined with the node customized desensitized data, the total data processing volume of each node is calculated using the formula:

[0236]

[0237] Among them, T i represents the total data processing capacity of node i, d ij is the amount of data received or sent by node i to node j, α ij is the data importance weight, adjusted based on the priority or sensitivity of the data, M is the number of nodes directly connected to node i, and the comprehensive data processing volume is obtained;

[0238] According to the comprehensive data processing volume, the graph theory is applied to build a network graph model, using the formula,

[0239]

[0240] Among them, P opt represents the optimized network path, E represents the edge set in the network, T i and T j is the data processing amount connecting nodes i and j, β is the smoothing factor to prevent the denominator from being zero, and the network topology is obtained;

[0241] To evaluate the data transmission efficiency of the network topology, the formula is used:

[0242]

[0243] Where τ is the average data transmission time, P opt,kis the length of the kth optimized path, γ is the adjustment factor used to adjust the impact of path length, and F is the total number of paths. The network topology structure evaluation result is obtained.

[0244] Assume that node i is directly connected to two nodes M = 2, and the data volume d i1 =100GB,d i2 =150GB, and weight α i1 =0.5,α i2 =1.5, then

[0245] T i =(0.5×100)+(1.5×150)=50+225=275GB

[0246] The total data processing volume of node i is 275 GB, which indicates the data flow density and importance of this node in the network.

[0247] Assume that there are three nodes in the network, and it is known that T1 = 275GB, T2 = 300GB, T3 = 250GB, and the smoothing factor β is 10. Substituting into the formula, we get:

[0248]

[0249] choose As the calculation basis of the optimal path, it reflects that the path selected by the minimum spanning tree algorithm is selected under the minimum possible total weight that ensures network connectivity.

[0250] Assuming F = 2 paths, the adjustment factor γ is 0.05, combined with the above calculation results, the P of the two paths is opt They are 0.00171 and 0.00179 respectively.

[0251]

[0252] The average data transmission time is 17.1, which shows that the optimized network topology can effectively reduce the data transmission time and improve the network efficiency.

[0253] See also Figure 7 , the specific steps for obtaining the optimized data transmission path are:

[0254] According to the network topology evaluation results, the key transmission paths are optimized using the formula:

[0255]

[0256] Among them, L ij is the new weight from node i to node j, H ijis the path length from node i to node j, g is the weight factor used to adjust the sensitivity of the weight, h is a constant used to prevent division by zero errors, and the new weight parameter is obtained;

[0257] Apply the new weight parameters and conduct data forwarding tests using the formula:

[0258]

[0259] Among them, I ij is the estimated transmission time from node i to node j, ξ is the data transmission rate factor, and the packet forwarding test result is obtained;

[0260] According to the data packet forwarding test results, adjust and optimize the data transmission path, using the formula,

[0261] Q=∑ (i,j)∈Y I ij ·η ij

[0262] Where Q is the total transmission time, Y is the total number of critical paths, and η ij is the usage frequency of path (i, j), and outputs the optimized data transmission path.

[0263] Assume that H ij =5 as the path length of a node pair, the weight factor g is 100, h is 0.01, and the formula is substituted to obtain:

[0264]

[0265] The calculated L ij ≈4 represents the new weight applied between this particular pair of nodes, which is used in the network routing algorithm to optimize data transmission efficiency.

[0266] Use the above weights to calculate the result L ij ≈4, assuming the data transmission rate factor ξ is 2Mbps, then,

[0267]

[0268] The calculated I ij ≈0.05001 seconds, which is the expected data transmission time under a given network weight and data transmission rate, used to evaluate network performance and stability.

[0269] Assume that there are three critical paths in the network (Y=3), and the transmission time of each path is I ij is 0.125, using frequency η ij is 0.5, then,

[0270]

[0271] The calculated total transmission time Q = 0.1875 seconds reflects the expected total transmission time of the entire network under the optimized configuration, which helps to further evaluate and adjust the network structure to achieve higher efficiency and lower latency.

[0272] See also Figure 8 , the steps to obtain the final anonymous multi-physics model simulation data are as follows:

[0273] Collect node data from the optimized data transmission path and recalculate the total amount of data for each node using the formula,

[0274]

[0275] Among them, A i represents the total amount of data at node i, d ij is the amount of data received by node i or sent to node j, c ij is the transmission stability coefficient, ∈ is a constant to prevent division by zero errors, k j is the connection strength coefficient between nodes, log(1+k j ) represents nonlinear scaling of the connection strength, q represents the total number of nodes, and outputs summary data;

[0276] The singular value decomposition method is applied to reduce the dimension of the summary data, extract key features and reduce the data dimension. The formula is used.

[0277]

[0278] in, Represents the data matrix A i Power law transformation is applied to enhance the sensitivity of data processing. a is the adjustment coefficient, S represents the singular value, U and V T Represents the orthogonal matrix obtained by singular value decomposition, and obtains the singular value decomposition result;

[0279] According to the results of singular value decomposition, the data structure is adjusted and the formula is used.

[0280]

[0281] Among them, S reduced represents the key singular value, b is the index used to adjust the influence of the singular value, B is a small value used to ensure numerical stability, and the adjusted data structure A is generated adj ;

[0282] Use the adjusted data structure A adj and the singular value decomposition of V T Perform data reconstruction using the formula,

[0283]

[0284] Where c is an exponent used to enhance anonymity and generate the final anonymized multi-physics model simulation data A final , ensuring the privacy of multi-physics model simulation data and preserving key physical properties.

[0285] Assume that node i receives data from two nodes j=1, 2. The specific values ​​are as follows:

[0286] d i1 =100GB, c i1 =0.9, k1=12

[0287] d i2 =200GB, c i2 =0.95, k2=13

[0288] Assume ∈ is 0.01, substitute into the formula to calculate:

[0289] A i =(100·0.9+0.01)·log(1+12)+(200·0.95+0.01)·log(1+13)

[0290] A i =(90.01)·log(13)+(190.01)·log(14)

[0291] A i ≈90.01·1.114+190.01·1.146

[0292] A i ≈100.27+217.75

[0293] A i ≈318.02GB

[0294] Total data volume A i ≈318.02GB indicates that node i processes the sum of data from two other nodes based on the optimized data transmission path.

[0295] Using the above result A i ≈318.02GB, assuming the adjustment factor a is 2, then,

[0296]

[0297] Perform singular value decomposition processing, usually done by software, here is The values ​​are sample inputs.

[0298] Assume S reduced=17.83, b is 1.5, B is 0.1, then:

[0299]

[0300] A adj =U·(75.29+0.1)

[0301] A adj =U·75.39

[0302] A adj Represents the adjusted data structure, an optimized data form prepared for data reconstruction.

[0303] Assuming the exponent c is 1.1, use A adj =U·75.39, if U and V T is a simple matrix, specifically:

[0304]

[0305] V T Each element of is raised to the power of c = 1.1, that is

[0306]

[0307] In A adj and Multiplying them, we get:

[0308]

[0309]

[0310] Get an A final The matrix represents the reconstruction result of the data after singular value decomposition and power operation. This reconstruction increases the privacy of the data by changing the structure of the original data so that it is difficult to directly associate with the original information, while retaining the key features of the data, which is suitable for further data analysis and model simulation.

[0311] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. Data desensitization system for multi-physics field model simulation based on distributed computing, It is characterized in that The system comprises: The data feature analysis module performs data sensitivity analysis based on multi-physics field model data, adjusts the regularization coefficient according to the physical properties and sensitivity assessment of the data, performs initial data desensitization operations based on the adjustment results, analyzes the data structure, extracts key information, and generates preliminary desensitized data packets; The desensitization strategy customization module is based on the preliminary desensitized data packet. The distributed nodes select the desensitization algorithm according to the local data characteristics and processing requirements, perform differential desensitization according to the selection results, perform data anonymization processing, adjust and optimize the data format, ensure data consistency between nodes, and generate node customized desensitized data; The data transmission path optimization module combines the node customized desensitized data to optimize the path redundancy, uses the graphical model theory to analyze the network structure between nodes, optimizes the data transmission path according to the analysis results, adjusts the connection weights between nodes, performs data packet forwarding tests, and generates an optimized data transmission path; The multidimensional data anonymization module performs data processing based on the optimized data transmission path, reduces the dimension of the data by using singular value decomposition, adjusts the data structure, performs data reconstruction, and generates the final anonymous multi-physics field model simulation data; The adjustment steps of the regularization coefficient are specifically as follows: Perform attribute analysis on multi-physics model data to extract the physical parameters of each data point , calculate the sensitivity score , using the formula, Generate a list of sensitivity scores where, represents the number of types of physical properties, Representative The weight of the physical attributes, Representative data points In Properties The value on Using the sensitivity score list, the regularization coefficient is set, using the formula, Get the initial regularization coefficient ,in, represents the total number of data points, Represents the sensitivity score of a single data point; Fine-tune the initial regularization coefficient using the formula, Generate adjusted regularization coefficients ,in, represents the adjustment parameter, which is used to fine-tune the regularization coefficient so that the regularization coefficient matches the actual sensitivity distribution of the data; The steps for obtaining the preliminary desensitized data packet are specifically as follows: The adjusted regularization coefficient is used to desensitize the data set, and the desensitized data points are calculated using the formula: in, Representative The data value after data desensitization, represents the value of the original data point, After adjusting the regularization coefficient, the desensitized data list is obtained; Perform structural analysis on the desensitized data list, apply clustering algorithm, and adopt formula, in, represents the key metrics of the data structure, represents the clustering sensitivity parameter, Represents the total number of data points, which is used to adjust the accuracy and response of clustering and identify the key structure of the data; Based on the data key structure, the key information of the data is extracted using the formula: . ； in, A comprehensive measure representing key information extracted from a data structure, represents the total number of data points, represents the weight of the information category, represent The data point values ​​in ; For data that needs to keep its original value, atomization technology is used to replace the value based on physical and structural parameter functions before sending the data to the distributed computing server to desensitize the data. Through the model: in, , , ( )and is an arbitrary functional representing the model and the physics, ( ) is a parameter, ( ) is the value to be calculated remotely, based on the regularization coefficient evaluation, replacing the function equation with a numerical value: The physical and structural parameters The value of will be hidden in the value ( ) and c, generate a preliminary desensitized data packet; The selection steps of the desensitization algorithm are specifically as follows: Based on the preliminary desensitized data packet, the mean and standard deviation of the data set are calculated using the formula, Generates a summary of statistical properties where represents a single data point, represents the average value, represents the standard deviation, represents the sample size; Using the statistical characteristic summary, the information gain evaluation method is applied to assess the relevance of data characteristics and processing requirements, using the formula, Determine a list of candidate desensitization algorithms, where: represents information gain, is a balancing factor used to adjust the sensitivity of the data; According to the candidate desensitization algorithm list, combined with the preset information gain threshold, the desensitization algorithm is selected using the formula: in, is the preset information gain threshold, and the desensitization algorithm selection result is obtained; The specific steps for obtaining the node customized desensitized data are: According to the selection results of the desensitization algorithm, the data is processed differently, and a dynamically adjusted standardization method is adopted, using the formula, in, represents the data points after differentiation processing, and represents the mean and standard deviation, To fine-tune parameters, used to enhance the flexibility of the model and obtain differentiated processing data; The differentially processed data is anonymized, and the hash function and salt value technology are used to enhance security. The formula is used. hash ; Among them, hash represents the hash function, represents the anonymized data point, is the data scaling factor, salt is the randomly generated salt value, and anonymized data sets are generated; Integrate the anonymized data set, unify the data format among all distributed nodes, and generate node-customized desensitized data; The analysis steps of the network structure between the nodes are specifically as follows: Combined with the node customized desensitized data, the total data processing volume of each node is calculated using the formula: in, Representative Node The total data processing volume, Is a node Receive or send to a node The amount of data, is the data importance weight, adjusted based on the priority or sensitivity of the data, Is a node The number of directly connected nodes is used to obtain the comprehensive data processing capacity; According to the comprehensive data processing volume, the graph theory is applied to construct a network graph model, using the formula, in, represents the most optimized network path, represents the edge set in the network, and Is a connection node and The amount of data processed, is a smoothing factor, which prevents the denominator from being zero and obtains the network topology; To evaluate the data transmission efficiency of the network topology, the formula is used: in, is the average data transfer time, It is The length of the optimized path, is the adjustment factor used to adjust for the path length effect, is the total number of paths, and the network topology evaluation result is obtained; The steps for obtaining the optimized data transmission path are specifically as follows: According to the network topology evaluation results, the key transmission path is optimized using the formula: in, Is a node To Node The new weight of Is a node To Node The path length, is the weight factor, which is used to adjust the sensitivity of the weight. is a constant used to prevent division by zero errors and obtain new weight parameters; Apply the new weight parameters to perform data forwarding test, using the formula, in, Is a node To Node The estimated transfer time is is the data transmission rate factor, and the packet forwarding test result is obtained; According to the data packet forwarding test results, the data transmission path is adjusted and optimized, using the formula, in, is the total transmission time, represents the total number of critical paths, Is the path The usage frequency is used to output the optimized data transmission path; The steps for obtaining the final anonymous multi-physics model simulation data are specifically as follows: Collect node data from the optimized data transmission path, recalculate the total amount of data for each node, and use the formula, in, Representation Node The total amount of data, Is a node Receive or send to a node The amount of data received, is the transmission stability coefficient, is a constant to prevent division by zero errors, is the connection strength coefficient between nodes, represents a nonlinear scaling of the connection strength, Indicates the total number of nodes and outputs summary data; The singular value decomposition method is applied to reduce the dimension of the summary data, extract key features and reduce the data dimension. The formula is used. in, Represents the data matrix Apply power law transformation to enhance the sensitivity of data processing. is the adjustment factor, represents singular values, and Represents the orthogonal matrix obtained by singular value decomposition, and obtains the singular value decomposition result; According to the singular value decomposition result, the data structure is adjusted, and the formula is adopted. in, represents the key singular value, is an exponent used to adjust the influence of singular values, Is a small value, used to ensure numerical stability, to generate an adjusted data structure ; Using the adjusted data structure and singular value decomposition Perform data reconstruction using the formula, in, is an index used to enhance anonymity and generate the final anonymized multiphysics model simulation data , ensuring the privacy of multi-physics model simulation data and preserving key physical properties.

Images