A power grid sensitive data implicit inference risk assessment method and related device
By constructing a multidimensional feature risk mapping model and optimizing it with a deep neural network, the problems of subjectivity of manual thresholds and low accuracy of assessment in implicit inference risk assessment of power grid sensitive data are solved, and more accurate risk assessment is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XI AN JIAOTONG UNIV
- Filing Date
- 2026-05-09
- Publication Date
- 2026-06-09
AI Technical Summary
Existing technologies for risk assessment of implicit inference of sensitive power grid data rely on manually setting fixed thresholds for correlation coefficients, resulting in high subjectivity and low assessment accuracy. Furthermore, a single statistical indicator is insufficient to fully characterize complex dependencies, making it prone to missed or incorrect assessments.
A multidimensional feature risk mapping model is constructed, and a loss function with L2 regularization constraint is introduced for joint optimization. The weight allocation of various statistical correlation coefficients is learned through a deep neural network, and a correlation coefficient threshold discrimination formula is generated to achieve adaptive learning and dynamic generation of the discrimination boundary.
It significantly improves the objectivity, consistency, and transferability of the assessment, comprehensively depicts the linear, nonlinear, and structural dependencies between power grid auxiliary characteristics and sensitive data, enhances the engineering authenticity and reliability of risk assessment, and avoids omissions and misjudgments caused by a single indicator.
Smart Images
Figure FT_1 
Figure FT_2 
Figure SMS_1
Abstract
Description
Technical Field
[0001] This invention belongs to the field of power system information security and data privacy protection technology, specifically relating to a method and related apparatus for risk assessment of implicit inference of sensitive power grid data. Background Technology
[0002] With the deepening construction of smart grids and the Internet of Things for power, massive amounts of power business data need to be continuously interacted and shared across all stages of generation, transmission, distribution, and consumption, as well as among different entities. Power system operation is influenced by complex constraints such as Kirchhoff's laws, power flow equations, and market clearing mechanisms. Different power variables typically exhibit significant linear correlations, nonlinear dependencies, and structural coupling relationships. For example, node voltage, injected power, line power flow, marginal electricity price, and their components often do not change in isolation but rather form complex relationships under the combined influence of physical constraints and operating conditions.
[0003] This complex relationship creates a hidden security problem in the process of opening, sharing and using power data: even if attackers cannot directly obtain highly sensitive data, they may use publicly available or less sensitive auxiliary data to analyze the statistical correlation between the two and build an inference model to reverse-engineer the target sensitive data, thus creating a hidden inference risk.
[0004] In existing inferential risk assessment methods, the correlation coefficient threshold is often used as an important basis for determining whether auxiliary data may constitute a risk. However, existing technologies still have significant shortcomings in determining the correlation coefficient threshold: 1. Traditional assessment methods typically rely on expert experience to set fixed thresholds for a single correlation coefficient. While this approach is simple to implement, it is highly subjective and difficult to adapt to risk assessment tasks under different power grid topologies, operating conditions, and data distributions. Furthermore, it is difficult to provide a unified explanation for the rationality of the threshold settings.
[0005] 2. A single statistical indicator is insufficient to fully characterize complex dependencies. For example, the Pearson correlation coefficient is better suited to reflecting linear relationships between variables; while mutual information, distance correlation coefficients, or kernel independence statistics can characterize more general nonlinear dependencies, their numerical ranges, dimensional characteristics, and sensitive objects are inconsistent. If a fixed threshold is still set for a single indicator, it is easy to miss some nonlinear risks or misjudge weak or redundant correlations.
[0006] 3. Relying solely on the magnitude of a correlation coefficient cannot reliably reflect the risk contribution of auxiliary data in the actual inference process. Some auxiliary features may not show significant correlation when viewed alone, but may generate strong inference power when combined with other variables; others may have high values for a certain statistical indicator, but their contribution to actual inference is limited. Therefore, there is no simple one-to-one correspondence between the magnitude of the correlation coefficient and the strength of inference risk.
[0007] Therefore, a method is needed to construct the judgment boundary by comprehensively considering multidimensional correlation coefficients in order to solve the problems of high subjectivity and low accuracy of manually set thresholds. Summary of the Invention
[0008] The purpose of this invention is to provide a method and related apparatus for implicit inference risk assessment of power grid sensitive data, overcoming the problems of high subjectivity and low assessment accuracy caused by relying on manually setting fixed thresholds for correlation coefficients in existing technologies for inferring risk assessment of power grid sensitive data. This invention constructs a multi-dimensional feature risk mapping model, introduces a loss function with L2 regularization constraints for joint optimization, learns the weight allocation of various statistical correlation coefficients, and thus generates a correlation coefficient threshold discrimination formula.
[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for implicit inference risk assessment of power grid sensitive data, comprising the following steps: Obtain the statistical correlation coefficients corresponding to the auxiliary characteristics of the power grid to be evaluated; The obtained statistical correlation coefficient is used as the input to the correlation coefficient threshold discrimination boundary model. The correlation coefficient threshold discrimination boundary model is used to evaluate whether the power grid auxiliary features constitute implicit inference risk, wherein: A correlation coefficient threshold discrimination boundary model is constructed using the parameters of a deep neural network inference model. The parameters include a feature weight vector and a bias scalar.
[0010] Preferably, the expression for the correlation coefficient threshold discrimination boundary model is:
[0011] in, The feature weight vector; It is a biased scalar; This is the aligned statistical correlation coefficient vector; This is the matrix transpose.
[0012] Preferably, the method for constructing a deep neural network inference model is as follows: Obtain the target sensitive data vector and the corresponding power auxiliary data matrix; Calculate six statistical correlation coefficients between the target sensitive data vector and each auxiliary feature vector in the power auxiliary data matrix; The deep neural network inference model is trained using each auxiliary feature vector and the corresponding six statistical correlation coefficients to obtain the trained deep neural network inference model. The trained deep neural network inference model is optimized by using a composite loss function that includes a mean squared error term and an L2 regularization term, resulting in an optimized deep neural network inference model.
[0013] Preferably, the deep neural network inference model includes a risk mapping layer, a weighted feature layer, and an inference prediction layer, wherein: The risk mapping layer is used to perform nonlinear mapping on the input statistical correlation coefficient vector to obtain the probability of risk occurrence; The weighted feature layer is used to multiply the probability of risk occurrence with the power grid auxiliary feature vector to obtain weighted features; The inference and prediction layer takes weighted features as input and outputs the inferred and predicted value of the target sensitivity vector corresponding to the power grid auxiliary feature vector through inter-layer nonlinear reconstruction.
[0014] Preferably, the expression for the composite loss function, which includes a mean squared error term and an L2 regularization term, is as follows:
[0015] in, The total number of samples; These are the first and second parts of the target sensitive data vector. The inferred predicted value and the actual observed value of a sample; The L2 regularization coefficient; For the first The weighted components of the statistical correlation coefficient; This is a composite loss function.
[0016] Preferably, the six statistical correlation coefficients are: Pearson correlation coefficient reflecting linear association, Spearman rank correlation coefficient reflecting the degree of monotonic association, Kendall rank correlation coefficient reflecting the consistency of sample pairs, normalized mutual information reflecting the amount of shared information between variables, distance correlation coefficient reflecting the overall dependence between variables, and kernel independence statistic reflecting nonlinear physical coupling relationship.
[0017] Secondly, the present invention provides a risk assessment system for implicit inference of power grid sensitive data, comprising: The correlation coefficient acquisition unit is used to obtain the statistical correlation coefficients corresponding to the auxiliary characteristics of the power grid to be evaluated. The risk inference unit uses the obtained statistical correlation coefficient as input to the correlation coefficient threshold discrimination boundary model, and evaluates whether the power grid auxiliary features constitute implicit inferred risks through the correlation coefficient threshold discrimination boundary model, wherein: A correlation coefficient threshold discrimination boundary model is constructed using the parameters of a deep neural network inference model. The parameters include a feature weight vector and a bias scalar.
[0018] Thirdly, the present invention provides an electronic device including a processor and a memory, wherein the memory stores computer instructions, and when the computer instructions are executed by the processor, the electronic device performs the aforementioned method for implicit inference risk assessment of power grid sensitive data.
[0019] Fourthly, the present invention provides a computer program product containing computer-executable instructions, which, when executed, implement the aforementioned method for implicit inference risk assessment of power grid sensitive data.
[0020] Fifthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the aforementioned method for implicit inference risk assessment of power grid sensitive data.
[0021] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a method for implicit inference risk assessment of sensitive power grid data. Addressing the issues of high subjectivity and low accuracy in existing technologies that rely on manually setting fixed thresholds for correlation coefficients, this invention obtains the statistical correlation coefficients corresponding to the auxiliary features of the power grid to be assessed and inputs them into a correlation coefficient threshold discrimination boundary model constructed using parameters (including feature weight vectors and bias scalars) of a deep neural network inference model. This achieves adaptive learning and dynamic generation of the discrimination boundary, thus avoiding the subjectivity and scenario adaptability problems caused by manual thresholds, significantly improving the objectivity, consistency, and transferability of the assessment. This method uses statistical correlation coefficients as model inputs and can integrate multi-dimensional statistical correlation information, rather than relying solely on statistical correlation coefficients. Limited by a single statistical indicator, this method more comprehensively characterizes the linear, nonlinear, and structural dependencies between power grid auxiliary features and sensitive data, overcoming the problems of missed and false judgments that are easily caused by a single indicator. At the same time, since this discrimination boundary model directly utilizes the parameters of a deep neural network inference model, which itself is trained with the actual inference task (i.e., predicting target sensitive data through auxiliary features) as its objective, the discrimination boundary constructed by its parameters reflects the mapping relationship between the statistical correlation coefficient and the true implicit inferred risk, rather than a simple comparison of the magnitude of the correlation. This effectively solves the engineering pain point that "high correlation does not equal high risk, and low correlation does not equal no risk," making the assessment results closer to the actual risk level and improving the engineering authenticity and reliability of risk assessment.
[0022] Furthermore, by introducing a risk mapping layer with L2 regularization constraints, this invention achieves a unified mapping from multiple statistical correlation coefficients to the implicit inferred risk strength, reducing the subjectivity caused by manually setting a single correlation coefficient threshold. Attached Figure Description
[0023] Figure 1 This is an overall flowchart of the present invention; Figure 2 This is a schematic diagram of the structure of a deep neural network inference model that includes a risk mapping layer. Detailed Implementation
[0024] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.
[0025] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.
[0026] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.
[0027] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."
[0028] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0029] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.
[0030] Example 1 As 1 to Figure 2 As shown in this embodiment, a method for implicit inference risk assessment of power grid sensitive data includes the following steps: Step 1: Obtain a power auxiliary data matrix containing multiple samples. With target sensitive data vector ; Regarding the power auxiliary data matrix The Middle Auxiliary feature vectors Calculate its relationship with the target sensitive data vector. The six statistical correlation coefficients between them were used to construct the first... The six-dimensional statistical correlation coefficient vector corresponding to each auxiliary feature vector The system employs a standardization method for domain alignment and outputs an aligned statistical correlation coefficient vector. ; Step 2, the auxiliary feature vector and its corresponding aligned statistical correlation coefficient vector As input, the data is fed into a deep neural network inference model containing a risk mapping layer and an inference prediction layer for training, and a composite loss function including a mean squared error term and an L2 regularization term is used. The deep neural network inference model is optimized and its parameters are updated until convergence. Step 3: Extract the globally shared feature weight vector from the risk mapping layer after training convergence. With bias scalar We construct a threshold discrimination boundary inequality for the correlation coefficient, and then complete the implicit inference risk assessment of power data.
[0031] In step 1 of this embodiment, the six statistical correlation coefficients specifically include: Pearson correlation coefficient reflecting linear association, Spearman rank correlation coefficient reflecting the degree of monotonic association, Kendall rank correlation coefficient reflecting the consistency of sample pairs, normalized mutual information reflecting the amount of shared information between variables, distance correlation coefficient reflecting the overall dependence between variables, and kernel independence statistic reflecting nonlinear physical coupling relationship.
[0032] In step 2 of this embodiment, the risk mapping layer is set at the input end of the deep neural network inference model, and the parameters of the risk mapping layer include feature weight vectors. With bias scalar This layer receives the aligned statistical correlation coefficient vector. As a feature input, and after non-linear mapping using the sigmoid function, the probability of the risk occurring is output. The calculation formula is:
[0033] in, This is the matrix transpose. It is an exponential function.
[0034] Subsequently, the deep neural network inference model will use the original auxiliary feature vector. The probability of the aforementioned risk occurring Perform multiplication to obtain weighted feature input. The weighted features are then input into the inference and prediction layer, which consists of multiple fully connected layers, and a target-sensitive data vector is output through inter-layer nonlinear reconstruction. inferred predicted value .
[0035] In step 2 of this embodiment, the composite loss function The definition of is:
[0036] in, The total number of samples; These are the target sensitive data vectors. The Middle The inferred predicted value and the actual observed value of a sample; The L2 regularization coefficient; Feature weight vector The first in The weighted components of the statistical correlation coefficient.
[0037] The composite loss function The mean squared error term is used to constrain the inference error of the network on sensitive data and maintain the inference accuracy of the model; the L2 regularization term is used to adjust the feature weight vector. Apply numerical penalty constraints.
[0038] During model training and optimization, the two form a mutually restrictive balance mechanism: the L2 regularization term forces all weight parameters to shrink towards zero, while increasing the weight values will incur additional loss penalties; therefore, the weights are only adjusted when a certain statistical correlation coefficient can significantly reduce the inference error. Only then can it maintain a high level during backpropagation.
[0039] Conversely, for non-critical statistical indicators that contribute little to the inference, contain redundant information, or have weak correlations, their corresponding weights will adaptively shrink to near zero during training iterations. Through joint optimization of the aforementioned composite loss function, the model can drive the weights corresponding to various statistical correlation coefficients to form a differentiated distribution, naturally filtering out invalid statistical dimensions while maintaining overall inference accuracy. The final converged feature weight vector... This approach can objectively reflect the true contribution of each statistical indicator to implicit inference risk, thus providing a quantitative basis for constructing an accurate correlation coefficient threshold discrimination boundary. Simultaneously, the numerical constraints imposed by the L2 regularization term effectively prevent individual weights from being abnormally amplified due to fitting local noise, suppressing model overfitting and ensuring the generalization performance of the discrimination boundary under different power grid data scenarios.
[0040] In step 3 of this embodiment, the method for constructing the correlation coefficient threshold discrimination boundary inequality is as follows: The threshold for the risk mapping layer is set as follows: Based on the monotonically increasing property of the Sigmoid function and its mathematical property of taking a value of 0.5 at the origin, when the output probability... At that time, it is equivalent to the input terms of the Sigmoid function. By simplifying and extracting the linear discriminant boundary, an explicit correlation coefficient threshold discriminant boundary inequality is constructed as follows:
[0041] in, Represents the feature weight vector; Represents a bias scalar; This represents the aligned six-dimensional statistical correlation coefficient vector, i.e., the... A set of six correlation coefficient indices between auxiliary feature vectors and target sensitive data after Z-Score standardization.
[0042] Example 2 This embodiment provides a method for implicit inference risk assessment of power grid sensitive data. The test data comes from real-time electricity data for the Mid-Atlantic region provided by the PJM (Pennsylvania-New Jersey-Maryland) energy market. The dataset covers a complete one-year period from January 1, 2024 to January 1, 2025, containing 8,762 records. The original data has 59 dimensions, including generation, system load information, and various nodal marginal price (LMP) components. Target setting: The core sensitive data "voltage node power consumption" is selected as the target object to be protected, i.e., the target sensitive data vector. The remaining 58 dimensions of data serve as potential sources of auxiliary inference, namely the power auxiliary data matrix. .
[0043] like Figure 1 As shown, the method in this embodiment mainly includes the following three core steps: Step S1: Calculate the power auxiliary data matrix Multiple statistical correlation coefficients between each auxiliary feature vector and the target sensitive data vector are used to construct a multidimensional correlation feature vector. Then, the multidimensional correlation feature vector is subjected to Z-Score standardization to eliminate dimensional differences and achieve domain alignment.
[0044] The specific implementation process is as follows: Step S101: For each auxiliary feature vector among the 58 auxiliary features Calculate its relationship with the target sensitive data vector respectively. The six statistical correlation coefficients constitute a six-dimensional statistical correlation coefficient vector. ,in: .
[0045] in, Represents the normalized mutual information value; This represents the Spearman rank correlation coefficient value; This represents the Pearson correlation coefficient value; This represents the Kendall rank correlation coefficient value; This represents the value of the distance correlation coefficient; This represents the numerical value of the nuclear independence statistic. The specific calculation principle and formula are as follows: (1) Normalized mutual information (NMI, denoted as ): Considering the characteristics of electric auxiliary The main characteristic is that it is a continuous numerical variable, therefore the mutual information regression method is used to estimate it. and mutual information And the histogram discretization method is used to estimate respectively and Empirical information entropy and A minimum entropy normalization mechanism is introduced, and the calculation formula is as follows:
[0046] in, Indicate auxiliary features sensitive data of the target Mutual information between them is used to characterize the amount of information shared between them; Indicate auxiliary features Empirical information entropy, Indicates target sensitive data The empirical information entropy is used to measure the amount of uncertainty information contained in the corresponding variable itself; This indicates that the smaller of the two information entropies is used as the normalization benchmark; outer layer and This is used to restrict the results to the [0,1] interval to enhance numerical stability and comparability.
[0047] In addition, when When this value is defined as 0, it indicates that the normalized denominator does not possess effective informational meaning, therefore... Defined as 0 to avoid denominator anomalies or numerical instability.
[0048] (2) Spearman rank correlation coefficient (Spearman, denoted as...) ): Auxiliary feature vector and target sensitive data vector Convert them into rank variables respectively and The calculation formula is:
[0049] in, and Both represent rank variables. That is, the auxiliary feature vector... and target sensitive data vector After sorting by size in their respective sample sequences, the rank number of the data's position; and Representing the rank sequence and The sample mean.
[0050] (3) Pearson correlation coefficient (hereinafter referred to as Pearson) The formula for measuring the strength of a linear correlation is:
[0051] in, Representing the Auxiliary feature variables The average of all sample observations; Represents sensitive target data The average of all sample observations.
[0052] (4) Kendall's rank correlation coefficient (Kendall, denoted as ): The number of identical pairs changing in the same direction. The number of opposite-order pairs with inverse changes Let the sample size be... The calculation formula is:
[0053] (5) Distance Correlation (hereinafter referred to as) ): Constructing auxiliary feature vectors and The pairwise Euclidean distance matrix is obtained and double-centered. The distance covariance is calculated. and distance variance The calculation formula is:
[0054] in, Represents auxiliary feature vectors and The sample distance covariance between the variables is used to measure the degree of deviation between the two variables based on their independence from each other based on their pairwise distance. Represents auxiliary feature vectors The sample distance variance; : Represents the target sensitive data vector The sample distance variance; For others.
[0055] (6) Nuclear Independence Statistic (HSIC, denoted as ): Selecting a Gaussian radial basis kernel function to map the data to the reproducing kernel Hilbert space, and calculating the kernel matrix respectively. and ,in, Represents auxiliary feature vectors The kernel matrix obtained under the action of the kernel function can be represented by the following matrix elements: ; Represents the target sensitive data vector The kernel matrix obtained under the action of the kernel function can be represented by the following matrix elements: Introducing a centralized matrix The calculation formula is:
[0056] in, Indicates the number of samples; The trace operation represents the sum of the elements on the main diagonal of a square matrix. By centering the kernel matrix and calculating the trace value, the statistical dependence between the auxiliary feature vector and the target sensitive data vector in the high-dimensional kernel space can be measured. The larger the value, the stronger the nonlinear correlation between the two.
[0057] The absolute values of each coefficient are extracted, arranged in order, and concatenated to form an initial six-dimensional statistical correlation coefficient vector. .
[0058] Step S102: Perform domain alignment using the Z-Score normalization method. For the first... The standardization process for the statistical correlation coefficient is as follows:
[0059] in, and For each of the auxiliary feature vectors at the th The mean and standard deviation of the statistical correlation coefficient; Representing the The initial six-dimensional statistical correlation coefficient vector calculated from the auxiliary features The first in The specific values of the statistical correlation coefficients (where) The value ranges from 1 to 6.
[0060] Each indicator forms an aligned statistical correlation coefficient vector. .
[0061] Step S2: Use the auxiliary feature vector and its corresponding aligned statistical correlation coefficient vector as input to train and update the model parameters of the deep neural network inference model containing the risk mapping layer and the inference prediction layer.
[0062] In this step, the auxiliary feature vector and its corresponding aligned statistical correlation coefficient vector As input.
[0063] (1) Risk mapping layer processing: This layer is set at the model input and contains feature weight vectors. and bias scalar ,in, The set of real numbers. This layer receives... Then, a non-linear mapping is performed using the Sigmoid function to output the probability of the risk occurring. :
[0064] (2) Inference prediction layer processing: The deep neural network inference model will process the auxiliary feature vector Corresponding probability of risk occurrence Perform element-wise multiplication to obtain weighted feature inputs. Then the weighted features are input. The input is a prediction layer consisting of multiple fully connected layers, and the output is a target-sensitive data vector. inferred predicted value .
[0065] (3) Model optimization training: A composite loss function including mean squared error term and L2 regularization term is adopted. Update the parameters:
[0066] in, The total number of samples; These are the target sensitive data vectors. The Middle The inferred predicted value and the actual observed value of a sample; The L2 regularization coefficient; Feature weight vector The first in The weighted components of the statistical correlation coefficient.
[0067] During training, the mean squared error term is used to constrain the network's inference error on sensitive data, ensuring that the model can reconstruct sensitive features; the L2 regularization term is used to adjust the feature weight vector. elements in Numerical constraints are applied. This measure encourages the weights corresponding to non-critical correlation coefficients to adaptively shrink during training and strengthens the expression of risk contribution of important statistical indicators. At the same time, it can effectively prevent individual weights from being abnormally amplified due to overfitting noise, thereby suppressing model overfitting and improving the generalization performance of the decision boundary under different power grid scenarios.
[0068] Step S3: Extract the weights and bias scalars after training convergence, construct the correlation coefficient threshold discrimination boundary inequality, and then complete the implicit inference risk assessment of power data.
[0069] After the model training converges, extract the globally shared feature weight vector from the risk mapping layer. With bias scalar According to the mathematical properties of the Sigmoid function, when its input terms... When the value is greater than 0, the output value is... Greater than 0.5. The risk assessment threshold is set as follows: This means that the auxiliary feature is considered to have a significant implicit inference risk. By simplifying the mapping probability formula and removing the nonlinear mapping function, an explicit correlation coefficient threshold discrimination boundary inequality is constructed:
[0070] Based on the framework of this embodiment, hierarchical evaluation is performed: due to the bias scalar In a mathematical sense, this symbolizes the rigor of risk assessment. Therefore, a set of parameters can be extracted to set the primary risk boundary before the model fully converges, and the parameters can be extracted to set the secondary risk boundary after the model fully converges. First-level risk boundary (preliminary screening): Before the model fully converges (e.g., extracting parameters from the 80th training iteration), the bias term at this stage... It is approximately -0.4. Substituting this into the inequality... Thirteen risk features were selected from 58 auxiliary features. These 13 features were used solely for inference and prediction, and their coefficients of determination were used to evaluate the goodness of fit and reconstruction accuracy of the inference model. The coefficient of determination reached 0.9784. The calculation formula is:
[0071] In the formula, The sample mean of the actual observed values of the target sensitive data. The closer the value is to 1, the higher the accuracy of the inference model in reconstructing the target sensitive data, meaning the stronger the implicit inference risk constituted by this set of auxiliary features. As a comparative test, if only the remaining 45 removed auxiliary features are used for inference and prediction, its... The value is only 0.1706. This comparison result shows that the primary risk boundary successfully retains the main inferred risk sources with a relatively small parameter threshold and effectively filters out redundant features without risk association, thus playing a good preliminary screening role.
[0072] Secondary risk boundary (core feature localization): Extract the parameters when the model training is in a convergent state (e.g., the 200th round), at which point the feature weight vector... and bias scalar The change is stable, and the bias term is biased. The expression converges to -1.2. Substituting this into the inequality... Because the judgment threshold is stricter, only four core auxiliary features are retained for judgment. Sensitive data is predicted using only these four dimensions of data. It remains at 0.9607; as a comparative test, if only the remaining 54 auxiliary features that failed the judgment are used for inference and prediction, its... The value is only 0.4085. This comparison result shows that the second-level risk boundary can accurately pinpoint the core risk sources constituting implicit inferred risk while reducing the feature size.
[0073] Experimental results show that this method successfully constructs a dynamic threshold for multidimensional correlation coefficients by utilizing L2 regularization constraints and a logical mapping layer. This mechanism effectively overcomes the subjective shortcomings of traditional assessments that rely on single statistical indicators and manually set fixed thresholds. Through adaptive learning and convergence of model parameters, it achieves an objective and accurate quantitative assessment of implicit inference risks in sensitive power grid data.
[0074] Example 3 This embodiment provides a power grid sensitive data implicit inference risk assessment system, including: The correlation coefficient acquisition unit is used to obtain the statistical correlation coefficients corresponding to the auxiliary characteristics of the power grid to be evaluated. The risk inference unit uses the obtained statistical correlation coefficient as input to the correlation coefficient threshold discrimination boundary model, and evaluates whether the power grid auxiliary features constitute implicit inferred risks through the correlation coefficient threshold discrimination boundary model, wherein: A correlation coefficient threshold discrimination boundary model is constructed using the parameters of a deep neural network inference model. The parameters include a feature weight vector and a bias scalar.
[0075] Example 4 This embodiment also provides a computing device. The computing device includes a bus, a processor, a memory, and a communication interface. The processor, memory, and communication interface communicate with each other via the bus. The computing device can be a server or a terminal device. It should be understood that this application does not limit the number of processors and memory in the computing device.
[0076] A bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, a bus can include a path for transmitting information between various components of a computing device (e.g., memory, processor, communication interfaces).
[0077] The processor may include any one or more of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), Tensor Processing Unit (TPU), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), Microprocessor (MP), or Digital Signal Processor (DSP).
[0078] The memory may include volatile memory, such as random access memory (RAM). The processor may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).
[0079] The memory stores executable program code, which the processor executes to implement the functions of the aforementioned units, thereby achieving, for example, the method described in Embodiment 1. That is, the memory may store instructions for the methods and functions relating to the computing device in any of the above embodiments.
[0080] The communication interface uses transceiver modules such as, but not limited to, network interface cards and transceivers to enable communication between computing devices and other devices or communication networks.
[0081] Example 5 This embodiment also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, cause the processor to perform the methods and functions of the computing device involved in any of the above embodiments.
[0082] Generally, the various embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software, which can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of this disclosure are shown and described as block diagrams, flowcharts, or represented using some other illustration, it should be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as, as non-limiting examples, in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.
[0083] Example 6 This embodiment provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods as described above with reference to the accompanying drawings. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.
[0084] Computer program code used to implement the methods of this disclosure may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, such that when executed by the computer or other programmable data processing apparatus, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be performed. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.
[0085] In the context of this disclosure, computer program code or related data may be carried on any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, etc. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.
[0086] Computer-readable media can be any tangible medium that contains or stores programs for or relating to an instruction execution system, apparatus, or device, or a data storage device such as a data center containing one or more available media. Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. More detailed examples of computer-readable storage media include electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical storage devices, magnetic storage devices, or any suitable combination thereof.
[0087] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A method for implicit inference risk assessment of sensitive power grid data, characterized in that, Includes the following steps: Obtain the six statistical correlation coefficients corresponding to the auxiliary characteristics of the power grid to be evaluated; The six statistical correlation coefficients obtained are used as inputs to the correlation coefficient threshold discrimination boundary model. The correlation coefficient threshold discrimination boundary model is used to evaluate whether the power grid auxiliary features constitute implicit inference risks, wherein: A correlation coefficient threshold discrimination boundary model is constructed using the parameters of a deep neural network inference model. The parameters include a feature weight vector and a bias scalar.
2. The method for implicit inference risk assessment of power grid sensitive data according to claim 1, characterized in that, The expression for the correlation coefficient threshold discrimination boundary model is as follows: in, The feature weight vector; It is a biased scalar; This is the aligned statistical correlation coefficient vector; This is the matrix transpose.
3. The method for implicit inference risk assessment of power grid sensitive data according to claim 1, characterized in that, The method for constructing a deep neural network inference model is as follows: Obtain the target sensitive data vector and the corresponding power auxiliary data matrix; Calculate six statistical correlation coefficients between the target sensitive data vector and each auxiliary feature vector in the power auxiliary data matrix; The deep neural network inference model is trained using each auxiliary feature vector and the corresponding six statistical correlation coefficients to obtain the trained deep neural network inference model. The trained deep neural network inference model is optimized by using a composite loss function that includes a mean squared error term and an L2 regularization term, resulting in an optimized deep neural network inference model.
4. A method for implicit inference risk assessment of power grid sensitive data according to claim 1 or 3, characterized in that, The deep neural network inference model includes a risk mapping layer, a weighted feature layer, and an inference prediction layer, wherein: The risk mapping layer is used to perform nonlinear mapping on the six input statistical correlation coefficients to obtain the probability of risk occurrence. The weighted feature layer is used to multiply the probability of risk occurrence with the power grid auxiliary feature vector to obtain weighted features; The inference and prediction layer takes weighted features as input and outputs the inferred and predicted value of the target sensitivity vector corresponding to the power grid auxiliary feature vector through inter-layer nonlinear reconstruction.
5. The method for implicit inference risk assessment of power grid sensitive data according to claim 3, characterized in that, The expression for the composite loss function, which includes a mean squared error term and an L2 regularization term, is as follows: in, The total number of samples; These are the first and second parts of the target sensitive data vector. The inferred predicted value and the actual observed value of a sample; The L2 regularization coefficient; For the first The weighted components of the statistical correlation coefficient; This is a composite loss function.
6. The method for implicit inference risk assessment of power grid sensitive data according to claim 1, characterized in that, The six statistical correlation coefficients are: Pearson correlation coefficient, which reflects linear association; Spearman rank correlation coefficient, which reflects the degree of monotonic association; Kendall rank correlation coefficient, which reflects the consistency of sample pairs; normalized mutual information, which reflects the amount of shared information between variables; distance correlation coefficient, which reflects the overall dependence between variables; and kernel independence statistic, which reflects nonlinear physical coupling.
7. A risk assessment system for implicit inference of sensitive power grid data, characterized in that, include: The correlation coefficient acquisition unit is used to obtain six statistical correlation coefficients corresponding to the auxiliary characteristics of the power grid to be evaluated; The risk inference unit uses the obtained six statistical correlation coefficients as input to the correlation coefficient threshold discrimination boundary model. The model assesses whether the power grid auxiliary features constitute implicit inferred risk. A correlation coefficient threshold discrimination boundary model is constructed using the parameters of a deep neural network inference model. The parameters include a feature weight vector and a bias scalar.
8. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer instructions, which, when executed by the processor, cause the electronic device to perform a method for implicit inference risk assessment of power grid sensitive data as described in any one of claims 1 to 6.
9. A computer program product, characterized in that, The computer program product contains computer-executable instructions, which, when executed, implement the implicit inference risk assessment method for sensitive power grid data according to any one of claims 1 to 6.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the implicit inference risk assessment method for sensitive power grid data according to any one of claims 1 to 6.