Identification method of sensitive data for power grid regulation

Abstract

The present invention specifically relates to a method for identifying sensitive data of power grid regulation, and belongs to the field of smart grid security protection technology. The method comprises the following steps: first collecting data that needs to be sensitively identified, and then quantizing the unstructured features and structured features of the data; calculating the relative sensitivity and absolute sensitivity of the unstructured features and structured features of the data that needs to be sensitively identified based on the results of vectorization; and identifying whether the power grid regulation data is sensitive based on the relative sensitivity and absolute sensitivity. The present invention adopts the Transformer model to extract, analyze and process the unstructured features of the power grid regulation data, and at the same time calculates the absolute sensitivity and relative sensitivity of the unstructured features as the basis for determining whether the power grid regulation data is sensitive, which greatly improves the processing of unstructured features in the power grid regulation data, and solves the problem of low accuracy of sensitivity identification of power grid regulation data with a large proportion of unstructured features in the existing technology.

Description

Technical Field

[0001] The present invention specifically relates to a method for identifying sensitive data of power grid regulation, and belongs to the technical field of smart grid security protection. Background Art

[0002] Grid control data refers to the various data used to monitor and manage power systems. This data typically includes real-time and historical data, such as voltage, current, frequency, load, and power market operations. It comes from diverse sources, has varying degrees of structure, is massive, time-sensitive, and highly socially valuable. With the acceleration of the digital transformation of power grids, the variety and volume of grid control data are rapidly increasing, making data security issues more prominent and complex, and posing unprecedented challenges to data security.

[0003] Grid control data contains a large amount of sensitive information, including grid control information, control strategies, and user privacy. Failure to implement effective access control for sensitive information or to effectively protect it from tampering can lead to information leaks, control errors, and even major social and national security incidents, impacting social stability and economic development. Therefore, identifying sensitive grid control data during transmission has become a key issue in smart grid security.

[0004] Most existing technologies use traditional machine learning algorithms to identify sensitive data of power grid regulation. Although they can identify the information sensitivity of power grid regulation data to a certain extent, due to the strong dependence of traditional machine learning algorithms on feature engineering, limited generalization ability, large consumption of computing resources, inflexible model updates, and insufficient ability to process unstructured features of power grid regulation data, the accuracy of sensitivity identification of power grid regulation data with a large proportion of unstructured features is not high, and there is a great risk of failing to identify sensitive data of power grid regulation. Summary of the Invention

[0005] The technical problem to be solved by the present invention is: how to increase the recognition accuracy of sensitive data for power grid regulation during the transmission of power grid regulation data.

[0006] The technical solution proposed in the present invention is: a method for identifying sensitive data of power grid regulation, comprising the following steps:

[0007] Step 1: Collect a grid control data that needs to be sensitively identified, the grid control data includes x structured features and y unstructured features; x and y are both natural numbers; vectorize the x structured features and y unstructured features of the grid control data, and collect the x structured feature vectors after vectorization to form a structured feature vector set V, V = {v1, v2, ..., vx}, v1 is the first structured feature vector after vectorization among the x structured features of the grid control data; v2 is the second structured feature vector after vectorization among the x structured features of the grid control data; v x is the xth vectorized structured feature vector among the x structured features of the power grid control data; the vectorized y unstructured feature vectors are collected to form an unstructured feature vector set S, S = {s1, s2, ..., s y}, s1 is the first vectorized unstructured feature vector of the y unstructured features of the power grid control data; s2 is the second vectorized unstructured feature vector of the y unstructured features of the power grid control data; s y is the yth unstructured feature vector after vectorization among the y unstructured features of the power grid control data;

[0008] Step 2: Substitute the x structured vectors in the structured feature vector set V into the following formula (1) to calculate the x structured absolute sensitivities of the power grid control data,

[0009]

[0010] In formula (1), i is a natural number greater than or equal to 1 and less than or equal to x; is the structured absolute sensitivity calculated for the i-th structured feature vector; P is the set of structured sensitive vectors; p1 is the first structured sensitive vector; p2 is the second structured sensitive vector; is the n1th structured sensitive vector;

[0011] Collect the calculated x structured absolute sensitivities to form the structured absolute sensitivity set M v , It is the first structured absolute sensitivity obtained by calculating the sensitivity of the first structured eigenvector; It is the second structured absolute sensitivity obtained by calculating the sensitivity of the second structured eigenvector; is the xth structured absolute sensitivity obtained by sensitivity calculation of the xth structured eigenvector;

[0012] Substitute the y unstructured vectors in the unstructured feature vector set S into the following formula (2) to calculate the y unstructured absolute sensitivities of the power grid control data,

[0013]

[0014] In formula (2), j is a natural number greater than or equal to 1 and less than or equal to y; is the unstructured absolute sensitivity calculated from the jth unstructured eigenvector; C is the set of unstructured sensitive vectors; c1 is the first unstructured sensitive vector; c2 is the second unstructured sensitive vector; is the n2th unstructured sensitive vector;

[0015] Collect the calculated y unstructured absolute sensitivities to form the unstructured absolute sensitivity set M s , It is the first unstructured absolute sensitivity obtained by calculating the sensitivity of the first unstructured eigenvector; It is the second unstructured absolute sensitivity obtained by calculating the sensitivity of the second unstructured eigenvector; is the yth unstructured absolute sensitivity obtained by sensitivity calculation of the yth unstructured eigenvector;

[0016] Step 3: Substitute the x structured vectors in the structured feature vector set V into the following formula (3) to calculate the x structured relative sensitivities of the power grid control data,

[0017]

[0018] In formula (1), is the structured absolute sensitivity calculated from the i-th structured feature vector; P is the set of structured sensitive vectors; o n is the prototype vector of non-sensitive data; o u is the prototype vector of sensitive data;

[0019] The prototype vector o of the non-sensitive data in formula (3) is n and prototype vector o of sensitive data u The learning vector quantization (LVQ) method is used to obtain the value. The process is as follows:

[0020] Step 3.1: Construct a manually labeled dataset D = {(w1,y1),(w2,y2),...,(w n ,y m )}, where y i=0 represents v i For non-sensitive data, y i =1 represents w i For sensitive data; initialize the prototype vector o n and o u is a vector of random numbers.

[0021] Step 3.2: Calculate the prototype vector o of non-sensitive data according to the following formula (4): n and prototype vector o of sensitive data u To the i-th data w in the dataset D i Euclidean distance

[0022]

[0023] Step 3.3: Loop and calculate the i-th data w in the dataset D i The corresponding L n,i , L u,i The size of the prototype vector o of non-sensitive data is updated n and prototype vector o of sensitive data u ; If the i-th data w in the data set D i The annotation type y i =0 and L n,i <L u,i Or the i-th data w in the dataset D i The annotation type y i =1 and L n,i >L u,i , then update the prototype vector o of non-sensitive data according to the following formula (5): n and prototype vector o of sensitive data u

[0024]

[0025] In formula (5), o n ' is (); o n * is();o u ' is (); o u * is();ηis();

[0026] Otherwise, update the prototype vector o of non-sensitive data according to the following formula (6): n and prototype vector o of sensitive data u

[0027]

[0028] If in the previous update round || w i- o n *If the average value of ||2 is below 0.05 or the maximum update round n is reached, stop updating o n 、o u ; The prototype vector o of the non-sensitive data at this time n and prototype vector o of sensitive data u Substitute into formula (3);

[0029] Collect the calculated x structured relative sensitivities to form the structured relative sensitivity set Q v , It is the first structured relative sensitivity obtained by calculating the sensitivity of the first structured eigenvector; It is the second structured relative sensitivity obtained by calculating the sensitivity of the second structured eigenvector; is the xth structured relative sensitivity obtained by sensitivity calculation of the xth structured eigenvector;

[0030] Step 4: Substitute the y unstructured vectors in the unstructured feature vector set S into the Transformer model in sequence to calculate the y unstructured relative sensitivities of the power grid control data, as shown in the following formula (7):

[0031]

[0032] In formula (7), is the jth unstructured eigenvector s j The calculated unstructured relative sensitivity; LayerNorm() is layer normalization, Linear() is the fully connected layer, FeedForward() is the feedforward layer, CrossAttention() is the cross attention module, MaskedAttention() is the masked attention module, MultiHeadAttention() is the multi-head attention module, T j is the output of the Transformer model.

[0033] Collect the calculated y unstructured absolute sensitivities to form the unstructured relative sensitivity set Q s , It is the first unstructured relative sensitivity obtained by calculating the sensitivity of the first unstructured eigenvector; It is the second unstructured relative sensitivity obtained by calculating the sensitivity of the second unstructured eigenvector; is the yth unstructured relative sensitivity obtained by sensitivity calculation of the yth unstructured eigenvector;

[0034] Step 5: Collect the x structured absolute sensitivities and relative sensitivities and y unstructured absolute sensitivities and relative sensitivities of the power grid control data obtained in steps 2, 3, and 4, and select the maximum value as the total sensitivity value of the power grid control data. If the total sensitivity value is greater than the empirical threshold Y, the power grid control data is considered to be sensitive data; if the total sensitivity value is not greater than the empirical threshold Y, the power grid control data is considered to be normal data.

[0035] Furthermore, the specific process of calculating the unstructured relative sensitivity of the unstructured vector by the Transformer model in step 4 is as follows:

[0036] Substitute the jth unstructured vector in the unstructured feature vector set S into the encoder of the Transformer model, calculate it through the multi-head attention module, and then normalize the calculation result of the multi-head attention module to obtain the output E of the encoder j ; The encoder output E j In the input decoder, the output θ of the encoder at the previous moment is processed by the mask attention module to obtain the mask result R j ; The encoder output E j And the mask result R j The calculation is performed together through the cross attention module in the decoder, and then the calculation results are processed through the feedforward layer and the fully connected layer to obtain the final decoder output result T j Finally, the decoder output is calculated according to the first term in formula (7) to obtain the unstructured relative sensitivity of the j-th unstructured vector input to the Transformer model.

[0037] Beneficial effects: The present invention provides a method for identifying sensitive data of power grid regulation, which uses a Transformer model to extract, analyze and process the unstructured features of power grid regulation data. At the same time, the present invention also calculates the absolute sensitivity and relative sensitivity of the unstructured features as a basis for whether the unstructured features of a power grid regulation data are sensitive, greatly improving the processing of unstructured features in power grid regulation data, and solving the problem of low accuracy in sensitivity identification of power grid regulation data with a large proportion of unstructured features in the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of the method for identifying sensitive data of power grid regulation according to the present invention. DETAILED DESCRIPTION

[0039] The following further describes a method for identifying sensitive data of power grid regulation according to the present invention in conjunction with the accompanying drawings and specific embodiments.

[0040] The present invention provides a method for identifying sensitive data of power grid regulation, such as Figure 1 As shown, the following steps are included:

[0041] Step 1: Collect a grid control data that needs to be sensitively identified. The grid control data includes x structured features and y unstructured features; x and y are natural numbers; vectorize the x structured features and y unstructured features of the grid control data, and collect the vectorized x structured feature vectors to form a structured feature vector set V, where V = {v1, v2, ..., v x}, v1 is the first vectorized structured feature vector of the x structured features of the power grid control data; v2 is the second vectorized structured feature vector of the x structured features of the power grid control data; v x is the xth vectorized structured feature vector among the x structured features of the power grid control data; the y vectorized unstructured feature vectors are collected to form an unstructured feature vector set S, S = {s1, s2, ..., s y}, s1 is the first vectorized unstructured feature vector of the y unstructured features of the power grid control data; s2 is the second vectorized unstructured feature vector of the y unstructured features of the power grid control data; y It is the yth unstructured feature vector after vectorization among the y unstructured features of the power grid control data;

[0042] Step 2: Substitute the x structured vectors in the structured feature vector set V into the following formula (1) to calculate the x structured absolute sensitivities of the power grid control data,

[0043]

[0044] In formula (1), i is a natural number greater than or equal to 1 and less than or equal to x; is the structured absolute sensitivity calculated for the i-th structured feature vector; P is the set of structured sensitive vectors; p1 is the first structured sensitive vector; p2 is the second structured sensitive vector; is the n1th structured sensitive vector;

[0045] Collect the calculated x structured absolute sensitivities to form the structured absolute sensitivity set M v , It is the first structured absolute sensitivity obtained by calculating the sensitivity of the first structured eigenvector; It is the second structured absolute sensitivity obtained by calculating the sensitivity of the second structured eigenvector; is the xth structured absolute sensitivity obtained by sensitivity calculation of the xth structured eigenvector;

[0046] Substitute the y unstructured vectors in the unstructured feature vector set S into the following formula (2) to calculate the y unstructured absolute sensitivities of the power grid control data,

[0047]

[0048] In formula (2), j is a natural number greater than or equal to 1 and less than or equal to y; is the unstructured absolute sensitivity calculated from the jth unstructured eigenvector; C is the set of unstructured sensitive vectors; c1 is the first unstructured sensitive vector; c2 is the second unstructured sensitive vector; is the n2th unstructured sensitive vector;

[0049] Collect the calculated y unstructured absolute sensitivities to form the unstructured absolute sensitivity set M s , It is the first unstructured absolute sensitivity obtained by calculating the sensitivity of the first unstructured eigenvector; It is the second unstructured absolute sensitivity obtained by calculating the sensitivity of the second unstructured eigenvector; is the yth unstructured absolute sensitivity obtained by sensitivity calculation of the yth unstructured eigenvector;

[0050] Step 3: Substitute the x structured vectors in the structured feature vector set V into the following formula (3) to calculate the x structured relative sensitivities of the power grid control data,

[0051]

[0052] In formula (1), is the structured absolute sensitivity calculated from the i-th structured feature vector; P is the set of structured sensitive vectors; o n is the prototype vector of non-sensitive data; o u is the prototype vector of sensitive data;

[0053] The prototype vector o of non-sensitive data in formula (3) n and prototype vector o of sensitive data u The learning vector quantization (LVQ) method is used to obtain the value. The process is as follows:

[0054] Step 3.1: Construct a manually labeled dataset D = {(w1,y1),(w2,y2),...,(w n ,y m )}, where y i =0 represents v i For non-sensitive data, y i =1 represents w i For sensitive data; initialize the prototype vector o n and o u is a vector of random numbers.

[0055] Step 3.2: Calculate the prototype vector o of non-sensitive data according to the following formula (4): n and prototype vector o of sensitive data u To the i-th data w in the dataset D i Euclidean distance

[0056]

[0057] Step 3.3: Loop and calculate the i-th data w in the dataset D i The corresponding L n,i , L u,i The size of the prototype vector o of non-sensitive data is updated n and prototype vector o of sensitive data u ; If the i-th data w in the data set D i The annotation type y u =0 and L n,i <L u,i Or the i-th data w in the dataset D i The annotation type y i =1 and L n,i >L u,i , then update the prototype vector o of non-sensitive data according to the following formula (5): n and prototype vector o of sensitive data u

[0058]

[0059] In formula (5), o n ' is (); o n * is();o u ' is (); o u * is();ηis();

[0060] Otherwise, update the prototype vector o of non-sensitive data according to the following formula (6): n and prototype vector o of sensitive data u

[0061]

[0062] If in the previous update round || w i- o n * If the average value of ||2 is below 0.05 or the maximum update round n is reached, stop updating o n 、o u ; The prototype vector o of the non-sensitive data at this time n and prototype vector o of sensitive data u Substitute into formula (3);

[0063] Collect the calculated x structured relative sensitivities to form the structured relative sensitivity set Q v , It is the first structured relative sensitivity obtained by calculating the sensitivity of the first structured eigenvector; It is the second structured relative sensitivity obtained by calculating the sensitivity of the second structured eigenvector; is the xth structured relative sensitivity obtained by sensitivity calculation of the xth structured eigenvector;

[0064] Step 4: Substitute the y unstructured vectors in the unstructured feature vector set S into the Transformer model in turn to calculate the y unstructured relative sensitivities of the power grid control data, as shown in the following formula (7):

[0065]

[0066] In formula (7), is the jth unstructured eigenvector s j The calculated unstructured relative sensitivity; LayerNorm() is layer normalization, Linear() is the fully connected layer, FeedForward() is the feedforward layer, CrossAttention() is the cross attention module, MaskedAttention() is the masked attention module, MultiHeadAttention() is the multi-head attention module, T j is the output of the Transformer model;

[0067] The specific process of calculating the unstructured relative sensitivity of the unstructured vector by the Transformer model is as follows:

[0068] Substitute the jth unstructured vector in the unstructured feature vector set S into the encoder of the Transformer model, calculate it through the multi-head attention module, and then normalize the calculation result of the multi-head attention module to obtain the output E of the encoder j ; The encoder output E j In the input decoder, the output θ of the encoder at the previous moment is processed by the mask attention module to obtain the mask result R j ; The encoder output E j And the mask result R j The calculation is performed together through the cross attention module in the decoder, and then the calculation results are processed through the feedforward layer and the fully connected layer to obtain the final decoder output result T j Finally, the decoder output is calculated according to the first term in formula (7) to obtain the unstructured relative sensitivity of the j-th unstructured vector input to the Transformer model.

[0069] Collect the calculated y unstructured absolute sensitivities to form the unstructured relative sensitivity set Q s , It is the first unstructured relative sensitivity obtained by calculating the sensitivity of the first unstructured eigenvector; It is the second unstructured relative sensitivity obtained by calculating the sensitivity of the second unstructured eigenvector; is the yth unstructured relative sensitivity obtained by sensitivity calculation of the yth unstructured eigenvector;

[0070] Step 5: Collect the x structured absolute sensitivities and relative sensitivities and y unstructured absolute sensitivities and relative sensitivities of the power grid control data obtained in steps 2, 3, and 4, and select the maximum value as the total sensitivity value of the power grid control data. If the total sensitivity value is greater than the empirical threshold Y, the power grid control data is considered to be sensitive data; if the total sensitivity value is not greater than the empirical threshold Y, the power grid control data is considered to be normal data.

[0071] The following is a sensitive identification of a specific power grid control data. Some power grid control data types are shown in Table 1 below:

[0072] Data Type Amount of structured data Amount of unstructured data Voltage data 12 1 Current data 12 1 Single point load 25 5 … … … Single point control instructions 40 6 Intranet communication 4 1

[0073] Table 1

[0074] Vectorize the structured and unstructured data in the above data types to obtain structured data feature vectors and unstructured data feature vectors. Calculate the absolute sensitivity and relative sensitivity of the structured and unstructured feature vectors, respectively. Calculate the sensitivity of each piece of data.

[0075] Based on 10,000 pieces of power grid control data, 50 of which are sensitive due to unstructured features, and 25 of which are sensitive due to structured features, the recognition results of the existing recognition method are compared with the recognition results of the method of the present invention, as shown in Table 2 below:

[0076]

[0077] Table 2

[0078] It can be clearly seen from Table 2 that the recognition accuracy of the present invention for sensitive data on power grid regulation caused by unstructured features is 96%, which is significantly higher than the accuracy of existing machine learning of 54%, reflecting the processing ability of the method in the present invention for the unstructured features of power grid regulation data.

Claims

1. A method for identifying sensitive data for power grid regulation, characterized by: The following steps are involved: Step 1: Collect a grid control data that needs to be sensitively identified, the grid control data includes x structured features and y unstructured features; x and y are natural numbers; vectorize the x structured features and y unstructured features of the grid control data, and collect the vectorized x structured feature vectors to form a structured feature vector set V, V = {v1, v2, ..., v x }, v1 is the first vectorized structured feature vector of the x structured features of the power grid control data; v2 is the second vectorized structured feature vector of the x structured features of the power grid control data; v x is the xth vectorized structured feature vector among the x structured features of the power grid control data; the vectorized y unstructured feature vectors are collected to form an unstructured feature vector set S, S = {s1, s2, ..., s y }, s1 is the first vectorized unstructured feature vector of the y unstructured features of the power grid control data; s2 is the second vectorized unstructured feature vector of the y unstructured features of the power grid control data; s y is the yth unstructured feature vector after vectorization among the y unstructured features of the power grid control data; Step 2: Substitute the x structured vectors in the structured feature vector set V into the following formula (1) to calculate the x structured absolute sensitivities of the power grid control data, In formula (1), i is a natural number greater than or equal to 1 and less than or equal to x; is the structured absolute sensitivity calculated for the i-th structured eigenvector; P is a set of structured sensitive vectors; p1 is the first structured sensitive vector; p2 is the second structured sensitive vector; is the n1th structured sensitive vector; Collect the calculated x structured absolute sensitivities to form the structured absolute sensitivity set M v , It is the first structured absolute sensitivity obtained by calculating the sensitivity of the first structured eigenvector; It is the second structured absolute sensitivity obtained by calculating the sensitivity of the second structured eigenvector; is the xth structured absolute sensitivity obtained by sensitivity calculation of the xth structured eigenvector; Substitute the y unstructured vectors in the unstructured feature vector set S into the following formula (2) to calculate the y unstructured absolute sensitivities of the power grid control data, In formula (2), j is a natural number greater than or equal to 1 and less than or equal to y; is the unstructured absolute sensitivity calculated from the jth unstructured eigenvector; C is the set of unstructured sensitive vectors; c1 is the first unstructured sensitive vector; c2 is the second unstructured sensitive vector; is the n2th unstructured sensitive vector; Collect the calculated y unstructured absolute sensitivities to form the unstructured absolute sensitivity set M s , It is the first unstructured absolute sensitivity obtained by calculating the sensitivity of the first unstructured eigenvector; It is the second unstructured absolute sensitivity obtained by calculating the sensitivity of the second unstructured eigenvector; is the yth unstructured absolute sensitivity obtained by sensitivity calculation of the yth unstructured eigenvector; Step 3: Substitute the x structured vectors in the structured feature vector set V into the following formula (3) to calculate the x structured relative sensitivities of the power grid control data, In formula (1), is the structured absolute sensitivity calculated for the i-th structured eigenvector; P is a set of structured sensitive vectors; n is the prototype vector of non-sensitive data; o u is the prototype vector of sensitive data; The prototype vector o of the non-sensitive data in formula (3) is n and prototype vector o of sensitive data u The learning vector quantization (LVQ) method is used to obtain the value. The process is as follows: Step 3.1: Construct a manually labeled dataset D = {(w1,y1),(w2,y2),...,(w n ,y m )}, where y i =0 represents v i For non-sensitive data, y i =1 represents w i For sensitive data; initialize the prototype vector o n and o u is a random number vector; Step 3.2: Calculate the prototype vector o of non-sensitive data according to the following formula (4): n and prototype vector o of sensitive data u To the i-th data w in the dataset D i Euclidean distance Step 3.3: Loop and calculate the i-th data w in the dataset D i The corresponding L n,i , L u,i The size of the prototype vector o of non-sensitive data is updated n and prototype vector o of sensitive data u ; If the i-th data w in the data set D i The annotation type y i =0 and L n,i <L u,i Or the i-th data w in the dataset D i The annotation type y i =1 and L n,i >L u,i , then update the prototype vector o of non-sensitive data according to the following formula (5): n and prototype vector o of sensitive data u In formula (5), o n ' is (); o n * is();o u ' is (); o u * yes(); η is (); Otherwise, update the prototype vector o of non-sensitive data according to the following formula (6): n and prototype vector o of sensitive data u If in the previous update round || w i- o n * If the average value of ||2 is below 0.05 or the maximum update round n is reached, stop updating o n 、o u ; The prototype vector o of the non-sensitive data at this time n and prototype vector o of sensitive data u Substitute into formula (3); Collect the calculated x structured relative sensitivities to form the structured relative sensitivity set Q v , It is the first structured relative sensitivity obtained by calculating the sensitivity of the first structured eigenvector; It is the second structured relative sensitivity obtained by calculating the sensitivity of the second structured eigenvector; is the xth structured relative sensitivity obtained by sensitivity calculation of the xth structured eigenvector; Step 4: Substitute the y unstructured vectors in the unstructured feature vector set S into the Transformer model in sequence to calculate the y unstructured relative sensitivities of the power grid control data, as shown in the following formula (7): In formula (7), is the jth unstructured eigenvector s j The calculated unstructured relative sensitivity; LayerNorm() is layer normalization, Linear() is the fully connected layer, FeedForward() is the feedforward layer, CrossAttention() is the cross attention module, MaskedAttention() is the masked attention module, MultiHeadAttention() is the multi-head attention module, T j is the output of the Transformer model; Collect the calculated y unstructured absolute sensitivities to form the unstructured relative sensitivity set Q s , It is the first unstructured relative sensitivity obtained by calculating the sensitivity of the first unstructured eigenvector; It is the second unstructured relative sensitivity obtained by calculating the sensitivity of the second unstructured eigenvector; is the yth unstructured relative sensitivity obtained by sensitivity calculation of the yth unstructured eigenvector; Step 5: Collect the x structured absolute sensitivities and relative sensitivities and y unstructured absolute sensitivities and relative sensitivities of the power grid control data obtained in steps 2, 3, and 4, and select the maximum value as the total sensitivity value of the power grid control data. If the total sensitivity value is greater than the empirical threshold Y, the power grid control data is considered to be sensitive data; if the total sensitivity value is not greater than the empirical threshold Y, the power grid control data is considered to be normal data.

2. The method for identifying sensitive data for power grid regulation according to claim 1, characterized in that: The specific process of calculating the unstructured relative sensitivity of the unstructured vector by the Transformer model in step 4 is as follows: Substitute the jth unstructured vector in the unstructured feature vector set S into the encoder of the Transformer model, calculate it through the multi-head attention module, and then normalize the calculation result of the multi-head attention module to obtain the output E of the encoder j ; The encoder output E j In the input decoder, the output θ of the encoder at the previous moment is processed by the mask attention module to obtain the mask result R j ; The encoder output E j And the mask result R j The calculation is performed together through the cross attention module in the decoder, and then the calculation results are processed through the feedforward layer and the fully connected layer to obtain the final decoder output result T j Finally, the decoder output is calculated according to the first term in formula (7) to obtain the unstructured relative sensitivity of the j-th unstructured vector input to the Transformer model.

Images