Controllable load and distributed power access network privacy data review method

Abstract

The application discloses a controllable load and distributed power access network privacy data examination method, determines sensitive words by the frequency of relevant word items in the privacy data, improves the accuracy of the sensitive word demarcation, and reduces the omission and loss caused by the artificial determination of the sensitive words. By calculating the coordinate vectors between the sensitive words and the context word items, the association degree between the specific content of the privacy data information center and the sensitive words is determined to determine whether the sensitive words and sensitive information are involved in the privacy data information. By establishing the association between the content and the sensitive words, the problems existing in the examination of only the sensitive words are avoided, the processed sensitive words in the privacy information can be accurately identified, the privacy information can be better examined, and the examination loopholes are avoided. Meanwhile, the relationship between the sensitive word feature vector and the sensitive word coordinate vector is established by using the neural network, and the accuracy and efficiency of the controllable load and distributed power access network privacy data examination are also improved.

Description

Technical Field

[0001] This invention relates to the field of privacy data security review, and in particular to a privacy data review method applied to controllable load and distributed power access networks. Background Technology

[0002] The integration of controllable loads and distributed power sources is an inevitable trend in the development of distribution networks. Distribution network automation sends control commands to various field terminals via communication, while simultaneously transmitting data such as the operating status of field equipment to the control center, thereby enabling flexible scheduling of energy resources. This process relies on the power communication network, through which numerous control commands monitor physical equipment, forming a mutually dependent "dual composite system"—a power-information-physical fusion system—comprising the power system and the information system. This deep integration of energy and information flows in the distribution network, while bringing more flexible and diverse control methods, also introduces information security risks. The lack of remote controllability of distributed energy resources by the power supplier and the absence of information security protection measures on the user side make controllable loads and distributed power sources susceptible to information leakage or attacks by harmful information. This could lead to malicious control of the system by attackers, resulting in large-scale simultaneous commissioning and decommissioning or frequent commissioning and decommissioning, thus threatening the safe and stable operation of the distribution network. The information security issues of controllable load and distributed power access networks differ from those of traditional information technology systems. From a system characteristic perspective, controllable load and distributed power access networks belong to cyber-physical systems. Their security assurance includes both traditional functional safety and the security of output and input information. These two levels of security have different focuses but are intertwined. In other words, the information security of controllable load and distributed power access networks must consider both traditional information domain security performance and the functionality of the controllable load and distributed power access network system. Relaxing the review of private data within the system will lead to a decline in system security, resulting in internal information leakage or the entry of harmful external information, thus becoming an entry point for external attackers to intrude into the control system.

[0003] Current methods for reviewing privacy data in controlled load and distributed power access networks generally have the problem of missing detections. In particular, sensitive words that appear in the output and input information of controlled load and distributed power access networks through means such as veiled or substituted words but not within the defined scope of sensitive words are often not effectively reviewed, resulting in the leakage of privacy data or the input of harmful information. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for reviewing privacy data in controllable load and distributed power supply access networks, characterized by comprising the following steps:

[0005] S1 selects information that has been marked as private data during the operation of controllable loads and distributed power sources connected to the network as training data;

[0006] S2 removes stop words and meaningless words from the privacy data, then segments the privacy data into terms, performs coordinate processing on the individual terms, and counts the N most frequent coordinates, and uses the terms corresponding to them as sensitive words for judging privacy data.

[0007] S3 obtains the coordinates of the two words preceding and following the sensitive word in the training data, calculates the distance between them and the coordinates of the sensitive word, and uses the four distance values ​​as feature values ​​to establish the feature vector of the sensitive word. If the two words preceding and following the sensitive word contain blank words, the distance between the coordinates of the sensitive word and the zero coordinate is used instead.

[0008] S4 uses the sensitive word feature vector as input and the sensitive word coordinate vector as output to train the privacy data review neural network.

[0009] S5 utilizes a trained privacy data review neural network to review privacy data in controlled load and distributed power access networks. It acquires information generated during the operation of these networks, removes stop words and meaningless words, segments the information into terms, and then coordinates each term. It obtains the distances between each term and the coordinates of its two preceding and two following terms, using these four distances as feature values ​​to create a feature vector for the term to be reviewed. If the preceding and following terms contain blank terms, the distance between the term's coordinates and the zero coordinate is used instead. The feature vector of the term to be reviewed is then input into the trained privacy data review neural network, and the resulting output coordinate vector is determined. If the distance between the output coordinate vector and the sensitive word's coordinate vector is less than a set value, the information is considered privacy information.

[0010] Furthermore, in S3, the distance between the coordinate values ​​of the two terms preceding and following the sensitive word is calculated using the Minkowski distance, where the p value is 7.

[0011] Furthermore, the privacy data review neural network in S4 is a sparse autoencoder network, which includes an input layer, a hidden layer, and an output layer. The number of neurons in the output layer is the same as the number of sensitive words. The hidden layer has three layers, each with a learning rate of 0.1 and an iteration count of 600. The weight decay parameter of the autoencoder is 3 × 10⁻⁶. -5The sparsity parameter is set to 0.1, the weight of the sparsity penalty term is 3, and the weight matrix and bias term of the autoencoder are randomly generated.

[0012] The privacy data review method for controllable load and distributed power access networks provided by this invention determines sensitive words by analyzing the frequency of relevant terms in the privacy data, improving the accuracy of sensitive word identification and reducing omissions and deficiencies caused by manual determination. By calculating the coordinate vectors between sensitive words and context terms, the correlation between the specific content of the privacy data information center and the sensitive words is determined, thereby identifying whether the privacy data information involves sensitive words and sensitive information. By establishing the association between content and sensitive words, the problems existing in reviewing only sensitive words are avoided. Processed sensitive words in privacy information can be accurately identified, thus improving the review of privacy information and preventing review loopholes. Simultaneously, using neural networks to establish the relationship between sensitive word feature vectors and sensitive word coordinate vectors also improves the accuracy and efficiency of privacy data review for controllable load and distributed power access networks. Attached Figure Description

[0013] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0014] Figure 1 This is a flowchart of the method for reviewing privacy data in a controllable load and distributed power supply access network according to the present invention. Detailed Implementation

[0015] To enhance understanding of the present invention, the invention will be further described in detail below with reference to embodiments and accompanying drawings.

[0016] like Figure 1 As shown, the circuit breaker condition diagnosis method in this embodiment includes the following steps:

[0017] S1 selects information that has been marked as private data during the operation of controllable loads and distributed power sources connected to the network as training data;

[0018] S2 removes stop words and meaningless words from the privacy data, then segments the privacy data into terms, performs coordinate processing on the individual terms, and counts the N most frequent coordinates, and uses the terms corresponding to them as sensitive words for judging privacy data.

[0019] S3 obtains the coordinates of the two words preceding and following the sensitive word in the training data, calculates the distance between them and the coordinates of the sensitive word, and uses the four distance values ​​as feature values ​​to establish the feature vector of the sensitive word. If the two words preceding and following the sensitive word contain blank words, the distance between the coordinates of the sensitive word and the zero coordinate is used instead.

[0020] S4 uses the sensitive word feature vector as input and the sensitive word coordinate vector as output to train the privacy data review neural network.

[0021] S5 utilizes a trained privacy data review neural network to review privacy data in controlled load and distributed power access networks. It acquires information generated during the operation of these networks, removes stop words and meaningless words, segments the information into terms, and then coordinates each term. It obtains the distances between each term and the coordinates of its two preceding and two following terms, using these four distances as feature values ​​to create a feature vector for the term to be reviewed. If the preceding and following terms contain blank terms, the distance between the term's coordinates and the zero coordinate is used instead. The feature vector of the term to be reviewed is then input into the trained privacy data review neural network, and the resulting output coordinate vector is determined. If the distance between the output coordinate vector and the sensitive word's coordinate vector is less than a set value, the information is considered privacy information.

[0022] In S3, the distance between the coordinates of the two terms preceding and following a sensitive word is calculated using the Minkowski distance, where the p-value is 7.

[0023] The privacy data review neural network in S4 is a sparse autoencoder network, which includes an input layer, hidden layers, and an output layer. The number of neurons in the output layer is the same as the number of sensitive words. The hidden layer has three layers, each with a learning rate of 0.1 and 600 iterations. The weight decay parameter of the autoencoder is 3 × 10⁻⁶. -5 The sparsity parameter is set to 0.1, the weight of the sparsity penalty term is 3, and the weight matrix and bias term of the autoencoder are randomly generated.

[0024] Using the privacy data review method for controllable load and distributed power access networks provided in this embodiment, privacy data review was performed on 1000 pieces of information generated during the operation of controllable load and distributed power access networks. The review accuracy rate was 100% for the 500 pieces of information marked as privacy data. Of the 500 pieces of information not marked as privacy data, 8 were incorrectly identified as privacy data, resulting in a review accuracy rate of 98.4%. This demonstrates that the privacy data review method for controllable load and distributed power access networks provided by this invention has high accuracy.

[0025] The above are merely specific embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A controllable load and distributed generation access network privacy data review method, characterized in that, The method comprises the following steps: S1 selecting information marked as privacy data in the operation process of controllable load and distributed power access network as training data; S2 deleting stop words and meaningless words in the information of privacy data, then performing word segmentation on the information of privacy data, performing coordinate processing on the obtained single word, simultaneously counting the N most frequent coordinates, and taking the corresponding word as a sensitive word for judging privacy data; S3 obtaining the coordinate values of the two words before and after the sensitive word in the training data, calculating the distance between the coordinate values and the coordinate value of the sensitive word, taking the obtained four distance values as feature values, establishing a sensitive word feature vector, and if the two words before and after the sensitive word contain blank words, replacing the distance between the coordinate of the sensitive word and zero coordinate; S4 training a privacy data review neural network by taking the sensitive word feature vector as the input value and the sensitive word coordinate vector as the output value; S5 using the trained privacy data review neural network to review the privacy data of controllable load and distributed power access network, obtaining information generated in the operation process of controllable load and distributed power access network, deleting stop words and meaningless words in the information, then performing word segmentation on the information, and performing coordinate processing on the obtained single word; obtaining the distance between each word and the coordinate values of the two words before and after it, taking the obtained four distance values as feature values, establishing a feature vector of the word to be reviewed, if the two words before and after the word to be reviewed contain blank words, replacing the distance between the coordinate of the word to be reviewed and zero coordinate; taking the feature vector of the word to be reviewed as the input value, inputting the trained privacy data review neural network, and obtaining the output coordinate vector, when the distance between the output coordinate vector and the sensitive word coordinate vector is less than a set value, it is determined that the information is privacy information.

2. The controllable load and distributed generation access network privacy data review method of claim 1, wherein, In S3, the distance between the coordinate values of the two words before and after the sensitive word is calculated, and the distance used is the Minkowski distance, and the p value in the Minkowski distance is 7.

3. The controllable load and distributed generation access network privacy data review method of claim 1, wherein, The privacy data review neural network in the S4 is a sparse auto-encoder network, which comprises an input layer, a hidden layer and an output layer, the number of neurons of the output layer is the same as the number of sensitive words, the number of layers of the hidden layer is 3, the learning rate of each layer of auto-encoder is 0.1, the number of iterations is 600 times, the weight decay parameter of the auto-encoder is 3*10 -5 , the sparsity parameter is 0.1, the weight of the sparsity penalty term is 3, and the auto-encoder weight matrix and the bias term are randomly generated.

Images