A power information security transmission method using data to represent desensitization

Abstract

The application discloses a kind of power information security transmission methods for desensitization using data representation, comprising: the non-sensitive input data is inputted into variational autoencoder for conversion, to obtain mean and variance data, construct analog Gaussian distribution, randomly sample from Gaussian distribution to obtain representation feature, divide the representation feature into zx part and za part, use the representation feature to fit the transmission information, use za part data to fit sensitive input data, train through generative adversarial network, to improve the independence of zx part and za part, input the output of zx part after training into regression network, and then obtain individual rating, and judge whether it meets the preset requirements by efficiency review. The power information security transmission method provided by the application uses the data representation method in deep learning to process the information data features to be transmitted, so that the attacker cannot reverse the key sensitive information in the original data through the processed data.

Description

Technical Field

[0001] This invention relates to the field of information security technology, and in particular to a method for secure transmission of power information using data characterization and desensitization. Background Technology

[0002] With the rapid development of information technology, the connection between people and computer technology is becoming increasingly close. However, during the operation of computer networks, there are security risks of information network attacks. Once user information security is not guaranteed, they will face huge risks to their privacy and property. In the power distribution network system, once user information is stolen by attackers, they can obtain sensitive information such as users' electricity needs, personal information, and home addresses, causing potential harm. Furthermore, by intercepting feedback information from users to the system, attackers can send incorrect information about user electricity usage to the power distribution provider, causing data contradictions and even power distribution chaos. Improving network information security can ensure the normal operation of an organization or institution and avoid problems such as the leakage of important internet information. Information attacks specifically refer to the use of data interception and related illegal software to obtain important information in transmitted data during the interaction of information in different network environments, resulting in the theft of core information of enterprises and institutions. The core of data transmission security is to encrypt data during sending, transmission, and receiving, making it impossible for attackers to easily obtain the content of the information, thus achieving the goal of data security.

[0003] Currently, the most common method to improve information security is data encryption. Even if information is obtained by malicious attackers, they still cannot effectively decipher the inherent meaning of the data, protecting the data of institutions or organizations from further infringement. However, attackers can indirectly crack encrypted information through differential analysis, data mining, and other methods. Sensitive information of a system is the most in need of protection; if an attacker fully obtains and understands the sensitive data of an institution or organization, it will cause even greater losses. Even if non-sensitive information is stolen by attackers, they can use technical means to deduce the sensitive information of the data owner from other data, posing a significant security risk. Furthermore, encrypting all transmitted data requires a large amount of processing time, thus increasing the overall data transmission time and reducing overall efficiency, failing to meet the efficiency requirements of power data transmission. Moreover, some data in power information is non-sensitive, and the security requirements for sensitive and non-sensitive data differ. Therefore, there is an urgent need to design a method that ensures the security of data transmission while fully meeting the requirements of high efficiency and near real-time operation in power transmission.

[0004] The above background information is provided only to assist in understanding the inventive concept and technical solution of this invention. It does not necessarily belong to the prior art of this patent application, nor does it necessarily provide technical teaching. In the absence of clear evidence that the above information was disclosed before the filing date of this patent application, the above background information should not be used to evaluate the novelty and inventiveness of this application. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides a method for secure transmission of power information using data characterization and desensitization. The specific technical solution is as follows:

[0006] On the one hand, a method for secure transmission of power information using data characterization and desensitization is provided, including the following steps:

[0007] A deep learning model is built using machine learning to process the transmitted information, and the corresponding model input data (x, a, y) and model output data are provided. The definition categorizes transmitted information into non-sensitive input data and sensitive input data, where x represents non-sensitive input data and a represents sensitive input data. x and a are distinguished by information sensitivity categories predefined by user privacy, and y represents the task label to be predicted. The individual rating values ​​given by the enterprise through the deep learning model;

[0008] In the deep learning model, the insensitive input data is input into a variational autoencoder for transformation to obtain mean and variance data. A simulated Gaussian distribution is constructed using the obtained mean and variance. Random sampling is performed from the Gaussian distribution to obtain representational features, denoted as z. These representational features are divided into two parts, denoted as zx and za. The representational features are used to fit the transmitted information, and the za part is used to fit the sensitive input data. A generative adversarial network is used for training to improve the independence of the zx and za parts. The zx part output after training is used as the encrypted feature (with sensitive input removed) and input into a regression network to obtain... right Efficiency is evaluated to determine whether the preset requirements are met. If the preset requirements are not met, the deep learning model needs to be further optimized until the preset requirements are met. The optimized zx part is then transmitted as encrypted information.

[0009] Furthermore, the efficiency assessment includes an assessment of the availability of encrypted information and an assessment of data containing sensitive information. The assessment of the availability of encrypted information is performed using the following formula:

[0010]

[0011] In the formula, A tIt is the efficiency of encrypting information to restore individual t, y t These are individual ratings predicted by companies based on actual information. It is an individual rating value predicted by the enterprise using encrypted information provided by a deep learning model;

[0012] The data contains sensitive information indicators, which are examined using the following formula.

[0013]

[0014] In the formula, It is an indicator of the ability to compress data and eliminate sensitive attributes; It is the evaluation result made by the enterprise based on the encrypted information representation given by the deep learning model. var(x) is the variance statistic of x; T is the total number of regions; k is the sensitive input type; k t The distribution of the sensitivity attribute k values ​​for individual t; is the average value of x, and m is the specified amount of sensitive information.

[0015] Furthermore, the variational autoencoder includes a mean encoder and a variance encoder. Both the mean encoder and the variance encoder define two-layer neural networks, both use the ReLU function as the inter-layer activation function, and both use the sigmoid function as the tail layer activation function.

[0016] The conversion formula corresponding to the mean encoder is as follows:

[0017] u = sigmoid(W u2 ×relu(W u1 ×x+b u1 )+b u2 )

[0018] The conversion formula corresponding to the variance encoder is as follows:

[0019] v = sigmoid(W) v2 ×relu(W v1 ×x+b v1 )+b v2 )

[0020] In the formula, u and v represent the mean and variance of the data, respectively, and W *i This represents the weight of the i-th network layer in the processing layer *, where b *i This represents the offset of the i-th network layer being processed, where * is either u or v.

[0021] Furthermore, the transmitted information is fitted using the aforementioned characterization features, and the corresponding fitting formula is as follows:

[0022]

[0023] The fitting formula for using Za partial data to fit sensitive input data is as follows:

[0024]

[0025] In the formula, For the fitted transmission information, For the fitted sensitive input data, W *i This represents the weight of the i-th network layer in the processing layer *, where b *i This represents the offset of the i-th network layer being processed, where * is either x or a.

[0026] Furthermore, the aforementioned representational features are used to fit the transmitted information to a neural network that reconstructs the output without being sensitive to input information. The loss function of the neural network that reconstructs the output without being sensitive to input information is calculated as follows:

[0027]

[0028] The z-partial data is used to fit a neural network that reconstructs the sensitive input information at the output, and the loss function of the neural network for reconstructing the sensitive input information is calculated as follows:

[0029]

[0030] The loss function that makes the zx and za parts independent is calculated as follows:

[0031]

[0032] In the formula, p(x|z) is the decoding distribution, q(z|x) is the encoding distribution, p(z) is the prior representation of the features, p(a|za) is the decoding distribution of the sensitive information, and za k It is the k-th predictive sensitive input representation dimension, and the relationship between z, zx, and za is z = [zx, za].

[0033] Based on the above three loss functions, the overall loss function for encrypted representation can be derived as follows:

[0034] Loss = Loss VAE +Loss a +λ×Loss z

[0035] In the formula, λ is a parameter that controls the degree of sensitive information elimination.

[0036] Furthermore, if the availability of encrypted information meets the requirements but the data containing sensitive information does not, the value of λ needs to be increased for further learning and optimization. If the data containing sensitive information meets the requirements but the availability of encrypted information does not, the value of λ needs to be decreased for further learning and optimization. The value of λ ranges from [0,1].

[0037] Furthermore, the regression network is denoted as f, and its corresponding loss function is as follows:

[0038]

[0039] in, It is the regression threshold.

[0040] Furthermore, the transmitted information is communication data between the power supply company and the user. The non-sensitive input data includes user electricity consumption information, electricity consumption time and account number, while the sensitive input data includes user name, user ID number, user gender, user address and user contact information.

[0041] In another aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the secure transmission method for power information.

[0042] In another aspect, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the power information secure transmission method described in any one of the claims.

[0043] Compared with the prior art, the present invention has the following advantages: it uses the data representation method in deep learning to process the features of the information data to be transmitted, so that attackers cannot deduce the key sensitive information in the original data from the processed data. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the variational automatic coding data processing process in the power information security transmission method using data characterization desensitization provided in the embodiments of the present invention;

[0045] Figure 2 This is an overall training flowchart of the power information security transmission method using data representation desensitization provided in the embodiments of the present invention;

[0046] Figure 3 This is a flowchart of the sensitive information removal process in the power information security transmission method using data characterization for desensitization provided in the embodiments of the present invention. Detailed Implementation

[0047] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0048] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0049] Deep learning is a machine learning method that uses multi-layered neural networks for data fitting, prediction, and analysis. Representation learning is a data processing method within deep learning that encodes features, extracting latent information from the data by transforming its mapping space. Variational autoencoders are unsupervised data generation models that encode input data into a representation space and then reconstruct the data from that space. The encoding process not only encrypts the data but also yields more independent representational features between columns.

[0050] The purpose of this invention is to improve data encryption methods, which can eliminate sensitive elements in information while encrypting it. This makes it impossible for either the data owner or the data attacker to deduce the data owner's sensitive information from the encrypted information, thus greatly improving the security of information systems.

[0051] In one embodiment of the present invention, a method for secure transmission of power information using data characterization desensitization is provided, comprising the following steps:

[0052] Step 1: Mathematical Modeling

[0053] A deep learning model is built using machine learning to process the transmitted information, and the corresponding model input data (x, a, y) and model output data are provided. The definition categorizes transmitted information into non-sensitive input data and sensitive input data, where x represents non-sensitive input data and a represents sensitive input data. x and a are distinguished by information sensitivity categories predefined by user privacy, and y represents the task label to be predicted. The individual rating values ​​given by the enterprise through the deep learning model;

[0054] For example, the power distribution management system can be used to provide the corresponding model input data;

[0055] Non-sensitive input information x: Describes owner information and their electricity usage information. It does not include defined identity-sensitive input information, and mainly includes relevant data such as user electricity consumption information, electricity usage time, and account number;

[0056] Sensitive input information a: Sensitive information of users in the power distribution system, including user name, user ID number, user gender, user address, user contact information, user regional information (Northeast China, North China, South China, etc.), electricity consumption type (industrial electricity, commercial electricity, residential electricity, rice paddy irrigation and drainage electricity), user discount, etc.

[0057] The label y for the task to be predicted: This label is optional. It can be used to verify the usability of encrypted information. The encryption process is only meaningful if the encrypted information can be used normally. For example, in a real-world electricity consumption scenario, based on encrypted information from a user's historical electricity consumption data, the user's electricity consumption for the next month can be predicted. The predicted amount is then compared with the actual future electricity consumption to evaluate the usability of the encrypted information. Evaluation can also be done without a prediction task. The model can provide a corresponding decoder to restore the encrypted information. Comparing the matching degree between the data before and after encryption can also evaluate the usability of encryption. This embodiment uses the label prediction process as an example to evaluate the model's usability. There are two main indicators for evaluating deep learning information encryption models. First, the usability of the encrypted information. The encrypted information must be able to be restored to a form that the enterprise can understand. Through the encrypted information, the enterprise can evaluate individuals. It must match the actual y; secondly, the security of encrypted information is paramount. It cannot restore the sensitive information contained in the original data. In other words, the original information itself may not be sensitive, but it can be used to perform differential attacks to infer the sensitivity, which is undesirable. Therefore, encrypted information must ensure the confidentiality of sensitive information.

[0058] Step 2: Model Network Construction

[0059] In deep learning models, see Figure 1 and Figure 2 The insensitive input data is input into a variational autoencoder for transformation to obtain mean and variance data. The obtained mean and variance are used to construct a simulated Gaussian distribution. Random sampling is performed from the Gaussian distribution to obtain a representation feature, denoted as z. The representation feature is divided into two parts, denoted as zx and za, respectively. The representation feature is used to fit the transmitted information, and the za part data is used to fit the sensitive input data. The network is trained by a generative adversarial network to improve the independence of the zx and za parts.

[0060] Specifically, variational autoencoders and generative adversarial networks are used to desensitize and informatize the data feature representations. First, an autoencoder model is built, and formulas (3)-(5) describe the transformation process of the data through the variational autoencoder. The variational autoencoder consists of a mean encoder and a variance encoder. Both encoders define a two-layer neural network, using the ReLU function as the inter-layer activation function and the sigmoid function as the tail activation function. The predicted mean and variance are used to construct a simulated distribution, and the final representation features are randomly sampled from the distribution.

[0061] The obtained representation is divided into two parts, zx and za. The original data is fitted with [zx+za], and the sensitive input data a is fitted with za. Equations (6)-(8) describe this adversarial learning process. Finally, the independence of zx and za is improved. The final output zx is used as the encrypted feature with the sensitive input removed.

[0062] u = sigmoid(W u2 ×relu(W u1 ×x+b u1 )+b u2 (1)

[0063] v = sigmoid(W) v2 ×relu(W v1 ×x+b v1 )+b v2 (2)

[0064]

[0065] zx,za=split(z) (4)

[0066]

[0067]

[0068] In the formula, u and v represent the mean and variance of the data, respectively. For the fitted transmission information, For the fitted sensitive input data, W *i This represents the weight of the i-th network layer in the processing layer *, where b *i This represents the offset of the i-th network layer being processed, where * is u, v, x, or a. The above operations are for desensitization representation, using the variational autoencoder and adversarial network together as the representation model. Therefore, the overall representation model includes the following sub-networks: a neural network for outputting the mean of the data, a neural network for outputting the variance of the data, a neural network for restoring the output without sensitive input information, and a neural network for restoring the output with sensitive input information.

[0069] Step 3: Model Optimization

[0070] The zx part of the final output is used as an encrypted feature with sensitive inputs removed and input into the regression network to obtain the predicted power demand value. The efficiency of the obtained predicted power demand value is examined to determine whether it meets the preset requirements. If it does not meet the preset requirements, the deep learning model needs to be further optimized until the preset requirements are met.

[0071] Specifically, model optimization includes the following:

[0072] In variational autoencoders, each feature column is sampled from a Gaussian distribution generated by its own unique predicted mean and variance. By constraining the relationship between the generated Gaussian distribution and the standard Gaussian distribution, the representation features become more independent.

[0073] The aforementioned representational features are used to fit the transmitted information to a neural network that reconstructs the output without sensitive input information. A simple mean squared error loss function is used to constrain the information carried in the representation. The loss function of the neural network that reconstructs the output without sensitive input information is calculated as follows:

[0074]

[0075] The z-part data is used to fit a neural network that reconstructs the sensitive input information at the output, and the information carried by the z-part in the representation is constrained by a simple mean squared error loss function. The loss function of the neural network for reconstructing the sensitive input information at the output is calculated as follows:

[0076]

[0077] The loss function that makes the zx and za parts independent is calculated as follows:

[0078]

[0079] In the formula, p(x|z) is the decoding distribution, q(z|x) is the encoding distribution, p(z) is the prior representation of the features, p(a|za) is the decoding distribution of the sensitive information, and za k It is the k-th predictive sensitive input representation dimension, and the relationship between z, zx, and za is z = [zx, za].

[0080] Based on the above three loss functions, the overall loss function for encrypted representation can be derived as follows:

[0081] Loss = Loss VAE +Loss a +λ×Loss z

[0082] In the formula, λ is a parameter that controls the degree to which the representation eliminates input information, and the value of λ ranges from [0,1]. The larger λ is, the deeper the optimization of desensitization. If the representation improves desensitization, then the useful information it contains will be reduced. Therefore, choosing an appropriate λ as the optimization balance parameter is also very important.

[0083] Using the final obtained representation zx, an individual rating value is output through a regression network f, i.e. The loss function for the regression network f is as follows:

[0084]

[0085] in, It is the regression threshold.

[0086] The obtained electricity demand forecast is evaluated for efficiency to determine whether it meets preset requirements. The efficiency evaluation includes an evaluation of the availability of encrypted information and an evaluation of the data containing sensitive information.

[0087] The availability assessment of the encrypted information is performed using the following formula.

[0088]

[0089] In the formula, A t This represents the efficiency of encrypting information to recover individual t. The larger the value, the lower the information loss rate of the encrypted information and the more efficient the encrypted transmission. It is the individual rating value predicted by the enterprise using encrypted information provided by a deep learning model; y t It is an individual rating value predicted by the company based on actual information.

[0090] The data contains sensitive information indicators, which are examined using the following formula.

[0091]

[0092] In the formula, This is an indicator of the ability to compress data and eliminate sensitive attributes. The higher the value, the less sensitive information is in the compressed data. It is the evaluation result made by the enterprise based on the encrypted information representation given by the deep learning model (i.e. ); var(x) is the variance statistic of x; T is the total number of individuals; k is the sensitive input type; k t The distribution of the sensitivity attribute k values ​​for individual t; is the average value of x; m is the specified amount of sensitive information.

[0093] If the availability of encrypted information meets the requirements but the data containing sensitive information does not, the value of λ needs to be increased for further learning and optimization. If the data containing sensitive information meets the requirements but the availability of encrypted information does not, the value of λ needs to be decreased for further learning and optimization. Through continuous adjustment and optimization, the requirements for both the availability of encrypted information and the data containing sensitive information are met simultaneously.

[0094] In this embodiment of the invention, step one mainly involves mathematically modeling the problem; step two mainly involves building the various network frameworks of the model, where the output of the representation is achieved using a variational autoencoder, information separation of the representation is achieved by strengthening the independence between different parts, and information preservation of the representation is achieved using a decoding network; step three mainly involves iteratively optimizing different neural networks using their respective optimization objectives. In a preferred embodiment of the invention, see [link to specific embodiment]. Figure 3 After mathematical modeling, the network module is initialized, and then the representation model is trained, namely, the variational autoencoder and the adversarial network are trained. By adjusting the setting λ, the expected information shielding ability is achieved. Then, the regression model is further trained. By adjusting the setting λ and training the representation model, the regression model is made to meet the expected accuracy requirements, so as to finally output the representation model and the regression model. The input data of the model is passed through the representation model and the regression model in sequence to obtain the predicted value of electricity demand.

[0095] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned method for secure transmission of power information. The concept of this device embodiment is the same as the working process of the detection method in the above embodiments. Therefore, the entire content of the above detection method embodiments is incorporated into this device embodiment by means of full reference, and will not be repeated here.

[0096] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned method for secure transmission of power information. The concept of this storage medium embodiment is the same as the working process of the detection method in the above embodiments. Therefore, the entire content of the above detection method embodiments is incorporated into this storage medium embodiment by means of full reference, and will not be repeated here.

[0097] The power information security transmission method provided by this invention utilizes representation techniques from deep learning to desensitize and digitize transmitted data, thereby achieving secure information transmission and preventing differential attacks. This invention represents data based on variational autoencoders and generative adversarial networks, and uses deep learning models to encrypt information, enhancing the protection of sensitive information and preventing more serious information intrusions.

[0098] The power information security transmission method provided by this invention uses variational automatic encoders and generative adversarial networks to reduce the correlation between sensitive information and transmitted information, separates the information in the transmitted data, so that sensitive information in the data does not enter the transmission process, and requires that non-sensitive information in the transmission cannot be used to infer sensitive information.

[0099] The above description is merely a preferred embodiment of the present invention and does not limit its patent scope. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, whether directly or indirectly applied to other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for secure transmission of power information using data characterization and desensitization, characterized in that, Includes the following steps: A deep learning model for transmitting information is established based on machine learning, and the corresponding model input data (x, a, y) and model output data are provided. The definition categorizes transmitted information into non-sensitive input data and sensitive input data, where x represents non-sensitive input data and a represents sensitive input data. x and a are distinguished by information sensitivity categories predefined by user privacy, and y represents the task label to be predicted. The individual rating values ​​given by the enterprise through the deep learning model; In the deep learning model, the insensitive input data is input into a variational autoencoder for transformation to obtain mean and variance data. A simulated Gaussian distribution is constructed using the obtained mean and variance. Random sampling is performed from the Gaussian distribution to obtain representational features, denoted as z. These representational features are divided into two parts, denoted as zx and za. The representational features are used to fit the transmitted information, and the za part is used to fit the sensitive input data. A generative adversarial network is used for training to improve the independence of the zx and za parts. The zx part output after training is used as the encrypted feature (with sensitive input removed) and input into a regression network to obtain... right Efficiency is evaluated to determine whether the preset requirements are met. If the preset requirements are not met, the deep learning model needs to be further optimized until the preset requirements are met. The optimized zx part is then transmitted as encrypted information. The efficiency evaluation includes an evaluation of the availability of encrypted information and an evaluation of data containing sensitive information. The evaluation of the availability of encrypted information is performed using the following formula. In the formula, A t It is the efficiency of encrypting information to restore individual t, y t These are individual ratings predicted by companies based on actual information. It is an individual rating value predicted by the enterprise using encrypted information provided by a deep learning model; The data contains sensitive information indicators, which are examined using the following formula. In the formula, It is an indicator of the ability to compress data and eliminate sensitive attributes; It is the evaluation result made by the enterprise based on the encrypted information representation given by the deep learning model. var(x) is the variance statistic of x; T is the total number of individuals; k is the sensitivity attribute; k t The distribution of the sensitivity attribute k values ​​for individual t; is the average value of x, and m is the specified number of sensitive attributes.

2. The method for secure transmission of power information according to claim 1, characterized in that, The variational autoencoder includes a mean encoder and a variance encoder. Both the mean encoder and the variance encoder define a two-layer neural network, using the ReLU function as the inter-layer activation function and the sigmoid function as the tail layer activation function. The conversion formula corresponding to the mean encoder is as follows: u=sigmoid(W u2 ×relu(W u1 ×x+b u1 )+b u2 ) The conversion formula corresponding to the variance encoder is as follows: v=sigmoid(W v2 ×relu(W v1 ×x+b v1 )+b v2 ) In the formula, u and v represent the mean and variance of the data, respectively, and W *i This represents the weight of the i-th network layer in the processing layer *, where b *i This represents the offset of the i-th network layer being processed, where * is either u or v.

3. The method for secure transmission of power information according to claim 1, characterized in that, The transmitted information is fitted using the aforementioned characterization features, and the corresponding fitting formula is as follows: The fitting formula for using the Za subset of data to fit the sensitive input data is as follows: In the formula, For the fitted transmission information, For the fitted sensitive input data, W *i This represents the weight of the i-th network layer in the processing layer *, where b *i This represents the offset of the i-th network layer being processed, where * is either x or a.

4. The method for secure transmission of power information according to claim 3, characterized in that, The aforementioned representational features are used to fit the transmitted information to a neural network that reconstructs the output without being sensitive to input information. The loss function of the neural network that reconstructs the output without being sensitive to input information is calculated as follows: The z-partial data is used to fit a neural network that reconstructs the sensitive input information at the output, and the loss function of the neural network for reconstructing the sensitive input information is calculated as follows: The loss function that makes the zx and za parts independent is calculated as follows: In the formula, p(x|z) is the decoding distribution, q(z|x) is the encoding distribution, p(z) is the prior representation of the features, p(a|za) is the decoding distribution of the sensitive information, and za k It is the k-th predictive sensitive input representation dimension, and the relationship between z, zx, and za is z = [zx, za]. Based on the above three loss functions, the overall loss function for encrypted representation can be derived as follows: Loss=Loss VAE +Loss a +λ×Loss z In the formula, λ is a parameter that controls the degree of sensitive information elimination.

5. The method for secure transmission of power information according to claim 4, characterized in that, If the availability of encrypted information meets the requirements but the data containing sensitive information does not meet the requirements, the value of λ needs to be increased for further learning and optimization. If the data containing sensitive information meets the requirements but the availability of encrypted information does not meet the requirements, the value of λ needs to be decreased for further learning and optimization. The value of λ ranges from [0,1].

6. The method for secure transmission of power information according to claim 1, characterized in that, The regression network is denoted as f, and its corresponding loss function is as follows: in, It is the regression threshold.

7. The method for secure transmission of power information according to claim 1, characterized in that, The transmitted information is communication data between the power company and the user. The non-sensitive input data includes the user's electricity consumption information, electricity consumption time, and account number. The sensitive input data includes the user's name, user's ID number, user's gender, user's address, and user's contact information.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the power information security transmission method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the power information security transmission method according to any one of claims 1 to 7.

Images