Electric power big data cross-domain model construction method based on security federal learning

Abstract

The invention discloses an electric power big data cross-domain model construction method based on security federated learning, and relates to the technical field of electric power system data security and federated learning, and the method comprises the steps: implementing security initialization, generating and distributing a secret key and parameters to an edge node and a cloud server, the node obtains the secret fragments of the private key and transmits the secret fragments through a key negotiation algorithm; the node calculates a random vector and adds the random vector into transmission data, and the cloud server verifies the identity of the node and decrypts parameter data; aggregation work is completed through a cloud server, and iterative training is performed on a federated learning-based cross-domain model; according to the method, the power big data cross-domain model is constructed and security federation learning is carried out, so that the security and reliability of the power big data are improved, and the high efficiency and accuracy of the cross-domain aggregation model are ensured; and data protection is enhanced by using secret sharing, a key protocol and a random vector, so that the security and privacy of the data in the transmission process are ensured.

Description

Technical Field

[0001] The present invention relates to the field of power system data security and federated learning technology, and specifically to a method for constructing a cross-domain model of power big data based on secure federated learning. Background Art

[0002] With the rapid development of information technology, the power industry is undergoing a data revolution. The continuous improvement of intelligence and automation levels has led to the continuous generation and storage of massive amounts of power data. These data cover real-time operation data of the power grid, detailed electricity consumption statistics, equipment operation and maintenance records, etc., and contain huge value. However, these valuable data resources are often scattered among different systems, departments and platforms, forming "data islands". Their integration and utilization face huge challenges. In order to break the deadlock of data islands, cross-domain joint modeling of power big data has emerged. This technology integrates data from different fields and sources through advanced technical means to achieve data interconnection. However, cross-domain sharing and joint modeling of power big data face great data privacy and security issues. Traditional data sharing solutions, such as data encryption, access control, data anonymization, data auditing, etc., although they can protect data security to a certain extent, have not yet formed an effective security framework and cannot meet the complex and changeable cross-domain information security needs.

[0003] Federated learning is a distributed machine learning technology. Its core idea is to build a global model under virtual fusion data by conducting distributed model training among multiple data sources with local data, without exchanging local individual or sample data, only by exchanging model parameters or intermediate results. This method can achieve data sharing and joint modeling on the basis of ensuring data privacy security and legal compliance, thereby solving the "data island" problem. Based on this background, the present invention proposes an improved cross-domain model of power big data based on secure federated learning and the need for cross-domain information sharing of power big data, which is used to improve the privacy and security of power big data during cross-domain sharing. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the traditional power data transmission method and data access sharing method cannot provide effective privacy security for power data and it is difficult to transmit data efficiently and accurately when facing large and complex data sets.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for constructing a cross-domain model of power big data based on secure federated learning, including: implementing secure initialization, generating and distributing keys and parameters to edge nodes and cloud servers, in the secret sharing stage, the node obtains the secret fragments of the private key and transmits them through the key negotiation algorithm; in the random vector stage, the node calculates the random vector and adds it to the transmitted data, in the identity authentication stage, the cloud server verifies the node identity and decrypts the parameter data; in the secure aggregation stage, the aggregation work is completed through the cloud server, and the cross-domain model based on federated learning is iteratively trained.

[0007] As a preferred solution of the method for constructing a cross-domain model of power big data based on secure federated learning described in the present invention, wherein: the generation and distribution of keys and parameters to edge nodes and cloud servers includes setting the security of the integrated cross-domain model of power big data to k, the number of edge nodes in different regions to n, the sharing threshold of data information secrets to t, and the shared public value between each node and the power cloud server to a u , the secret sharing space is F, the public parameter pp is generated by KA.param(k), and the key pair generated by the edge node u is expressed as:

[0008]

[0009] in, is the key pair, is the public key in the key pair of node u, is the signature key in the key pair of node u, PK is the public key, SK is the private key, which is used for encryption and decryption of power data in the cross-domain model. After the key is generated, the public key is sent to the power cloud server through the edge node. The number of public keys collected by the power cloud server is h, and the edge nodes are counted as a group, marked as U1. The power cloud server will store the public key information of the same group of edge nodes Sent to edge nodes.

[0010] As a preferred solution of the method for constructing a cross-domain model of power big data based on secure federated learning described in the present invention, wherein: the transmission through the key negotiation algorithm includes the edge node u receiving the public key information After that, a piece of information bu is randomly selected as the seed of PRG, and two different edge nodes u and v are generated The shared fragments of information are represented as:

[0011]

[0012] Node u generates a shared fragment about bu for node v, which is expressed as:

[0013]

[0014] For the edge nodes in the same group, the key encryption algorithm is used to calculate the corresponding ciphertext, which is expressed as:

[0015]

[0016] Among them, e u,v is the ciphertext, To negotiate the key, The generated ciphertext is sent to the power cloud server for the power data to be encrypted, and the ciphertext is counted as a group, marked as U2. The power cloud server counts the ciphertext of the same group of edge nodes as Sent to edge nodes.

[0017] As a preferred solution of the method for constructing a cross-domain model of power big data based on secure federated learning described in the present invention, wherein: the calculation of the random vector includes the node u receiving from the power cloud server After that, calculate the protocol key s of the nodes in the same group u,v , expressed as:

[0018]

[0019] According to s u,v The random vector representation is calculated as:

[0020] p u,v =Δ u,v ·PRG(s u,v )

[0021] Among them, p u,v is the random vector from node u to node v, Δ u,v is the distance between node u and node v, PRG(s u,v ) is the number of power transmissions from node u to node v;

[0022] At the same time, the random vector p of node u is calculated u =PRG(b u ), and the random vector y added to the power transfer number is obtained u , expressed as:

[0023]

[0024] Among them, p u is the random vector of node u, b u is the power transmission number of node u, y u Add a random vector to node u, x u For power data;

[0025] Adding Gaussian noise based on random vectors and performing homomorphic encryption algorithm is expressed as:

[0026] z u =Enc pk (y u +noise(ψ,t'))

[0027] Among them, z u is the homomorphic encrypted ciphertext of node u, noise(ψ,t') is Gaussian noise, Enc pk For the homomorphic encryption algorithm, the power cloud server counts the encryption methods of the same group of edge nodes into one group, marked as U3;

[0028] The power cloud server aggregates all edge nodes of a group and expresses it as:

[0029] z'=z1*z2*z3*...*z n

[0030] After z' is calculated, it is sent to the edge nodes of the same group through the power cloud server.

[0031] As a preferred solution of the method for constructing a cross-domain model of power big data based on secure federated learning described in the present invention, wherein: the verification of node identity and decryption of parameter data include setting the identity certificate as a u , the verification method is expressed as:

[0032]

[0033] Among them, a' u For identity proof u Verification information

[0034] Power Cloud Server via a u and a' u Verify the identity information of the node, node u receives z' from the power cloud server and calculates the data decryption method z" = Dec sk (z’), z’ and z” are sent to the power cloud server and marked as U4. The edge nodes in the same group share the encryption and decryption methods in U4.

[0035] As a preferred solution of the method for constructing a cross-domain model of power big data based on secure federated learning described in the present invention, the aggregation work includes: node u receiving {u}u ∈U4 , and participate in learning the encrypted ciphertext in U2, expressed as:

[0036]

[0037] in, for u to s v,u The secret fragment of b v,u For u to bv Secret fragments of

[0038] The edge nodes will With b v,u Sent to the power server, the power server reorganizes the edge nodes that exited midway and calculates the private key It is expressed as:

[0039]

[0040] Based on the private key of sufficient nodes, use PRG to calculate the random vector p for node u∈U4 v,u , calculate the private key random vector b of the reorganized edge node u , expressed as:

[0041]

[0042] Based on the random vector p of edge nodes v,u and the private key random vector b u The aggregation result of the cross-domain model is obtained, which is expressed as:

[0043]

[0044] Among them, X is the aggregation result.

[0045] As a preferred solution of the method for building a cross-domain model of power big data based on secure federated learning described in the present invention, the iterative training includes building and executing a cross-domain model training structure, the power system central server sends the initial parameters of the global model to each edge node participating in the training, each edge node updates the model parameters on the local training data, and transmits the updated parameters to the power central server. After the central server receives the parameters transmitted by all nodes, it trains the model parameters and updates the global model through repeated iterations until the global model converges or the number of iterations is reached.

[0046] Another object of the present invention is to provide a cross-domain model construction system for electric power big data based on secure federated learning, which can solve the problem of high computational overhead of existing models by constructing a cross-domain model of electric power big data and performing secure federated learning.

[0047] As a preferred solution of the method for building a cross-domain model of power big data based on secure federated learning described in the present invention, it includes: an initialization module; a data protection module, and an aggregation training module.

[0048] The initialization module is used to implement security initialization, generate and distribute keys and parameters to edge nodes and cloud servers. In the secret sharing stage, the node obtains the secret fragment of the private key and transmits it through the key negotiation algorithm.

[0049] The data protection module is used for the node to calculate the random vector in the random vector phase and add it to the transmission data. In the identity authentication phase, the cloud server verifies the node identity and decrypts the parameter data.

[0050] The aggregation training module is used to complete the aggregation work through the cloud server in the security aggregation stage and perform iterative training on the cross-domain model based on federated learning.

[0051] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that the processor executes the computer program to implement a step of a method for building a cross-domain model of power big data based on secure federated learning.

[0052] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of a method for constructing a cross-domain model of power big data based on secure federated learning are implemented.

[0053] Beneficial effects of the present invention: The method for constructing a cross-domain model of power big data based on secure federated learning provided by the present invention realizes encryption and decryption of data and privacy protection by using variants of secret sharing technology and key protocol, adding random vectors to add Gaussian noise and performing homomorphic encryption, cloud server node identity authentication, multiple data desensitization, anti-tampering and access control data security measures, thereby improving the confidentiality of power big data in the cross-domain model of power big data; by using a federated learning cross-domain aggregation algorithm and constructing a cross-domain model training structure based on federated learning, not only the local power data of the edge node is protected to prevent information leakage in the model learning process, but also data isolation and privacy protection between edge nodes are realized, further effectively improving the security and privacy of the data; by iterative training of the power center server and updating of global model parameters during the cross-domain model training process, the efficiency and accuracy of the cross-domain model are ensured, and the present invention achieves better results in terms of efficiency, accuracy and security. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0055] Figure 1 An overall flow chart of a method for building a cross-domain model of power big data based on secure federated learning provided in the first embodiment of the present invention.

[0056] Figure 2The time taken for different models to transmit 100 copies of data in a method for building a cross-domain model of power big data based on secure federated learning provided in the second embodiment of the present invention.

[0057] Figure 3 The second embodiment of the present invention provides a method for building a cross-domain model of power big data based on secure federated learning, which provides the computational overhead of encrypting different model data.

[0058] Figure 4 A system overall flow chart of a method for building a cross-domain model of power big data based on secure federated learning provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0059] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0060] Example 1

[0061] Reference Figure 1 , is an embodiment of the present invention, and provides a cross-domain model of power big data based on secure federated learning, including:

[0062] S1: Implement security initialization, generate and distribute keys and parameters to edge nodes and cloud servers. In the secret sharing phase, the node obtains the secret fragment of the private key and transmits it through the key agreement algorithm.

[0063] Furthermore, generating and distributing keys and parameters to edge nodes and cloud servers includes setting the security of the integrated power big data cross-domain model as k, the number of edge nodes in different regions as n, the sharing threshold of data information secrets as t, and the shared public value between each node and the power cloud server as a u , the secret sharing space is F, the public parameter pp is generated by KA.param(k), and the key pair generated by the edge node u is expressed as:

[0064]

[0065] in, is the key pair, is the public key in the key pair of node u, is the signature key in the key pair of node u, PK is the public key, SK is the private key, which is used for encryption and decryption of power data in the cross-domain model. After the key is generated, the public key is sent to the power cloud server through the edge node. The number of public keys collected by the power cloud server is h, and the edge nodes are counted as a group, marked as U1. The power cloud server will store the public key information of the same group of edge nodes Sent to edge nodes.

[0066] It should be noted that the transmission through the key agreement algorithm includes the edge node u receiving the public key information After that, a piece of information bu is randomly selected as the seed of PRG, and two different edge nodes u and v are generated The shared fragments of information are represented as:

[0067]

[0068] Node u generates a shared fragment about bu for node v, which is expressed as:

[0069]

[0070] For the edge nodes in the same group, the key encryption algorithm is used to calculate the corresponding ciphertext, which is expressed as:

[0071]

[0072] Among them, e u,v is the ciphertext, To negotiate the key, The generated ciphertext is sent to the power cloud server for the power data to be encrypted, and the ciphertext is counted as a group, marked as U2. The power cloud server counts the ciphertext of the same group of edge nodes as Sent to edge nodes.

[0073] It should also be noted that the generation and distribution of keys and parameters during the initialization process is to generate and distribute keys and parameters for edge nodes and cloud servers, providing a security foundation for subsequent data transmission and model training. In the secret sharing stage, the secret fragments of the private key are securely distributed to each node through secret sharing. Through the key negotiation algorithm, a secure communication channel is established, which effectively prevents the leakage of private keys and ensures the security of data during transmission.

[0074] S2: In the random vector phase, the node calculates the random vector and adds it to the transmitted data. In the authentication phase, the cloud server verifies the node identity and decrypts the parameter data.

[0075] Furthermore, the calculation of the random vector includes the node u receiving from the power cloud server After that, calculate the protocol key s of the nodes in the same groupu,v , expressed as:

[0076]

[0077] According to s u,v The random vector representation is calculated as:

[0078] p u,v =Δ u,v ·PRG(s u,v )

[0079] Among them, p u,v is the random vector from node u to node v, Δ u,v is the distance between node u and node v, PRG(s u,v ) is the number of power transmissions from node u to node v

[0080] At the same time, the random vector p of node u is calculated u =PRG(b u ), and the random vector y added to the power transfer number is obtained u , expressed as:

[0081]

[0082] Among them, p u is the random vector of node u, b u is the power transmission number of node u, y u Add a random vector to node u, x u For power data;

[0083] Adding Gaussian noise based on random vectors and performing homomorphic encryption algorithm is expressed as:

[0084] z u =Enc pk (y u +noise(ψ,t'))

[0085] Among them, z u is the homomorphic encrypted ciphertext of node u, noise(ψ,t') is Gaussian noise, Enc pk For the homomorphic encryption algorithm, the power cloud server counts the encryption methods of the same group of edge nodes into one group, marked as U3;

[0086] The power cloud server aggregates all edge nodes of a group and expresses it as:

[0087] z'=z1*z2*z3*...*z n

[0088] After z' is calculated, it is sent to the edge nodes of the same group through the power cloud server.

[0089] It should be noted that verifying the node identity and decrypting the parameter data includes setting the identity certificate to a u , the verification method is expressed as:

[0090]

[0091] Among them, a' u For identity proof u Verification information;

[0092] Power Cloud Server via a u and a' u Verify the identity information of the node, node u receives z' from the power cloud server and calculates the data decryption method z" = Dec sk (z’), z’ and z” are sent to the power cloud server and marked as U4. The edge nodes in the same group share the encryption and decryption methods in U4.

[0093] It should also be noted that in the random vector stage, adding random vectors can disrupt the original structure of the data, masking the true value of the data, making the data more difficult to analyze and identify, thereby preventing data leakage. At the same time, adding random vectors can prevent data between different nodes from being directly associated, preventing attackers from obtaining sensitive information through association analysis. Further adding Gaussian noise to the random vector can further mask the true value of the data, making the data more difficult to analyze and identify. At the same time, Gaussian noise can meet the requirements of the homomorphic encryption algorithm, so that the encrypted data can be operated, thereby realizing the training of the security model.

[0094] S3: In the security aggregation phase, the aggregation work is completed through the cloud server, and the cross-domain model based on federated learning is iteratively trained.

[0095] Furthermore, the aggregation work involves node u receiving And participate in learning the encrypted ciphertext in U2, expressed as:

[0096]

[0097] in, for u to s v,u The secret fragment of b v,u For u to b v Secret fragments of

[0098] The edge nodes will With b v,u Sent to the power server, the power server reorganizes the edge nodes that exited midway and calculates the private key It is expressed as:

[0099]

[0100] Based on the private key of sufficient nodes, use PRG to calculate the random vector p for node u∈U4 v,u , calculate the private key random vector b of the reorganized edge node u , expressed as:

[0101]

[0102] Based on the random vector p of edge nodes v,u and the private key random vector b u The aggregation result of the cross-domain model is obtained, which is expressed as:

[0103]

[0104] Among them, X is the aggregation result.

[0105] It should be noted that iterative training includes building and executing a cross-domain model training structure. The power system central server sends the initial parameters of the global model to each edge node participating in the training. Each edge node updates the model parameters based on the local training data and transmits the updated parameters to the power central server. After the central server receives the parameters transmitted by all nodes, it trains the model parameters and updates the global model through repeated iterations until the global model converges or the number of iterations is reached.

[0106] It should also be noted that the federated learning cross-domain aggregation algorithm is the core of building a cross-domain model of power big data based on secure federated learning. It provides strong support for building efficient, accurate and secure cross-domain models by protecting data privacy, improving model accuracy and improving model efficiency. The cross-domain aggregation algorithm does not need to exchange the original data of each node, but performs model training locally and only uploads updated model parameters or intermediate results, which can effectively protect data privacy. The cross-domain aggregation algorithm can also ensure that the data between each node is isolated from each other, effectively preventing data leakage and abuse.

[0107] Example 2

[0108] Reference Figure 2-4 , which is an embodiment of the present invention, provides a method for building a cross-domain model of power big data based on secure federated learning. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0109] In order to analyze the effectiveness of the federated learning cross-domain aggregation algorithm constructed by the institute, first, the basic hardware and software environment required for algorithm testing is constructed, as shown in Table 1 below:

[0110] Table 1. Hardware and software configuration of the experimental environment

[0111]

[0112] After configuring the experimental environment, the Electricity Consumption Public Dataset and SmartGrid Dataset were selected for training in the federated learning cross-domain aggregation algorithm constructed in the study. After the power data of each area in the dataset is trained, it is sent to the next area for training through the algorithm's unique encryption method, and this cycle is repeated to achieve cross-domain secure sharing of dataset information. Among them, the Electricity Consumption Public Dataset is a public dataset of electricity consumption, and the Smart Grid Dataset is a smart grid dataset.

[0113] In order to verify the superiority of the algorithm constructed in this study, a centralized differential privacy model, a knowledge graph convolutional network model, and an unsupervised learning model were selected for comparison. Reference

[16] is a centralized differential privacy model, Reference

[17] is a knowledge graph convolutional network model, Reference

[18] is an unsupervised learning model, and the research model is a cross-domain model of power big data based on secure federated learning.

[0114] Secondly, the leakage rate of data transmitted by different models was tested. After the electricity data in the Electricity Consumption Public Dataset was trained with different models, the leakage rate of data transmission was low. However, our invention had the lowest leakage rate, about 0.5%. The data leakage rate of the centralized differential privacy model was about 1.0%, while the leakage rate of the knowledge graph convolutional network model was about 1.2%, and the leakage rate of the unsupervised learning model was about 2.8%. The amount of electricity data in the Smart Grid Dataset is more and more complex than that in the Electricity Consumption Public Dataset. Therefore, the leakage rate of the four models has increased slightly, but it is still the lowest data leakage rate of our invention, about 0.8%. This shows that the federated learning cross-domain aggregation model of the present invention has strong data privacy protection capabilities when processing large-scale and complex data sets. Among them, Model type is the model type, and Data leakage rate is the data leakage rate.

[0115] Third, test the efficiency of data transmission of different models, such as Figure 2 shown.

[0116] Depend on Figure 2It can be seen that when the data volume of the data set is low, the time taken by the four models to transmit 100 copies of data is not much different. Whether in the Electricity Consumption Public Dataset or in the Smart Grid Dataset, when the data volume exceeds 300, the time taken by the four models to transmit data begins to change significantly. Our invented model takes 1.2s and 2.3s to transmit 100 copies of data, which is the least time among the four models, while the unsupervised learning model takes the most time, about 4.3s and 4.6s respectively. The centralized differential privacy model and the knowledge graph convolutional network model are between the research model and the unsupervised learning model. The data transmission time shows that the federated learning cross-domain aggregation model of the present invention improves the data transmission efficiency compared with the prior art. Data volume is the data volume and Transmission time is the transmission time.

[0117] Finally, the encryption overhead of four models is tested, such as Figure 3 shown.

[0118] Depend on Figure 3 (a), (b), and (c) show that when cross-domain data messages are sent, the computational overhead of the centralized differential privacy model, the knowledge graph-based convolutional network model, and the unsupervised learning model increases with the increase in the size of the encrypted ciphertext and the number of attributes. Figure 3 (d) It can be seen that the computational overhead of our invention remains stable even when the encrypted ciphertext size and the number of attributes increase, which reflects that our invention has good robustness in data scalability and attribute diversity, and is superior to the other three models. It shows that our invention can handle and cope with high data loads and a high number of attributes without increasing computational overhead for no reason, and has excellent scalability and stability. The above three experimental data show that our invention has high privacy security and reliability for cross-domain data transmission, where Ciphertext size is the ciphertext size, Number of attributes in an access structure is the number of attributes in the access structure, and Consumption (seconds) is the computational overhead.

[0119] Example 3

[0120] Reference Figure 4 , is an embodiment of the present invention, and provides a method for building a cross-domain model of power big data based on secure federated learning, including: an initialization module; a data protection module, and an aggregation training module.

[0121] The initialization module is used to implement secure initialization, generate and distribute keys and parameters to edge nodes and cloud servers. In the secret sharing phase, the node obtains the secret fragment of the private key and transmits it through the key negotiation algorithm; the data protection module is used in the random vector phase, the node calculates the random vector and adds it to the transmitted data. In the identity authentication phase, the cloud server verifies the node identity and decrypts the parameter data; the aggregation training module is used to complete the aggregation work through the cloud server in the secure aggregation phase, and iteratively train the cross-domain model based on federated learning.

[0122] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0123] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0124] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0125] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0126] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for constructing a cross-domain model of power big data based on secure federated learning, characterized in that: include: Implement secure initialization, generate and distribute keys and parameters to edge nodes and cloud servers. In the secret sharing phase, nodes obtain secret fragments of private keys and transmit them through the key agreement algorithm; In the random vector phase, the node calculates the random vector and adds it to the transmitted data. In the authentication phase, the cloud server verifies the node identity and decrypts the parameter data. In the security aggregation stage, the aggregation work is completed through the cloud server, and the cross-domain model based on federated learning is iteratively trained.

2. The method for constructing a cross-domain model of power big data based on secure federated learning according to claim 1, characterized in that: The generation and distribution of keys and parameters to edge nodes and cloud servers includes setting the security of the integrated power big data cross-domain model to be k, the number of edge nodes in different regions to be n, the sharing threshold of data information secrets to be t, and the shared public value between each node and the power cloud server to be a u , the secret sharing space is F, the public parameter pp is generated by KA.param(k), and the key pair generated by the edge node u is expressed as: in, is the key pair, is the public key in the key pair of node u, is the signature key in the key pair of node u, PK is the public key, SK is the private key, which is used for encryption and decryption of power data in the cross-domain model. After the key is generated, the public key is sent to the power cloud server through the edge node. The number of public keys collected by the power cloud server is h, and the edge nodes are counted as a group, marked as U1. The power cloud server will store the public key information of the same group of edge nodes Sent to edge nodes.

3. The method for constructing a cross-domain model of power big data based on secure federated learning as claimed in claim 2, characterized in that: The transmission through the key agreement algorithm includes edge node u receiving public key information After that, a piece of information bu is randomly selected as the seed of PRG, and two different edge nodes u and v are generated The shared fragments of information are represented as: Node u generates a shared fragment about bu for node v, which is expressed as: For the edge nodes in the same group, the key encryption algorithm is used to calculate the corresponding ciphertext, which is expressed as: Among them, e u,v is the ciphertext, To negotiate the key, The generated ciphertext is sent to the power cloud server for the power data to be encrypted, and the ciphertext is counted as a group, marked as U2. The power cloud server counts the ciphertext of the same group of edge nodes as Sent to edge nodes.

4. The method for constructing a cross-domain model of power big data based on secure federated learning as claimed in claim 3, characterized in that: The calculation of the random vector includes receiving the node u from the power cloud server After that, calculate the protocol key s of the nodes in the same group u,v , expressed as: According to s u,v The random vector representation is calculated as: p u,v =Δ u,v ·PRG(s u,v ) Among them, p u,v is a random vector from node u to node v, Δ u,v is the distance between node u and node v, PRG(s u,v ) is the number of power transmissions from node u to node v; At the same time, the random vector p of node u is calculated u =PRG(b u ), and the random vector y added to the power transfer number is obtained u , expressed as: Among them, p u is the random vector of node u, b u is the power transmission number of node u, y u Add a random vector to node u, x u For power data; Adding Gaussian noise based on random vectors and performing homomorphic encryption algorithm is expressed as: z u =Enc pk (y u +noise(ψ,t,)) Among them, z u is the homomorphic encrypted ciphertext of node u, noise(ψ,t,) is Gaussian noise, Enc pk For the homomorphic encryption algorithm, the power cloud server counts the encryption methods of the same group of edge nodes into one group, marked as U3; The power cloud server aggregates all edge nodes of a group and expresses it as: z'=z1*z2*z3*...*z n After z' is calculated, it is sent to the edge nodes of the same group through the power cloud server.

5. The method for constructing a cross-domain model of power big data based on secure federated learning according to claim 4, characterized in that: The verification of node identity and decryption of parameter data includes setting the identity certificate to a u , the verification method is expressed as: Among them, a' u For identity proof u Verification information Power Cloud Server via a u and a' u Verify the identity information of the node, node u receives z' from the power cloud server and calculates the data decryption method z" = Dec sk (z,), z’ and z” are sent to the power cloud server and marked as U4. The edge nodes in the same group share the encryption and decryption methods in U4.

6. The method for constructing a cross-domain model of power big data based on secure federated learning according to claim 5, characterized in that: The aggregation work includes node u receiving And participate in learning the encrypted ciphertext in U2, expressed as: in, for u to s v,u The secret fragment of b v,u For u to b v Secret fragments of The edge nodes will With b v,u Sent to the power server, the power server reorganizes the edge nodes that exited midway and calculates the private key It is expressed as: Based on the private key of sufficient nodes, use PRG to calculate the random vector p for node u∈U4 v,u , calculate the private key random vector b of the reorganized edge node u , expressed as: Based on the random vector p of edge nodes v,u and the private key random vector b u The aggregation result of the cross-domain model is obtained, which is expressed as: Among them, X is the aggregation result.

7. The method for constructing a cross-domain model of power big data based on secure federated learning according to claim 6, characterized in that: The iterative training includes constructing and executing a cross-domain model training structure. The power system central server sends the initial parameters of the global model to each edge node participating in the training. Each edge node updates the model parameters based on the local training data and transmits the updated parameters to the power central server. After the central server receives the parameters transmitted by all nodes, it trains the model parameters and updates the global model through repeated iterations until the global model converges or the number of iterations is reached.

8. A system using the method for building a cross-domain model of power big data based on secure federated learning as described in any one of claims 1 to 7, characterized in that: Includes initialization module, data protection module, and aggregation training module; The initialization module is used to implement security initialization, generate and distribute keys and parameters to edge nodes and cloud servers. In the secret sharing stage, the node obtains the secret fragment of the private key and transmits it through the key agreement algorithm; The data protection module is used for the node to calculate the random vector in the random vector phase and add it to the transmission data, and the cloud server to verify the node identity and decrypt the parameter data in the identity authentication phase; The aggregation training module is used to complete the aggregation work through the cloud server in the security aggregation stage and perform iterative training on the cross-domain model based on federated learning.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for building a cross-domain model of power big data based on secure federated learning as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for building a cross-domain model of power big data based on secure federated learning as described in any one of claims 1 to 7 are implemented.

Images