CN113901412A - Power quality disturbance detection method and device, electronic device, storage medium - Google Patents
Power quality disturbance detection method and device, electronic device, storage medium Download PDFInfo
- Publication number
- CN113901412A CN113901412A CN202111208979.8A CN202111208979A CN113901412A CN 113901412 A CN113901412 A CN 113901412A CN 202111208979 A CN202111208979 A CN 202111208979A CN 113901412 A CN113901412 A CN 113901412A
- Authority
- CN
- China
- Prior art keywords
- task
- ciphertexts
- client
- global model
- power quality
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/30—Authentication, i.e. establishing the identity or authorisation of security principals
- G06F21/31—User authentication
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/30—Authentication, i.e. establishing the identity or authorisation of security principals
- G06F21/45—Structures or tools for the administration of authentication
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/602—Providing cryptographic facilities or services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Computer Security & Cryptography (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Hardware Design (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Bioethics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a power quality disturbance detection method and device, electronic equipment and a storage medium, which relate to the technical field of information security and comprise the following steps: a private block chain based on intelligent contracts and federal learning is constructed, and model training under the condition that local electric energy quality data of a client side is not shared is achieved; the credibility problem of the client side for collecting the power quality data is solved through the equipment identity verification based on certificateless encryption, and the equipment identity verification method is used for ensuring the safety and credibility of the client side nodes participating in the task; the gradient updating parameters are encrypted, so that the risk of model training parameter leakage is solved, and the safety of the gradient updating parameters of the global model is ensured; and the accuracy of the model is ensured by disturbing and detecting the model through the constructed power quality. And finally, under the condition that the equipment is credible, the data of each energy subsidiary company is shared under the condition of protecting privacy, so that accurate and efficient power quality disturbance detection is completed.
Description
Technical Field
The present application relates to the field of information security technologies, and in particular, to a method and an apparatus for detecting power quality disturbance, an electronic device, and a storage medium.
Background
The second industrial revolution in the beginning of the twentieth century pushes human beings to enter the pipelined 'electrical era' of mass production, with the arrival of the industrial 4.0 era, the social and economic conditions and the living standard of people are greatly leaped, a system based on the fusion of big data and sensors is used in large scale in production, and the global demand for electric energy is increased dramatically. At present, renewable energy sources such as photovoltaic energy, wind energy and the like are largely integrated in a power system to generate electricity so as to deal with the global energy crisis.
However, the use of renewable energy sources still presents some problems: firstly, because the environment has random fluctuation, the power generation process is greatly influenced by external factors (such as lightning stroke, short circuit caused by severe weather and the like); secondly, the introduction of nonlinear switches such as a PWM inverter and the like is easy to cause power quality disturbance, and has serious influence on the stable operation of a power grid. Therefore, real-time identification and classification evaluation of power quality disturbance in a complex environment are needed, a basis can be provided for power grid scheduling decision at a source end, the operation reliability of a power system is effectively improved, electronic instrument equipment of a user can be protected at a load end, the service life of components is prolonged, and the risk of equipment damage is reduced.
Although a great amount of detection work of power quality disturbance is proposed, the detection work under a real complex environment still cannot be dealt with, which is mainly shown in the following steps: firstly, in data use, the existing work often focuses on data analysis, the used experimental data come from an open-source data set or are collected by self, the data are converged to a central server for model training, and subsequent abnormality judgment is executed. However, in a national energy scene, the sensing data of the power system representing the quality of electric energy is the information of the national key infrastructure, and if an adversary can break through the cloud server storing the data, the operation state in the power system can be easily deduced, and a plan is further made to endanger the national infrastructure. Secondly, in an energy environment, network topologies of different factories, different mechanisms and different environments are complex, data acquisition nodes (such as a wind power generator, a photovoltaic generator and the like) are easily interfered by the outside, and original equipment has a credibility problem. For example, an attacker accesses an unknown device to conduct a data corruption attack. In addition, the acquired data may also be noisy due to environmental influences. The above-mentioned untrusted devices and noisy data greatly increase the difficulty of model training. Thirdly, in the process of detecting the power quality disturbance, data/parameter interaction related to operation cannot be performed under the condition of encryption, and the data cannot be available and invisible. Fourthly, the existing power quality disturbance identification method mainly depends on feature extraction and machine learning models, for example, statistical features are extracted according to rules defined by experts in advance, and then model training is carried out by utilizing a machine learning method. However, as the complexity of energy and power systems increases, feature libraries must be updated periodically, which undoubtedly increases the amount of human labor, and lacks automation and intelligence.
Disclosure of Invention
In view of this, embodiments of the present application provide a power quality disturbance detection method and apparatus, an electronic device, and a storage medium, which can perform robust model training without locally extracting original data of a client, protect model parameters, make data available and invisible, and perform trusted authentication on the device itself, so as to ensure privacy, availability, security, and accuracy of a power quality disturbance detection system.
According to a first aspect of the embodiments of the present application, there is provided a power quality disturbance detection method applied to a task initiator, including:
constructing a private block chain based on an intelligent contract and federal learning, wherein the private block chain consists of a task initiator, an aggregator and a client;
verifying the identity of the client by adopting a non-certificate-based encryption algorithm, wherein the client can participate in the global model training task after the verification is passed;
creating a public and private task key for the global model training task, and broadcasting and distributing the public task key to all clients participating in the global model training task;
sending a global model to all clients participating in a global model training task, so that after receiving the global model, the clients train the global model by using local power quality data to obtain gradient update parameters of model training in the current round, encrypting the gradient update parameters by using the public task key to obtain ciphertext, sending the ciphertext to the aggregator, so that after the aggregator collects the ciphertexts uploaded by all the clients, multiplying the ciphertexts of different clients by each other to obtain multiplied ciphertexts, and sending all the multiplied ciphertexts to the task initiator;
receiving all multiplied ciphertexts transmitted by the aggregator, decrypting all multiplied ciphertexts to obtain respective corresponding plaintext addition values, quantizing and verifying the promotion of the global model by using the plaintext addition values corresponding to each multiplied cipher text based on a training quality certification of a consensus mechanism, and providing rewards for clients generating effective gradient update parameters;
updating the global model by using a federal learning aggregation algorithm according to the effective gradient updating parameters to obtain a power quality disturbance detection model;
and detecting the power quality disturbance data to be detected by using the power quality disturbance detection model.
Further, the identity of the client is verified by adopting a non-certificate-based encryption algorithm, and the method comprises the following sub-steps:
and verifying the signature by using a public key generated by each client and the identity of the client, wherein the signature is the result output by encrypting information by using a private key by using the client, the private key is established by using a part of private key and a secret value of the client, a public key is created according to the secret value and the public key is disclosed, and the part of private key is created by a key generation center according to the identity of the client.
Further, after providing the reward to the client generating the effective gradient update parameter, the method further includes:
and verifying the received ciphertext, packaging a block, and finally chaining the block.
According to a second aspect of the embodiments of the present application, there is provided a power quality disturbance detection method applied to a client, including:
constructing a private block chain based on an intelligent contract and federal learning, wherein the private block chain consists of a task initiator, an aggregator and a client;
receiving the task to initiate identity authentication by adopting a certificateless encryption algorithm, and participating in a global model training task after the authentication is passed;
receiving a public and private task key broadcasted by the task initiator, wherein the public and private task key is created for a global model training task by the task initiator;
receiving a global model sent by the task initiator;
training the global model by using local power quality data to obtain gradient updating parameters of the model training of the current round;
encrypting the gradient updating parameters by using a public task key to obtain a ciphertext;
sending the ciphertext to an aggregator, so that after the aggregator collects the ciphertexts uploaded by all the clients, the cipher texts of different clients are multiplied by each other to obtain multiplied ciphertexts, and all the calculated multiplied ciphertexts are transmitted to a task initiator so that after the task initiator receives all the multiplied ciphertexts transmitted by the aggregator, decrypting all the multiplied ciphertexts to obtain plaintext addition values, quantizing and verifying the promotion of the global model by using the plaintext addition value corresponding to each multiplied cipher text based on the training quality certification of the consensus mechanism, providing rewards for a client side generating effective gradient updating parameters, updating the global model by using a federal learning aggregation algorithm according to the effective gradient updating parameters to obtain a power quality disturbance detection model, and detecting power quality disturbance data to be detected by using the power quality disturbance detection model.
According to a third aspect of the embodiments of the present application, there is provided a power quality disturbance detection method applied to an aggregator, including:
constructing a private block chain based on an intelligent contract and federal learning, wherein the private block chain consists of a task initiator, an aggregator and a client;
receiving ciphertexts sent by all clients, wherein the clients initiate identity verification by adopting a certificateless encryption algorithm for receiving the tasks, and can participate in the global model training task after the verification is passed; the cipher text is obtained by encrypting a gradient updating parameter by the client by using a public task key, the public task key is created by the task initiator for a global model training task, and the gradient updating parameter is obtained by the client receiving a global model sent by the task initiator and training the global model by using local power quality data;
carrying out multiplication operation on the ciphertexts of different clients pairwise to obtain multiplied ciphertexts;
and transmitting all the calculated multiplication ciphertexts to a task initiator so that the task initiator decrypts all the multiplication ciphertexts after receiving all the multiplication ciphertexts to obtain a plain text added value, quantizing and verifying the lifting of the global model by using a plain text added value corresponding to each multiplication cipher text by using a training quality certificate based on a consensus mechanism, providing rewards for a client generating an effective gradient updating parameter, updating the global model by using a federal learning aggregation algorithm according to the effective gradient updating parameter to obtain an electric energy quality disturbance detection model, and detecting electric energy quality disturbance data to be detected by using the electric energy quality disturbance detection model.
According to a fourth aspect of the embodiments of the present application, there is provided a power quality disturbance detection apparatus, which is applied to a task initiator, and includes:
the system comprises a first construction module, a second construction module and a third construction module, wherein the first construction module is used for constructing a private block chain based on an intelligent contract and federal learning, and the private block chain consists of a task initiator, an aggregator and a client;
the first verification module is used for verifying the identity of the client by adopting a non-certificate-based encryption algorithm, and the client after verification can participate in the global model training task;
the system comprises a creating module, a searching module and a processing module, wherein the creating module is used for creating a public and private task key for a global model training task and broadcasting and distributing the public task key to all clients participating in the global model training task;
the first sending module is used for sending a global model to all clients participating in a global model training task, so that after receiving the global model, the clients train the global model by using local electric energy quality data to obtain gradient updating parameters of the model training in the current round, encrypt the gradient updating parameters by using the public task key to obtain ciphertexts, send the ciphertexts to the aggregator, so that after the aggregator collects the ciphertexts uploaded by all the clients, multiply the ciphertexts of different clients pairwise to obtain multiplied ciphertexts, and transmit all the multiplied ciphertexts to the task initiator;
the encryption verification module is used for receiving all multiplied ciphertexts transmitted by the aggregator, decrypting all multiplied ciphertexts to obtain respective corresponding plaintext addition values, quantizing and verifying the promotion of the global model by using the plaintext addition values corresponding to each multiplied cipher text based on a training quality certification based on a consensus mechanism, and providing rewards for clients generating effective gradient update parameters;
the updating module is used for updating the global model by utilizing a federal learning aggregation algorithm according to the effective gradient updating parameters to obtain a power quality disturbance detection model;
and the detection module is used for detecting the power quality disturbance data to be detected by using the power quality disturbance detection model.
According to a fifth aspect of the embodiments of the present application, there is provided a power quality disturbance detection apparatus, applied to a client, including:
the second construction module is used for constructing a private block chain based on intelligent contracts and federal learning, and the private block chain consists of a task initiator, an aggregator and a client;
the second verification module is used for receiving the task to initiate identity verification by adopting a certificateless encryption algorithm and participating in the global model training task after the verification is passed;
a first receiving module, configured to receive a public-private task key broadcasted by the task initiator, where the public-private task key is created by the task initiator for a global model training task;
the second receiving module is used for receiving the global model sent by the task initiator;
the training module is used for training the global model by using local power quality data to obtain gradient updating parameters of the model training of the current round;
the encryption module is used for encrypting the gradient updating parameters by using the public task key to obtain a ciphertext;
a second sending module, configured to send the ciphertext to an aggregator, so that after the aggregator collects ciphertexts uploaded by all clients, the aggregator multiplies two ciphertexts of different clients by two to obtain multiplied ciphertexts, and sends all calculated multiplied ciphertexts to a task initiator, so that after the task initiator receives all multiplied ciphertexts sent by the aggregator, all multiplied ciphertexts are decrypted to obtain a plaintext sum value, a training quality certificate based on a consensus mechanism is used to quantify and verify the improvement of the global model for the plaintext sum value corresponding to each multiplied ciphertext, so as to provide a reward for the client that generates an effective gradient update parameter, and according to the effective gradient update parameter, the global model is updated by using a federal-learned aggregation algorithm to obtain an electric energy quality disturbance detection model, and detecting the power quality disturbance data to be detected by using the power quality disturbance detection model.
According to a sixth aspect of the embodiments of the present application, there is provided a power quality disturbance detection apparatus applied to an aggregator, including:
the third building module is used for building a private block chain based on intelligent contracts and federal learning, and the private block chain consists of a task initiator, an aggregator and a client;
the third receiving module is used for receiving all ciphertexts sent by the client, and the client initiates identity verification based on a certificateless encryption algorithm for receiving the task and can participate in the global model training task after the verification is passed; the cipher text is obtained by encrypting a gradient updating parameter by the client by using a public task key, the public task key is created by the task initiator for a global model training task, and the gradient updating parameter is obtained by the client receiving a global model sent by the task initiator and training the global model by using local power quality data;
the multiplication operation module is used for carrying out multiplication operation on the ciphertexts of different clients pairwise to obtain multiplied ciphertexts;
and the third sending module is used for transmitting all the calculated multiplied ciphertexts to the task initiator so that the task initiator decrypts all the multiplied ciphertexts after receiving all the multiplied ciphertexts to obtain a plain text added value, quantifies and verifies the promotion of the global model by using the plain text added value corresponding to each multiplied cipher text based on a training quality certificate of a consensus mechanism, provides rewards for a client generating an effective gradient updating parameter, updates the global model by using a federated learning aggregation algorithm according to the effective gradient updating parameter to obtain an electric energy quality disturbance detection model, and detects the electric energy quality disturbance data to be detected by using the electric energy quality disturbance detection model.
According to a seventh aspect of embodiments of the present application, there is provided an electronic apparatus, comprising:
one or more processors;
a memory for storing one or more programs;
when executed by the one or more processors, cause the one or more processors to implement a method as described in the first aspect.
According to an eighth aspect of embodiments herein, there is provided a computer-readable storage medium having stored thereon computer instructions, characterized in that the instructions, when executed by a processor, implement the steps of the method according to the first aspect.
The technical scheme provided by the embodiment of the application can have the following beneficial effects:
according to the embodiment, the federate learning is adopted, so that the risk of data leakage of the client is overcome, and model training under the condition that local data of the client is not shared is further realized. Identity authentication based on a certificateless encryption algorithm is adopted, so that the reliability problem of data acquisition equipment is solved, and the safety and reliability of a client participating in model training are further ensured. And encrypting the gradient updating parameters by using the common task key, so that the risk of model training parameter leakage is overcome, and the gradient security of deep learning is further realized. Experimental results under the truly built micro-grid show that the method can give consideration to privacy, usability, safety and accuracy, and provides useful reference for building an intelligent energy power system.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and together with the description, serve to explain the principles of the application.
Fig. 1 is a flow chart illustrating a method of power quality disturbance detection in accordance with an exemplary embodiment 1.
Fig. 2 is a schematic structural diagram illustrating a power quality disturbance detection apparatus according to an exemplary embodiment 1.
Fig. 3 is a flow chart illustrating a method of power quality disturbance detection in accordance with an exemplary embodiment 2.
Fig. 4 is a schematic structural diagram illustrating a power quality disturbance detection apparatus according to an exemplary embodiment 2.
Fig. 5 is a flow chart illustrating a method of power quality disturbance detection in accordance with an exemplary embodiment 3.
Fig. 6 is a schematic structural diagram illustrating a power quality disturbance detection apparatus according to an exemplary embodiment 3.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this application and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.
In an energy scene, a plurality of energy sub-companies want to cooperate to perform power quality disturbance detection, but the power data circulation of the sub-companies has privacy problems, each sub-company manages a plurality of power stations, each power station is provided with a plurality of motors, the motors collect data locally and have certain computing capacity, the power stations are provided with gateways, and the gateways have data input and output capacity and certain computing capacity. In the embodiment of the invention, the motor of the power station is taken as a client, and the gateway of the power station is taken as an aggregator. The final purpose is to share data of each energy subsidiary company (model training) under the condition of protecting privacy under the condition that the equipment is credible, so that accurate and efficient power quality disturbance detection is completed.
Example 1:
fig. 1 is a flowchart illustrating a power quality disturbance detection method according to an exemplary embodiment, and referring to fig. 1, a power quality disturbance detection method provided by an embodiment of the present invention, which is applicable to a task initiator, may include the following steps:
step S11, constructing a private block chain based on intelligent contracts and federal learning, wherein the private block chain consists of a task initiator, an aggregator and a client;
step S12, verifying the identity of the client by adopting a non-certificate-based encryption algorithm, wherein the client after verification can participate in the global model training task;
step S13, creating a public and private task key for the global model training task, and broadcasting and distributing the public task key to all clients participating in the global model training task;
step S14, sending a global model to all clients participating in a global model training task, so that after receiving the global model, the clients train the global model by using local power quality data to obtain gradient update parameters of the model training in the current round, encrypting the gradient update parameters by using the common task key to obtain ciphertexts, sending the ciphertexts to the aggregator, so that after the aggregator collects the ciphertexts uploaded by all the clients, multiplying the ciphertexts of different clients by two to obtain multiplied ciphertexts, and sending all the multiplied ciphertexts to the task initiator;
step S15, receiving all the multiplied ciphertexts transmitted by the aggregator, decrypting all the multiplied ciphertexts to obtain respective corresponding plaintext addition values, quantizing and verifying the global model promotion by using the plaintext addition values corresponding to each multiplied cipher text based on the training quality certification of the consensus mechanism, and providing rewards for the client end generating the effective gradient update parameters;
step S16, updating the global model by using a federal learning aggregation algorithm according to the effective gradient updating parameters to obtain a power quality disturbance detection model;
and step S17, detecting the power quality disturbance data to be detected by using the power quality disturbance detection model.
According to the embodiment, the federate learning is adopted, so that the risk of data leakage of the client is overcome, and model training under the condition that local data of the client is not shared is further realized. Identity authentication based on a certificateless encryption algorithm is adopted, so that the reliability problem of data acquisition equipment is solved, and the safety and reliability of a client participating in model training are further ensured. And encrypting the gradient updating parameters by using the common task key, so that the risk of model training parameter leakage is overcome, and the gradient security of deep learning is further realized. Experimental results under the truly built micro-grid show that the method can give consideration to privacy, usability, safety and accuracy, and provides useful reference for building an intelligent energy power system.
In the specific implementation of the step S11, constructing a private block chain based on intelligent contracts and federal learning, wherein the private block chain is composed of a task initiator, an aggregator and a client;
it should be noted that the private blockchain constructed based on the intelligent contracts and the federal learning can be constructed by any one of the task initiator, the aggregator and the client, and is generally constructed by the task initiator.
In the specific implementation of the step S12, the identity of the client is verified by using a non-certificate-based encryption algorithm, and the client after verification can participate in the global model training task;
specifically, before the global model starts to be trained, all clients in the private block chain need to be verified, and the clients after verification can participate in the global model training task; the invention adopts the encryption algorithm based on no certificate to realize the verification of the participating client. Namely, a signature is verified by using a public key generated by each client and the identity of the client, wherein the signature is a result output by encrypting information by the client by using a private key, the private key is established by the client by using a part of private key and a secret value of the client, a public key is created according to the secret value and the public key is disclosed, and the part of private key is created by a key generation center according to the identity of the client.
More specifically, client a issues an SK with its private keyASigned transaction and its public key PKAAnd a unique identity IDAAppended to the transaction. The remaining clients, the task initiator and the aggregator, except client a, can check: 1) the transaction is actually performed with PKAAssociated private key signature, and 2) PKAIndeed belonging to IDA. In this way, it can be easily verified whether the transaction is made by having an IDACreated by the client device of (1). Establishing a public key PK for a client AAAnd a private key SKAThe method comprises the following steps:
Setup(1λ)→(K,MSK)
(a) the setup algorithm takes the security parameter lambda and returns the system parameter K and a secret master key MSK. The algorithm is run by the key generation center KGC, only KGC knows the value of MSK.
PSkeyGen(K,IDA,MSK)→(PSKA)
(b) Part of the private key generation algorithm adopts a system parameter K and the identity ID of the client AA∈{0,1}*And the master key MSK, and outputPartial private key PSKA. The algorithm is operated by KGC, and the output PSKAWill be transmitted to client a.
SValGen(K,IDA)→(XA)
(c) The secret value generation algorithm adopts a system parameter K and an identity ID of a client AAOutputting secret value XA. The algorithm is run by the user, XAWill be used to convert part of the private key to the private key. The algorithm is run by the client.
SKeyGen(K,PSKA,XA)→(SKA)
(d) Private key generation algorithm based on system parameter K and partial private key PSKAAnd a secret value XAFor input, return private key SKA. The algorithm is run by the client, and only the client has the private key.
PKeyGen(K,XA)→(PKA)
(e) Public key generation algorithm with system parameter K and secret value XAConstructing a public key PKA. The algorithm is run by the client, PKAWill be broadcast to the public.
In the stage of equipment identity authentication:
Sign(K,M∈M,SKA)→Sig
(a) the algorithm adopts a system parameter K, a message M needing to be signed and a private key SK of a clientAAnd outputs a signature.
Ver(M∈M,Sig,IDA,PKA)→Valid∨Invalid∨⊥
(b) The algorithm adopts message M, signature Sig and user IDAAnd a user private key SKAAnd outputs whether the signature is valid, i.e., determines that the transaction is indeed used with the PKAThe associated private key signature.
VerID(IDA,PKA,K)→Valid∨Invalid∨⊥
(c) PK assessmentAIndeed belonging to IDA。
In the specific implementation of step S13, a public-private task key is created for the global model training task, and the public-private task key is broadcast and distributed to all clients participating in the global model training task;
specifically, the task initiator needs to generate a pair of public and private task keys pk and sk for each global model training task (each iteration), and disclose the public task key pk in the task specification, so that each client participating in the task can obtain the public task key through a public task requirement or a block chain broadcast.
The method adopts homomorphic encryption based on the Paillier cryptosystem. Homomorphic encryption is a method that can perform computation on encrypted data without decryption, can obtain the same result as computing on original data, and prevents selected ciphertext from being attacked by using a proxy re-encryption technique. Paillier is a cryptographic system with addition homomorphism, namely:
for arbitrary plaintext m1，m2∈ZNAnd any r1，
Corresponding ciphertext c1＝E[m1,r1]， c2＝E[m2,r2]Satisfies the following conditions:
after decryption, obtaining:
D[c1·c2]＝D[E[m1,r1]·E[m2,r2]mod N2]＝m1+m2mod N2
the process of the task initiator generating the key based on Paillier is as follows:
1) two large prime numbers p and q are selected to ensure that gcd (pq, (p-1) (q-1)) ═ 1
2) Calculation of n ═ pq,. lambda. ═ lcm (p-1, q-1)
3) Definition of
Note that: here, the division means division
4) Randomly selecting one less than n2And μ ═ L (g) is presentλmod n2))-1mod n
5) The public key is (n, g)
6) The private key is (lambda, mu)
In case of the same key length, the key can be generated quickly:
g＝n+1,λ＝Φ(n),μ＝Φ(n)-1mod n,
Φ (n) means the Euler function, here equal to (p-1) × (q-1)
In the specific implementation of step S14, sending the global model to all clients participating in the global model training task, so that after receiving the global model, the clients train the global model using the local power quality data to obtain the gradient update parameters of the current model training, encrypt the gradient update parameters using the common task key to obtain a ciphertext, send the ciphertext to the aggregator, so that after the aggregator collects the ciphertexts uploaded by all clients, multiply the ciphertexts of different clients by two to obtain multiplied ciphertexts, and transmit all multiplied ciphertexts to the task initiator;
specifically, the local power quality data of the client includes normal data and power quality disturbance data, and the data may be acquired from an IEEE PES database or generated by designing a microgrid platform by itself. Aiming at the micro-grid platform, the micro-grid platform simulating a real power system can be built by combining with an actual application scene. The platform has a flexible topology, including various lines, distributed power supplies, and line impedance simulators (resistive, inductive, and capacitive loads) by which different line faults can be repeated through logic program control. Wherein the distributed power supply comprises a photovoltaic roof concentrated inverter, a 90kw photovoltaic simulator, a 30kw diesel generator and a 45kw wind generator simulator. Loads of the microgrid include architectural lighting, three-phase asynchronous induction motors, electric vehicle charging piles and programmable electronic loads of 100 kW. The bus voltage of the microgrid is 410V and 50 Hz. The micro-grid platform utilizes a line impedance simulator and simulates line faults under different scenes such as voltage sag, bump, flicker, short circuit, harmonic wave and the like in a real environment through programmable logic control. In other words, the platform can provide a programmable control environment to simulate the operating characteristics of the microgrid under different scenes. By utilizing the platform, 12000 signals can be generated, which correspond to 9000 normal data and 3000 power quality disturbance data.
For each client P participating in the tasktDownloading a global model m provided by a task initiator via a network is requirediWherein
t denotes the t-th client among all clients that are involved in the task, and i denotes the number of communication iterations. For each selected client PtAfter receiving the global model, preprocessing the local power quality data, continuing to train the model and updating the gradient updating parameters by using the local power quality data, reducing the training time by using the GPU of the node and other hardware accelerating devices during the period, and finally obtaining the gradient updating parameters of the client under the current communication iteration
The example trains the global model by using local power quality data and an LSTM algorithm; specifically, the VMD algorithm is used to perform noise filtering and signal reconstruction on the local power quality data, and the essence of the VMD algorithm is a process of solving the variational problem. This process involves the construction and solution of a variational problem, which is expressed as: the original signal f is decomposed into a plurality of IMF characteristic mode functions, and each characteristic mode function is assumed to have a limited bandwidth and a different center frequency. In the VMD algorithm decomposition calculation process, the penalty factor alpha and the modal decomposition quantity K influence the decomposition result of the VMD. The number of modes K needs to be given before VMD decomposition, and when the value of K is too small, which results in insufficient decomposition of the original signal, the original multiple modes of the signal may be aliased in one mode component, even resulting in that a certain mode cannot be estimated. Conversely, when the K value is too large to cause the signal to be decomposed excessively, some modal component of the decomposed signal appears in a plurality of modal components, so that the modal center frequencies obtained by the decomposition overlap. The smaller the penalty factor alpha value is, the larger the bandwidth of each intrinsic mode component obtained after decomposition is; the larger the penalty factor alpha value is, the smaller the bandwidth of each eigenmode component obtained after decomposition is. Noise filtering and signal reconstruction can be effectively performed by selecting proper alpha and K. Selecting a parameter alpha of 2500 on the basis of an experimental test;
training the enhanced local power quality data by using the LSTM, and taking the time sequence signals of the K modes reserved in the previous step as the input of the LSTM, wherein the LSTM neural network has the following structure:
an input layer: n × K × L, the IMF signal obtained by VMD decomposition is used as input to LSTM. Since the final output is either tag 1 (representing normal) or tag 0 (representing abnormal), the input to the LSTM must contain both normal and abnormal data. Where K represents the number of modes determined by the VMD, N represents the number of input signals (including normal and abnormal), and L represents the length of each input (duration of the signal);
hiding the layer: the layer parameters are obtained by tuning, and the layer parameters comprise 2 hidden layers in total, and each hidden layer has 32 nodes;
an output layer: and the output layer adopts a softmax function to obtain a classification result in a training stage.
In order to protect the privacy of the user, the data of the client may be highly personalized, and not all users want to upload the data to the data center, and a distributed machine learning method is used to implement the learning task, so that the model should be trained without transmitting the original training data.
The VMD algorithm is used for carrying out noise filtration and signal reconstruction on the local electric energy quality data, and then the LSTM is used for training the enhanced local electric energy quality data, so that the defects of coverage rate and accuracy rate when the traditional statistical characteristics are used for modeling are overcome, the labor cost is reduced, and the automation level and accuracy of the fault diagnosis of the electric power system are improved.
Using a common task key PskEncrypting the gradient updating parameters to obtain a ciphertext
The specific encryption process is as follows:
1) the plaintext m is a positive integer greater than or equal to 0 and less than n
2) Randomly selecting r to satisfy 0<r<n and
(a sufficient condition is that r, n are coprime):
means r is at n2Is that the multiplication inverse exists
3) Calculating cipher text c ═ gmrnmod n2
And then the client uploads the encrypted gradient update parameter ciphertext
And the client generates an authentication public key and a Timestamp to the aggregator based on the unlicensed encryption.
Collecting ciphertext c uploaded by clients participating in tasks at an aggregator1,c2,c3…,cnThen, at this time, since all the gradient update parameters M are encrypted, the aggregator cannot know the specific gradient update parameters uploaded by the other clients. At the moment, the aggregator multiplies the collected ciphertexts pairwise to obtain n pairs of multiplied ciphertexts cm1,2,cm2,3,cm3,4,…,cmn,1And passes it to the task originator.
In the specific implementation of step S15, all the multiplied ciphertexts transmitted by the aggregator are received, all the multiplied ciphertexts are decrypted to obtain respective corresponding plaintext addition values, the plaintext addition value corresponding to each multiplied cipher text is quantized and verified by using the training quality certification based on the consensus mechanism to promote the global model, and a reward is provided to the client that generates the effective gradient update parameter;
specifically, after receiving the multiplied ciphertexts uploaded by the aggregator, the task initiator decrypts all the multiplied ciphertexts one by using the private task key sk which is not disclosed to obtain a plaintext addition value, namely D (cm)t,t+1,sk)＝mt+mt+1. Adding the values of all the multiplied ciphertexts after decryption operation and dividing by two to finally obtain the sum of the gradient update parameters provided by all the clients in the current iteration, wherein the specific decryption process is as follows:
for ciphertext
And (4) calculating to obtain a plaintext m:
at this time, even if the task initiator has the private key generated by the Paillier cryptosystem, the specific parameter data of the gradient update provided by the client cannot be directly obtained, and only the data obtained by adding two gradient parameters can be obtained, so that the security of the gradient update parameters is protected to the greatest extent. In the whole transmission of the gradient updating parameters, under the condition of reasonably keeping the public and private task keys, no third party can obtain the specific gradient updating parameter value of a certain client except for knowing the specific parameter value of the client during local training, so that the privacy protection is realized to the maximum extent.
And the task initiator performs Proof of Training Quality (PoQ) based on a consensus mechanism on each pair of added gradient parameters to quantize and verify the improvement of the global model by the added value of the plaintext corresponding to each multiplied ciphertext. In particular, in classification during training, accuracy is represented by the fraction of correctly classified records, measured by Mean Absolute Error (MAE):
wherein f (x)i) Is a model miPredicted value of (a), yiIs the true value of the record, model miThe lower the MAE of (a), the higher the accuracy of m. The task initiator marks the gradient updating parameters of which the MAE is higher than a set threshold value as invalid models, otherwise marks the gradient updating parameters as valid models, and resets the trust b of the two clients according to the proportion provided by the historical valid models of the nodes.
An ethernet award may be provided to the client that generated the valid gradient update parameter.
It should be noted that, regardless of the MAE result, all the models uploaded by the clients in the current iteration are recorded, and as the number of iterations increases, the MAE dynamically decreases.
After providing the reward to the client generating the valid local model, further comprising: packing a block, uplink-linking the block, and broadcasting to inform all clients, wherein the block at least comprises a hash value of a local model parameter provided by the client.
Specifically, the block may include the following: the size of the block is (h + δ N)D) Where h is the chunk header, including the hash value of the previous chunk, the timestamp generated by the chunk, etc., NDThe number of mobile devices providing a local model is represented, δ is a nine-tuple (task id, communication iteration round number i, device identifier id, hash value h of model parameters provided by the node, MAE corresponding to the model provided by the node, whether the model is an identifier m of an effective model, reward p of the node, trust degree b of the node, and uploading timestamp of the model of the node), wherein the device identifier is a public key of different devices based on unlicensed encryption, and a task initiator verifies and packages received transaction information and finally enters a chain.
In a specific implementation of step S16, updating the global model using a federated learning aggregation algorithm according to the effective gradient update parameter;
the aggregation algorithm of the present example may employ a FedAvg aggregation algorithm, a Practical Secure aggregation algorithm, a NIKE-based Fast aggregation algorithm, and a PrivFL algorithm, and the like.
The embodiment provides a federated learning method to realize model training, and federated learning can allow a user to jointly obtain the benefit of training a shared model from rich data, and meanwhile, the user does not need to store the data by using a central cloud. In other words, the data of the user does not leave the local mobile terminal, so that the privacy security of the user can be protected to the maximum extent. The training task is solved by a loose federation of clients coordinated by the task initiator. Each client has a local training data set that is never uploaded to the server, and each client computes an update to the current global model maintained by the server and only communicates this update. In addition, the method is different from the traditional distributed method, and high-level model training can still be realized under the conditions that the user data are not independently distributed and the user data are distributed in a large scale. This method is implemented using the federal Averaging (FedAvg) algorithm: the federal mean algorithm integrates a plurality of deep learning models using a stochastic gradient descent algorithm into a global model. Similar to standalone machine learning, the goal of federal learning is also experience risk minimization, namely:
where n is the sample volume, siDenotes the ith sample, f (x, s)i) Representing the loss function over the model. Suppose there are K local patterns, PkA set of sequence numbers representing sample individuals owned by the kth model. n isk＝|PkI to rewrite the objective function to:
it should be noted that, since the data of each client cannot represent global data, it cannot be considered that
Like f (x), that is, any one local model cannot be used as a global model.
Further, the steps S12 to S16 may be performed through multiple iterations until the global model converges or meets the requirement, and finally the purpose of model training is achieved.
This embodiment refers to one parameter update of the local model as one iteration. B denotes a batch, then the kth local model iteration formula is:
the overall approach taken is summarized as follows: the training process is divided into a plurality of rounds, and C x K (C is more than or equal to 0 and less than or equal to 1) local models are selected from each round to learn the data. The number of epochs of the kth local model in one round is E, the size of batch is B, and the iteration number is Enkand/B. After one round is finished, the parameters of all the local models participating in learning are weighted and averaged to obtain a global model.
It should be noted that the local model here represents a new model obtained by each client training the global model with its own local power quality data in each iteration, that is, the local model, and of course, the local model is not returned here, and the local gradient parameter update (and the local model parameter minus the global model parameter) is returned here.
In the specific implementation of the step S17, the updated global model is used to detect the power quality disturbance data to be detected.
Specifically, the power quality disturbance data to be detected is input into the updated global model, and a detection result is obtained.
Corresponding to the embodiment of the power quality disturbance detection method, the application also provides an embodiment of an information display device.
Fig. 2 is a block diagram illustrating a power quality disturbance detection apparatus according to an exemplary embodiment. Referring to fig. 2, the apparatus is applied to a task initiator, and includes:
the first building module 21 is used for building a private block chain based on intelligent contracts and federal learning, wherein the private block chain consists of a task initiator, an aggregator and a client;
the first verification module 22 is configured to verify the identity of the client by using a non-certificate-based encryption algorithm, and the client after verification can participate in the global model training task;
the creating module 23 is configured to create a public-private task key for the global model training task, and broadcast and distribute the public-private task key to all clients participating in the global model training task;
the first sending module 24 is configured to send a global model to all clients participating in a global model training task, so that the clients receive the global model, train the global model by using local power quality data to obtain gradient update parameters of the current model training, encrypt the gradient update parameters by using the common task key to obtain ciphertexts, send the ciphertexts to the aggregator, so that the aggregator collects the ciphertexts uploaded by all the clients, multiply the ciphertexts of different clients by two to obtain multiplied ciphertexts, and send all the multiplied ciphertexts to the task initiator;
the encryption verification module 25 is configured to receive all the multiplied ciphertexts transmitted by the aggregator, decrypt all the multiplied ciphertexts to obtain respective corresponding plaintext addition values, quantize and verify the lifting of the global model by using the plaintext addition value corresponding to each multiplied cipher text based on a training quality certification based on a consensus mechanism, and provide rewards for clients generating effective gradient update parameters;
an updating module 26, configured to update the global model according to the effective gradient update parameter by using a federal learning aggregation algorithm, so as to obtain a power quality disturbance detection model;
and the detection module 27 is configured to detect the power quality disturbance data to be detected by using the power quality disturbance detection model.
With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
Example 2:
fig. 3 is a flowchart illustrating a power quality disturbance detection method according to an exemplary embodiment, and referring to fig. 3, a power quality disturbance detection method provided by an embodiment of the present invention is applied to a client, and includes:
step S31, constructing a private block chain based on intelligent contracts and federal learning, wherein the private block chain consists of a task initiator, an aggregator and a client;
step S32, receiving the task to initiate identity authentication based on a certificateless encryption algorithm, and participating in a global model training task after the authentication is passed;
step S33, receiving a public and private task key broadcasted by the task initiator, wherein the public and private task key is created by the task initiator for a global model training task;
step S34, receiving the global model sent by the task initiator;
step S35, training the global model by using local power quality data to obtain gradient updating parameters of the model training of the current round;
step S36, encrypting the gradient updating parameter by using a public task key to obtain a ciphertext;
step S37, sending the ciphertext to an aggregator, so that after the aggregator collects the ciphertexts uploaded by all the clients, the aggregator multiplies the ciphertexts of different clients by each other to obtain multiplied ciphertexts, and all the calculated multiplied ciphertexts are transmitted to a task initiator so that after the task initiator receives all the multiplied ciphertexts transmitted by the aggregator, decrypting all the multiplied ciphertexts to obtain plaintext addition values, quantizing and verifying the promotion of the global model by using the plaintext addition value corresponding to each multiplied cipher text based on the training quality certification of the consensus mechanism, providing rewards for a client side generating effective gradient updating parameters, updating the global model by using a federal learning aggregation algorithm according to the effective gradient updating parameters to obtain a power quality disturbance detection model, and detecting power quality disturbance data to be detected by using the power quality disturbance detection model.
For the detailed description of the above steps S31-S37, reference may be made to steps S11-S17 in embodiment 1, which are not repeated here.
Corresponding to the embodiment of the power quality disturbance detection method, the application also provides an embodiment of an information display device.
Fig. 4 is a block diagram illustrating a power quality disturbance detection device in accordance with an exemplary embodiment. Referring to fig. 4, the apparatus is applied to a client, and includes:
a second construction module 41, configured to construct a private block chain based on an intelligent contract and federal learning, where the private block chain is composed of a task initiator, an aggregator, and a client;
the second verification module 42 is configured to accept that the task initiates identity verification using a non-certified encryption algorithm, and can participate in the global model training task after the verification passes;
a first receiving module 43, configured to receive a public-private task key broadcasted by the task initiator, where the public-private task key is created by the task initiator for a global model training task;
a second receiving module 44, configured to receive the global model sent by the task initiator;
the training module 45 is used for training the global model by using local power quality data to obtain gradient updating parameters of the model training of the current round;
the encryption module 46 is configured to encrypt the gradient update parameter by using a public task key to obtain a ciphertext;
a second sending module 47, configured to send the ciphertext to the aggregator, so that after the aggregator collects the ciphertexts uploaded by all the clients, the aggregator multiplies every two ciphertexts of different clients to obtain multiplied ciphertexts, and sends all the calculated multiplied ciphertexts to the task initiator, so that after the task initiator receives all the multiplied ciphertexts sent by the aggregator, all the multiplied ciphertexts are decrypted to obtain a plaintext sum value, the increase of the global model is quantified and verified by using a training quality certificate based on a consensus mechanism to add the plaintext sum value corresponding to each multiplied ciphertext, so as to provide a reward for the client that generates an effective gradient update parameter, and according to the effective gradient update parameter, the global model is updated by using a federal learned aggregation algorithm to obtain an electric energy quality disturbance detection model, and detecting the power quality disturbance data to be detected by using the power quality disturbance detection model.
With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
Example 3:
fig. 5 is a flowchart illustrating a power quality disturbance detection method according to an exemplary embodiment, and referring to fig. 5, a power quality disturbance detection method provided by an embodiment of the present invention is applied to an aggregator, and includes:
step S51, constructing a private block chain based on intelligent contracts and federal learning, wherein the private block chain consists of a task initiator, an aggregator and a client;
step S52, receiving all ciphertexts sent by the client, wherein the client initiates identity verification based on a certificateless encryption algorithm for receiving the task, and can participate in a global model training task after the verification is passed; the cipher text is obtained by encrypting a gradient updating parameter by the client by using a public task key, the public task key is created by the task initiator for a global model training task, and the gradient updating parameter is obtained by the client receiving a global model sent by the task initiator and training the global model by using local power quality data;
step S53, carrying out multiplication operation on the ciphertexts of different clients pairwise to obtain multiplied ciphertexts;
and step S54, transmitting all the calculated multiplication ciphertexts to a task initiator, so that the task initiator decrypts all the multiplication ciphertexts after receiving all the multiplication ciphertexts to obtain plain text added values, quantizing and verifying the promotion of the global model by using the plain text added values corresponding to each multiplication cipher text based on a training quality certificate of a consensus mechanism, providing rewards for clients generating effective gradient update parameters, updating the global model by using a federated learning aggregation algorithm according to the effective gradient update parameters to obtain an electric energy quality disturbance detection model, and detecting the electric energy quality disturbance data to be detected by using the electric energy quality disturbance detection model.
For the detailed description of the above steps S51-S54, reference may be made to steps S11-S17 in embodiment 1, which are not repeated here.
Corresponding to the embodiment of the power quality disturbance detection method, the application also provides an embodiment of an information display device.
Fig. 6 is a block diagram illustrating a power quality disturbance detection apparatus according to an exemplary embodiment. Referring to fig. 6, the apparatus is applied to an aggregator comprising:
the third building module 61 is used for building a private block chain based on intelligent contracts and federal learning, and the private block chain consists of a task initiator, an aggregator and a client;
a third receiving module 62, configured to receive ciphertext sent by all the clients, where the clients initiate authentication using a non-certified encryption algorithm to receive the task, and can participate in the global model training task after the authentication passes; the cipher text is obtained by encrypting a gradient updating parameter by the client by using a public task key, the public task key is created by the task initiator for a global model training task, and the gradient updating parameter is obtained by the client receiving a global model sent by the task initiator and training the global model by using local power quality data;
a multiplication operation module 63, configured to multiply the ciphertexts of different clients by two to obtain multiplied ciphertexts;
the third sending module 64 is configured to transmit all the calculated multiplied ciphertexts to the task initiator, so that the task initiator decrypts all the multiplied ciphertexts after receiving all the multiplied ciphertexts to obtain a plaintext addition value thereof, quantizes and verifies the lifting of the global model by using the plaintext addition value corresponding to each multiplied cipher text based on a training quality certificate of a consensus mechanism, provides a reward to a client that generates an effective gradient update parameter, updates the global model by using a federal learning aggregation algorithm according to the effective gradient update parameter to obtain an electric energy quality disturbance detection model, and detects electric energy quality disturbance data to be detected by using the electric energy quality disturbance detection model.
With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
For the device embodiments, since they substantially correspond to the method embodiments, reference may be made to the partial description of the method embodiments for relevant points. The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of the application. One of ordinary skill in the art can understand and implement it without inventive effort.
Correspondingly, the present application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; when executed by the one or more processors, cause the one or more processors to implement a power quality disturbance detection method as described above.
Accordingly, the present application also provides a computer readable storage medium having stored thereon computer instructions, wherein the instructions, when executed by a processor, implement a power quality disturbance detection method as described above.
Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.
It will be understood that the present application is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the application is limited only by the appended claims.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111208979.8A CN113901412A (en) | 2021-10-18 | 2021-10-18 | Power quality disturbance detection method and device, electronic device, storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111208979.8A CN113901412A (en) | 2021-10-18 | 2021-10-18 | Power quality disturbance detection method and device, electronic device, storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113901412A true CN113901412A (en) | 2022-01-07 |
Family
ID=79192668
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111208979.8A Pending CN113901412A (en) | 2021-10-18 | 2021-10-18 | Power quality disturbance detection method and device, electronic device, storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113901412A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114529016A (en) * | 2022-02-22 | 2022-05-24 | 光大科技有限公司 | Model aggregation processing method and device, storage medium and electronic device |
| CN114580011A (en) * | 2022-01-29 | 2022-06-03 | 国网青海省电力公司电力科学研究院 | Power facility security situation sensing method and system based on federal privacy training |
| CN114662618A (en) * | 2022-05-23 | 2022-06-24 | 湖南工商大学 | A fault diagnosis method, device and related equipment based on federated learning |
| CN114707606A (en) * | 2022-04-11 | 2022-07-05 | 中国电信股份有限公司 | Data processing method and device based on federal learning, equipment and storage medium |
| CN116307138A (en) * | 2023-03-01 | 2023-06-23 | 中国科学院上海微系统与信息技术研究所 | Electrical fire early warning method, device, terminal and medium based on federal learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112861152A (en) * | 2021-02-08 | 2021-05-28 | 北京航空航天大学 | Federal learning incentive method and system based on permit chain |
| WO2021120676A1 (en) * | 2020-06-30 | 2021-06-24 | 平安科技（深圳）有限公司 | Model training method for federated learning network, and related device |
| CN113344222A (en) * | 2021-06-29 | 2021-09-03 | 福建师范大学 | Safe and credible federal learning mechanism based on block chain |
| CN113434873A (en) * | 2021-06-01 | 2021-09-24 | 内蒙古大学 | Federal learning privacy protection method based on homomorphic encryption |
-
2021
- 2021-10-18 CN CN202111208979.8A patent/CN113901412A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021120676A1 (en) * | 2020-06-30 | 2021-06-24 | 平安科技（深圳）有限公司 | Model training method for federated learning network, and related device |
| CN112861152A (en) * | 2021-02-08 | 2021-05-28 | 北京航空航天大学 | Federal learning incentive method and system based on permit chain |
| CN113434873A (en) * | 2021-06-01 | 2021-09-24 | 内蒙古大学 | Federal learning privacy protection method based on homomorphic encryption |
| CN113344222A (en) * | 2021-06-29 | 2021-09-03 | 福建师范大学 | Safe and credible federal learning mechanism based on block chain |
Non-Patent Citations (2)
| Title |
|---|
| 于华楠: "基于改进堆叠去噪自动编码器的电能质量扰动分类方法", 电力信息与通信技术, 30 September 2021 (2021-09-30), pages 2 - 3 * |
| 王健宗: "深入浅出联邦学习 原理与实践", 31 May 2021, 机械工业出版社 , pages: 53 - 54 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114580011A (en) * | 2022-01-29 | 2022-06-03 | 国网青海省电力公司电力科学研究院 | Power facility security situation sensing method and system based on federal privacy training |
| CN114529016A (en) * | 2022-02-22 | 2022-05-24 | 光大科技有限公司 | Model aggregation processing method and device, storage medium and electronic device |
| CN114529016B (en) * | 2022-02-22 | 2025-05-30 | 光大科技有限公司 | Model aggregation processing method, device, storage medium and electronic device |
| CN114707606A (en) * | 2022-04-11 | 2022-07-05 | 中国电信股份有限公司 | Data processing method and device based on federal learning, equipment and storage medium |
| CN114707606B (en) * | 2022-04-11 | 2023-12-22 | 中国电信股份有限公司 | Data processing method and device based on federal learning, equipment and storage medium |
| CN114662618A (en) * | 2022-05-23 | 2022-06-24 | 湖南工商大学 | A fault diagnosis method, device and related equipment based on federated learning |
| CN116307138A (en) * | 2023-03-01 | 2023-06-23 | 中国科学院上海微系统与信息技术研究所 | Electrical fire early warning method, device, terminal and medium based on federal learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhuang et al. | Blockchain for cybersecurity in smart grid: A comprehensive survey | |
| Ferrag et al. | Cyber security for fog-based smart grid SCADA systems: Solutions and challenges | |
| CN113901412A (en) | Power quality disturbance detection method and device, electronic device, storage medium | |
| Khan et al. | An elliptic curve cryptography based mutual authentication scheme for smart grid communications using biometric approach | |
| Jiang et al. | Energy big data: A survey | |
| Yan et al. | Distributed consensus-based economic dispatch in power grids using the paillier cryptosystem | |
| Wang et al. | Identity‐based remote data possession checking in public clouds | |
| Luo et al. | Cloud-based information infrastructure for next-generation power grid: Conception, architecture, and applications | |
| Bao et al. | A new differentially private data aggregation with fault tolerance for smart grid communications | |
| CN113691380B (en) | Multidimensional private data aggregation method in smart power grid | |
| CN114338045A (en) | Information data verifiability safety sharing method and system based on block chain and federal learning | |
| Ahn et al. | Synchronized aggregate signatures: new definitions, constructions and applications | |
| KR101437033B1 (en) | Method for authenticating low efficiency device | |
| Wang et al. | An efficient privacy‐preserving aggregation and billing protocol for smart grid | |
| Iyer | Cyber security for smart grid, cryptography, and privacy | |
| Kim et al. | Device authentication protocol for smart grid systems using homomorphic hash | |
| Natanzi et al. | Secure distributed controllers in SDN based on ECC public key infrastructure | |
| Sui et al. | An efficient signcryption protocol for hop-by-hop data aggregations in smart grids | |
| Jiang et al. | Lightweight data security protection method for AMI in power Internet of Things | |
| CN117614604A (en) | Power grid data privacy protection and sharing system and method based on blockchain technology | |
| Aggarwal et al. | Smart grid | |
| Huang et al. | A lightweight and fault-tolerable data aggregation scheme for privacy-friendly smart grids environment | |
| Wang et al. | An ID-based online/offline signature scheme without random oracles for wireless sensor networks | |
| Ibrahem et al. | Privacy-preserving, lightweight, and decentralized load forecasting in smart grid ami networks | |
| Wang et al. | Simulatable and secure certificate‐based threshold signature without pairings |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |