A power grid segmentation learning method and system based on a V2G network

Abstract

The present application relates to V2G network optimization technology, disclose a kind of based on V2G network's power grid segmentation learning method, comprising: client obtains real-time transaction information, and according to sensitivity function, real-time transaction information is carried out sensitivity classification, obtain multiple hierarchical sensitivity transaction information;Sensitivity attribute in multiple hierarchical sensitivity transaction information is obtained, and noise adding operation is carried out to sensitive attribute, and the transaction information after adding noise is stored to cloud storage server;Cloud storage server receives the transaction information after adding noise, and parameter optimization training is carried out to LSTM-FCNN multi-task model, and target LSTM-FCNN multi-task model is obtained;Cloud storage server and aggregator utilize segmentation learning method to target LSTM-FCNN multi-task model collaborative training, obtain based on V2G network's power grid multi-task model.The present application also proposes a kind of based on V2G network's power grid segmentation learning system.The present application can improve the optimization efficiency of real-time transaction information in V2G network and improve the stability of V2G network.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of V2G network optimization, and particularly relates to a power grid segmentation learning method and system based on a V2G network. BACKGROUND

[0002] The existing V2G (Vehicle-to-Grid) network is a vehicle-grid interaction technology, which allows electric vehicles (EVs) to send power back to the power grid through a bidirectional charger. The payment transaction of the electric vehicle pays more attention to ensuring the safety of user sensitive information while realizing data sharing, so most of the current payment mechanisms are based on blockchains.

[0003] The electric vehicle accesses the V2G network through special facilities in the parking lot and provides auxiliary services to the smart grid under the control of the local aggregator. The electric vehicle has three states: charging, discharging, and distributed discharging. Charging is initiated by the electric vehicle, and power is transmitted from the smart grid to the electric vehicle. The electric vehicle needs to pay electricity charges to the smart grid. Discharge is initiated by the smart grid when the power grid is overloaded during the user demand peak period. The electric vehicle can alleviate power shortages by discharging to the grid and obtain economic rewards from the smart grid. Distributed discharging is initiated by one or more electric vehicles when the battery capacity is below a certain threshold. Unlike the charging state, the electric vehicle with sufficient power will temporarily act as a power supplier for the electric vehicle with insufficient power, and the latter needs to pay electricity charges to the supplier.

[0004] In the actual payment mechanism system, the account registration process mainly submits the user-generated account to the RA (Registered Account) for authentication and registration. The user creates a pair of keys and then sends them to the RA together with the identity information. After checking the identity, the RA replies to the user with a signature generated by the private key of the RA. Then the RA generates a legal account for the user. Each user can use the public key of the RA to verify the account. The payment execution process is actually a blockchain transaction, that is, verifying and writing the transaction into the global ledger. The payee can use different accounts for each transaction, then sends the registered account and unit price to the payer, the payer checks the global ledger, calculates the total price, and after payment, sends the transaction id to the payee. If the transaction is correct, the payee starts to supply power.

[0005] In the current V2G transaction process, due to the complexity of multi-party cooperation and information exchange, sensitive information such as user identity, transaction amount, charging location, etc. is vulnerable to attack or accidental disclosure in transmission or storage, and there is a lack of effective hierarchical protection measures. And in the process of information interaction and data processing of the existing V2G network, inefficient optimization strategies and privacy protection technologies are usually used, and the multi-party devices in the V2G network do not use the split learning method for multi-task model collaborative training, which is prone to cause the real-time transaction information in the V2G network to run inefficiently due to the low efficiency of model training in multi-party devices and the exposure of sensitive information, thereby affecting the stability and the integrity of the life cycle of the entire network. SUMMARY

[0006] The application provides a power grid split learning method and system based on a V2G network, which can improve the optimization efficiency of real-time transaction information in the V2G network and improve the stability of the V2G network.

[0007] To achieve the above-mentioned purpose, the application provides a power grid split learning method based on a V2G network, which comprises:

[0008] The client obtains real-time transaction information from multiple information sources, and classifies the real-time transaction information according to a sensitivity function to obtain multiple transaction information with hierarchical sensitivity, wherein the real-time transaction information includes charging time t c , power consumption E, transaction amount A, price fluctuation P, power grid load L, user ID (UID), charging station number (SID), and transaction ID (TID);

[0009] Sensitive attributes in the multiple transaction information with hierarchical sensitivity are obtained, and a noise addition operation is performed on the sensitive attributes, and the transaction information after adding noise is stored in a cloud storage server;

[0010] The cloud storage server receives the transaction information after adding noise, and uses the transaction information after adding noise to perform parameter optimization training on the LSTM-FCNN multi-task model to obtain a target LSTM-FCNN multi-task model;

[0011] The cloud storage server and the aggregator use a split learning method to collaboratively train the target LSTM-FCNN multi-task model to obtain a power grid multi-task model based on a V2G network.

[0012] Optionally, before classifying the real-time transaction information according to the sensitivity function, the real-time transaction information is subjected to data standardization processing and feature extraction, wherein the data standardization processing and feature extraction of the real-time transaction information comprises:

[0013] The redundant information and error information in the real-time transaction information are removed, and different dimensional implementation transaction information is unified to the same dimensional range by using a data standardization formula.

[0014] The amount features and mode features in the real-time transaction information are extracted, wherein the amount features include transaction amount features and electricity price fluctuation features, and the mode features include charging duration features, electricity consumption features, power grid load features, user charging frequency features, charging habit features, transaction time features and data acquisition frequency features.

[0015] Optionally, the sensitivity grading of the real-time transaction information according to the sensitivity function comprises:

[0016] The comprehensive sensitivity of each feature in the real-time transaction information is calculated by using the sensitivity function.

[0017] The real-time transaction information is graded according to the comprehensive sensitivity and a preset membership function.

[0018] The real-time transaction information is graded according to the sensitivity grading, to obtain transaction information with multiple graded sensitivities.

[0019] Optionally, the comprehensive sensitivity of each feature in the real-time transaction information is calculated by using the sensitivity function, and the method comprises:

[0020] The comprehensive sensitivity S(R) of each feature is calculated by using the following formula:

[0021]

[0022] wherein w i is a feature weight of the i th feature, i is a subscript index of the feature, S(f i ) is a sensitivity score, wherein n is a total number of features, IG(f i ) is a contribution of the feature to the sensitivity classification, IG(f i ) = H(S) - H(S|f i ), H(S) is an entropy of the comprehensive sensitivity, and H(S|f i ) is an entropy of the comprehensive sensitivity given the feature.

[0023] Optionally, the real-time transaction information is graded according to the sensitivity grading, to obtain transaction information with multiple graded sensitivities, and the method comprises:

[0024] The real-time transaction information is graded into low-sensitivity information, medium-sensitivity information and high-sensitivity information by using the sensitivity grading.

[0025] The low-sensitivity information is stored in a blockchain.

[0026] Use smart contracts to set first access permissions for medium-sensitivity information, and store the medium-sensitivity information with first access permissions in the blockchain;

[0027] Highly sensitive information is generated using zero-knowledge proof generation technology, and a second access permission is set using a smart contract. The medium-sensitivity information under the second access permission is then stored in the blockchain.

[0028] Optionally, the step of performing a noise-adding operation on sensitive attributes and storing the noise-added transaction information to a cloud storage server includes:

[0029] Calculate the global sensitivity of sensitive attributes and set the privacy budget parameter ε;

[0030] Noise is added to sensitive properties using the Laplace mechanism; the specific steps for adding noise to sensitive properties using the Laplace mechanism are as follows:

[0031] For the original transaction information, noise is generated. Where Δf is the global sensitivity;

[0032] Calculate the protected transaction information x ′ =x + η;

[0033] The transaction information after adding noise x ′ The original transaction information x is replaced to obtain the transaction information with noise added, and the transaction information with noise added is uploaded and stored on the cloud storage server.

[0034] Optionally, the cloud storage server and aggregator utilize a segmentation learning method to collaboratively train the target LSTM-FCNN multi-task model, including:

[0035] The cloud storage server uses the gradient descent update algorithm to update the network parameters of the input layer and LSTM layer in the target hybrid LSTM-FCNN model, and then sends the target hybrid LSTM-FCNN model to the aggregator;

[0036] The aggregator uses the gradient descent update algorithm to update the network parameters of the fully connected layer and the output layer in the target hybrid LSTM-FCNN model. After the model parameters are updated, a power grid multi-task model based on V2G network is obtained.

[0037] To address the aforementioned issues, this invention also provides a system for a power grid segmentation learning method based on V2G networks, the system comprising: a cloud storage server, an aggregator, and a client.

[0038] Optionally, the aggregator determines sensitive attributes of the transaction information, and updates model parameters of full connection layers and output layers in the hybrid LSTM-FCNN model using a gradient descent update algorithm.

[0039] Optionally, the client communicates with the electric vehicle device in the V2G network in a bidirectional manner, the client communicates with the cloud storage server in a bidirectional manner, and the cloud storage server provides cloud services to the electric vehicle device in the V2G network.

[0040] According to the sensitivity function, the real-time transaction information is classified according to the sensitivity, the information with different sensitivities in the transaction information can be classified and protected, the management and processing efficiency of the transaction information can be improved, in addition, the sensitive attributes in the transaction information are obtained, and the noise adding operation is performed on the sensitive attributes, so that the data security of the transaction information in the transmission process can be realized, furthermore, the hybrid LSTM-FCNN model is trained to obtain the power grid multi-task model based on the V2G network, so that the optimization efficiency of the real-time transaction information in the V2G network can be realized, and the stability of the V2G network can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 A flowchart of the power grid segmentation learning method based on the V2G network provided by an embodiment of the present application is shown in the figure.

[0042] Figure 2 An information sensitivity classification diagram of the power grid segmentation learning method based on the V2G network provided by an embodiment of the present application is shown in the figure.

[0043] Figure 3 A noise adding transmission diagram of the power grid segmentation learning method based on the V2G network provided by an embodiment of the present application is shown in the figure.

[0044] Figure 4 A cloud storage server differential privacy learning diagram of the power grid segmentation learning method based on the V2G network provided by an embodiment of the present application is shown in the figure.

[0045] Figure 5 A cloud service providing diagram of the power grid segmentation learning method based on the V2G network provided by an embodiment of the present application is shown in the figure.

[0046] Figure 6 A system diagram of the power grid segmentation learning device based on the V2G network provided by an embodiment of the present application is shown in the figure.

[0047] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0048] It should be understood that the specific embodiments described herein are merely intended to explain the present application, and are not intended to limit the present application.

[0049] The embodiment of the present application provides a power grid segmentation learning method based on a V2G network. The execution subject of the power grid segmentation learning method based on the V2G network includes but is not limited to at least one of electronic devices such as a server, a terminal and the like that can be configured to execute the method provided by the embodiment of the present application. In other words, the power grid segmentation learning method based on the V2G network can be executed by software or hardware installed in a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to a single server, a server cluster, a cloud server or a cloud server cluster and the like. The server can be a stand-alone server, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0050] Referring to Figure 1 Fig. 1 is a flowchart of a power grid segmentation learning method based on a V2G network provided by an embodiment of the present application. In the embodiment, the power grid segmentation learning method based on the V2G network includes the following steps.

[0051] S1, the client obtains real-time transaction information from a plurality of information sources, and classifies the real-time transaction information according to a sensitivity function to obtain a plurality of transaction information with classified sensitivity, wherein the real-time transaction information includes charging time t c , power consumption E, transaction amount A, price fluctuation P, power grid load L, user ID (UID), charging station number (SID), and transaction ID (TID).

[0052] In the embodiment of the present application, the information source includes but is not limited to an electric vehicle sensor, a power grid load monitoring system, environmental data and the like that can collect transaction information.

[0053] In the embodiment of the present application, the sensitivity function refers to a function for defining the sensitivity score of a feature, and the score range is [0, 1], and 1 represents high sensitivity. For example, in the embodiment of the present application, f i is a feature, and x i is a feature value.

[0054] When f i is an ID type, if f i = UID, the sensitivity function is S(f i , x i ) = 1; otherwise, the sensitivity function S(f i , x i ) = 0.2.

[0055] When f i is the amount type, the sensitivity function S(f i ,x i ) = 0.9

[0056] When f i is the time type, the sensitivity function S(f i ,x i ) = 0.8

[0057] When f i is the power resource type (electricity price, electricity quantity, load), the sensitivity function S(f Wherein, f max is the maximum value of eigenvalue, f min is the minimum value of eigenvalue.

[0058] As an embodiment of the present application, before the real-time transaction information is graded according to the sensitivity function, the real-time transaction information is further subjected to data standardization processing and feature extraction.

[0059] Further, the data standardization processing and feature extraction of the real-time transaction information include:

[0060] Removing redundant information and error information in the real-time transaction information, and using a data standardization formula to unify the different dimension implementation transaction information to the same dimension range;

[0061] Extracting the amount features and mode features in the real-time transaction information, wherein the amount features include transaction amount features and electricity price fluctuation features, and the mode features include charging duration features, electricity consumption features, power grid load features, user charging frequency features, charging habit features, transaction time features and data acquisition frequency features.

[0062] As an embodiment of the present application, the real-time transaction information is graded according to the sensitivity function to obtain a plurality of transaction information with graded sensitivity, including:

[0063] Using the sensitivity function to calculate the comprehensive sensitivity of each feature in the real-time transaction information;

[0064] Grading the real-time transaction information according to the comprehensive sensitivity and a preset membership function;

[0065] Grading the real-time transaction information according to the sensitivity grading to obtain a plurality of transaction information with graded sensitivity.

[0066] Further, the preset membership function of the embodiment of the present application adopts the following membership function:

[0067]

[0068] Wherein, threshold a is the upper limit of low sensitivity, threshold b is the core value of medium sensitivity, and threshold c is the lower limit of high sensitivity.

[0069] Further, the comprehensive sensitivity of each feature in the real-time transaction information is calculated by using the sensitivity function, including:

[0070] The comprehensive sensitivity S(R) of each feature is calculated by using the following formula:

[0071]

[0072] Wherein, w i is the feature weight of the i th feature, i is the index of the feature, S(f i ) is the sensitivity score, Wherein, n is the total number of features, IG(f i ) is the contribution of the feature to the sensitivity classification, IG(f i ) = H(S) - H(S|f i ), H(S) is the entropy of the comprehensive sensitivity, and H(S|f i ) is the entropy of the comprehensive sensitivity given the feature.

[0073] The embodiment of the application also includes calculating the information entropy of each feature, and the following formula is used for calculation:

[0074]

[0075] Wherein, H(f i ) is the feature information entropy, p k is the probability of the feature f i taking the value k.

[0076] Further, the real-time transaction information is classified according to the sensitivity classification, and a plurality of classified sensitive transaction information is obtained, including:

[0077] The real-time transaction information is classified into low sensitivity information, medium sensitivity information and high sensitivity information by using the sensitivity classification;

[0078] The low sensitivity information is stored in the blockchain;

[0079] The medium sensitivity information is set with a first access permission by using the smart contract, and the medium sensitivity information with the first access permission is stored in the blockchain;

[0080] The high sensitivity information is generated by using zero-knowledge proof generation technology to generate high sensitivity zero-knowledge proof, and a second access permission is set by using the smart contract, and the medium sensitivity information with the second access permission is stored in the blockchain.

[0081] S2, acquire sensitive attributes in the transaction information of multiple hierarchical sensitivities, and perform a noise adding operation on the sensitive attributes, and store the transaction information after adding the noise to a cloud storage server.

[0082] In the embodiment of the application, the sensitive attributes refer to charging behavior patterns, charging peak time, power grid load, and price change correlation, etc.

[0083] As an embodiment of the application, the noise adding operation is performed on the sensitive attributes, and the transaction information after adding the noise is stored to the cloud storage server, comprising:

[0084] calculating the global sensitivity of the sensitive attributes, and setting a privacy budget parameter ε;

[0085] adding noise to the sensitive attributes by using a Laplace mechanism; wherein the specific steps of adding noise to the sensitive attributes by using the Laplace mechanism are as follows:

[0086] generating noise for the original transaction information wherein Δf is the global sensitivity;

[0087] calculating the protected transaction information x ′ =x+η;

[0088] substituting the transaction information x ′ after adding the noise for the original transaction information x, obtaining the transaction information after adding the noise, and uploading and storing the transaction information after adding the noise in the cloud storage server.

[0089] Exemplarily, the global sensitivity of the sensitive attributes can be acquired by using the following implementation method:

[0090] For example, D and D' are adjacent data sets;

[0091] Method one, proportional function: wherein count A (D) is the number of records satisfying condition A, count B (D) is the number of records satisfying condition B. The global sensitivity wherein c A =count A (D'), c B =count B (D'), and the global sensitivity

[0092] Method two, maximum value function: The global sensitivity Δf=max|max(D)-max(D')|=Δx;

[0093] Method three, median or percentile function: f(D) = mid(D), global sensitivity where Δx is the maximum range of variation of a single record, and n is the dataset size;

[0094] Method four, distance function: Global sensitivity Δf = Δd, where Δd is the maximum distance variation;

[0095] Method five, complex aggregate function: f(D) = ∑ x∈D g(x), g(x) is a nonlinear function of a single record x, global sensitivity

[0096] In the embodiments of the present application, the privacy budget parameter epsilon can control the strength of privacy protection, and a smaller epsilon value provides stronger privacy protection.

[0097] S3, the cloud storage server receives the transaction information added with noise, and uses the transaction information added with noise to perform parameter optimization training on the LSTM-FCNN multi-task model to obtain a target LSTM-FCNN multi-task model.

[0098] In the embodiments of the present application, the LSTM-FCNN multi-task model is a deep learning model combining a long short-term memory network (LSTM) and a fully convolutional neural network (FCN), and the LSTM-FCNN multi-task model includes an input layer, 2 LSTM layers, 2 fully connected layers, and an output layer, and can realize functions such as charging demand prediction, power grid load prediction, charging behavior classification, and power grid state classification.

[0099] In the embodiments of the present application, an Adam optimizer with adaptive learning rate can be used to perform parameter optimization training on the LSTM-FCNN multi-task model, wherein the Adam optimizer with adaptive learning rate calculates its own learning rate for each parameter, so that Adam can have different learning speeds on different parameters, thereby accelerating the convergence speed of the LSTM-FCNN multi-task model.

[0100] S4, the cloud storage server and the aggregator use a split learning method to collaboratively train the target LSTM-FCNN multi-task model to obtain a power grid multi-task model based on a V2G network.

[0101] As an embodiment of the present application, the cloud storage server and the aggregator use a split learning method to collaboratively train the target LSTM-FCNN multi-task model, including:

[0102] The cloud storage server updates the network parameters of the input layer and the LSTM layer in the target hybrid LSTM-FCNN model using a gradient descent update algorithm, and sends the target hybrid LSTM-FCNN model to the aggregator.

[0103] The aggregator updates the network parameters of the full connection layer and the output layer in the target hybrid LSTM-FCNN model by using a gradient descent update algorithm, and obtains the power grid multitask model based on the V2G network after the model parameter update.

[0104] The embodiment of the application updates the input layer and the LSTM layer parameters in the target hybrid LSTM-FCNN model by using a gradient descent update algorithm, and can adopt the following implementation steps:

[0105] Gradient clipping: clip the gradient of each sample to a preset norm range to prevent a single sample from having an excessive impact on the gradient, wherein the preset norm range is a pre-set fixed value, denoted as O, and a suitable O value can be selected through experiments and cross validation;

[0106] Adding noise: adding noise ζ ~ N(0, σ 2 I) to the accumulated gradient, wherein σ controls the noise intensity, and I is a unit matrix.

[0107] Exemplarily, the input layer and the LSTM layer parameter update in the target hybrid LSTM-FCNN model can adopt the following gradient descent update algorithm update:

[0108] Let W c be the weight of the input layer and the LSTM layer parameters in the client, b c be the bias of the input layer and the LSTM layer parameters in the client, g c be the gradient received by the client with noise, η 1 be the first learning rate. The update formula is as follows: wherein is the weight gradient, is the bias gradient.

[0109] Exemplarily, the aggregator can update the model parameters of the full connection layer and the output layer in the target hybrid LSTM-FCNN model by using the following gradient descent update algorithm:

[0110] Let W s be the weight of the full connection layer and the output layer in the cloud storage server, b s be the bias of the full connection layer and the output layer in the cloud storage server, g s be the gradient calculated by the cloud storage server. η 2 be the second learning rate. U is the gradient clipping threshold, and ζ is the added noise.

[0111] The update formula is as follows:

[0112] where clip(g s , U) denotes that the gradient g s is limited within U, is the clipped and noisy weight gradient, is the clipped and noisy bias gradient.

[0113] In the embodiments of the present application, the input layer and the LSTM layer are responsible for preliminary feature extraction and time series data processing, and their functions are feature extraction, dimension compression, and data transmission.

[0114] In the embodiments of the present application, the full connection layer and the output layer are responsible for further feature fusion and final prediction, and their functions are feature fusion, final prediction, loss calculation, and gradient transmission.

[0115] After the aggregator updates the model parameters of the full connection layer and the output layer in the target hybrid LSTM-FCNN model by using the gradient descent update algorithm, the present application further comprises receiving the model parameters of the full connection layer and the output layer by using the cloud storage server, and completing the prediction output of the model by using the power grid multi-task model based on the V2G network, which can adopt the following implementation steps:

[0116] Full connection layer 1→full connection layer 2→output layer→loss calculation (the latter half of the processing generates a prediction output, and the difference between the prediction output and the real label is calculated by using the selected loss function according to the task type)

[0117] Calculate the loss function and calculate the gradient by back propagation (calculate the gradient based on the loss function, perform gradient clipping and add noise, and finally update the model parameters)

[0118] The calculation of the loss function can determine the loss function according to the task type:

[0119] I. Loss function for regression task (such as charging demand prediction and power grid load prediction): where N is the number of samples, y z is the real value of the zth sample, is the prediction value of the zth sample.

[0120] II. Loss function for classification task (such as charging behavior classification and power grid state classification):

[0121] 1. Binary classification:

[0122] 2. Multi-classification:

[0123] where C is the number of classes, y z,cis the true label of the z-th sample in the c-th class (usually 0 or 1), is the probability of the c-th class predicted by the model.

[0124] In the embodiments of the present application, the power grid multitask model based on the V2G network can realize functions such as charging demand prediction, power grid load prediction, charging behavior classification, and power grid state classification.

[0125] After obtaining the power grid multitask model based on the V2G network, the embodiments of the present application further include cloud service application of the power grid multitask model based on the V2G network, which is implemented by the following implementation steps:

[0126] Step 1: Cloud service platform architecture design.

[0127] Multi-layer architecture design:

[0128] Data layer: store differential privacy data from electric vehicles and smart grids.

[0129] Service layer: provide data processing, analysis, and machine learning model training functions.

[0130] Interface layer: provide a unified API interface for electric vehicles and power grids, supporting data upload, model invocation, and result acquisition.

[0131] Security layer: integrate differential privacy, identity verification, and permission management mechanisms to ensure the security of data and services.

[0132] Step 2: Service interface and API design.

[0133] Unified interface standard: design RESTful API to provide standardized service invocation methods; define clear request and response formats to support multiple data types and operations.

[0134] In the embodiments of the present application, RESTful API is designed: HTTP methods (GET, POST, PUT, DELETE); request uses JSON format, containing necessary parameters and data; response adopts a unified JSON structure, including status code, message, and data.

[0135] Identity authentication and authorization: integrate OAuth 2.0 authentication mechanism to ensure that only authorized electric vehicles and power grid devices can access the cloud service.

[0136] Implementation steps: 1, set up an authorization server; 2, client authentication (get access token); 3, resource access; 4, token verification.

[0137] Service Catalog and Documentation: Provide detailed API documentation and usage guidelines for developers and device integrators to integrate cloud services; establish a service catalog listing available model services, data analysis services, etc.

[0138] Step 3: Cloud Service Application for Electric Vehicles and Power Grid

[0139] Electric Vehicle End Application:

[0140] (1) Intelligent Charging Recommendation:

[0141] (i) Electric vehicles obtain optimal charging time and location recommendations by invoking cloud storage server model services.

[0142] (ii) Provide personalized charging solutions based on real-time electricity prices, charging station load, and user habits.

[0143] (2) Energy Consumption Optimization:

[0144] Optimize electric vehicle energy consumption management strategies using cloud storage server analysis results to improve battery usage efficiency.

[0145] Power Grid End Application:

[0146] (1) Load Forecasting and Dispatching:

[0147] (i) The power grid invokes cloud services to obtain electric vehicle charging demand forecasts, optimizing power dispatching.

[0148] (ii) Anticipate charging peaks to balance grid load and avoid overload.

[0149] (2) Demand Response Management:

[0150] (i) Implement demand response strategies based on cloud storage server analysis to guide electric vehicles to charge during off-peak hours.

[0151] (ii) Provide incentives to encourage users to participate in grid regulation, improving overall energy utilization efficiency.

[0152] Step 4: User Feedback and Service Improvement

[0153] User Feedback Collection:

[0154] (1) Provide channels (such as API feedback interfaces, customer support) to collect opinions and suggestions from electric vehicle and power grid users.

[0155] (2) Classify and analyze feedback to identify service issues and improvement points.

[0156] Service Optimization:

[0157] (1) Continuously optimize cloud service functions and performance based on user feedback and data analysis results.

[0158] (2) Introduce new technologies and algorithms to improve the intelligence and convenience of services.

[0159] Community and ecological construction:

[0160] (1) Establish a developer community to encourage third-party developers to develop new applications based on cloud services.

[0161] (2) Open some data and model interfaces to promote the flourishing development of the V2G network ecosystem.

[0162] The present application can realize the hierarchical protection of information with different sensitivity in transaction information according to the sensitivity function, can improve the management and processing efficiency of transaction information, in addition, the sensitive attributes in the transaction information are obtained, and the noise adding operation is performed on the sensitive attributes, which can realize the data security of the transaction information in the transmission process, furthermore, the mixed LSTM-FCNN model is trained to obtain the power grid multi-task model based on the V2G network, which can realize the optimization efficiency of real-time transaction information in the V2G network, and improve the stability of the V2G network.

[0163] As shown in Figure 2 , it is an information sensitivity grading diagram of the power grid segmentation learning method based on the V2G network provided by an embodiment of the present application.

[0164] As shown in Figure 3 , it is a noise transmission diagram of the power grid segmentation learning method based on the V2G network provided by an embodiment of the present application.

[0165] As shown in Figure 4 , it is a cloud storage server differential privacy learning diagram of the power grid segmentation learning method based on the V2G network provided by an embodiment of the present application.

[0166] As shown in Figure 5 , it is a schematic diagram of providing cloud services of the power grid segmentation learning method based on the V2G network provided by an embodiment of the present application.

[0167] As shown in Figure 6 , it is a system diagram of a power grid segmentation learning device based on the V2G network provided by an embodiment of the present application.

[0168] As an embodiment of the present application, the system comprises a cloud storage server, an aggregator and a client.

[0169] Further, the aggregator determines the sensitive attributes of the transaction information, and updates the model parameters of the full connection layer and the output layer in the target mixed LSTM-FCNN model by using the gradient descent update algorithm.

[0170] Further, the client communicates with the electric vehicle device in the V2G network in a bidirectional manner, the client communicates with the cloud storage server in a bidirectional manner, and the cloud storage server provides cloud services to the electric vehicle device in the V2G network.

[0171] It should be understood that the embodiments are only illustrative, and the patent application scope is not limited by the structure.

[0172] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0173] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any additional reference signs in the claims should not be considered as limiting the claims involved.

[0174] The blockchain referred to in the present application is a new application mode of distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm and other computer technologies. Blockchain, in essence, is a decentralized database, which is a series of data blocks associated using cryptographic methods, each data block contains a batch of network transaction information, used to verify the validity (anti-fake) of the information and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer.

[0175] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. Among them, artificial intelligence (Artificial Intelligence, AI) is to use digital computers or digital computer-controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Theory, method, technology and application system.

[0176] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The plurality of units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, not any specific order.

[0177] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A grid partition learning method based on a V2G network, characterized in that, The method comprises: The client obtains real-time transaction information from multiple information sources, and classifies the real-time transaction information according to a sensitivity function to obtain multiple classified sensitivity transaction information, wherein the real-time transaction information includes charging duration , power consumption E, transaction amount A, electricity price fluctuation P, power grid load L, user ID, charging station number, transaction ID, wherein the sensitivity classification of the real-time transaction information according to the sensitivity function to obtain multiple classified sensitivity transaction information includes: calculating the comprehensive sensitivity of each feature in the real-time transaction information by using the sensitivity function; classifying the real-time transaction information according to the comprehensive sensitivity and a preset membership function; classifying the real-time transaction information according to the sensitivity classification to obtain multiple classified sensitivity transaction information; obtaining sensitive attributes in transaction information of multiple hierarchical sensitivities, and performing a noise addition operation on the sensitive attributes, and storing the transaction information after adding noise to a cloud storage server; the cloud storage server receives the transaction information after adding noise, and uses the transaction information after adding noise to perform parameter optimization training on the LSTM-FCNN multi-task model to obtain a target LSTM-FCNN multi-task model; the cloud storage server and the aggregator use a split learning method to collaboratively train the target LSTM-FCNN multi-task model to obtain a power grid multi-task model based on a V2G network, wherein the cloud storage server and the aggregator use a split learning method to collaboratively train the target LSTM-FCNN multi-task model, which comprises: the cloud storage server uses a gradient descent update algorithm to update the network parameters of the input layer and the LSTM layer in the target hybrid LSTM-FCNN model, and sends the target hybrid LSTM-FCNN model to the aggregator; the aggregator uses a gradient descent update algorithm to update the network parameters of the fully connected layer and the output layer in the target hybrid LSTM-FCNN model, and obtains a power grid multi-task model based on a V2G network after the model parameter update is completed; Wherein, the input layer and the LSTM layer parameter of the target hybrid LSTM-FCNN model are updated by the following gradient descent update algorithm: let be the weight of the model input layer and the LSTM layer parameter in the client, be the bias of the model input layer and the LSTM layer parameter in the client, be the gradient with noise received by the client, be the first learning rate, and the update formula is: , wherein is the weight gradient, is the bias gradient; Wherein, the aggregator uses gradient descent update algorithm to update the model parameters of the full connection layer and the output layer in the target hybrid LSTM-FCNN model, and uses the following gradient descent update algorithm to update: let be the weight of the full connection layer and the output layer in the model in the cloud storage server, be the bias of the full connection layer and the output layer in the model in the cloud storage server, be the gradient calculated by the cloud storage server, be the second learning rate, and U be the gradient clipping threshold, be the added noise; the update formula is as follows: , , wherein, represents that the gradient is limited within U, is the clipped and noisy weight gradient, is the clipped and noisy bias gradient. 2.The V2G network-based power grid segmentation learning method of claim 1, wherein, Before classifying the real-time transaction information according to the sensitivity function, the real-time transaction information is also subjected to data standardization processing and feature extraction, wherein the data standardization processing and feature extraction of the real-time transaction information comprises: removing redundant information and error information in the real-time transaction information, and using a data standardization formula to unify the implementation transaction information of different dimensions to the same dimension range; extracting the amount feature and the mode feature in the real-time transaction information, wherein the amount feature comprises transaction amount feature and price fluctuation feature, and the mode feature comprises charging duration feature, power consumption feature, power grid load feature, user charging frequency feature, charging habit feature, transaction time feature and data acquisition frequency feature. 3.The V2G network-based power grid segmentation learning method of claim 1, wherein, The use of the sensitivity function to calculate the comprehensive sensitivity of each feature in the real-time transaction information comprises: The overall sensitivity of each feature is calculated using the following formula : wherein, is a feature weight of a feature of the th feature, is a subscript index of the feature, is a sensitivity score, wherein, is a total number of features, is a contribution of the feature to the sensitivity classification, , is an entropy of the overall sensitivity, is an entropy of the overall sensitivity for a given feature. 4.The V2G network-based power grid segmentation learning method of claim 1, wherein, The classification of the real-time transaction information according to the sensitivity classification comprises: classifying the real-time transaction information into low sensitivity information, medium sensitivity information and high sensitivity information by using the sensitivity classification; storing the low sensitivity information into the blockchain; setting a first access permission for the medium sensitivity information by using the smart contract, and storing the medium sensitivity information with the first access permission into the blockchain; generating high sensitivity zero-knowledge proof for the high sensitivity information by using zero-knowledge proof generation technology, setting a second access permission by using the smart contract, and storing the medium sensitivity information with the second access permission into the blockchain. 5.The V2G network-based power grid segmentation learning method of claim 1, wherein, The noise addition operation on the sensitive attributes and the storage of the transaction information after adding noise to the cloud storage server comprise: Computing a global sensitivity of a sensitive attribute and setting a privacy budget parameter ; adding noise to the sensitive attributes by using Laplace mechanism; wherein the specific steps of adding noise to the sensitive attributes by using Laplace mechanism are as follows: For raw transaction information, generate noise ), wherein, is the global sensitivity; Computing protected transaction information ; transaction information to which noise is added instead of the original transaction information transaction information to which noise is added is obtained, and the transaction information to which noise is added is uploaded and stored in the cloud storage server. 6.The V2G network-based power grid segmentation learning method of claim 1, wherein, The cloud storage server and the aggregator utilize a split learning method to collaboratively train a target LSTM-FCNN multi-task model, including: The cloud storage server updates the network parameters of the input layer and the LSTM layer in the target hybrid LSTM-FCNN model using the gradient descent update algorithm, and sends the target hybrid LSTM-FCNN model to the aggregator; The aggregator updates the network parameters of the fully connected layer and the output layer in the target hybrid LSTM-FCNN model using the gradient descent update algorithm, and obtains a power grid multi-task model based on the V2G network after the model parameter update is completed.

7. A system for implementing the V2G network-based grid partition learning method according to any one of claims 1 to 6, the system comprising: The cloud storage server, the aggregator, and the client.

8. The system of claim 7, wherein the V2G network-based power grid split learning method comprises: The aggregator determines the sensitive attributes of the transaction information, and updates the model parameters of the fully connected layer and the output layer in the target hybrid LSTM-FCNN model using the gradient descent update algorithm.

9. The system of claim 7, wherein the V2G network-based power grid split learning method comprises: The client communicates with the electric vehicle equipment in the V2G network in a bidirectional manner, the client communicates with the cloud storage server in a bidirectional manner, and the cloud storage server provides cloud services to the electric vehicle equipment in the V2G network.

Images