An engineering knowledge federation learning and trusted sharing method and system for cross-organization collaboration

Abstract

The application discloses an engineering knowledge federation learning and trusted sharing method and system facing cross-organization collaboration. The method comprises the following steps: each organization unit configures a client and converts local engineering data into a local knowledge graph based on a domain ontology; the client sends the parameters of a shared representation layer after training to a federation server; the federation server deploys a global shared model; the federation server calculates a global cosine similarity matrix, identifies abnormal clients from the global cosine similarity matrix by using a density clustering algorithm; the non-abnormal clients are used as to-be-updated clients, the weights of the to-be-updated clients are determined, and the global shared model parameters are updated; the to-be-updated clients update the shared representation layer according to the updated global shared model; and the updated global shared model, the shared representation layer of the client and the local knowledge graph are stored in a blockchain. The application realizes efficient, safe and trusted knowledge collaborative utilization under the premise of protecting data sovereignty.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method and system for federated learning and trusted sharing of engineering knowledge for cross-organizational collaboration. Background Technology

[0002] In large-scale engineering projects, numerous stakeholders (such as design institutes, contractors, and suppliers) need to collaborate. However, due to concerns about trade secrets and data sovereignty, their data often exists in silos, severely hindering the flow of knowledge and the development of intelligent applications. Traditional centralized data sharing platforms require all parties to upload data, which fails to fundamentally alleviate the privacy concerns of stakeholders, resulting in low collaboration efficiency.

[0003] Federated learning is an emerging distributed machine learning paradigm that allows for collaborative model training without exchanging raw data, offering new insights into solving data silos and privacy issues. However, applying federated learning to engineering fields faces numerous challenges. First, engineering knowledge possesses complex graph structures, making it difficult for simple models to effectively represent it. Second, the vast differences in data distribution among different participants can severely degrade the performance of standard federated learning algorithms. Third, open federated network environments are vulnerable to malicious attacks, such as model poisoning. Finally, there is a lack of a reliable mechanism to govern and audit the entire knowledge-sharing process.

[0004] Therefore, there is an urgent need for a systematic solution that can simultaneously address the four major challenges of privacy protection, data heterogeneity, security threats, and trusted governance. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for federated learning and trusted sharing of engineering knowledge for cross-organizational collaboration, which can solve the technical problems of trust deficiency, difficulty in ensuring data privacy, poor model training effect caused by data heterogeneity, and lack of secure and trusted sharing and auditing mechanisms in cross-organizational collaboration.

[0006] To solve the above-mentioned technical problems, the present invention is implemented as follows.

[0007] A method for federated learning and trusted sharing of engineering knowledge for cross-organizational collaboration includes: Step S1: Construct a globally unified domain ontology; each organizational unit configures a client, and the organizational unit transforms local engineering data into a local knowledge graph based on the domain ontology; Step S2: The client deploys the graph neural network locally and decouples it into a shared representation layer and a personalized task layer. The shared representation layer on the client has the same structure. The client trains the graph neural network based on the local knowledge graph and sends the parameters of the trained shared representation layer to the federated server. Step S3: The federated server deploys a global shared model, the structure of which is the same as that of the shared representation layer; the federated server calculates the global cosine similarity matrix, uses density clustering algorithm to identify abnormal clients from the global cosine similarity matrix, assigns a reputation score for the current round to the client, and the client adds the reputation score for the current round to its total reputation score; Step S4: Select non-abnormal clients as clients to be updated, determine the weights of the clients to be updated, and update the global shared model parameters; the clients to be updated update the shared representation layer according to the updated global shared model; Step S5: Store the updated global shared model, the client's shared representation layer, and the hash value and data digest of the local knowledge graph into the blockchain.

[0008] Preferably, the rows and columns of the global cosine similarity matrix are all client numbers, and the elements of the global cosine similarity matrix represent the cosine similarity of the parameters of the shared representation layer of the client corresponding to the row and column of that element; when a client is identified as an abnormal client, the reputation score for this round is negative; when a client is identified as a normal client, the reputation score for this round is positive.

[0009] Preferably, in step S1, after generating the local knowledge graph, the client k calculates the topological complexity of the local knowledge graph. and semantic distribution feature vector ; Topological complexity The calculation formula is as follows:

[0010] in, For the set of nodes in the local knowledge graph of client k, | | represents the number of nodes in the node set. For the edge set of the local knowledge graph of client k, | | represents the number of edges in the edge set. Let v be the degree of node v; Semantic distribution feature vector Where pi represents the i-th type of engineering semantic relation in the local knowledge graph edge set The proportion it accounts for.

[0011] Preferably, in step S2, the graph neural network is deployed locally on the client and decoupled into a shared representation layer and a personalized task layer, including: Client k receives semantic distribution feature vectors through a gating network G(·). , thus obtaining continuous gated vectors Differentiable gating is obtained by continuous relaxation of Gumbel-Sigmoid, and the differentiable gating is discretized into a binary mask vector. ; This represents element-wise multiplication broadcast by channel, on the client side. Local shared parameters and local personalized parameters Defined respectively as [ ], where d represents the channel dimension of the hidden representation in the graph neural network. These are the parameters of the graph neural network.

[0012] Preferably, the graph neural network training is completed when the total number of local training cycles on the client reaches a preset total number of local training cycles, or when the graph neural network loss function converges; after the graph neural network training is completed, privacy enhancement processing is performed on the parameters of the shared representation layer, including: updating the parameters of the shared representation layer. Gradient sparsity processing is performed to retain the top values ​​with the largest magnitude or absolute value. Set the shared representation layer parameters to zero, and set the rest of the shared representation layer parameters to zero. Adaptive differential privacy noise following a Gaussian distribution is added to the retained shared representation layer parameters, where the variance of the noise is... Update based on shared presentation layer parameters Norm and preset privacy budget Dynamic adjustments are made; the retained shared representation layer parameters after noise is added are sent to the federated server as the parameters of the trained shared representation layer.

[0013] Preferably, the local target loss function of client k is

[0014] Where c is the engineering component category number, and C is the total number of engineering component categories. For the true category label of node v, The class probabilities predicted by the graph neural network. is the sparse regularization coefficient.

[0015] Preferably, in step S4, non-abnormal clients are selected as clients to be updated, and the weights of the clients to be updated are determined and the globally shared model parameters are updated, including: The federated server determines the weight of each client to be updated based on its reputation score and the amount of local knowledge graph data it possesses. Based on the weights of each client, the server updates the weights of the clients and the corresponding shared representation layer parameters. Perform a weighted average and update the globally shared model parameters; Weight of the client to be updated The calculation formula is as follows: in, This represents the amount of local knowledge graph data for the client to be updated. The reputation score for the client to be updated. The topological complexity of the client to be updated. For hyperparameters, For the set of clients to be updated, The client number to be updated For the number The amount of local knowledge graph data in the client to be updated. For the number The reputation score of the client to be updated For the number The topological complexity of the client to be updated.

[0016] A federated learning and trusted sharing system for engineering knowledge oriented towards cross-organizational collaboration includes: Local initialization module: configured to build a globally unified domain ontology; each organizational unit is configured with a client, and the organizational unit transforms local project data into a local knowledge graph based on the domain ontology; Local training module: Configured for local deployment of graph neural networks on the client side, and decouples the graph neural network into a shared representation layer and a personalized task layer. The shared representation layer structure is the same on the client side. The client trains the graph neural network based on the local knowledge graph and sends the parameters of the trained shared representation layer to the federated server. Global module: Configured to deploy a global shared model on the federated server. The structure of the global shared model is the same as that of the shared representation layer. The federated server calculates the global cosine similarity matrix, uses density clustering algorithm to identify abnormal clients from the global cosine similarity matrix, assigns a reputation score for the current round to the client, and the client adds the reputation score for the current round to its total reputation score. Update module: Configured to treat non-abnormal clients as clients to be updated, determine the weight of the clients to be updated and update the global shared model parameters; the clients to be updated update the shared representation layer according to the updated global shared model; Storage module: Configured to store the updated global shared model, the client's shared representation layer, and the local knowledge graph into the blockchain.

[0017] The present invention provides an electronic device, characterized in that the electronic device comprises: A processor is used to execute multiple instructions; Memory, used to store multiple instructions; The plurality of instructions are to be stored in the memory and loaded and executed by the processor as described above.

[0018] The present invention has the following beneficial effects: (1) This invention is a data processing and sharing method and system that integrates federated learning, graph neural networks and blockchain technologies, and is particularly suitable for industries such as architecture, engineering and construction that require multi-party collaboration and are highly data-sensitive. This invention solves the knowledge sharing problem caused by data silos, privacy concerns and lack of trust in cross-organizational engineering collaboration, and realizes efficient, secure and reliable collaborative utilization of knowledge under the premise of protecting data sovereignty.

[0019] (2) This invention ensures that the original data of each participant does not leave their local area through the paradigm of federated learning where the data is stationary but the model is dynamic, thus fundamentally solving concerns about data privacy and sovereignty.

[0020] (3) The present invention uses personalized model decoupling and alternating optimization strategies to train a model that takes into account both the commonalities of global knowledge and the characteristics of local data, thus significantly improving model performance.

[0021] (4) The contribution-aware robust aggregation mechanism of this invention can effectively identify and resist model poisoning attacks by malicious clients, thus ensuring the security of the federated learning process.

[0022] (5) This invention establishes an end-to-end trusted collaboration mechanism, based on blockchain dynamic access control and trusted auditing, providing transparent and tamper-proof governance capabilities for the entire lifecycle of knowledge sharing, establishing technical trust between mutually distrustful entities, and promoting deeper collaboration. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the process of the federated learning and trusted sharing method for engineering knowledge across organizations, as described in this invention. Detailed Implementation

[0024] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] like Figure 1 As shown, this invention proposes a method for federated learning and trusted sharing of engineering knowledge for cross-organizational collaboration, including: Step S1: Constructing a globally unified domain ontology; Each organizational unit is configured with a client, and the organizational unit transforms local engineering data into a local knowledge graph based on the domain ontology; Step S2: The client deploys the graph neural network locally and decouples it into a shared representation layer and a personalized task layer. The shared representation layer on the client has the same structure. The client trains the graph neural network based on the local knowledge graph and sends the parameters of the trained shared representation layer to the federated server. Step S3: The federated server deploys a global shared model, the structure of which is the same as that of the shared representation layer; the federated server calculates the global cosine similarity matrix, uses density clustering algorithm to identify abnormal clients from the global cosine similarity matrix, assigns a reputation score for the current round to the client, and the client adds the reputation score for the current round to its total reputation score; In this invention, the rows and columns of the global cosine similarity matrix are client numbers, and the elements of the global cosine similarity matrix represent the cosine similarity of the parameters of the shared representation layer of the client corresponding to the row and column of that element. When a client is identified as an abnormal client, the reputation score for this round is negative; when a client is identified as a normal client, the reputation score for this round is positive. This invention ensures that the final reputation score is non-negative through a non-negative mapping function.

[0026] Step S4: Select non-abnormal clients as clients to be updated, determine the weights of the clients to be updated, and update the global shared model parameters; the clients to be updated update the shared representation layer according to the updated global shared model; Step S5: Store the updated global shared model, the client's shared representation layer, and the hash value and data digest of the local knowledge graph into the blockchain.

[0027] Furthermore, in step S1, after generating the local knowledge graph, the client k calculates the topological complexity of the local knowledge graph. and semantic distribution feature vector ; Topological complexity The calculation formula is as follows:

[0028] in, For the set of nodes in the local knowledge graph of client k, | | represents the number of nodes in the node set. For the edge set of the local knowledge graph of client k, | | represents the number of edges in the edge set. Let v be the degree of node v. The larger the value, the closer the connections between nodes in the local knowledge graph and the more complex the topology.

[0029] Semantic distribution feature vector ,in p i Indicates the firsti Semantic relationships of various engineering projects in the edge set of the local knowledge graph The proportion it accounts for.

[0030] For example, engineering semantic relationships include spatial inclusion, temporal sequence, and logical dependency.

[0031] This invention abandons the traditional strategy of hard decoupling by network layer and proposes a dynamic model decoupling mechanism guided by engineering semantics.

[0032] Further, in step S2, the client locally deploys the graph neural network and decouples the graph neural network into a shared representation layer and a personalized task layer, including: Client k receives semantic distribution feature vectors through a gating network G(·). , thus obtaining continuous gated vectors Differentiable gating is obtained by continuous relaxation of Gumbel-Sigmoid, and the differentiable gating is discretized into a binary mask vector. ; This represents element-wise multiplication broadcast by channel, on the client side. Local shared parameters and local personalized parameters Defined respectively as [ ], where d represents the channel dimension of the hidden representation in the graph neural network. These are the parameters of the graph neural network.

[0033] In this invention, by setting a threshold (e.g., 0.5) or employing the Gumbel-Softmax technique, a continuous mask vector with the same dimension as the number of hidden channels in the graph neural network is output. Using the mask matrix Parameters of a graph neural network At the channel level, they are adaptively decoupled to share representation layer parameters. and personalized task layer parameters Specifically, for the j-th channel in the weight matrix, if Then the channel parameter is classified as ;like Then it is classified as This mechanism enables GNNs to dynamically determine which feature channels participate in the federated sharing of global knowledge and which channels are reserved for local personalized adaptation, based on the specific semantic needs of the local project, such as whether to focus on schedule management or geometric collision.

[0034] Further, in step S2, the client trains a graph neural network based on the local knowledge graph and sends the trained parameters of the shared representation layer to the federated server. The training employs an alternating parameter freezing method, including: optimizing the parameters in two phases within each local training cycle. In the first stage, the parameters of the personalized task layer are frozen, the local knowledge graph is used as training data, and the parameters of the shared representation layer are updated in the first preset number of steps using the gradient descent algorithm. In the second stage, the parameters of the shared representation layer are frozen, the local knowledge graph is used as training data, and the gradient descent algorithm is used to update the parameters of the personalized task layer for a second preset number of steps.

[0035] Furthermore, the graph neural network training is complete when the total number of local training cycles on the client reaches the preset total number of local training cycles, or when the graph neural network loss function converges. After the graph neural network training is complete, privacy enhancement processing is performed on the parameters of the shared representation layer, including: updating the parameters of the shared representation layer. Gradient sparsity processing is performed to retain the top values ​​with the largest magnitude or absolute value. Set the shared representation layer parameters to zero, and set the rest of the shared representation layer parameters to zero. Adaptive differential privacy noise following a Gaussian distribution is added to the retained shared representation layer parameters, where the variance of the noise is... Update based on shared presentation layer parameters Norm and preset privacy budget Dynamic adjustments are made; the retained shared representation layer parameters after noise is added are sent to the federated server as the parameters of the trained shared representation layer.

[0036] Furthermore, the local objective loss function for client k is: The local objective loss function of client k is

[0037] Where c is the engineering component category number, and C is the total number of engineering component categories. For the true category label of node v, The class probabilities predicted by the graph neural network. represents the sparse regularization coefficient. The first term is the node classification loss, and the second term is the shared channel sparse regularization term, which controls the number of feature channels assigned to the shared sub-model to avoid the global shared part from over-absorbing the local semantics of specific clients.

[0038] The gradient calculation formula for the shared representation layer is as follows: The updated formula is as follows .

[0039] The gradient calculation formula for the personalized task layer is as follows: The updated formula is as follows .in, This is the learning rate.

[0040] Further, in step S4, non-abnormal clients are selected as clients to be updated, and the weights of the clients to be updated are determined and the globally shared model parameters are updated, including: The federated server determines the weight of each client to be updated based on its reputation score and the amount of local knowledge graph data it possesses. Based on the weights of each client, the server updates the weights of the clients and the corresponding shared representation layer parameters. Perform a weighted average and update the globally shared model parameters; Weight of the client to be updated The calculation formula is as follows: in, This represents the amount of local knowledge graph data for the client to be updated. The reputation score for the client to be updated. The topological complexity of the client to be updated. For hyperparameters, For the set of clients to be updated, The client number to be updated For the number The amount of local knowledge graph data in the client to be updated. For the number The reputation score of the client to be updated For the number The topological complexity of the client to be updated.

[0041] In step S4, this invention introduces dynamic aggregation weights that are aware of the graph topology. Traditional federated aggregation relies solely on data volume or parameter similarity, while this invention argues that the more complex the topological structure of the engineering knowledge graph, the greater the contribution of its extracted gradient updates to the generalization ability of the global model. For example, Using this weight Update the parameters of the shared presentation layer for all clients to be updated. Perform a weighted average and update the globally shared model: The updated global shared model is then sent to the clients awaiting updates.

[0042] This invention establishes a trusted access control and full lifecycle audit mechanism based on graph evolution. The updated globally shared model hash value, dynamic feature mask matrix information for each client, and local knowledge graph access strategies are deployed to the smart contract of the consortium blockchain.

[0043] In this invention, owners of knowledge assets can define fine-grained, dynamic access control policies through smart contracts. Specifically, the smart contracts record the evolution snapshot hashes of the global engineering knowledge graph. The access control contracts not only verify the identity and permissions of the requester but also automatically verify the current lifecycle stage of the project by comparing on-chain graph snapshots. For example, certain core construction process knowledge is only decrypted and shared with specific subcontractors when the graph evolves to the "main structure construction" stage. All critical operations, including permission changes, access records, model update hashes, and topology summaries, are permanently recorded on the blockchain as transactions, forming an immutable audit log.

[0044] This invention also provides a federated learning and trusted sharing system for engineering knowledge across organizations, comprising: Local initialization module: configured to build a globally unified domain ontology; each organizational unit is configured with a client, and the organizational unit transforms local project data into a local knowledge graph based on the domain ontology; Local training module: Configured for local deployment of graph neural networks on the client side, and decouples the graph neural network into a shared representation layer and a personalized task layer. The shared representation layer structure is the same on the client side. The client trains the graph neural network based on the local knowledge graph and sends the parameters of the trained shared representation layer to the federated server. Global module: Configured to deploy a global shared model on the federated server. The structure of the global shared model is the same as that of the shared representation layer. The federated server calculates the global cosine similarity matrix, uses density clustering algorithm to identify abnormal clients from the global cosine similarity matrix, assigns a reputation score for the current round to the client, and the client adds the reputation score for the current round to its total reputation score. Update module: Configured to treat non-abnormal clients as clients to be updated, determine the weight of the clients to be updated and update the global shared model parameters; the clients to be updated update the shared representation layer according to the updated global shared model; Storage module: Configured to store the updated global shared model, the client's shared representation layer, and the local knowledge graph into the blockchain.

[0045] The specific embodiments described above only illustrate the design principles of the present invention. The shapes and names of the components in this description may differ and are not limited. Therefore, those skilled in the art can modify or make equivalent substitutions to the technical solutions described in the foregoing embodiments; and these modifications and substitutions do not depart from the inventive spirit and technical solutions of the present invention, and should all fall within the protection scope of the present invention.

Claims

1. A method for federated learning and trusted sharing of engineering knowledge for cross-organizational collaboration, characterized in that, include: Step S1: Construct a globally unified domain ontology; Each organizational unit is configured with a client, and the organizational unit transforms local engineering data into a local knowledge graph based on the domain ontology; Step S2: The client deploys the graph neural network locally and decouples it into a shared representation layer and a personalized task layer. The shared representation layer on the client has the same structure. The client trains the graph neural network based on the local knowledge graph and sends the parameters of the trained shared representation layer to the federated server. Step S3: The federated server deploys a global shared model, the structure of which is the same as that of the shared representation layer; the federated server calculates the global cosine similarity matrix, uses density clustering algorithm to identify abnormal clients from the global cosine similarity matrix, assigns a reputation score for the current round to the client, and the client adds the reputation score for the current round to its total reputation score; Step S4: Select non-abnormal clients as clients to be updated, determine the weights of the clients to be updated, and update the global shared model parameters; the clients to be updated update the shared representation layer according to the updated global shared model; Step S5: Store the updated global shared model, the client's shared representation layer, and the hash value and data digest of the local knowledge graph into the blockchain.

2. The method as described in claim 1, characterized in that, The rows and columns of the global cosine similarity matrix are all client numbers. The elements of the global cosine similarity matrix represent the cosine similarity of the parameters of the shared representation layer of the client corresponding to the row and column of that element. When a client is identified as an abnormal client, the reputation score for this round is negative; when a client is identified as a normal client, the reputation score for this round is positive.

3. The method as described in claim 2, characterized in that, In step S1, after generating the local knowledge graph, client k calculates the topological complexity of the local knowledge graph. and semantic distribution feature vector ; Topological complexity The calculation formula is as follows: in, For the set of nodes in the local knowledge graph of client k, | | represents the number of nodes in the node set. For the edge set of the local knowledge graph of client k, | | represents the number of edges in the edge set. Let v be the degree of node v; Semantic distribution feature vector Where pi represents the i-th type of engineering semantic relation in the local knowledge graph edge set The proportion it accounts for.

4. The method according to any one of claims 1-3, characterized in that, In step S2, the client locally deploys a graph neural network and decouples the graph neural network into a shared representation layer and a personalized task layer, including: the client k receives semantic distribution feature vectors through a gating network G(·). , thus obtaining continuous gated vectors Differentiable gating is obtained by continuous relaxation using Gumbel-Sigmoid, and the differentiable gating is discretized into a binary mask vector. ; This represents element-wise multiplication broadcast by channel, on the client side. Local shared parameters and local personalized parameters Defined respectively as [ ], where d represents the channel dimension of the hidden representation in the graph neural network. These are the parameters of the graph neural network.

5. The method as described in claim 4, characterized in that, The graph neural network (Graph Neural Network) training is complete when the total number of local training cycles on the client reaches the preset total number of local training cycles, or when the graph neural network loss function converges. After the graph neural network training is complete, privacy enhancement processing is performed on the parameters of the shared representation layer, including: updating the parameters of the shared representation layer. Gradient sparsity processing is performed to retain the top values ​​with the largest magnitude or absolute value. Set the shared representation layer parameters to zero, and set the rest of the shared representation layer parameters to zero. Adaptive differential privacy noise following a Gaussian distribution is added to the retained shared representation layer parameters, where the variance of the noise is... Update based on shared presentation layer parameters Norm and preset privacy budget Dynamic adjustments are made; the retained shared representation layer parameters after noise is added are sent to the federated server as the parameters of the trained shared representation layer.

6. The method as described in claim 5, characterized in that, The local objective loss function for client k is: Where c is the engineering component category number, and C is the total number of engineering component categories. For the true category label of node v, The class probabilities predicted by the graph neural network. is the sparse regularization coefficient.

7. The method as described in claim 6, characterized in that, In step S4, non-abnormal clients are selected as clients to be updated. The weights of the clients to be updated are determined and the globally shared model parameters are updated, including: The federated server determines the weight of each client to be updated based on its reputation score and the amount of local knowledge graph data it possesses. Based on the weights of each client, the server updates the weights of the clients and the corresponding shared representation layer parameters. Perform a weighted average and update the globally shared model parameters; Weight of the client to be updated The calculation formula is as follows: in, This represents the amount of local knowledge graph data for the client to be updated. The reputation score for the client to be updated. The topological complexity of the client to be updated. For hyperparameters, For the set of clients to be updated, The client number to be updated For the number The amount of local knowledge graph data in the client to be updated. For the number The reputation score of the client to be updated For the number The topological complexity of the client to be updated.

8. A federated learning and trusted sharing system for engineering knowledge oriented towards cross-organizational collaboration, characterized in that, include: Local initialization module: configured to build a globally unified domain ontology; Each organizational unit is configured with a client, and the organizational unit transforms local engineering data into a local knowledge graph based on the domain ontology; Local training module: Configured for local deployment of graph neural networks on the client side, and decouples the graph neural network into a shared representation layer and a personalized task layer. The shared representation layer structure is the same on the client side. The client trains the graph neural network based on the local knowledge graph and sends the parameters of the trained shared representation layer to the federated server. Global module: Configured to deploy a global shared model on the federated server. The structure of the global shared model is the same as that of the shared representation layer. The federated server calculates the global cosine similarity matrix, uses density clustering algorithm to identify abnormal clients from the global cosine similarity matrix, assigns a reputation score for the current round to the client, and the client adds the reputation score for the current round to its total reputation score. Update module: Configured to treat non-abnormal clients as clients to be updated, determine the weight of the clients to be updated and update the global shared model parameters; the clients to be updated update the shared representation layer according to the updated global shared model; Storage module: Configured to store the updated global shared model, the client's shared representation layer, and the local knowledge graph into the blockchain.

9. A computer-readable storage medium storing a plurality of instructions, characterized in that, The plurality of instructions are used by the processor to load and execute the method as described in any one of claims 1-7.

10. An electronic device, characterized in that, include: A processor is used to execute multiple instructions; Memory, used to store multiple instructions; The plurality of instructions are to be stored in the memory and loaded by the processor and executed as described in any one of claims 1-7.

Images