A blockchain-based federated learning cross-border trade incentive mechanism method

Abstract

The application provides a blockchain-based federal learning cross-border trade incentive mechanism method, which comprises the following steps: when a local training node receives a local model training task, a local model is trained according to a last round reputation value and a preset utility function, and a local training model is obtained; the local training model and a local training cost parameter are uploaded to a verification node; the verification node verifies the local training model, updates the last round reputation value of the local training node according to a verification result, obtains a current round reputation value, and calculates a reward value of the verification node in the current round training process; a reward value of the local training node is calculated according to the verification result and the current round reputation value; a model aggregation node aggregates the verified models, obtains a global model, and calculates a reward value of the model aggregation node in the current round training process; and the global model is stored in a block, and a model ledger is updated. Through implementation of the scheme, data contribution can be improved from the source, and the reliability of the collaborative training process can be strengthened.

Description

Technical Field

[0001] This invention belongs to the field of blockchain federated learning technology for cross-border trade, specifically involving a blockchain-based federated learning incentive mechanism method for cross-border trade. Background Technology

[0002] Cross-border trade, as a means of exchanging goods and services between countries, encompasses the entire process of cross-border import and export arrangements, trade models, and trade settlement. It facilitates economic and technological interaction through the import and export of products. Currently, cross-border trade is primarily regulated digitally through cross-border trade platforms. Given the inherent risks in cross-border trade, effective risk monitoring of transaction data on these platforms is crucial. Therefore, establishing a federated learning incentive mechanism for risk protection and data sharing through shared data models is particularly important.

[0003] Federated learning is a privacy-preserving machine learning technique that provides a solution for protecting privacy and building high-quality models. Data owners can directly train models using local data, breaking down data silos through model interaction and fully utilizing the value of private data. However, federated learning is vulnerable to security threats such as single point-of-failure attacks, poisoning attacks, and free-riding attacks. Furthermore, the lack of incentive mechanisms in federated learning leads to unreasonable and unfair distribution of model benefits, resulting in low motivation for data owners to participate in training and low data quality. Introducing blockchain technology can achieve decentralized learning in federated learning, and its anonymity, immutability, and verifiability provide a new solution for this approach.

[0004] Among related technologies, a dual-blockchain data sharing and reputation management method based on federated learning has been implemented, in which the reputation incentive blockchain ( RIchain ) and model quality blockchain ( MQchain Interaction between blockchains to achieve data sharing and reputation management based on dual blockchains suffers from high computational resource overhead. Therefore, a high-quality federated learning method based on blockchain and reputation mechanisms is proposed. This method uses a single blockchain to complete reputation management and federated learning. The method only uses fair value game theory to assess the impact of nodes on global model aggregation, triggering a reputation update mechanism to update the data owner's reputation value, thus increasing the data owner's reward and the proportion of local model fusion. However, the data used by the data owner for training is a static value directly reported by nodes and subject to reputation constraints, leaving the data owner in a management blind spot and making it difficult to guarantee the quality of the training data provided by the data owner. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a blockchain-based incentive mechanism for cross-border trade in federated learning, in order to meet the need to improve the quality of federated learning.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides a blockchain-based federated learning cross-border trade incentive mechanism method, comprising: when a local training node receives a local model training task issued by the blockchain network, it performs the current round of training of the local model based on the previous round reputation value and a preset utility function to obtain a local training model; the local training node uploads the local training model and the current round training cost parameters to a verification node; the verification node verifies the local training model, updates the previous round reputation value of the local training node based on the verification result to obtain the current round reputation value, and calculates the reward value of the verification node in the current round of training based on the workload of the verification process; the verification node calculates the reward value of the local training node in the current round of training based on the verification result, the current round reputation value, and the current round training cost parameters; the verification node submits the verified local training model to the model aggregation node; the model aggregation node aggregates the verified models to obtain a global model and calculates the reward value of the model aggregation node in the current round of training; and the global model is stored in a block to update the model ledger.

[0008] This invention provides a blockchain-based federated learning cross-border trade incentive mechanism. On one hand, it utilizes blockchain and reputation contribution values ​​for calculation, achieving secure and fair reward distribution. Addressing the uneven reward distribution problem in traditional federated learning, it constructs a multi-dimensional contribution evaluation model for heterogeneous nodes and employs a reputation-weighted algorithm to dynamically allocate training rewards, establishing a balanced distribution mechanism for node rewards. On the other hand, local training nodes are no longer passively reporting values ​​to satisfy constraints, but rather actively making optimal decisions after weighing costs and benefits, better reflecting the rational behavior of nodes in real-world scenarios. This is achieved by introducing local training... epoch Using the quantity as a variable, the contribution amount is directly linked to the training effort, which incentivizes nodes to increase training input to improve effectiveness and improve the quality of data contribution from the source.

[0009] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0010] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration:

[0011] Figure 1 This is a schematic diagram of an application system for a blockchain-based federated learning cross-border trade incentive mechanism method in this invention.

[0012] Figure 2 This is a flowchart illustrating a specific example of a blockchain-based federated learning cross-border trade incentive mechanism method according to the present invention.

[0013] Figure 3 This is an interaction diagram between nodes in a blockchain-based federated learning cross-border trade incentive mechanism method of the present invention;

[0014] Figure 4 This is a flowchart of the verification node for evaluating the quality of the model in this invention. Detailed Implementation

[0015] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can also refer to the internal connection of two components; and they can refer to a wireless connection or a wired connection. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0017] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0018] This invention provides a blockchain-based federated learning cross-border trade incentive mechanism method, applicable to, for example... Figure 1 The system shown comprises a training layer, a blockchain layer, and a reputation reward layer. The training layer includes local training nodes, each using its own dataset to train the downloaded global model locally, resulting in a local trained model. The blockchain layer generates blocks, and verification nodes validate the local trained models uploaded by the local training nodes, using this data as the data source for the reputation reward layer for reputation and reward management. The system selects... PoSBased on the blockchain consensus protocol, the PoS consensus algorithm is extended to the trust maintenance of validator nodes, which meets the framework's requirements for node security and task efficiency. Before the local model is uploaded to the blockchain, validator nodes verify the quality of the locally uploaded model and score it by voting. Only after generating a reputation value according to the score is the model and reputation value uploaded to the blockchain. After multiple rounds of iteration, an aggregated global model is obtained.

[0019] Specifically, a blockchain-based federated learning method for cross-border trade incentive mechanisms can be applied to cross-border trade scenarios, such as... Figure 2 As shown, it includes:

[0020] S101, When the local training node receives the local model training task issued by the blockchain network, it performs the training of the local model in this round according to the reputation value of the previous round and the preset utility function to obtain the local training model.

[0021] S102, the local training node uploads the locally trained model and the training cost parameters for this round to the verification node;

[0022] S103, the verification node verifies the locally trained model, updates the previous reputation value of the local training node based on the verification result, obtains the reputation value of the current round, and calculates the reward value of the verification node in the current training process based on the workload in the verification process.

[0023] S104, The verification node calculates the reward value of the local training node in this training process based on the verification result, the reputation value of this round, and the training cost parameters of this round;

[0024] S105, the verification node submits the verified local training model to the model aggregation node;

[0025] S106, the model aggregation node aggregates the validated models to obtain the global model, and calculates the reward value of the model aggregation node in this round of training;

[0026] S107, store the global model into a block and update the model ledger.

[0027] For example, this embodiment describes the interaction method of each node, such as... Figure 3As shown, firstly, the task publisher publishes a task to the blockchain network. This task can be a global model training task. When a blockchain node receives the task, it broadcasts the task parameters to its local training node. These parameters can include local model training parameters. The local training node receives the local model training task from the blockchain network and, based on its previous reputation value and a preset utility function, performs the current round of training for its local model, resulting in a local training model. The previous reputation value influences the upper limit of the amount of data a node can provide during this round of training. The preset utility function is the game objective for each local training node to maximize its reward value, i.e., maximizing node efficiency, which can be understood as net profit and is the node's ultimate maximization goal.

[0028] Specifically, based on the reputation value of the previous round and the preset utility function, the local model is trained in this round to obtain the local training model. This includes: limiting the amount of data contribution in the current round of training based on the reputation value of the local training node in the previous round and the amount of data contribution in the previous round; determining the amount of data contribution in the current round of training within the range of the amount of data contribution provided in the current round of training, with the goal of maximizing the preset utility function; and training the local model in this round based on the amount of data contribution in the current round of training to obtain the training model in this round.

[0029] The amount of data and model quality provided by local training nodes determine their contribution rewards in each round. Introducing local training nodes increases the number of local training epochs, allowing trainers to gain more benefits and attracting data owners to participate in federated learning training. To continuously reduce the participation of malicious nodes, the training nodes... j The amount of data provided in each round cannot exceed the reputation-weighted contribution of the previous round. For ease of derivation, the contribution of discrete data is approximated as a continuous value. Taking the current training round as round t as an example, the training nodes... j exist t Data contribution during training rounds The range is represented as:

[0030]

[0031] in, Represents training nodes j exist t-1 Reputation value in the wheel, Represents training nodes j exist t-1 The data contribution during each training round is a quantified value of the substantial value input provided by the training node, such as the quantified value of providing 1000 high-quality labeled data points or completing 50 effective model training iterations. It can be 0, or it can be any other set threshold. When it is less than Nodes are not allowed to participate in tasks. That is, nodes that engage in malicious behavior will be punished and barred from participating in training, effectively preventing data poisoning and free-riding attacks.

[0032] The default utility function is:

[0033]

[0034] The essence of a utility function lies in the difference between the benefits derived from data contributions and the costs of resource consumption. Training node j exist t The computational resource consumption in each round of training mainly includes the computational cost of round t. , communication cost and storage costs , that is , Represents training nodes j exist t The reputation value within the round. For other nodes, this also includes transaction verification costs, storage resources consumed, etc.

[0035] The utility function is defined as the benefit derived from data contribution minus the cost of resource consumption, and the amount of data contribution... This can be reflected in reputation; therefore, it can be simplified to calculating the proportion of a node's reward among all nodes using reputation and training participation. j In the t Data contribution during rounds of problem solving The objective function can be expressed as:

[0036]

[0037] in, R This represents the total budget in federalized learning, which is also the total reward value. Indicates the first t wheel node j The number of local training epochs (i.e., training participation). j This indicates the total number of local training nodes.

[0038] Since all nodes participate in reward distribution, reward competition can be viewed as a non-cooperative game, where each node can obtain more rewards. In a non-cooperative game, to maximize their interests, each node will calculate the optimal strategy. The benefits of each node are limited by contribution and cost; the most direct way is to increase contribution to obtain higher rewards. A utility function can clearly calculate the expected revenue of each node to measure whether the model has achieved its expected results. A reward evaluation method based on reputation-weighted contribution can fairly and reasonably distribute rewards among nodes with different roles in the system, maintaining the fairness of the federated learning training system, encouraging data owners to actively participate in federated learning, and preventing malicious nodes from engaging in behaviors that harm the global model. The revenue of malicious nodes is constrained by costs and incentive mechanisms. In addition to the cost of each training cycle, they also face penalties, especially when a node performs malicious operations after accumulating a high reputation, in which case the penalties will be more severe. This makes it difficult for malicious nodes to achieve their goals due to the high cost of deception. Combining model verification mechanisms and incentive mechanisms can effectively reduce malicious behavior and improve the security of the federated learning system. The optimal strategy obtained through the above game theory is used to train a local model, resulting in a locally trained model.

[0039] After obtaining the locally trained model, the local training node uploads the local training model and local training cost parameters to the validation node. These local training cost parameters include data contribution, training participation, and other factors. The validation node validates the local training model and updates its reputation value based on the validation results, thus obtaining its reputation value for this round. Simultaneously, based on the workload during this validation process, the validation node calculates its reward value for this round of training.

[0040] Next, the validator node calculates the reward value for the local training node in this training round based on the verification results, the reputation value of this round, and the training cost parameters of this round. Then, the validator node submits the validated local training model to the model aggregation node. The model aggregation node verifies the signature of the local training node in the system. If the signature passes verification, it aggregates the local model parameters with the voting results and puts them into a privately constructed candidate block, adding it to the blockchain. The model aggregation node attempts to merge the voting results with the corresponding local model in the blockchain and stores it in the next consensus block. Members of the model aggregation organization cooperate with each other and are jointly responsible for model aggregation. Rewarding the model aggregation node makes the entire system more complete and can attract node users with computing power but no data to join the training. Finally, the blockchain network broadcasts the latest model to all nodes.

[0041] This invention provides a blockchain-based federated learning cross-border trade incentive mechanism. On one hand, it utilizes blockchain and reputation contribution values ​​for calculation, achieving secure and fair reward distribution. Addressing the uneven reward distribution problem in traditional federated learning, it constructs a multi-dimensional contribution evaluation model for heterogeneous nodes and employs a reputation-weighted algorithm to dynamically allocate training revenue, establishing a balanced distribution mechanism for node revenue. On the other hand, local training nodes are no longer passively reporting values ​​to meet constraints, but rather actively making optimal decisions after weighing costs and benefits, better reflecting the rational behavior of nodes in real-world scenarios. By introducing the number of local training epochs as a variable, the contribution amount is directly linked to the training effort, incentivizing nodes to increase training investment to improve utility and enhance data contribution quality from the source. The reputation constraint mechanism effectively curbs node speculative behavior and strengthens the reliability of the collaborative training process.

[0042] As an optional implementation, the verification node verifies the locally trained model, updates the previous round reputation value of the local training node based on the verification result, and obtains the current round reputation value, including:

[0043] The verification node performs signature verification on the locally trained model uploaded by the local training node. If the signature verification passes, the following steps are executed:

[0044] The verification node trains a pre-stored first local model using the target dataset to obtain a second local model. The target dataset is a dataset whose distribution similarity to the dataset trained by the local training node is higher than a first threshold. The first local model is a model whose similarity to the local model distributed to the local training node by the blockchain network is higher than a second threshold. The verification node determines the honesty threshold based on the first test result index of the second local model. Each verification node tests the local training model to obtain the second test result index of the local training model. Multiple verification nodes compare the second test result index with the honesty threshold to determine the corresponding honesty judgment result of the local training node. Multiple verification nodes vote on the local training node based on the honesty judgment result to obtain the voting result as the verification result. Based on the verification result of the local training node, the reputation value of the local training node in this round is obtained.

[0045] For example, the verification node verification and evaluation model quality process is as follows: Figure 4 As shown, specifically: First, the system initializes a reputation value for each node. ,exist tDuring the federated learning process, it is assumed that the datasets possessed by the local training nodes and the verification nodes have similar distributions. Specifically, the similarity between the target dataset of the verification node and the dataset trained by the local training node is higher than a first threshold. The first local model of the verification node is not significantly different from the local model of the local training node. In other words, the first local model of the verification node is a model whose similarity to the local model distributed to the local training node by the blockchain network is higher than a second threshold. Therefore, the verification node executes local model training, specifically by executing one... epoch The training round yields a second local model. The first test result metric of the second local model, such as its accuracy, is used as the honesty threshold. Similarly, for the locally trained model uploaded by the local training node, the same dataset is used for testing to determine the second test result metric, such as its accuracy. The node accuracy is calculated by the verification node; a higher accuracy indicates more honest training behavior. The verification node then judges the honesty of the local training node based on the honesty threshold and votes based on the judgment result. Nodes with an accuracy greater than the threshold are considered honest and receive a positive vote; otherwise, they are considered malicious and receive a negative vote. The final vote count is used as the final verification result. The system then updates the global reputation value and determines whether to terminate. If not, it proceeds to the next round of federated learning. In round t, the reputation value of the local training node for this round is obtained based on the verification results. Specifically, the number or percentage of positive votes can be used as the reputation value for this round; this embodiment does not limit this.

[0046] This invention provides a blockchain-based federated learning incentive mechanism for cross-border trade. Multiple verification nodes obtain a second test result index by testing the locally trained model. This index is compared with an honesty threshold and then voted on. The current reputation value is obtained by combining the previous reputation value of the local training node. This approach considers both the real-time feedback of the node's current training performance and the cumulative impact of historical reputation. This allows the reputation value to dynamically and comprehensively reflect the node's long-term contribution capability and short-term training quality, enhancing the dynamism and rationality of reputation value updates. Furthermore, through signature verification, model testing, and multi-node voting, verification nodes can accurately identify nodes that upload malicious models and punish them by lowering their reputation value. Honest nodes, on the other hand, receive reputation boosts upon passing verification. This forms a two-way mechanism of honesty incentives and malicious constraint, effectively curbing malicious node behavior and enhancing system security.

[0047] As an optional implementation, the reward value of the local training node in this round of training is calculated based on the verification results, the reputation value of this round, and the training cost parameters of this round, including:

[0048] When the number of verification nodes voting as honest nodes is greater than or equal to the number of verification nodes voting as malicious nodes in the verification results, the total reward budget is obtained. Based on the total reward budget, the reputation value of this round, the training cost parameters of this round, and the preset reward rules, the reward value of the local training node in this round of training is determined.

[0049] When the number of verification nodes voting as honest nodes is less than the number of verification nodes voting as malicious nodes in the verification results, the local training node will receive a reward of 0 during this round of training.

[0050] For example, local training nodes dynamically weight their participation based on data throughput, task completeness, parameter optimization, and validation set accuracy, and the system allocates on-chain incentive points according to their contribution weights. Based on this, the specific formula is as follows:

[0051] (3)

[0052] in, Let be the reward value for local training node j in the t-th round of training. This represents the total reward received by the local training node from the overall reward pool. This portion consists of three parts: reputation, data contribution, and training participation. In other words, the more actively a local training node contributes, the higher its reward. That is, cost. This represents the net reward. This represents the number of validators who voted as honest nodes during round t of training. This indicates the number of validator nodes whose votes were malicious. R This indicates the total budget for rewards in federal learning. Indicates the first t The number of local training epochs of node j during the training round (i.e., training participation). j This indicates the total number of local training nodes.

[0053] This invention provides a blockchain-based federated learning cross-border trade incentive mechanism method, which effectively curbs the participation of malicious nodes and further improves the security and reliability of federated learning.

[0054] As an optional implementation, the reward value of the verification node in this round of training is calculated based on the workload in the verification process, including: counting the number of local training nodes that pass the signature verification during the verification process as a first workload representation; counting the number of votes cast by the verification node during the verification process as a second workload representation; obtaining the total reward budget and the reputation value of the verification node; and obtaining the reward value of the verification node in this round of training based on the first workload representation, the second workload representation, the total reward budget, and the reputation value of the verification node.

[0055] For example, the verification node prevents malicious local model updates during training from affecting the global model accuracy. By introducing a reputation value λ as a control coefficient, consistently reliable and honest nodes receive higher rewards, preventing malicious nodes from engaging in behaviors that harm the global model. In each iteration, the verification node verifies the transactions received from the training node. The signature of}, and from the verified signature { Extract the local model { } Evaluate and vote on it, that is, generate { In return, they receive rewards. These rewards for validating nodes provide a certain level of security for the entire system. v Contribution Rewards Represented as:

[0056]

[0057] in, This represents the number of local training nodes that passed signature verification during the verification process, and serves as a representation of the first workload. This represents the number of votes cast during the verification process, serving as a second measure of workload. Represents training nodes j In the t The reputation value of the wheel. If { The signature of} failed the verification by the verification node, and the verification node will not vote for the unverified local training node.

[0058] As an optional implementation, the reward value of the model aggregation node in this round of training is calculated, including:

[0059]

[0060] in, This represents the reward value of the aggregation node. This represents the number of nodes in the local training node set that passed the signature vote.

[0061] The following is the pseudocode for the algorithm described above:

[0062]

[0063] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A blockchain-based federated learning method for cross-border trade incentive mechanisms, characterized in that, include: When a local training node receives a local model training task from the blockchain network, it performs the current round of training of the local model based on the reputation value of the previous round and the preset utility function to obtain the local training model. The local training node uploads the locally trained model and the training cost parameters for this round to the verification node; The verification node verifies the locally trained model. Based on the verification results, it updates the reputation value of the local training node in the previous round to obtain the reputation value in the current round. Based on the workload in the verification process, it calculates the reward value of the verification node in the current training process. The verification node calculates the reward value for the local training node in this training round based on the verification results, the reputation value of this round, and the training cost parameters of this round. The verification node submits the locally trained model that has passed verification to the model aggregation node; The model aggregation node aggregates the validated models to obtain the global model and calculates the reward value of the model aggregation node in this round of training. Store the global model into blocks and update the model ledger; Based on the reputation value from the previous round and the preset utility function, perform the current round of training on the local model to obtain the locally trained model, including: Based on the reputation value and data contribution of the local training node in the previous round, the data contribution in the current training process is limited; Within the range of data contribution provided during this training round, the data contribution during this training round is determined with the goal of maximizing the preset utility function; Based on the data contribution during this training round, perform this round of training on the local model to obtain the trained model for this round; The reward value of the model aggregation node during this training round is calculated, including: in, This represents the reward value for the aggregation node. R represents the number of nodes that voted in favor of the signature in the set of local training nodes; R represents the total reward budget in federated learning. Preset utility function for: in, This represents the number of local training epochs at node j in round t. R represents the reputation value of training node j in round t, and R represents the total reward budget. It represents the computational resource consumption of training node j in training round t, and N represents the total number of local training nodes.

2. The blockchain-based federated learning cross-border trade incentive mechanism method according to claim 1, characterized in that, The validation node validates the locally trained model. Based on the validation results, it updates the previous round reputation value of the local training node to obtain the current round reputation value, which includes: The verification node performs signature verification on the locally trained model uploaded by the local training node. If the signature verification passes, the following steps are executed: The verification node trains a pre-stored first local model using the target dataset to obtain a second local model. The target dataset is a dataset whose distribution similarity to the dataset trained by the local training node is higher than a first threshold. The first local model is a model whose similarity to the local model distributed to the local training node by the blockchain network is higher than a second threshold. The verification node determines the honesty threshold based on the first test result index of the second local model; Each verification node tests the locally trained model and obtains the second test result metric of the locally trained model. Multiple verification nodes compare the second test result index with the honesty threshold to determine the honesty judgment result of the corresponding local training node. Multiple verification nodes vote on the local training nodes based on the honesty judgment results, and the voting results are used as the verification results. Based on the verification results of the local training node, the reputation value of the local training node in this round is obtained.

3. The blockchain-based federated learning cross-border trade incentive mechanism method according to claim 2, characterized in that, Based on the verification results, the reputation value of this round, and the training cost parameters of this round, calculate the reward value of the local training node in this training round, including: When the number of verification nodes voting as honest nodes is greater than or equal to the number of verification nodes voting as malicious nodes in the verification results, the total reward budget is obtained. Based on the total reward budget, the reputation value of this round, the training cost parameters of this round, and the preset reward rules, the reward value of the local training node in this round of training is determined. When the number of verification nodes voting as honest nodes is less than the number of verification nodes voting as malicious nodes in the verification results, the local training node will receive a reward of 0 during this round of training.

4. The blockchain-based federated learning cross-border trade incentive mechanism method according to claim 2, characterized in that, Based on the workload during the verification process, the reward value for the verification node in this round of training is calculated, including: The number of local training nodes that pass signature verification during the verification process is used as the first representation of workload. The number of votes cast during the verification process at the statistical verification node is used as a second representation of the workload. Obtain the total reward budget and the reputation value of the validator nodes; The reward value of the verification node in this round of training is obtained based on the first workload representation, the second workload representation, the total reward budget, and the reputation value of the verification node.

5. The blockchain-based federated learning cross-border trade incentive mechanism method according to claim 3, characterized in that, Based on the total reward budget, the reputation value of this round, the training cost parameters of this round, and the preset reward rules, determine the reward value of the local training node in this round of training, including: in, This represents the reward value of the local training node. Represents a node j exist t Data contribution during training rounds This represents the number of local training epochs at node j in round t. Represents training nodes j exist t The reputation value in the round, where R represents the total reward budget. Training node j exist t The computational resource consumption in each round of training tasks, where N represents the total number of local training nodes.

6. The blockchain-based federated learning cross-border trade incentive mechanism method according to claim 4, characterized in that, Based on the first workload representation, the second workload representation, the total reward budget, and the reputation value of the validation node, the reward value of the validation node in this round of training is obtained, including: in, This represents the reward value for the verification node. This represents the number of local training nodes that passed signature verification during the verification process, and serves as a representation of the first workload. R represents the number of votes cast during the verification process, serving as a secondary representation of workload; R represents the total reward budget. Represents training nodes j In the t The reputation value of the wheel.

Images