Federal relay transmission method, federal relay transmission system and terminal equipment

Abstract

The invention discloses a federal relay transmission method and system and terminal equipment, and the transmission method comprises the steps: obtaining a model increment generated by each client in a local training model, and carrying out the fragmentation and encryption processing, and obtaining a data unit set; obtaining a path list currently issued by the aggregator, distributing a transmission path for each data unit in the data unit set, and packaging routing information; multi-hop forwarding is carried out on the data unit through the relay node, link index information is recorded and reported to the aggregator, and the aggregator updates the path list based on the link index information; and verifying, decrypting and recombining the received data units in the aggregator, and updating the global model based on the model increment recombined by all the clients. The method is applied to the field of encryption transmission, safe, reliable and efficient transmission of model increment in federated learning can be effectively achieved, and data privacy is protected and the method adapts to a complex network environment through end-to-end encryption, fragmentation parallel transmission and dynamic multi-hop path optimization.

Description

Technical Field

[0001] This invention relates to the field of encrypted transmission technology, specifically a federated relay transmission method, system, and terminal equipment. Background Technology

[0002] Federated learning is a distributed machine learning framework where clients train models locally, generate model increments, and upload them to an aggregator for global model updates. During transmission, data privacy (to prevent leaks), efficiency (for fast transmission), and reliability (to cope with network instability) must be ensured. Traditional methods typically transmit the entire model increment directly via a single path or use simple relay nodes for forwarding, but these face security and efficiency challenges and suffer from the following drawbacks: Low transmission efficiency: 1) Single-path transmission leads to long latency for large model increments, especially in low-bandwidth or high-latency networks; 2) Lack of parallel transmission mechanism fails to fully utilize network concurrency. Security shortcomings: 1) Although end-to-end encryption is provided, the single path is vulnerable to traffic analysis attacks; 2) Relay nodes may become single points of attack targets, lacking risk diversification mechanisms. Poor reliability: 1) Single path is susceptible to congestion, packet loss or node failure, and there is no redundant path support; 2) Retransmission mechanism increases latency and is not optimized for out-of-order or partial loss scenarios. Weak network adaptability: 1) Static path selection cannot adapt to dynamic network changes (such as relay node load, link congestion); 2) Lack of real-time path monitoring and load balancing mechanisms.

[0003] In addition, recent studies have explored encrypted transmission (such as differential privacy and encryption protocols) and multipath transmission, but there is still a lack of comprehensive solutions that take into account security, efficiency and adaptability to complex networks. Summary of the Invention

[0004] To address the shortcomings of the existing technologies, this invention provides a federated relay transmission method, system, and terminal device, which can effectively achieve secure, reliable, and efficient transmission of model increments in federated learning. Through end-to-end encryption, fragmented parallel transmission, and dynamic multi-hop path optimization, it protects data privacy and adapts to complex network environments.

[0005] To achieve the above objectives, the present invention provides a federated relay transmission method, comprising the following steps: Step 1: Obtain the model increment generated by each client during local training, and perform sharding and encryption processing on the model increment to obtain the data unit set corresponding to the model increment. Step 2: Obtain the list of paths currently issued by the aggregator, and assign a transmission path to each data unit in the data unit set, encapsulating routing information; Step 3: The data unit is forwarded via multiple hops through the relay node. At the same time, the link indicator information is recorded and reported to the aggregator. The aggregator updates the path list based on the link indicator information. Step 4: The aggregator verifies, decrypts, and reassembles the received data units, and incrementally updates the global model based on the reassembled models from all clients.

[0006] In one embodiment, step 1 involves the process of fragmenting and encrypting the model increment as follows: Increment the model Cut into Each segment of equal size Assign a unique index to each fragment This indicates its increment in the model. The position in the middle; Using symmetric keys Encrypt each fragment Generate ciphertext and for each ciphertext Generate authentication labels ; index Embedded encrypted routing header The inner layer ultimately outputs the model increment. The corresponding data unit set is .

[0007] In one embodiment, in step 2, a path is selected from the path list based on link quality or a randomization strategy. A transmission path is assigned, and at least one transmission path is allocated to each data unit in the data unit set, with the transmission path appended to the encrypted routing header of each data unit. The outer layer generates hop-by-hop routing information for each relay node on the transmission path, where... .

[0008] In one embodiment, step 3, the process of multi-hop forwarding of the data unit through the relay node, is as follows: Relay nodes receive data units sent by clients or previous relay nodes. Then, parse the encrypted routing header. The outermost layer uses its private key to decrypt and obtain the address of the next-hop relay node or aggregator, then forwards the current data unit. To the next hop node address, where keep , constant.

[0009] In one embodiment, in step 3, the link metric information includes bandwidth, latency, and packet loss rate; The process by which the aggregator updates the path list based on the link metric information is as follows: After filtering out abnormal data, calculate the value of each transmission path. , , Calculate its overall performance score ,for:

[0010] The transmission paths are sorted from highest to lowest score to generate a path list.

[0011] In one embodiment, step 4, the process of verifying, decrypting, and reassembling the received data unit includes: For the received data unit set Each data unit is verified, i.e., the authentication label is verified. Is it equal to If verification fails, the corresponding data unit is discarded and an error log is recorded. For data units that pass verification, use the key. Decrypting the encrypted routing header Inner layer, extract index Use key Decryption Obtain plaintext fragments And fragment the plaintext Stored in the corresponding slot of the buffer; Check if the buffer has been collected. One valid plaintext fragment: If complete, then concatenate the model increments in index order. ; Otherwise, discard the model increment currently transmitted by the client. And record the error log.

[0012] In one embodiment, step 4, the process of incrementally updating the global model based on the reorganized model from all clients, specifically involves:

[0013] in, This is the weight matrix after the global model update. The weight matrix before the global model update. For the first The model increment generated by the locally trained model on each client. For the first The weighting coefficients for each client.

[0014] To achieve the above objectives, the present invention also provides a federated relay transmission system, which performs federated relay transmission using the above method. The federated relay transmission system includes: The sharding encryption unit is used to shard and encrypt the model increment generated by each client training the model locally, so as to obtain the data unit set corresponding to the model increment. The path allocation unit is used to obtain the list of paths currently issued by the aggregator, allocate a transmission path to each data unit in the data unit set, and encapsulate routing information. Relay nodes are used for multi-hop forwarding of data units, while recording link indicator information and reporting it to the aggregator; Aggregators are used to verify, decrypt, and reassemble received data units; The global update unit is used to incrementally update the global model based on the reorganized model from all clients.

[0015] To achieve the above objectives, the present invention also provides a terminal device, wherein the terminal device is provided with: Memory, used to store programs; A processor is configured to execute the program stored in the memory, and when the program is executed, the processor is configured to perform the method as described above.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: 1. This invention shortens transmission time and effectively improves transmission efficiency by using fragmentation and multi-path parallel transmission; 2. This invention effectively enhances transmission security by fragmenting risks, using layered EncHeaders to protect data privacy, and resisting traffic analysis. 3. This invention addresses packet loss and out-of-order delivery through multi-path redundancy and index-driven reassembly, and ensures reliable delivery of fragments through a timeout retransmission mechanism, thereby improving reliability; 4. This invention effectively optimizes network adaptability by dynamically monitoring path performance (latency, bandwidth, packet loss rate), adjusting the path list, balancing the load, and adapting to complex networks. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the principle architecture of the federated relay transmission method in this embodiment of the invention. Figure 2 This is a flowchart of the federated relay transmission method in an embodiment of the present invention; Figure 3 This is a structural block diagram of the federal relay transmission system in an embodiment of the present invention; Figure 4 This is a structural block diagram of the terminal device in an embodiment of the present invention.

[0019] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0022] Example 1 This embodiment discloses a federated relay transmission method aimed at achieving secure, reliable, and efficient transmission of model increments in federated learning. Through end-to-end encryption, fragmented parallel transmission, and dynamic multi-hop path optimization, it protects data privacy and adapts to complex network environments. The method mainly comprises three core components: a client, relay nodes, and an aggregator, which collaboratively complete the entire process from local training to global model updates. Design goals include high security (preventing data leakage), high reliability (fault tolerance and redundancy), and high efficiency (parallel transmission and load balancing).

[0023] The client consists of distributed devices such as mobile devices, IoT terminals, and edge servers, responsible for local model training and data fragment transmission. Local model training refers to preprocessing local data and calculating model increments. Fragmentation and Encryption: The data is divided into M segments, encrypted, and labeled with integrity tags and indexes. Path selection and encapsulation: Based on the path list provided by the aggregator, multi-hop paths are selected and data units are encapsulated.

[0024] Local data can include electricity consumption data, sales data, etc. For example, electricity consumption data could be:

[0025] Relay nodes, such as edge gateways and ISP backbone nodes deployed in critical network locations, are responsible for forwarding encrypted data and have no access to plaintext. Data forwarding refers to forwarding encrypted segments along a specified path. In addition, relay nodes also perform link monitoring functions, periodically collecting metrics such as bandwidth, latency, and packet loss rate to generate status reports, and reporting link status and forwarding statistics to the aggregator.

[0026] The aggregator, whether centralized or distributed, is responsible for global aggregation, path optimization, and data reassembly. Path management refers to maintaining the network topology and dynamically generating and distributing path lists. Shard verification and reassembly refers to verifying the integrity of fragments, decrypting and reassembling the model increments generated by each client's locally trained model in index order. Global aggregation refers to weighted averaging of the model increments from all clients to update the global model.

[0027] refer to Figure 1 , Figure 2 The federated relay transmission method in this embodiment specifically includes the following steps: Step 1: Obtain the model increment generated by each client during local training, and perform sharding and encryption on the model increment to obtain the data unit set corresponding to the model increment. Here, the model increment refers to the neural network parameter matrix. Step 2: Obtain the list of paths currently issued by the aggregator, assign a transmission path to each data unit in the data unit set, and encapsulate routing information; Step 3: The data unit is forwarded via multiple hops through the relay node. At the same time, the link indicator information is recorded and reported to the aggregator. The aggregator updates the path list based on the link indicator information. Step 4: The aggregator verifies, decrypts, and reassembles the received data units, and incrementally updates the global model based on the reassembled models from all clients.

[0028] In the specific implementation of step 1, the process of sharding and encrypting the model increment is as follows: First, increment the model Cut into Each segment of equal size Assign a unique index to each fragment This indicates its increment in the model. The position in the middle; Then, using a symmetric key Encrypt each fragment Generate ciphertext and for each ciphertext Generate authentication labels ; index Embedded encrypted routing header The inner layer ultimately outputs the model increment. The corresponding data unit set is .

[0029] In this embodiment, the encrypted routing header employs a layered encryption design, similar to onion routing, consisting of an outer layer and an inner layer. The outer layer of the encrypted routing header primarily contains hop-by-hop routing information, including the next-hop address of the current relay node. For example, if the current relay node is relay node 1, the next address is relay node 2. The outer layer uses hop-by-hop encryption, generating an encryption layer for each relay node on the path. This means that the relay node uses its private key to decrypt and only obtains the next-hop address, allowing it to quickly determine the forwarding target without accessing sensitive data. The inner layer of the encrypted routing header uses end-to-end encryption, meaning only the aggregator can decrypt it. It mainly contains the shard index. This includes complete path information, etc. The inner layer of the encrypted routing header uses a symmetric key shared between the client and the aggregator. Encryption ensures that relay nodes cannot access the information, thereby protecting sensitive information (index). ), providing aggregators with the ability to reorganize model increments on demand.

[0030] The advantages of the layered encryption design of the encrypted routing header in this embodiment are mainly: high security, with inner-layer encryption ensuring index security. Sensitive information is readable only by the aggregator; high efficiency: the outer layer provides simple routing information, supporting fast forwarding; high flexibility: supports plaintext or hop-by-hop encryption to adapt to different security requirements. Furthermore, by fragmenting and encrypting the model increment generated by the client's locally trained model, not only is it impossible to reconstruct the model increment due to the leakage of a single fragment, thus dispersing the transmission risk, but the smaller fragments are also suitable for low-bandwidth networks and support parallel transmission.

[0031] In the specific implementation of step 2, the path list is selected based on link quality or randomization strategy. One transmission path, Link quality-based refers to selecting a path from the list based on a comprehensive performance score. highest A transmission path. For example, the client may select one of the following three paths: Path 1: Client -> Relay Node 1 -> Relay Node 2 -> Aggregator.

[0032] Path 2: Client -> Relay Node 3 -> Relay Node 4 -> Aggregator.

[0033] Path 3: Client -> Relay Node 5 -> Aggregator.

[0034] In selection After assigning a transmission path, allocate at least one transmission path to each data unit in the data unit set, and append the transmission path to the encrypted routing header of each data unit. The outer layer generates hop-by-hop routing information for each relay node on the transmission path. For example, the first hop of path 1 (relay node 1 -> relay node 2) is encrypted using the public key of relay node 1, and the second hop (relay node 2 -> aggregator) is encrypted using the public key of relay node 2. The inner layer remains unchanged. Preferably, the client can selectively add key segments within... Multiple redundant paths are set at the outer layer, for example This segment traverses both path 1 and path 2 simultaneously to handle packet loss. The final output is a set of data units with path information encapsulated in an encrypted routing header, which is then assigned to the specified path.

[0035] In the specific implementation of step 3, the process of multi-hop forwarding of data units through relay nodes is as follows: Relay nodes receive data units sent by clients or previous relay nodes. Then, parse the encrypted routing header. The outermost layer uses its private key to decrypt and obtain the address of the next-hop relay node or aggregator, then forwards the current data unit. To the next hop node address, where keep , constant.

[0036] The link metrics recorded by relay nodes include basic metrics: bandwidth, latency, and packet loss rate. The process by which the aggregator updates the path list based on these link metrics is as follows: First, filter out abnormal data, such as data with a packet loss rate >100%, to ensure accurate assessment; Then, calculate the transmission path. , , Calculate its overall performance score ,for:

[0037] For example, for path 1 (relay node 1 -> relay node 2, delay 100ms, bandwidth 50MB / s, packet loss rate 5%), substituting into the above formula yields... =0.59; Finally, the transmission paths are sorted from highest to lowest score to generate a path list, for example, path 3 > path 1 > path 2.

[0038] In the specific implementation of step 4, the process of verifying, decrypting, and reassembling the received data unit includes: For the received data unit set Each data unit is verified, i.e., the authentication label is verified. Is it equal to If verification fails, the corresponding data unit is discarded and an error log is recorded (e.g., client 1's shard 3 verification failed, suspected of being tampered with); in addition, abnormal data units, such as duplicate indexes, are also discarded. To prevent malicious injection from unknown clients, etc.; For data units that pass verification, use the key. Decrypting the encrypted routing header Inner layer, extract index Use key Decryption Obtain plaintext fragments And fragment the plaintext Stored in the corresponding slot of the buffer; Then, fragment reassembly is performed, and the buffer is checked to see if the fragments have been collected. One valid plaintext fragment: If complete, then concatenate the model increments in index order. ; Otherwise, to ensure the effectiveness and security of the aggregation model are not affected by the current client's model increment... Impact, discarding the current client-transmitted model increment And record error logs, such as incomplete model increment set for client 1.

[0039] After concatenating all client model increments, a weighted average is applied to all recombined client model increments to obtain an updated global model. The global model parameters are verified, the aggregation results are saved, and the final global model is generated. The specific process of updating the global model is as follows:

[0040] in, This is the weight matrix after the global model update. The weight matrix before the global model update. For the first The model increment generated by the locally trained model on each client. For the first The weighting coefficients for each client.

[0041] It is worth noting that, although this embodiment Figure 2 The steps are shown sequentially as indicated by the arrows, but they are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are performed; they can be executed in other orders. Figure 2At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0042] Example 2 Based on the federated relay transmission method in Embodiment 1, this embodiment discloses a federated relay transmission system, referencing... Figure 3 The federated relay transmission system includes fragmentation encryption units and path allocation units deployed at the client, as well as relay nodes, aggregators, and global update units, specifically: The sharding encryption unit is used to shard and encrypt the model increments generated by each client's local training model to obtain the data unit set corresponding to the model increment. The path allocation unit is used to obtain the list of paths currently issued by the aggregator, and to allocate a transmission path to each data unit in the data unit set, encapsulating routing information; Relay nodes are used to perform multi-hop forwarding of data units, while recording link indicator information and reporting it to the aggregator; Aggregators are used to verify, decrypt, and reassemble received data units; The global update unit is used to incrementally update the global model based on the reorganized model from all clients.

[0043] In this embodiment, the specific working process and working principle of the fragmentation encryption unit, path allocation unit, relay node, aggregator, and global update unit are the same as those in Embodiment 1, and therefore will not be described again in this embodiment. Each unit module can be implemented entirely or partially through software, hardware, or a combination thereof. Each unit module can be embedded in or independent of the processor in the computer device in hardware form, or it can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above unit modules.

[0044] Example 3 like Figure 4 The diagram illustrates a terminal device disclosed in this embodiment, comprising a transmitter, a receiver, a memory, and a processor. The transmitter transmits instructions and data, the receiver receives instructions and data, the memory stores computer-executed instructions, and the processor executes the computer-executed instructions stored in the memory to implement the method described in Embodiment 1 above.

[0045] It is important to note that the aforementioned memory can be either standalone or integrated with the processor. When the memory is set up independently, the terminal device also includes a bus for connecting the memory and the processor.

[0046] The above description is only a preferred embodiment of the present invention and does not limit the scope of protection of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the scope of protection of the present invention.

Claims

1. A federated relay transmission method, characterized in that, Includes the following steps: Step 1: Obtain the model increment generated by each client during local training, and perform sharding and encryption processing on the model increment to obtain the data unit set corresponding to the model increment. Step 2: Obtain the list of paths currently issued by the aggregator, and assign a transmission path to each data unit in the data unit set, encapsulating routing information; Step 3: The data unit is forwarded via multiple hops through the relay node. At the same time, the link indicator information is recorded and reported to the aggregator. The aggregator updates the path list based on the link indicator information. Step 4: The aggregator verifies, decrypts, and reassembles the received data units, and incrementally updates the global model based on the reassembled models from all clients.

2. The federated relay transmission method according to claim 1, characterized in that, In step 1, the process of sharding and encrypting the model increment is as follows: Increment the model Cut into Each segment of equal size Assign a unique index to each fragment This indicates its increment in the model. The position in the middle; Using symmetric keys Encrypt each fragment Generate ciphertext and for each ciphertext Generate authentication labels ; index Embedded encrypted routing header The inner layer ultimately outputs the model increment. The corresponding data unit set is .

3. The federated relay transmission method according to claim 2, characterized in that, In step 2, the path list is selected based on link quality or a randomization strategy. A transmission path is assigned, and at least one transmission path is allocated to each data unit in the data unit set, with the transmission path appended to the encrypted routing header of each data unit. The outer layer generates hop-by-hop routing information for each relay node on the transmission path, where... .

4. The federated relay transmission method according to claim 3, characterized in that, In step 3, the process of multi-hop forwarding of data units through relay nodes is as follows: Relay nodes receive data units sent by clients or previous relay nodes. Then, parse the encrypted routing header. The outermost layer uses its private key to decrypt and obtain the address of the next-hop relay node or aggregator, then forwards the current data unit. To the next hop node address, where keep , constant.

5. The federated relay transmission method according to claim 1, 2, 3, or 4, characterized in that, In step 3, the link metrics information includes bandwidth, latency, and packet loss rate; The process by which the aggregator updates the path list based on the link metric information is as follows: After filtering out abnormal data, calculate the value of each transmission path. , , Calculate its overall performance score ,for: The transmission paths are sorted from highest to lowest score to generate a path list.

6. The federated relay transmission method according to claim 2, 3, or 4, characterized in that, Step 4, the process of verifying, decrypting, and reassembling the received data units, includes: For the received data unit set Each data unit is verified, i.e., the authentication label is verified. Is it equal to If verification fails, the corresponding data unit is discarded and an error log is recorded. For data units that pass verification, use the key. Decrypting the encrypted routing header Inner layer, extract index Use key Decryption Obtain plaintext fragments And fragment the plaintext Stored in the corresponding slot of the buffer; Check if the buffer has been collected. One valid plaintext fragment: If complete, then concatenate the model increments in index order. ; Otherwise, discard the model increment currently transmitted by the client. And record the error log.

7. The federated relay transmission method according to claim 1, 2, 3, or 4, characterized in that, In step 4, the process of incrementally updating the global model based on the reorganized model from all clients is specifically as follows: in, This is the weight matrix after the global model update. The weight matrix before the global model update. For the first The model increment generated by the locally trained model on each client. For the first The weighting coefficients for each client.

8. A federal relay transmission system, characterized in that, Federated relay transmission is performed using the method described in any one of claims 1 to 7, wherein the federated relay transmission system comprises: The sharding encryption unit is used to shard and encrypt the model increment generated by each client training the model locally, so as to obtain the data unit set corresponding to the model increment. The path allocation unit is used to obtain the list of paths currently issued by the aggregator, allocate a transmission path to each data unit in the data unit set, and encapsulate routing information. Relay nodes are used for multi-hop forwarding of data units, while recording link indicator information and reporting it to the aggregator; Aggregators are used to verify, decrypt, and reassemble received data units; The global update unit is used to incrementally update the global model based on the reorganized model from all clients.

9. A terminal device, characterized in that, The terminal device is equipped with: Memory, used to store programs; A processor for executing the program stored in the memory, wherein when the program is executed, the processor is configured to perform the method as described in any one of claims 1 to 7.

Images