A data tag tracing system and method for 5G communication

Abstract

The present application relates to the technical field of communication security, and relates to a data label tracing system and method for 5G communication, which comprises the following steps: collecting original data; performing differential privacy processing on the original data to generate initial label data; inputting the initial label data into data sources in a federal learning system, and performing local processing on the initial label data by each data source to generate local label model parameters; collecting the local label model parameters of each data source, and generating a global label model by summarizing through the federal learning system; recording the path of each data transmission in the process of data transmission to generate path record data; and using the global label model in combination with the path record data to analyze the network environment and generate optimized path data. The present application protects the path record data by differential privacy, dynamically adjusts the transmission path, performs real-time verification and correction, and conducts comprehensive compliance inspection, so that the accurate tracing of the data flow path is realized under the premise of ensuring privacy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication security, and more particularly, to a data label tracing system and method for 5G communication. BACKGROUND

[0002] In the current 5G network data governance, such as in the coal intelligent equipment, power and Internet of Things industries, privacy protection and path tracing of data flow is a major problem. Data frequently flows between different systems, networks and regions, especially when processing multi-source heterogeneous data, there are challenges such as complex data types and increasing privacy risks. Although traditional data label identification technology can help record and trace data flow paths, it is often difficult to ensure the accuracy of tracing while ensuring privacy protection.

[0003] How to ensure the privacy and compliance of data flow has become a key problem in data governance. Therefore, federated learning technology is introduced as a solution. The Chinese patent with publication number CN115623433A discloses a whole-count photovoltaic data sharing and value-added system and method based on 5G and federated learning. This solution aggregates photovoltaic terminal data to a 5G power multi-in-one fusion terminal through a 5G power production control private network, and then uploads it to a power dispatch center, realizing accurate monitoring and regulation of photovoltaic terminals, effectively balancing power safety protection requirements and costs. At the same time, the power grid company coordinates the construction of a photovoltaic monitoring data sharing and value-added platform, constructs exclusive transmission channels between the platform and the management systems of each party, and realizes the safe sharing of photovoltaic data among multiple parties.

[0004] However, how to realize accurate tracing of data flow paths while ensuring privacy is still a technical difficulty to be solved. SUMMARY

[0005] To solve the above problems, the embodiments of the present application provide a data label tracing method for 5G communication, the method comprising:

[0006] Collecting original data;

[0007] Performing differential privacy processing on the original data to generate initial label data;

[0008] Transmitting the initial label data into data sources in a federated learning system, each data source performing local processing on the initial label data to generate local label model parameters; collecting the local label model parameters of each data source, and generating a global label model by summarizing through the federated learning system;

[0009] In the process of data transmission, record the path of each data transmission to generate path record data;

[0010] The global label model is combined with path record data to analyze a network environment and generate optimized path data.

[0011] The global label model is verified using the optimized path data to obtain a verification report, the verification report includes consistency and inconsistency, the global label model with inconsistent verification report is modified to generate updated label data.

[0012] Further, the step of differential privacy processing includes:

[0013] S101: Layering privacy budget and grading sensitivity on the collected original data;

[0014] The privacy budget includes the first layer privacy budget, the second layer privacy budget, the third layer privacy budget and the fourth layer privacy budget; the sensitivity includes the first level sensitivity, the second level sensitivity, the third level sensitivity and the fourth level sensitivity;

[0015] S102: Using Laplace distribution to generate noise for each layer of privacy budget data items in the original data, and injecting noise into the corresponding data items;

[0016] S103: Generate independent labels for each layer of privacy budget data items, and inject labels into data items to generate single-item label data;

[0017] S104: Input all single-item label data into a pre-built merging optimization model to output initial label data; the merging optimization model is a weighted aggregation model.

[0018] Further, the federated learning system includes a data source and a central server; the step of local processing includes:

[0019] S201: Distribute the initial label data to each data source in the federated learning system, and each data source independently receives the initial label data;

[0020] S202: Each data source trains a local model based on the initial label data to generate local label model parameters.

[0021] Further, the training method of the local model includes:

[0022] All initial label data are used as input of the local model, the local model takes the predicted local label model parameters for each group of initial label data as output, takes the actual local label model parameters corresponding to each group of initial label data as prediction target, and takes the sum of all first prediction accuracy of predicted local label model parameters as training target; the local model is trained until the sum of first prediction accuracy reaches convergence to stop training; the local model is a convolutional neural network model.

[0023] Further, the generation method of the global label model comprises:

[0024] The central server sets an initial global label model, the parameters of the initial global label model being initial randomized parameters, receives the local label model parameters of each data source, uses a federated learning algorithm to summarize, generates global label model parameters, updates the initial global label model based on the global label model parameters, generates a global label model, and the central server sends the global label model back to each data source, each data source continues to train and iterate the previous round of global label model based on the initial label data, until the global label model converges.

[0025] Further, the generation method of the optimized path data comprises:

[0026] The initial label data is input into the global label model of each data source, and the transmission requirement is output; the transmission requirement and the path record data are input into the pre-constructed traceability transmission optimization model, and the optimized path data is output.

[0027] Further, the training method of the traceability transmission optimization model comprises:

[0028] All transmission requirements and path record data are taken as inputs of the traceability transmission optimization model, the traceability transmission optimization model takes the optimized path data predicted for each set of transmission requirements and path record data as output, takes the actual optimized path data corresponding to each set of transmission requirements and path record data as a prediction target, and takes the sum of the second prediction accuracy of all predicted optimized path data as a training target;

[0029] The traceability transmission optimization model is trained until the sum of the second prediction accuracy reaches convergence, and the training is stopped; the traceability transmission optimization model is a Pareto optimization model.

[0030] Further, the verification method is to compare the transmission requirement in the global label model with the actual performance indicators in the optimized path data, the comparison comprising delay comparison, bandwidth comparison and packet loss rate comparison; the actual performance indicators comprising path delay, bandwidth and packet loss rate; the inconsistency comprising delay inconsistency, bandwidth inconsistency and packet loss rate inconsistency;

[0031] Delay comparison: if the path delay in the transmission requirement is greater than the path delay in the optimized path data, it is marked as delay inconsistency; otherwise, it is marked as consistent;

[0032] Bandwidth comparison: if the bandwidth in the transmission requirement is less than the bandwidth in the optimized path data, it is marked as bandwidth inconsistency; otherwise, it is marked as consistent;

[0033] Packet loss rate comparison: if the packet loss rate in the transmission requirement is greater than the packet loss rate in the optimized path data, it is marked as inconsistent; otherwise, it is marked as consistent.

[0034] Further comprising:

[0035] Input path record data and update label data into the compliance verification mechanism, output verification result, verification result includes compliance and non-compliance.

[0036] Further, the steps of the compliance verification mechanism are:

[0037] Step S301: Set the prover P and the verifier V;

[0038] Step S302: The prover P encrypts the path record data of the transmission path, generates a commitment value C, and submits the commitment value C to the verifier V;

[0039] Step S303: The verifier V initiates a challenge to the prover P according to the compliance requirement;

[0040] Step S304: The prover P calculates the response value using the known path record data according to the challenge proposed by V; the prover P submits the generated response value to the verifier V;

[0041] Step S305: The verifier V uses the response value provided by P and compares it with the commitment value C generated in step S302 to verify and generate a verification result.

[0042] A data label tracing system for 5G communication, the system comprising:

[0043] Collection module: collect raw data;

[0044] Privacy label module: differential privacy processing of raw data to generate initial label data;

[0045] Federal collaborative module: input initial label data into federal learning system, each data source performs local processing to generate local label model parameters; collect local label model parameters of each data source, and generate global label model by summarizing through federal learning system;

[0046] Tracing record module: record the path of each data transmission in the process of data transmission to generate path record data;

[0047] Tracing optimization module: use global label model combined with path record data to analyze network environment and generate optimized path data;

[0048] Check update module: use the optimized path data to check the global label model, obtain the check report, the check report includes consistent and inconsistent, modify the global label model which is inconsistent with the check report, generate updated label data;

[0049] Compliance verification module: input path record data and updated label data into compliance verification mechanism, output verification result, verification result includes compliance and non-compliance;

[0050] Compliance detection module: check the compliance of the updated label data corresponding to the compliance, the updated label data corresponding to the non-compliance is marked as non-compliance and directly recorded in the compliance report data.

[0051] The technical effects and advantages of the 5G communication data label tracing system and method provided by the application are as follows:

[0052] The application realizes efficient and reliable privacy protection and path tracing by differential privacy protection of path record data, dynamic adjustment of transmission path, real-time checking and correction and comprehensive compliance check, and realizes accurate tracing of data flow path under the premise of ensuring privacy. The application processes the data by differential privacy, divides the privacy budget and sensitivity level, and injects noise to ensure that high-sensitive data is strongly protected, while low-sensitive data retains more precision, laying a foundation for privacy protection for subsequent path recording and tracing; In the data transmission process, the transmission path of each data is recorded to generate path record data, realizing accurate tracking and recording of data transmission path, and providing detailed and traceable basis for subsequent optimization and compliance check of the path; Combined with the global label model and the path record data, the network environment is analyzed and the optimized path data is generated, which ensures the dynamic optimization of the transmission path on the basis of privacy protection, makes the path tracing more accurate, and improves the efficiency and reliability of the transmission; The optimized path data is checked in real time to ensure that the path transmission demand is consistent with the actual performance, and the checking and updating process is continuously fed back and corrected to make the path tracing more accurate and ensure that each transmission path meets the changes of real-time network environment; Combined with the updated path and data label information, comprehensive compliance check is performed to ensure that the path tracing, data transmission and other operations meet the regulations, and the compliance of the data flow path is accurately traced and checked under the premise of ensuring privacy protection, ensuring that the path tracing process meets the compliance requirements. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 It is a connection schematic diagram of a 5G communication data label tracing system in embodiment 1;

[0054] Figure 2 It is a connection schematic diagram of a 5G communication data label tracing system in embodiment 2;

[0055] Figure 3 A flowchart of a data tag tracing method for 5G communication. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0057] Embodiment 1:

[0058] Please refer to Figure 1 As shown in the figure, the 5G communication data tag tracing system described in the present embodiment includes a collection module, a privacy tag module, a federal collaboration module, a tracing record module, a tracing optimization module, a verification and update module, and an inversion analysis module. Each module is connected through a blockchain network to realize data transmission between modules.

[0059] Collection module: collect raw data;

[0060] Privacy tag module: differentially private processing of raw data to generate initial tag data;

[0061] Federal collaboration module: input initial tag data into data sources in the federal learning system, each data source locally processes initial tag data to generate local tag model parameters; collect local tag model parameters of each data source, and generate a global tag model by summarizing through the federal learning system;

[0062] Tracing record module: record the path of each data transmission in the process of data transmission to generate path record data;

[0063] Tracing optimization module: use the global tag model in combination with the path record data to analyze the network environment and generate optimized path data;

[0064] Verification and update module: use the optimized path data to verify the global tag model to obtain a verification report, the verification report includes consistency and inconsistency, and modify the corresponding global tag model for the inconsistent verification report to generate updated tag data.

[0065] The original data refers to multi-source heterogeneous data generated in the process of network transmission, system storage, application interaction, etc., including user interaction data, system generated data, business data, sensor data and network transmission data. The user interaction data such as login information, account operation record, personal identity information and transaction information can be obtained through user input or automatic log record of the system. The system generated data such as system running log, background monitoring data, network traffic record and application layer data can be automatically collected by the background system, monitored and collected network traffic, or obtained from the database regularly. The business data such as transaction record in the financial system, medical record data in the medical system and order data in the e-commerce system can be exported from the business application database or grabbed through the system interface. The sensor data such as environmental information and device state data are data from Internet of Things devices, which can be collected by the Internet of Things devices through transmission. The network transmission data such as message header information, data packet content and transmission protocol are various data packets transmitted from the client to the server or from one server to another server in the network, which can be collected by network monitoring tools and network packet capture tools.

[0066] The steps of the differential privacy processing include:

[0067] S101: Hierarchical privacy budget and sensitivity classification of the collected original data;

[0068] The privacy budget includes the first layer privacy budget, the second layer privacy budget, the third layer privacy budget and the fourth layer privacy budget. The sensitivity includes the first level sensitivity, the second level sensitivity, the third level sensitivity and the fourth level sensitivity. As shown in Table 1, it is a classification table of the original data.

[0069] Table 1 Grading Table

[0070]

[0071] S102: Use Laplace distribution to generate noise for each layer of privacy budget data item in the original data, and inject the noise into the corresponding data item. The data item is user interaction data, system generated data, business data, sensor data and network transmission data.

[0072] The noise injection formula is:

[0073] ;

[0074] In the formula, i is the level of the data item, is the noise of the i-th level, is the sensitivity of the i-th level data item, is the privacy budget of the i-th level data item.

[0075] S103: generating independent labels for each layer of the data item of the privacy budget, and injecting the labels into the data item to generate single-label data;

[0076] S104: inputting all single-label data into a pre-built merging optimization model to output initial label data.

[0077] The merging optimization model is a weighted aggregation model, and the single-label data is , the corresponding weight is , and the formula of the merging optimization model is as follows:

[0078] ;

[0079] In the formula, That is, the smaller the privacy budget is, the higher the weight is; is the initial label data.

[0080] The federated learning system includes data sources and a central server.

[0081] The local processing step includes:

[0082] S201: distributing the initial label data to each data source in the federated learning system, and each data source independently receives the initial label data; the data source may be, for example, a plurality of devices, servers, nodes, etc.

[0083] S202: each data source trains a local model based on the initial label data to generate local label model parameters.

[0084] Through the differential privacy processing step, the system performs hierarchical privacy budget and sensitivity classification processing on the collected multi-source heterogeneous data such as user interaction data, system generated data, and business data, and ensures the privacy protection of different levels of data through the Laplace noise injection mechanism. The generated single-label data is further optimized through a weighted aggregation model to form global initial label data. In the entire process, the privacy protection and data utility are effectively balanced, ensuring that highly sensitive data such as personal identity information and transaction records are strongly protected, while less sensitive data such as system logs and network traffic records retain more precision and analysis value. The final generated initial label data not only meets the privacy protection requirements, but also provides an efficient and secure data foundation for subsequent data analysis and transmission optimization, ensuring the stability and compliance of system operation.

[0085] The local model training method includes:

[0086] all initial label data as input, the local model outputs local label model parameters predicted for each set of initial label data, takes the actual local label model parameters corresponding to each set of initial label data as a prediction target, and takes the sum of first prediction accuracies of all predicted local label model parameters as a training target.

[0087] wherein the calculation formula of the first prediction accuracy is: , is the number of each set of initial label data, is the first prediction accuracy, is the number of the i-th set of initial label data, is the predicted local label model parameter corresponding to the initial label data, is the number of the i-th set of initial label data, is the actual local label model parameter corresponding to the i-th set of initial label data; the local model is trained until the sum of the first prediction accuracies reaches convergence, and the training is stopped; the local model is a convolutional neural network model.

[0088] The local label model parameters are weights of the local model trained by each data source.

[0089] The introduction of the local model realizes independent model training of each data source without sharing the original data. Each data source generates local label model parameters based on the initial label data, and the parameters are trained by a convolutional neural network (CNN) to gradually optimize the model accuracy by minimizing the prediction error, thereby realizing privacy protection and efficient collaborative learning, and providing protection for the overall performance and security of the federated learning system.

[0090] The method for generating the global label model comprises:

[0091] The central server sets an initial global label model, the parameters of the initial global label model are initial randomized parameters, receives the local label model parameters of each data source, uses a federated learning algorithm to aggregate and generate global label model parameters, updates the initial global label model based on the global label model parameters to generate a global label model, and sends the global label model back to each data source. Each data source continues to train and iterate the global label model of the previous round based on the initial label data until the global label model converges.

[0092] The calculation method of the federated learning algorithm is:

[0093] ;

[0094] wherein, is the global label model parameter; is the local label model parameter of the j-th data source; The sample quantity of the data source; is the total sample quantity of all data sources; N is the total number of data sources participating in training, and j=1, 2,..., N.

[0095] The generation method of the global label model achieves technical effects of privacy protection, efficient collaborative learning, improvement of model accuracy and generalization ability, and fast convergence through distributed local training and aggregation of the federated learning algorithm, and provides a safe and efficient solution for model construction in complex data scenarios.

[0096] The path record data includes source device, target device, intermediate node, transmission time, protocol information and transmission state; the source device is the starting point of recording data transmission, such as a client or a sensor device; the target device is the final destination of recording data transmission, such as a server or a database; the intermediate node is the network device and the number of hops through which the recorded data passes from the source to the destination, such as a router or a switch; the transmission time is the delay or interruption information of data transmission in the timestamp of each node; the protocol information records the transmission protocol used and its related information; the transmission state is whether the transmission is successfully completed, whether there is packet loss, delay, etc.; the path record data is used to trace and record the actual path of data transmission, ensuring that the specific details of each data transmission are tracked.

[0097] The generation method of the optimized path data includes:

[0098] The initial label data is input into the global label model of each data source, and the transmission demand is output; the transmission demand and the path record data are input into the pre-constructed traceability transmission optimization model, and the optimized path data is output.

[0099] The training method of the traceability transmission optimization model includes:

[0100] All transmission demands and path record data are taken as inputs of the traceability transmission optimization model, the traceability transmission optimization model takes the predicted optimized path data for each set of transmission demand and path record data as output, takes the actual optimized path data corresponding to each set of transmission demand and path record data as a prediction target, and takes the sum of the second prediction accuracy of all predicted optimized path data as a training target;

[0101] The calculation formula of the second prediction accuracy is: , wherein, is the number of each set of transmission demand and path record data, is the second prediction accuracy, is the first predicted optimized path data corresponding to the transmission demand and path record data, is the first The group transmission demand corresponds to actual optimization path data of the path record data; a provenance transmission optimization model is trained until the sum of the second prediction accuracy reaches convergence to stop training; and the provenance transmission optimization model is a Pareto optimization model.

[0102] The generation method of the optimization path data realizes efficient data transmission path selection, real-time network adaptability, optimized use of network resources, and high reliability of transmission through intelligent analysis and dynamic adjustment strategies, ensuring stable and fast data transmission of the system in a complex network environment, and providing strong support for improving overall network performance and user experience.

[0103] The optimization path data refers to the best path scheme for data transmission from the source to the target generated by the provenance transmission optimization model. It contains detailed path information, dynamic adjustment strategies, and protocol configuration suggestions to ensure the efficiency, stability, and reliability of data transmission. The detailed path information includes, for example, intermediate nodes, hop count, delay, bandwidth, and packet loss rate.

[0104] Examples:

[0105] Suppose the provenance transmission optimization model generates optimization path data for a data transmission, then the data is as follows:

[0106] Source device: 192.168.1.1; the source device refers to the client device;

[0107] Target device: 10.10.5.5; the target device refers to the server;

[0108] Intermediate node list:

[0109] Router A (IP: 172.16.10.1);

[0110] Switch B (IP: 172.16.20.1);

[0111] Gateway F (IP: 172.16.30.1);

[0112] Hop count: 3 hops;

[0113] Expected delay: 10 milliseconds;

[0114] Bandwidth allocation: 1 Gbps;

[0115] Packet loss rate: 0.01%;

[0116] Network congestion status: Router A has mild congestion, and it is recommended to avoid this path;

[0117] Backup path:

[0118] Switch D (IP: 172.16.40.1);

[0119] Gateway E (IP: 172.16.50.1);

[0120] Switching condition: If the delay of Router A exceeds 50 ms, switch to the backup path;

[0121] Transmission protocol: TCP protocol;

[0122] Protocol configuration suggestion: Set the TCP window size to 64 KB and enable the fast retransmission mechanism.

[0123] The verification method is to compare the transmission requirements in the global label model with the actual performance indicators in the optimized path data, including delay comparison, bandwidth comparison, and packet loss rate comparison; the actual performance indicators include path delay, bandwidth, and packet loss rate; the inconsistencies include delay inconsistency, bandwidth inconsistency, and packet loss rate inconsistency.

[0124] Delay comparison: If the path delay in the transmission requirements is greater than the path delay in the optimized path data, it is marked as delay inconsistency; otherwise, it is marked as consistent;

[0125] Bandwidth comparison: If the bandwidth in the transmission requirements is less than the bandwidth in the optimized path data, it is marked as bandwidth inconsistency; otherwise, it is marked as consistent;

[0126] Packet loss rate comparison: If the packet loss rate in the transmission requirements is greater than the packet loss rate in the optimized path data, it is marked as packet loss rate inconsistency; otherwise, it is marked as consistent.

[0127] The correction method includes delay correction, bandwidth correction, and packet loss rate correction;

[0128] Delay correction: Replace the transmission path;

[0129] Bandwidth correction: Increase the total bandwidth by parallel transmission of the path;

[0130] Packet loss rate correction: Use forward error correction or retransmission mechanism to reduce the impact of packet loss.

[0131] The updated label data includes the updated transmission requirements and the optimized path strategy;

[0132] The path strategy, such as starting forward error correction or retransmission mechanism.

[0133] Compliance check refers to the analysis of updated label data to ensure that data processing, transmission, and storage processes comply with relevant regulations, industry standards, and enterprise internal security and privacy policies. Common compliance check items include: regulatory compliance, industry standard compliance, internal policy compliance, data transmission compliance, and privacy protection compliance;

[0134] Regulatory compliance:

[0135] Systems need to check if the data being processed complies with current regulations, especially those related to data privacy and cross-border data transfer;

[0136] Industry standard compliance:

[0137] Some industries have specific standards for data processing and transmission, such as the Payment Card Industry Data Security Standard (PCI DSS) for the financial industry and the Information Security Management System (ISMS) standard for the telecommunications industry;

[0138] Compliance checks will ensure that the data transmission complies with these industry-specific standards.

[0139] Internal policy compliance:

[0140] Each enterprise may have its internal compliance policies that require data to meet internal security and privacy standards when stored, processed, and transmitted, such as:

[0141] Data retention policy: Check if only the necessary data is stored according to the company's internal regulations, and if the expired data is deleted.

[0142] Data access control: Check if unauthorized personnel access to sensitive data is limited.

[0143] Data transmission compliance:

[0144] Need to check if the data transmission complies with relevant transmission standards, common checks include:

[0145] Encrypted transmission: Ensure that data transmission uses a strong enough encryption mechanism, such as SSL / TLS.

[0146] Transmission protocol: Confirm the use of transmission protocols that meet compliance requirements, such as HTTPS, FTPS, and IPSec.

[0147] Privacy protection compliance:

[0148] Check if the original data has been processed with sufficient differential privacy to ensure that sensitive information is not leaked during transmission and processing; ensure that data anonymization or pseudo-anonymization processing meets relevant regulatory requirements.

[0149] Compliance report data is a detailed report generated after completing the compliance check, showing whether the updated label data complies with the above compliance requirements, used to record the results of the check, and provide compliance proof for audit, compliance departments or external regulatory agencies.

[0150] Example:

[0151] Assuming that the system conducts compliance checks on a certain data transmission, the transmitted data uses differential privacy processing, uses SSL / TLS encryption during transmission, and meets the cross-border data transmission requirements, but a certain type of data is stored beyond the period and is not deleted according to internal policy requirements, so the generated compliance report data may include the following contents:

[0152] Compliance check results:

[0153] Privacy protection compliance: passed.

[0154] Encryption transmission compliance: passed.

[0155] Data cross-border transmission compliance: passed.

[0156] Data storage compliance: failed.

[0157] Privacy and security protection details:

[0158] Transmission protocol: HTTPS (compliant)

[0159] Data encryption: SSL / TLS encryption (compliant)

[0160] Differential privacy processing: applied (compliant)

[0161] Data transmission path compliance:

[0162] Transmission path: source device (EU) → transit node (USA) → target device (Japan)

[0163] Check: meets cross-border transmission requirements.

[0164] Data storage and retention compliance:

[0165] Certain type of transaction data has been stored beyond the period and is recommended to be deleted within 30 days.

[0166] Embodiment 2

[0167] As shown in Figure 2 , this embodiment further improves the design based on embodiment 1, the difference is that although embodiment 1 can conduct transmission path compliance check in actual operation, the specific information of the path will be exposed during the check process, increasing the risk of transmission path privacy leakage, based on this, a 5G communication data tag traceability system further includes:

[0168] Compliance verification module: input path record data and update tag data to compliance verification mechanism, output verification result, verification result includes compliance and non-compliance;

[0169] Compliance detection module: The compliance corresponding update label data is checked for compliance, and the non-compliance corresponding update label data is marked as non-compliance and directly recorded in the compliance report data.

[0170] The steps of the compliance verification mechanism are:

[0171] Step S301: Set the question:

[0172] The prover P has specific information in the transmission path, including nodes, protocols, hop counts, and other records in the path record data.

[0173] The verifier V wants to verify whether the transmission path meets the compliance requirements, but does not want to know the specific transmission path details;

[0174] Step S302: The prover P first encrypts the path record data of the transmission path to generate a "commitment value", which is the encrypted summary of the path record data. The commitment value contains the information of the transmission path, but does not disclose the specific details.

[0175] The calculation formula of the commitment value C is:

[0176] ;

[0177] In the formula, H is an encryption function or a hash function, which is used to encrypt the path record data;

[0178] The path record data contains detailed information such as the source device, target device, intermediate nodes, transmission protocol, and hop count of data transmission, but these information will not be disclosed in the commitment phase.

[0179] The prover P submits the encrypted commitment value C to the verifier V as a preliminary declaration of proving the compliance of the path. At this time, the verifier V cannot know the specific transmission details from the commitment value, such as the IP address of the intermediate node and the transmission protocol.

[0180] Step S303: The verifier V will initiate a challenge to the prover P according to the compliance requirements, requiring the prover P to prove certain transmission path characteristics without requiring the prover P to disclose specific path details.

[0181] For example, the challenge that the verifier V may propose includes:

[0182] Compliance of path nodes: requiring the prover P to prove that the nodes in the path, such as certain intermediate nodes, meet the encryption transmission requirements.

[0183] Cross-border transmission compliance: requiring P to prove that the data transmission path does not pass through unauthorized countries or regions.

[0184] Step S304: The prover P uses the known path record data to calculate a response value according to the challenge raised by the verifier V; the calculation method of the response value is that P combines the encrypted hash values of all nodes into an overall response value, which is a single encrypted result, to prove the compliance of the path to the verifier; the response value can prove that the path meets the compliance requirements, but will not leak specific path details; the prover P submits the generated response value to the verifier V.

[0185] For example, P can generate a response value according to whether each intermediate node uses an encrypted transmission protocol to prove that the path meets the encryption requirements, but will not directly display the IP address or protocol details of these nodes.

[0186] Step S305: The verifier V uses the response value provided by P and compares it with the previously generated commitment value C to verify and generate a verification result. In this way, V can verify whether the path meets the compliance requirements raised by it;

[0187] If the response value and the commitment value match and meet the compliance requirements, V can be sure that the path meets the regulations;

[0188] If the response value and the commitment value do not match, V will know that P's claim is not true, or the path does not meet the regulations.

[0189] In this way, the verifier V verifies the path compliance without obtaining the specific information of the transmission path, and the privacy of the path is protected.

[0190] Embodiment 3

[0191] As shown in Figure 3 A data label traceability method for 5G communication, the method comprising:

[0192] Collecting raw data;

[0193] Performing differential privacy processing on the raw data to generate initial label data;

[0194] Transmitting the initial label data into data sources in a federated learning system, each data source performing local processing on the initial label data to generate local label model parameters; collecting the local label model parameters of each data source, and generating a global label model by summarizing through the federated learning system;

[0195] In the process of data transmission, record the path of each data transmission to generate path record data;

[0196] Using the global label model in combination with the path record data to analyze the network environment and generate optimized path data;

[0197] The global label model is validated using optimized path data, and a validation report is obtained. The validation report includes consistent and inconsistent results. The global label model corresponding to the inconsistent validation report is corrected, and updated label data is generated.

[0198] Input path record data and update label data to the compliance verification mechanism, and output the verification result, which includes compliance and non-compliance;

[0199] The updated tag data corresponding to compliance is checked for compliance and a compliance report is generated. The updated tag data corresponding to non-compliance is marked as non-compliant and recorded directly in the compliance report data.

[0200] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0201] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A data tag tracing method for 5G communication, characterized in that, The methods include: Collect raw data; Differential privacy processing is performed on the raw data to generate initial label data; The steps of differential privacy processing include: S101: Stratify the collected raw data according to privacy budget and sensitivity. S102: Use a Laplace distribution to generate noise for each privacy budget data item in the original data and inject the noise into the corresponding data item; S103: Generate independent labels for each data item in the privacy budget layer, and inject the labels into the data items to generate single-item label data; S104: Input all individual label data into the pre-built merge optimization model and output the initial label data; Initial label data is fed into the data source of the federated learning system. Each data source processes the initial label data locally to generate local label model parameters. The local label model parameters of each data source are collected and aggregated through the federated learning system to generate a global label model. During data transmission, the path of each data transmission is recorded, generating path record data; Using a global labeling model combined with path recording data, network environment is analyzed to generate optimized path data; the method for generating the optimized path data includes: The initial label data is input into the global label model of each data source, and the transmission requirements are output; the transmission requirements and path record data are input into the pre-built source tracing transmission optimization model, and the optimized path data is output. All transmission demand and path record data are used as input to the source tracing transmission optimization model. The source tracing transmission optimization model takes the optimized path data predicted for each set of transmission demand and path record data as output, takes the actual optimized path data corresponding to each set of transmission demand and path record data as prediction target, and takes minimizing the sum of the second prediction accuracies of all predicted optimized path data as training target. The source tracing transmission optimization model is trained until the sum of the second prediction accuracies converges, at which point training stops; the source tracing transmission optimization model is a Pareto optimization model. The global label model is validated using optimized path data, and a validation report is obtained. The validation report includes consistent and inconsistent results. Global label models with inconsistent validation reports are corrected, and updated label data is generated. Input path record data and update label data to the compliance verification mechanism, and output the verification result, which includes compliance and non-compliance; The updated label data corresponding to compliance will undergo compliance checks, while the updated label data corresponding to non-compliance will be marked as non-compliant and directly recorded in the compliance report data.

2. The data tag tracing method for 5G communication according to claim 1, characterized in that: The privacy budget includes a first-level privacy budget, a second-level privacy budget, a third-level privacy budget, and a fourth-level privacy budget; the sensitivity includes a first-level sensitivity, a second-level sensitivity, a third-level sensitivity, and a fourth-level sensitivity. The merging optimization model is a weighted aggregation model; The federated learning system includes a data source and a central server; The local processing steps include: S201: Each data source receives initial label data independently; S202: Each data source trains a local model based on the initial label data and generates local label model parameters.

3. The data tag tracing method for 5G communication according to claim 2, characterized in that, The training methods for the local model include: All initial label data are used as input to the local model. The local model outputs the local label model parameters predicted for each set of initial label data, uses the actual local label model parameters corresponding to each set of initial label data as the prediction target, and minimizes the sum of the first prediction accuracies of all predicted local label model parameters as the training objective. The local model is trained until the sum of the first prediction accuracies converges, at which point training stops. The local model is a convolutional neural network model.

4. The data tag tracing method for 5G communication according to claim 1, characterized in that, The method for generating the global label model includes: The central server sets up an initial global label model with initially randomized parameters. After receiving the local label model parameters from each data source, it uses a federated learning algorithm to aggregate them and generate global label model parameters. Based on these global label model parameters, the central server updates the initial global label model and generates a global label model. The central server then sends the global label model back to each data source, and each data source continues to train and iterate the previous round of global label model based on the initial label data until the global label model converges.

5. The data tag tracing method for 5G communication according to claim 1, characterized in that, The method for validating the global label model using optimized path data is as follows: The transmission requirements in the global label model are compared with the actual performance indicators in the optimized path data. The comparison includes latency comparison, bandwidth comparison, and packet loss rate comparison. The actual performance indicators include path latency, bandwidth, and packet loss rate. Inconsistencies include latency inconsistency, bandwidth inconsistency, and packet loss rate inconsistency. Delay comparison: If the path delay in the transmission requirement is greater than the path delay in the optimized path data, it is marked as a delay inconsistency; Conversely, it is marked as consistent; Bandwidth comparison: If the bandwidth in the transmission requirement is less than the bandwidth in the optimized path data, it is marked as bandwidth inconsistency; Conversely, it is marked as consistent; Packet loss rate comparison: If the packet loss rate in the transmission requirement is greater than the packet loss rate in the optimized path data, it is marked as inconsistent packet loss rate; Conversely, it is marked as consistent.

6. A data tag tracing method for 5G communication according to claim 5, characterized in that, The compliance verification mechanism consists of the following steps: Step S301: Define the prover P and the verifier V; Step S302: Proofer P encrypts the path record data of the transmission path to generate a commitment value C, and prover P submits the commitment value C to verifier V; Step S303: Verifier V will challenge prover P according to compliance requirements; Step S304: Prover P calculates the response value using the known path record data based on the challenge proposed by verifier V; The prover P submits the generated response value to the verifier V; Step S305: Verifier V uses the response value provided by prover P and compares prover P with the commitment value C generated in step S302 to generate a verification result.

7. A 5G communication data tag traceability system, characterized in that, The system is used to implement a data tag tracing method for 5G communication as described in any one of claims 1-6, the system comprising: Collection module: Collects raw data; Privacy Tagging Module: Performs differential privacy processing on the raw data to generate initial tag data; the differential privacy processing steps include: S101: Stratify the collected raw data according to privacy budget and sensitivity. S102: Use a Laplace distribution to generate noise for each privacy budget data item in the original data and inject the noise into the corresponding data item; S103: Generate independent labels for each data item in the privacy budget layer, and inject the labels into the data items to generate single-item label data; S104: Input all individual label data into the pre-built merge optimization model and output the initial label data; Federated Collaboration Module: Initial label data is input into the federated learning system, and each data source processes the data locally to generate local label model parameters; the local label model parameters of each data source are collected and aggregated through the federated learning system to generate a global label model; Traceability and Recording Module: During data transmission, it records the path of each data transmission and generates path record data; The traceability optimization module uses a global labeling model combined with path record data to analyze the network environment and generate optimized path data. The method for generating the optimized path data includes: The initial label data is input into the global label model of each data source, and the transmission requirements are output; the transmission requirements and path record data are input into the pre-built source tracing transmission optimization model, and the optimized path data is output. All transmission demand and path record data are used as input to the source tracing transmission optimization model. The source tracing transmission optimization model takes the optimized path data predicted for each set of transmission demand and path record data as output, takes the actual optimized path data corresponding to each set of transmission demand and path record data as prediction target, and takes minimizing the sum of the second prediction accuracies of all predicted optimized path data as training target. The source tracing transmission optimization model is trained until the sum of the second prediction accuracies converges, at which point training stops; the source tracing transmission optimization model is a Pareto optimization model. Verification and Update Module: Uses optimized path data to verify the global label model and obtains a verification report. The verification report includes consistent and inconsistent results. The global label model with inconsistent verification reports is corrected and updated label data is generated. Compliance verification module: Input path record data and update tag data to the compliance verification mechanism, and output verification results, including compliance and non-compliance; Compliance Detection Module: Performs compliance checks on the updated tag data corresponding to compliance, and marks the updated tag data corresponding to non-compliance as non-compliant and records it directly in the compliance report data.

Images