Cross-enterprise industrial internet identification analysis data sharing and prediction method based on federal learning

Abstract

The invention discloses a method for sharing and predicting cross-enterprise industrial internet identification analysis data based on federated learning. The method comprises the following steps: establishing a longitudinal federated learning environment; registering an identifier for enterprise equipment, performing standardization processing on local data generated by each piece of equipment, and performing data association on the obtained standardized data and the corresponding identifier; constructing an initial global model, issuing the initial global model to each enterprise node through a federated learning coordination server, and performing iterative training on the initial global model based on encryption gradient parameters fed back by each enterprise to determine a final global model; judging whether the equipment state is changed or not; and when the state of the equipment changes, inputting real-time operation data of the equipment into the final global model, and outputting state prediction data of the equipment. According to the invention, while efficient data sharing and data privacy protection are carried out, accurate prediction of the fault rate of enterprise equipment is improved.

Description

Technical Field

[0001] This application relates to the field of industrial internet identifier resolution technology, and more specifically, to a cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning. Background Technology

[0002] With the gradual improvement and development of the Industrial Internet, relying solely on the identifier resolution system can no longer meet the needs of high-efficiency, high-robust, and high-security industrial production scenarios. At the same time, due to the increasing importance of information security, the phenomenon of data silos is becoming more serious, making it difficult to achieve deep cross-enterprise collaboration.

[0003] Currently, traditional industry approaches include three main methods: one is data analysis and model training based on a centralized data warehouse; the other is for each company to build models using its own data; and the third is data access based on APIs, where companies open API interfaces to provide data access permissions to other companies, allowing external systems to query specific data.

[0004] The existing method has the following problems: Centralized data warehouses are highly relied upon and pose a single point of failure risk, while also carrying an extremely high risk of data leakage when collaborating across enterprises.

[0005] Enterprises that use their own data to build models independently may be limited by the single data source and the lack of data volume.

[0006] API-based data access faces significant development and maintenance costs, and data exposure issues persist even with weak authentication mechanisms.

[0007] Therefore, the existing technology has defects and urgently needs improvement. Summary of the Invention

[0008] In view of the above problems, the purpose of this invention is to provide a cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning, through the cross-combination of vertical federated learning and the industrial internet identifier system. It aims to achieve efficient data sharing and data privacy protection while improving the accuracy of predicting enterprise equipment failure rates, enhancing industrial supply chain collaboration, developing personalized and customized products, and ultimately improving the competitiveness of the entire supply chain.

[0009] The first aspect of this invention provides a cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning, comprising: Obtain enterprise needs; A vertical federated learning environment will be built based on the aforementioned enterprise needs. Register corresponding identifiers based on the identification codes assigned to the enterprise's equipment, standardize the local data generated by each device, and associate the standardized data with the corresponding identifiers; the local data includes equipment operation data and equipment maintenance data. An initial global model is constructed and distributed to each enterprise node through a federated learning coordination server. The initial global model is iteratively trained based on the cryptographic gradient parameters fed back by each enterprise to determine the final global model. The federated learning coordination server has a three-layer architecture, namely the identifier resolution layer, the federated learning layer, and the security layer. Obtain real-time operating data of the equipment; The real-time operating data of the device is analyzed to determine whether the device status has changed. When the device status changes, the real-time operating data of the device is input into the final global model, and the device status prediction data is output.

[0010] In this solution, the construction of a vertical federated learning environment based on the enterprise's needs includes: Based on the enterprise's needs, analyze the data sharing objectives, and formulate corresponding data sharing agreements; Assign identification codes to enterprise equipment based on the Industrial Internet Identifier Resolution System; Each enterprise sets up a federated learning client locally and configures it on the aggregation server of the federated learning coordinator to complete the construction of the vertical federated learning environment.

[0011] In this solution, the step of registering corresponding identifiers based on the identification codes assigned to enterprise equipment, standardizing the local data generated by each device, and associating the resulting standardized data with the corresponding identifiers includes: Based on the identifier resolution node, register the corresponding identifier according to the identifier code assigned to the device; Based on the industrial internet identifier resolution system, enterprise nodes parse the local data of devices into a standard format to obtain standardized data; Based on the identifier, the corresponding standardized data is stored in the local database, and a mapping relationship between the identifier and the storage location of the corresponding standardized data is established through the identifier resolution node.

[0012] In this scheme, the construction of an initial global model involves distributing the initial global model to each enterprise node via a federated learning coordination server, and iteratively training the initial global model based on the encrypted gradient parameters fed back by each enterprise to determine the final global model. This includes: The enterprise node analyzes the received initial global model to determine the identifiers used for model training; The corresponding standardized data is filtered from the local database using the identifiers used for model training; A local model is trained based on the corresponding standardized data, and gradient parameters are calculated; the local model is the global model received by the enterprise through the enterprise node. The gradient parameters are encrypted to determine the encrypted gradient parameters, and the encrypted gradient parameters are uploaded to the federated learning coordination server through the enterprise node. The initial global model is updated by aggregating the encrypted gradient parameters of each enterprise through a federated learning coordination server. The updated global model is then distributed to each enterprise node for iterative training until the global model converges, resulting in the final global model training.

[0013] In this solution, when the device state changes, the real-time operating data of the device is input into the final global model, and device state prediction data is output, including: The device's real-time operating data is used to obtain the device's operating data within a preset time interval; Based on the device's identifier, shared data within the scope of the data sharing protocol is obtained from the Industrial Internet Identifier Resolution System; The device operation data and the shared data within a preset time interval are aligned, and the aligned data is input into the final global model to output device status prediction data.

[0014] This plan also includes: The industrial internet identifier resolution system includes an identifier registration module, an identifier query module, and a data access control module.

[0015] In this scheme, encrypting the gradient parameters and determining the encrypted gradient parameters includes: Obtain the sensitivity classification of the sample ID corresponding to the gradient parameter; When the sensitivity level of the sample ID is high sensitivity, the gradient parameters are encrypted using homomorphic encryption; When the sensitivity level of the sample ID is low, the gradient parameters are encrypted using differential privacy.

[0016] This plan also includes: When the device status changes, the identifier resolution node notifies the federated learning coordination server through an event subscription mechanism, triggering global model iterative training. The federated learning coordination server dynamically adjusts the training cycle and hyperparameters of the global model based on the update status of the sample ID.

[0017] This plan also includes: The data access records in the identifier resolution, the parameter aggregation logs in the federated learning, and the encrypted hash values ​​of the prediction results are synchronously written to the blockchain for blockchain notarization.

[0018] This plan also includes: Before receiving the encrypted gradient parameters uploaded by each enterprise, the federated learning coordination server verifies the validity of the encrypted gradient parameters through identifier resolution.

[0019] This invention discloses a cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning. The method includes: building a vertical federated learning environment; registering identifiers for enterprise devices; standardizing the local data generated by each device; associating the standardized data with the corresponding identifiers; constructing an initial global model; distributing the initial global model to each enterprise node through a federated learning coordination server; iteratively training the initial global model based on the encrypted gradient parameters fed back by each enterprise to determine the final global model; determining whether the device status has changed; and when the device status changes, inputting the device's real-time operating data into the final global model and outputting device status prediction data. This invention improves the accuracy of predicting enterprise device failure rates while achieving efficient data sharing and data privacy protection.

[0020] 1. Efficient and accurate data sharing: Through the industrial internet identifier resolution system, automatic and accurate matching of cross-enterprise data is achieved, improving the accuracy of sample alignment and supporting full feature-tag association.

[0021] 2. High-performance joint modeling: By adopting longitudinal federated learning and dynamic gradient aggregation, the model prediction accuracy is improved and the convergence speed is accelerated.

[0022] 3. Strong privacy and security protection: Combining TEE secure computing with identifier-level encrypted access control, the risk of gradient leakage is extremely small, and blockchain notarization ensures that the entire process is auditable. Attached Figure Description

[0023] Figure 1 The flowchart illustrates a cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning provided by the present invention; Figure 2 The flowchart shown is for global model training using features Xa from local data of company A and features Xb from local data of company B, as provided by the present invention. Detailed Implementation

[0024] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0025] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0026] Figure 1 The flowchart illustrates a cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning provided by this invention.

[0027] like Figure 1 As shown, this invention discloses a cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning, including: S101, Obtain Enterprise Needs; S102, a vertical federated learning environment built based on enterprise needs; S103, Register corresponding identifiers based on the identification codes assigned to the enterprise's equipment, standardize the local data generated by each device, and associate the obtained standardized data with the corresponding identifiers; local data includes equipment operation data and equipment maintenance data; S104. Construct an initial global model. Distribute the initial global model to each enterprise node through the federated learning coordination server. Iteratively train the initial global model based on the encrypted gradient parameters fed back by each enterprise to determine the final global model. The federated learning coordination server has a three-layer architecture, namely the identifier resolution layer, the federated learning layer, and the security layer. S105, acquire real-time operating data of the device; S106, Analyze the real-time operating data of the equipment to determine whether the equipment status has changed; S107: When the device status changes, the real-time operating data of the device is input into the final global model, and the device status prediction data is output.

[0028] According to embodiments of the present invention, cross-enterprise industrial internet identifier resolution data sharing and prediction are achieved through an industrial internet identifier resolution system, a vertical federated learning environment, and a federated learning coordination server. The federated learning coordination server has a three-layer architecture, including an identifier resolution layer, a federated learning layer, and a security layer.

[0029] Identifier Resolution Layer: Assigns a unique identifier (such as Ecode, Handle) to each industrial entity such as equipment, product and component, and establishes a distributed identifier resolution network.

[0030] The identifier generation rules are based on the OID hierarchical structure of the ISO / IEC 9834 standard, or the hierarchical encoding rules of Ecode, to ensure the global uniqueness and scalability of identifiers. For example, the following Ecode encoding: ; The identifier resolution network supports resolution services, using recursive resolution algorithms (such as DNS-like protocols) to map identifiers to data storage locations, enabling cross-enterprise data access. Simultaneously, the identifier resolution nodes maintain an "identifier-data address-access permission" mapping table for subsequent data alignment and permission verification in federated learning.

[0031] Federated Learning Layer: Based on the identifier resolution system, it enables joint modeling for privacy protection of cross-enterprise data, including sample ID generation, automatic alignment of cross-enterprise features and expressions using sample IDs (such as equipment operating parameters and maintenance records), and global model training.

[0032] The sample ID generation algorithm combines industrial internet identifiers (such as Ecode) with metadata such as timestamps and geographic locations to generate a unique sample ID, which serves as a unique identifier for data in federated learning.

[0033] The sample ID generation algorithm is as follows: ; in, For sample ID, Ecode is a message authentication code based on the SHA256 hash function, and it is also known as an Industrial Internet identifier. For timestamps, For geographical location.

[0034] Security Layer: Combining access control from the identifier resolution system with privacy protection techniques from federated learning, end-to-end data security is achieved. At the security layer, encryption algorithms (such as homomorphic encryption or differential privacy noise injection) are dynamically selected based on the sensitivity classification of sample IDs. In cross-enterprise data sharing, public and private key pairs from the identifier resolution system are used to achieve multi-hop encrypted transmission (e.g., enterprise A → enterprise B → coordination server). A Trusted Execution Environment (TEE) (such as Intel SGX) is deployed in the coordination server to ensure that data is not tampered with or leaked during model aggregation. Trusted Execution Environment: A secure computing area created through hardware isolation, ensuring that even system administrators cannot access its internal data and computation processes.

[0035] In addition, the industrial internet identifier resolution system and federated learning coordination server can be used to achieve identifier lifecycle management and model version control.

[0036] The entire lifecycle (creation, update, deletion) of sample IDs is managed through the Industrial Internet Identifier Resolution System, and the federated learning coordination server dynamically adjusts the training strategy based on the lifecycle status.

[0037] For example, when a sample ID is marked as "deleted," the coordination server removes the relevant parameters from the global model to prevent historical data from affecting the prediction results. The identifier resolution node synchronizes Sample ID version information with the federated learning coordination server via gRPC or a REST API. If an anomaly is detected in a particular version of Sample ID data, the coordination server can roll back to a previous version of the global model.

[0038] According to an embodiment of the present invention, a vertical federated learning environment is built based on enterprise needs, including: Analyze the needs of enterprises, clarify data sharing objectives, and formulate corresponding data sharing agreements; Assign identification codes to enterprise equipment based on the Industrial Internet Identifier Resolution System; Each enterprise sets up a federated learning client locally and configures it on the aggregation server of the federated learning coordinator to complete the construction of the vertical federated learning environment.

[0039] It's important to note that building a vertical federated learning environment begins with understanding and defining the data sharing goals of each enterprise (e.g., quality prediction or supply chain optimization). This definition clarifies the data sharing objectives for each company. Simultaneously, corresponding data sharing protocols (e.g., the scope of data sharing, operator permissions, etc.) are developed for each data sharing objective. Then, a unique identifier is assigned to each enterprise's equipment (or product) using an industrial internet identifier resolution system (e.g., Handle, Ecode, or OID) to facilitate subsequent model training and data sharing. Finally, federated learning clients are built locally on each enterprise's website, supporting frameworks such as PySyft and TensorFlow. The federated learning coordinator server needs to employ a secure aggregation algorithm (e.g., FedAvg) and define hyperparameters such as the number of training epochs and the learning rate.

[0040] Among them, Handle is a distributed identifier resolution technology managed by the international organization DONA Foundation, which provides the Industrial Internet with globally unique digital object identifier management capabilities.

[0041] Ecode (Entity Code) is an IoT identification system proposed by China, belonging to the national standard object identification technology (GB / T 31866-2015). Its core is to assign a unique digital identity to physical / virtual objects through unified coding rules, and support multi-level resolution services to achieve cross-system data interconnection.

[0042] OID (Object Identifier): An international standard identification system (ISO / IEC 9834) jointly developed by ISO / IEC and ITU-T, which uses a hierarchical tree structure to provide a globally unique identifier for any type of object.

[0043] Pysyft: A privacy-preserving machine learning framework based on PyTorch, developed by the OpenMined community, that supports federated learning (FL), differential privacy (DP), and secure multi-party computation (MPC).

[0044] TensorFlow: As an open-source machine learning framework led by Google, it provides federated learning support through the TensorFlow Federated (TFF) module.

[0045] FedAvg (Federated Averaging): A benchmark algorithm for federated learning, implemented through multiple rounds of iteration.

[0046] According to an embodiment of the present invention, based on the identification code assigned to enterprise equipment, a corresponding identifier is registered, the local data generated by each device is standardized, and the obtained standardized data is associated with the corresponding identifier, including: Based on the identifier resolution node, register the corresponding identifier according to the identifier code assigned to the device; Based on the industrial internet identifier resolution system, enterprise nodes parse the local data of devices into a standard format to obtain standardized data; Based on the identifier, the corresponding standardized data is stored in the local database, and a mapping relationship between the identifier and the storage location of the corresponding standardized data is established through the identifier resolution node.

[0047] It should be noted that enterprises can use the Industrial Internet Identifier Resolution System to assign a unique identifier to equipment (or product) for identifier registration.

[0048] The data format of standardized data is a pre-set format by the system, usually JSON format, for example: { "identifier":"http: / / id.example.com / enterpriseA / device123", "timestamp":"2025-06-01T09:00:00Z", Temperature: 85.3 "vibration":0.12 }

[0049] Once a device registers an identifier, it will match the system's preset data sharing principles based on that identifier. Data generated by the device (such as device operating status and product quality parameters) will be stored in the enterprise's local database according to these matching principles. Simultaneously, the identifier resolution node also contains a mapping relationship between the identifier and the actual storage location. This allows the enterprise to directly determine the actual storage location of the data generated by the device through the identifier, facilitating enterprise queries.

[0050] According to an embodiment of the present invention, an initial global model is constructed, and the initial global model is distributed to each enterprise node through a federated learning coordination server. The initial global model is iteratively trained based on the encrypted gradient parameters fed back by each enterprise to determine the final global model, including: The enterprise node analyzes the received initial global model to determine the identifiers used for model training; The corresponding standardized data is filtered from the local database using identifiers used for model training; The local model is trained based on the corresponding standardized data, and the gradient parameters are calculated; the local model is the global model received by the enterprise through the enterprise node. The gradient parameters are encrypted, the encrypted gradient parameters are determined, and the encrypted gradient parameters are uploaded to the federated learning coordination server through the enterprise node. The encrypted gradient parameters of each enterprise are aggregated by the federated learning coordination server to update the initial global model; The updated global model is then distributed to each enterprise node for iterative training until the global model converges, resulting in the final global model training.

[0051] It should be noted that the initial global model is typically a neural network structure, which is distributed to each enterprise node via a federated learning coordination server. Each enterprise node stores the received initial global model locally, resulting in a local model (the local model is updated based on each received initial global model or the updated global model). The enterprise nodes first extract identifiers for model training from the received initial global model (or updated global model). Based on these identifiers, each enterprise further filters samples from its local database to obtain standardized data corresponding to the identifiers used for model training, which is then used to train its local model.

[0052] Using the local dataset D of enterprise k k For example, the gradient parameter calculation process is as follows: first, the local loss function of the local model is calculated: ; in, Let k be the local loss function for firm k. Let k be the number of samples in the local dataset of enterprise k. This means summing up the losses from all samples. Let be the sample loss function. For the features of the i-th sample, Let be the true label of the i-th sample, and be the model parameters. .

[0053] Then calculate the gradient using the obtained local function: ; in, Let k be the local gradient of firm k. Indicates to About all model parameters The gradient vector is obtained by taking the partial derivatives, where each component of the vector is the partial derivative of the function with respect to each parameter.

[0054] In the process of encrypting gradient parameters, the sensitivity level (including high sensitivity and low sensitivity) of the sample ID corresponding to the gradient parameter is first calculated, and the corresponding encryption method (including homomorphic encryption (HE) or differential privacy (DP)) is selected according to its sensitivity level.

[0055] The federated learning server aggregates the encrypted gradient parameters uploaded by each enterprise node using the FedAvg algorithm to generate a new global model, which then replaces the previous global model to obtain the updated global model. The method for aggregating the encrypted gradient parameters uploaded by each enterprise node using the FedAvg algorithm is expressed by the following formula: ; in, This represents the global model parameters in round t+1. The contributions from enterprise nodes from 1 to K are summed. These are the weighting coefficients. Let be the model parameters of the k-th enterprise node after the t-th round of updates.

[0056] Simultaneously, secure aggregation is performed within the Trusted Execution Environment (TEE): .

[0057] in, These are global model parameters within the TEE environment. For decryption operations in a trusted environment, Summation of the encryption gradient.

[0058] Finally, the federated learning aggregation server redistributes the updated global model to each enterprise node for iteration until convergence. For example, the process of training a global model using features Xa from enterprise A's local data and features Xb from enterprise B's local data is as follows: Figure 2 As shown, training parameters θa and θb are set respectively. The feature Xa is trained using training parameter θa to obtain the hidden layer result ha, and the feature Xb is trained using training parameter θb to obtain the hidden layer result hb. The hidden layer results ha and hb are transmitted to the coordination server for concatenation, and the shared label is calculated through a fully connected layer. The loss L(Y, is calculated by combining it with the initial shared label.) The loss is backpropagated to the hidden layers corresponding to features Xa of company A and Xb of company B, respectively, updating the corresponding training parameters θa and θb. Iterative training is then performed until the loss L(Y, The system meets the expected requirements, and the global model is determined to be converged.

[0059] In addition, during local model training, the weights of the model are updated in real time. The specific triggering mechanism for real-time weight updates and the formula for incremental training parameter updates are as follows: ; .

[0060] in, Trigger signal, This is an indicator function (returns True if the value inside the parentheses is true, and False if the value inside the parentheses is false). The L2 norm distance between two vectors. This is the preset norm distance threshold. and These are the model weights at time step t+1 and time step t, respectively. For learning rate, This represents the gradient of the loss function with respect to the weights, in the incremental dataset. The above calculations yielded the results.

[0061] According to an embodiment of the present invention, when the device state changes, real-time operating data of the device is input into the final global model, and device state prediction data is output, including: Obtain the device's operating data within a preset time interval based on the device's real-time operating data; Based on the device's identifier, shared data within the scope of the data sharing protocol is obtained from the Industrial Internet Identifier Resolution System; The system aligns the device's operating data and shared data within a preset time interval, inputs the aligned data into the final global model, and outputs device status prediction data.

[0062] It should be noted that equipment status includes operating temperature, humidity, and pressure, which are specifically set by those skilled in the art according to actual needs. After acquiring real-time operating data, each real-time operating parameter is compared with its corresponding system preset parameter threshold. When a real-time operating parameter does not meet the corresponding system preset parameter threshold (e.g., operating temperature is greater than the preset maximum temperature threshold or less than the preset minimum temperature threshold), the equipment status is determined to have changed. Shared data within the scope of the data sharing protocol includes equipment operating data and equipment maintenance data stored in other companies' local databases. The preset time interval is set by those skilled in the art according to actual needs, such as the past month or the past year. The equipment operating data within the preset time interval and shared inputs from other companies are input into the global model to obtain equipment status prediction data for a future period.

[0063] Equipment status prediction data is obtained through joint inference analysis, specifically: Construct the feature alignment matrix X: ; in, X represents the alignment condition of the feature alignment matrix, where X is a diagonal matrix generated based on the sample feature matrices of companies A and B respectively.

[0064] The prediction result y is generated using the feature alignment matrix X: ; in, These are the global model parameters obtained through federated learning. Indicates the use of a global model The input matrix X is forward propagated to obtain y.

[0065] The prediction results are decrypted to obtain the equipment status prediction data y. pred : ; in, It is a homomorphic encryption and decryption algorithm. The result of the previous encrypted prediction, This is the private key for company A. This is the plaintext prediction data for the decrypted device status.

[0066] For example, manufacturing company A holds the operating parameters (e.g., temperature, pressure, etc.) of its relevant production equipment for the past year, while maintenance service company B holds the historical maintenance records of the same equipment for the past year. If company A wants to predict the failure rate of the production equipment in the next month, it first obtains the most recent shared data from company B (which may contain homomorphically encrypted historical maintenance records) through an identifier resolution system. This shared data, along with company A's own recent equipment operating parameters, is then input into the federated learning coordination server. The federated learning coordination server automatically aligns and encrypts features with the same sample IDs, and then inputs the encrypted feature data into the global model for joint inference. After obtaining the prediction result, the federated learning coordination server returns the encrypted prediction value to the query initiator (company A). Company A then uses the agreed-upon decryption key to decrypt the data, thus obtaining the final prediction result (equipment status prediction data).

[0067] The alignment conditions and alignment precision for feature Xa in company A and feature Yb in company B are as follows: ; .

[0068] in, For the result set of the alignment operation, Is it a sample that satisfies ID equals sample Data pairs for ID. To determine the alignment precision, N represents the total number of queries in the alignment process. This is an indicator function (returns True if the value inside the parentheses is true, and False if the value inside the parentheses is false). This means that True is returned when the requester's sample ID is not equal to the responder's sample ID in the i-th alignment query.

[0069] According to an embodiment of the present invention, it further includes: The Industrial Internet Identifier Resolution System includes an identifier registration module, an identifier query module, and a data access control module.

[0070] It should be noted that the Industrial Internet Identifier Resolution System is the core infrastructure for achieving cross-enterprise data interconnection. By assigning unique, identifiable digital identifiers to industrial equipment (as well as industrial entities such as products and components) and establishing a distributed resolution network, it solves the problem of data silos and supports secure and accurate data alignment and sharing. The Industrial Internet Identifier Resolution System comprises three main modules: Identifier Registration: Enterprises apply for enterprise codes from the national node and register instance identifiers and their metadata in the local system. Metadata includes Industrial Internet identifiers (such as Ecode), timestamps, and geographical locations; Identifier Query: The identifier system returns the data address to the querying party through recursive resolution; Data Access Control: Identifier resolution follows strict data sharing protocols.

[0071] According to an embodiment of the present invention, encrypting gradient parameters and determining encrypted gradient parameters includes: Obtain the sensitivity classification of the sample ID corresponding to the gradient parameters; When the sensitivity level of the sample ID is high sensitivity, the gradient parameters are encrypted using homomorphic encryption. When the sensitivity level of the sample ID is low, the gradient parameters are encrypted using differential privacy.

[0072] It should be noted that the Industrial Internet identifier in the sample ID is generated from equipment-related data. Data involving user privacy or trade secrets (such as customer orders and equipment maintenance records) is identified as highly sensitive data; publicly available data (such as equipment model and production batch) is identified as low-sensitivity data. When the equipment-related data corresponding to the Industrial Internet identifier in the sample ID is highly sensitive data, the sensitivity level of the sample ID is determined to be high-sensitive; when the equipment-related data corresponding to the Industrial Internet identifier in the sample ID is low-sensitivity data, the sensitivity level of the sample ID is determined to be low-sensitivity. Based on the sensitivity level of the sample ID, the corresponding encryption method is selected for parameter encryption. .

[0073] The formula has two possible solutions: For Option 1 (high-sensitivity sample data): It is a homomorphic encryption algorithm. Let be the plaintext gradient of the k-th node. This is the public key.

[0074] For Option 2 (low-sensitivity sample data): Let be the plaintext gradient of the k-th node. Let be a noise sampled randomly from a Laplace distribution.

[0075] Additionally, when encrypting parameters solely through differential privacy, a larger Laplace noise (e.g., ε=0.1) is injected into the data for highly sensitive Sample IDs. A smaller noise (e.g., ε=1.0) is injected into the data for less sensitive Sample IDs to reduce model accuracy loss.

[0076] Among them, DP (Differential Privacy) is a mathematically rigorous privacy-preserving framework that ensures that statistical query results for a dataset are insensitive to changes in individual records. In federated learning, it is achieved through noise injection.

[0077] HE: Homomorphic encryption is a cryptographic technique that allows direct computation on encrypted data.

[0078] According to an embodiment of the present invention, it further includes: When the device status changes (such as abnormal temperature), the identifier resolution node notifies the federated learning coordination server through the event subscription mechanism, triggering global model iterative training. The federated learning coordination server dynamically adjusts the training cycle and hyperparameters of the global model based on the update status of the sample IDs.

[0079] It should be noted that the identifier resolution node integrates MQTT or Kafka message queues to push sample ID change events to the coordination server in real time. The Industrial Internet identifier resolution system records version changes for each sample ID (such as device replacement or parameter updates). The federated learning model selects corresponding historical data for training based on the version number, avoiding interference from outdated data. The training cycle and hyperparameters (such as learning rate and batch size) of the global model are dynamically adjusted by the system based on the actual situation during training.

[0080] According to an embodiment of the present invention, it further includes: The data access records in the identifier resolution, the parameter aggregation logs in the federated learning, and the encrypted hash values ​​of the prediction results are synchronously written to the blockchain for blockchain notarization.

[0081] It's important to note that blockchain-based evidence storage ensures the transparency and immutability of data flows. The blockchain used is either Hyperledger Fabric or an Ethereum consortium blockchain, supporting private deployment and access control. Smart contract design: A smart contract is written to record the following operations: identifier resolution requests (Sample ID query and data access), federated learning parameter upload and aggregation (hash value for gradient updates), and the distribution of the cryptographic hash value and decryption key of the prediction results. Simultaneously, a blockchain explorer interface is provided, allowing users to query the access history of sample IDs and model training logs.

[0082] According to an embodiment of the present invention, it further includes: Before receiving the encrypted gradient parameters uploaded by each enterprise, the federated learning coordination server verifies the validity of the encrypted gradient parameters through identifier resolution.

[0083] It should be noted that the validity of the cryptographic gradient parameters can be verified by checking the consistency between the sample ID and the identifier. When uploading the cryptographic gradient parameters, enterprise nodes attach a digital signature of the sample ID (based on an Ecode public / private key pair), and the coordination server verifies the signature's validity. If the identifier resolution record of the sample ID and the identifier is found to be inconsistent, the coordination server refuses to aggregate the cryptographic gradient parameters and marks the enterprise node as abnormal, thereby preventing malicious nodes from forging data.

[0084] All information (including but not limited to user equipment information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals (including but not limited to signals transmitted between user terminals and other devices) involved in this application have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the "enterprise needs" and "real-time operating data of the equipment" mentioned in this disclosure were obtained with full authorization.

[0085] This invention discloses a cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning. The method includes: building a vertical federated learning environment; registering identifiers for enterprise devices; standardizing the local data generated by each device; associating the standardized data with the corresponding identifiers; constructing an initial global model; distributing the initial global model to each enterprise node through a federated learning coordination server; iteratively training the initial global model based on the encrypted gradient parameters fed back by each enterprise to determine the final global model; determining whether the device status has changed; and when the device status changes, inputting the device's real-time operating data into the final global model and outputting device status prediction data. This invention improves the accuracy of predicting enterprise device failure rates while achieving efficient data sharing and data privacy protection.

[0086] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0087] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0088] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0089] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0090] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning, characterized in that, include: Obtain enterprise needs; A vertical federated learning environment will be built based on the aforementioned enterprise needs. Register corresponding identifiers based on the identification codes assigned to the enterprise's equipment, standardize the local data generated by each device, and associate the standardized data with the corresponding identifiers; the local data includes equipment operation data and equipment maintenance data. An initial global model is constructed and distributed to each enterprise node through a federated learning coordination server. The initial global model is iteratively trained based on the cryptographic gradient parameters fed back by each enterprise to determine the final global model. The federated learning coordination server has a three-layer architecture, namely the identifier resolution layer, the federated learning layer, and the security layer. Obtain real-time operating data of the equipment; The real-time operating data of the device is analyzed to determine whether the device status has changed. When the device status changes, the real-time operating data of the device is input into the final global model, and the device status prediction data is output.

2. The cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning according to claim 1, characterized in that, The construction of a vertical federated learning environment based on the enterprise's needs includes: Based on the enterprise's needs, analyze the data sharing objectives, and formulate corresponding data sharing agreements; Assign identification codes to enterprise equipment based on the Industrial Internet Identifier Resolution System; Each enterprise sets up a federated learning client locally and configures it on the aggregation server of the federated learning coordinator to complete the construction of the vertical federated learning environment.

3. The cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning according to claim 2, characterized in that, The process of registering corresponding identifiers based on the identification codes assigned to enterprise equipment, standardizing the local data generated by each device, and associating the resulting standardized data with the corresponding identifiers includes: Based on the identifier resolution node, register the corresponding identifier according to the identifier code assigned to the device; Based on the industrial internet identifier resolution system, enterprise nodes parse the local data of devices into a standard format to obtain standardized data; Based on the identifier, the corresponding standardized data is stored in the local database, and a mapping relationship between the identifier and the storage location of the corresponding standardized data is established through the identifier resolution node.

4. The cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning according to claim 1, characterized in that, The initial global model is constructed by distributing it to each enterprise node through a federated learning coordination server. Based on the encrypted gradient parameters fed back by each enterprise, the initial global model is iteratively trained to determine the final global model, including: The enterprise node analyzes the received initial global model to determine the identifiers used for model training; The corresponding standardized data is filtered from the local database using the identifiers used for model training; A local model is trained based on the corresponding standardized data, and gradient parameters are calculated; the local model is the global model received by the enterprise through the enterprise node. The gradient parameters are encrypted to determine the encrypted gradient parameters, and the encrypted gradient parameters are uploaded to the federated learning coordination server through the enterprise node. The initial global model is updated by aggregating the encrypted gradient parameters of each enterprise through a federated learning coordination server. The updated global model is then distributed to each enterprise node for iterative training until the global model converges, resulting in the final global model training.

5. The cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning according to claim 1, characterized in that, When the device state changes, the real-time operating data of the device is input into the final global model, and device state prediction data is output, including: The device's real-time operating data is used to obtain the device's operating data within a preset time interval; Based on the device's identifier, shared data within the scope of the data sharing protocol is obtained from the Industrial Internet Identifier Resolution System; The device operation data and the shared data within a preset time interval are aligned, and the aligned data is input into the final global model to output device status prediction data.

6. The cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning according to claim 1, characterized in that, Also includes: The industrial internet identifier resolution system includes an identifier registration module, an identifier query module, and a data access control module.

7. The cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning according to claim 4, characterized in that, The step of encrypting the gradient parameters and determining the encrypted gradient parameters includes: Obtain the sensitivity classification of the sample ID corresponding to the gradient parameter; When the sensitivity level of the sample ID is high sensitivity, the gradient parameters are encrypted using homomorphic encryption; When the sensitivity level of the sample ID is low, the gradient parameters are encrypted using differential privacy.

8. The cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning according to claim 1, characterized in that, Also includes: When the device status changes, the identifier resolution node notifies the federated learning coordination server through an event subscription mechanism, triggering global model iterative training. The federated learning coordination server dynamically adjusts the training cycle and hyperparameters of the global model based on the update status of the sample ID.

9. The cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning according to claim 1, characterized in that, Also includes: The data access records in the identifier resolution, the parameter aggregation logs in the federated learning, and the encrypted hash values ​​of the prediction results are synchronously written to the blockchain for blockchain notarization.

10. The cross-enterprise industrial internet identifier resolution data sharing and prediction method based on federated learning according to claim 1, characterized in that, Also includes: Before receiving the encrypted gradient parameters uploaded by each enterprise, the federated learning coordination server verifies the validity of the encrypted gradient parameters through identifier resolution.

Images