Cross-platform and cross-system configurable data interaction method and system

Abstract

The invention discloses a cross-platform and cross-system configurable data interaction method and system, and relates to the field of data integration and processing.The method comprises the steps that a bimodal metadata derivation model is used for processing a comprehensive data set, an initial mapping rule table is obtained, a global mapping model is established through federated learning, the initial mapping rule table is input into the global mapping model, and the global mapping model is established; obtaining an optimized mapping rule table, associating source system data of the target system with the twin to form a virtual-real fusion data stream, converting the virtual-real fusion data stream into a target format, encrypting and transmitting the virtual-real fusion data stream to the target system; according to the method, the efficiency and the flexibility of rule generation are greatly improved through automatic derivation and priority optimization, and particularly in a dynamically changing heterogeneous environment, the method can quickly adapt to the change of field semantics and load requirements, so that the real-time performance of data interaction is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of data integration and processing, and in particular to a cross-platform and cross-system configurable data interaction method and system. Background Art

[0002] With the rapid development of information and digital technologies, cross-platform and cross-system data interaction has become an important research direction in the fields of cloud computing, Internet of Things, Industrial Internet, and intelligent manufacturing. Early related technologies mainly relied on static data exchange protocols to achieve data transmission between heterogeneous systems through predefined interfaces.

[0003] However, the current popular methods for implementing cross-platform and cross-system data interaction still have many shortcomings. For example, traditional methods basically rely on static mapping rules or manual configuration, which makes it difficult to cope with frequent changes in the formats of heterogeneous data sources or dynamic adjustments in business scenarios, resulting in inefficient configuration and inability to optimize in real time. Moreover, the existing digital twin technology can achieve synchronization between physical devices and digital models, but when integrating data across systems, it often lacks dynamic verification of the reliability of mapping rules, which can easily lead to erroneous results due to data semantic deviations or outliers. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a cross-platform and cross-system configurable data interaction method to solve the problem of difficulty in coping with heterogeneous data sources or dynamic adjustment of business scenarios, resulting in low configuration efficiency and inability to optimize in real time.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a cross-platform and cross-system configurable data interaction method, which comprises: Access raw data from multiple heterogeneous data sources in real time, obtain metadata, traffic characteristics, and semantic analysis results of the raw data, and combine them into a comprehensive data set; The bimodal metadata inference model is used to process the comprehensive dataset to obtain the initial mapping rule table; Establish a global mapping model through federated learning, input the initial mapping rule table into the global mapping model, and obtain an optimized mapping rule table; Associate the source system data of the target system with the twin to form a virtual-real fusion data stream, convert the virtual-real fusion data stream into the target format and encrypt and transmit it to the target system; The target system receives data and returns feedback, and performs optimization based on the feedback results.

[0007] As a preferred solution of the cross-platform and cross-system configurable data interaction method of the present invention, wherein: the semantic analysis result refers to the semantic analysis of the field name of the metadata through the BERT model to generate a similarity score; The metadata, traffic characteristics, and similarity scores of the raw data are integrated into a comprehensive dataset in JSON format.

[0008] As a preferred solution of the cross-platform and cross-system configurable data interaction method of the present invention, wherein: using the bimodal metadata inference model to process the comprehensive data set to obtain the initial mapping rule table specifically includes the following steps: The bimodal metadata inference model deeply analyzes the field semantics based on the metadata and similarity scores, generates preliminary mapping candidates, processes the traffic characteristics, determines the matching priority of the fields in the metadata, and generates a field priority table; The preliminary mapping candidates are merged with the field priority table to generate an initial mapping rule table.

[0009] As a preferred solution of the cross-platform and cross-system configurable data interaction method of the present invention, wherein: a global mapping model is established through federated learning, and the initial mapping rule table is input into the global mapping model to obtain an optimized mapping rule table, which specifically includes the following steps: The federated learning framework distributes the initial mapping rule table to each training node, and each training node trains the MLP model based on local data; Upload the encrypted parameters to the asynchronous parameter server. After receiving the parameters of each training node, the server aggregates and generates a global mapping model. The initial mapping rule table is input into the global mapping model, and the optimized mapping rule table is output.

[0010] As a preferred solution of the cross-platform and cross-system configurable data interaction method of the present invention, wherein: the forming of the virtual-real fusion data stream specifically includes the following steps: The optimized mapping rule table is loaded through the digital twin mapping engine, real-time data is extracted from the source system, associated with the twin data, and the fields are bound through a two-way dynamic mapping mechanism to form a virtual-real fusion data flow.

[0011] As a preferred solution of the cross-platform and cross-system configurable data interaction method of the present invention, wherein: the virtual-reality fusion data stream is converted into a target format and encrypted and transmitted to the target system, specifically comprising the following steps: Use weighted decision-making technology to generate a comprehensive credibility score for virtual-real fusion data streams based on historical mapping records, real-time semantic analysis, and business priorities; A confidence threshold is set based on historical mapping records, and the comprehensive credibility score is compared with the confidence threshold. If it is lower than the confidence threshold, the process of re-deriving mapping rules is triggered; Convert the virtual-reality fusion data stream into a format that the target system can recognize, and attach a checksum to transmit it to the target system.

[0012] As a preferred solution of the cross-platform and cross-system configurable data interaction method of the present invention, the target system receives data and returns feedback, and optimizes according to the feedback result, specifically including the following steps: The target system parses the received virtual-reality fusion data stream, generates interactive feedback, and transmits it back to the adaptive mapping engine through the data bus; The adaptive mapping engine adjusts the current mapping rule table according to the interactive feedback, updates the global mapping model through incremental training of federated learning, synchronously adjusts the digital twin state, and switches the strategy to optimize the transmission.

[0013] In a second aspect, the present invention provides a cross-platform and cross-system configurable data interaction system, comprising: The collection module accesses raw data from multiple heterogeneous data sources in real time, obtains metadata, traffic characteristics, and semantic analysis results of the raw data, and combines them into a comprehensive data set; The processing module processes the comprehensive dataset using the bimodal metadata inference model to obtain an initial mapping rule table; The optimization module establishes a global mapping model through federated learning, inputs the initial mapping rule table into the global mapping model, and obtains an optimized mapping rule table; The transmission module associates the target system source system data with the twin according to the optimized mapping rule table to form a virtual-real fusion data stream, and converts the virtual-real fusion data stream into the target format and encrypts it for transmission to the target system; In the optimization module, the target system receives data and returns feedback, and performs optimization based on the feedback results.

[0014] The beneficial effects of the present invention are: the field mapping relationship is automatically derived by using a bimodal model, and the introduction of a priority table ensures that key fields are processed first when resources are limited, providing a preliminary rule framework for cross-system interaction, and greatly improving the efficiency and flexibility of rule generation through automated derivation and priority optimization. Especially in dynamically changing heterogeneous environments, it can quickly adapt to changes in field semantics and load requirements, thereby enhancing the real-time nature of data interaction; in addition, the source system data is integrated with the physical device status through digital twin technology, and the two-way mapping mechanism supports dynamic synchronization of data between virtual and real, providing a unified interactive data stream for the target system, achieving real-time consistency of virtual and real data, and being able to effectively support the monitoring and decision-making of complex systems, thereby significantly improving the integrity and practical value of data interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0016] Figure 1 This is a flow chart of the cross-platform and cross-system configurable data interaction method in Example 1.

[0017] Figure 2 This is a dual-mode processing flow chart in Example 1.

[0018] Figure 3 This is the federated learning training graph in Example 1.

[0019] Figure 4 This is the virtual-reality fusion flow chart in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, reference Figure 1 , Figure 2 , Figure 3 and Figure 4 , which is the first embodiment of the present invention, and provides a cross-platform and cross-system configurable data interaction method, comprising the following steps: S1. Access raw data from multiple heterogeneous data sources in real time, obtain metadata, traffic characteristics, and semantic analysis results of the raw data, and combine them into a comprehensive data set.

[0024] The specific steps include: Use the OPC UA protocol to collect raw data in real time from multiple heterogeneous data sources (such as trading systems running on Windows, IoT sensors on Linux, and cloud analysis platforms). The OPC UA protocol is a unified architecture for open platform communications.

[0025] The adapter traverses the output interface of the data source, reads the field name, such as TransactionID, DeviceID; reads the data type, such as string, floating point number; and constraints, such as length limit and value range; and generates metadata entries for each field, attaching a source identifier, such as source1, for subsequent tracing and processing; The monitoring component samples the transmission activity of the data source every second through the embedded probe, calculates the query rate per second and the average packet size, and uses the sliding window algorithm to smooth the results to generate stable traffic characteristic values, such as query rate per second = 200, packet size = 1KB, to reflect the load status and transmission characteristics of the data source.

[0026] All field names are input into the BERT model. The BERT model calculates the semantic vector distance between fields based on word embedding technology and generates a similarity score (ranging from 0 to 1). For example, TransactionID and DeviceID receive higher scores due to their semantic proximity. This analysis process is limited to the field name level to avoid processing the complete data to reduce computing overhead.

[0027] Use data serialization tools to organize metadata, traffic features, and similarity scores into a nested structure, such as placing metadata under the sources key, traffic features under the traffic key, and similarity scores under the semantics key. Generate the final comprehensive dataset.

[0028] Further explanation: It provides a multi-dimensional input foundation for cross-platform and cross-system data interaction. Metadata defines the structural characteristics of the data, traffic characteristics reflect the real-time operation status, and semantic analysis reveals the potential association between fields. The combination of the three provides comprehensive data support for subsequent mapping rule derivation and optimization. Especially in dynamically changing scenarios, it ensures the comprehensiveness and adaptability of the interaction process.

[0029] S2. Use the bimodal metadata inference model to process the comprehensive data set and obtain the initial mapping rule table.

[0030] The specific steps include: S2.1. The BERT model is called through the semantic analysis module of the bimodal inference engine, and the field name and similarity score are used as input. Combined with the predefined context dictionary (including cross-system terms, for example, ID represents identification fields, such as TransactionID and DeviceID, and Value represents numerical fields, such as Amount and Temperature), the semantic relevance between fields is recalculated through the word embedding technology of the BERT model. The consistency of field types is considered in the analysis process, for example, string type fields are easier to match, and preliminary mapping candidates are generated, such as TransactionID to DeviceID or Amount to Temperature. And an initial relevance weight is assigned to each pair of candidates.

[0031] S2.2. The traffic characteristics in the comprehensive data set are processed simultaneously through the traffic characteristics module of the dual-modal inference engine. First, the query rate and data packet size are read, and the load status is judged according to the preset load evaluation rules, such as query rate <100 is low load, 100-300 is medium load, and >300 is high load (specific rules can be customized according to usage scenarios and usage). Then, according to the pre-configured business importance rules, such as the priority of amount-related fields is higher than the identification field (specific rules can be customized according to usage scenarios and usage), the field priority is adjusted in combination with the load status, such as marking Amount as a high-priority field, and generating a field priority table, in which each field is accompanied by a priority coefficient (range 0-1).

[0032] S2.3, the bimodal inference engine runs the rule synthesizer, which aligns the preliminary mapping candidates with the field priority table, and removes candidate pairs with priority coefficients lower than 0.3 and relevance weights lower than 0.7 (the specific values ​​can be customized according to your own needs and business needs). Subsequently, an initial weight is assigned to each retained mapping rule. The initial weight can be obtained by integrating the relevance weight and priority coefficient, or it can be freely set based on experience, historical data, and business needs. And sort by weight to form an initial mapping rule table to ensure that high-priority fields are reflected first.

[0033] Preferably, compared with traditional static rule configuration or single semantic analysis methods, the present invention significantly improves the intelligence and flexibility of mapping rule generation through bimodal fusion, avoiding the inefficiency of manual intervention and the limitations of single-dimensional analysis. The introduction of the priority mechanism further optimizes resource allocation, allowing the rule table to dynamically adapt to business needs and loads, thereby improving the initial accuracy and execution efficiency of data interaction, and providing high-quality input for subsequent optimization.

[0034] S3. Establish a global mapping model through federated learning, input the initial mapping rule table into the global mapping model, and obtain an optimized mapping rule table.

[0035] The specific steps include: The federated learning framework runs the distribution manager and broadcasts the initial mapping rule table to each training node in JSON format through the TLS protocol. The distribution manager verifies the reception status of each training node. If the training node does not respond, it is marked as offline and logs are recorded to prepare a unified starting point for distributed training.

[0036] S3.1. Each training node loads local private data, such as the historical records of the trading system and the real-time readings of the sensors; the mapping pairs in the initial rule table are used as supervisory signals to train the MLP model (multi-layer perceptron) to adjust the rule weights. The number of iterations is set, and the mean square error loss function is used to update the MLP model parameters to generate local optimization parameters. The local optimization parameters of each node are also the parameters of each node. The MLP model is used as the mapping model because it is lighter and more efficient than the deep network.

[0037] S3.2, the node runs the encryption module, uses differential privacy technology to encrypt the local optimization parameters, and generates irreversible parameter update values. The encrypted local optimization parameters are uploaded through the TLS channel, and a 5-second timeout limit is set (in order to meet real-time requirements, the specific value can be customized according to personal needs and business). If the upload fails, retry once to ensure that the server receives the contribution of each training node.

[0038] S3.3. The server runs the aggregator, decrypts the received local optimization parameters, and uses the weighted average algorithm to fuse the local optimization parameters of each training node, removes outliers, such as parameters that deviate from the mean by 2 times the standard deviation, and generates global mapping model parameters.

[0039] S3.4, the global mapping model loads the initial rule table, recalculates the weight of each rule based on the trained parameters, for example, the weight of TransactionID to DeviceID may be adjusted from the initial value to the new value, and sorts the rule table in descending order of weight. Verify the consistency of the rules, such as checking whether there is a conflicting mapping. If a problem is found, roll back to the previous version parameters, and finally output the optimized mapping rule table.

[0040] Preferably, compared with the traditional centralized training method, the present invention avoids the risk of privacy leakage caused by uploading original data through federated learning, and only transmits encryption parameters to significantly improve data security. At the same time, distributed training and asynchronous aggregation mechanisms fully utilize the computing power of each node, speed up the rule optimization speed, and enable the optimized mapping rule table to more accurately adapt to the complex needs of multi-source heterogeneous environments, thereby improving the global accuracy and real-time adaptability of data interaction.

[0041] S4. Associate the target system source system data with the twin to form a virtual-reality fusion data stream, and convert the virtual-reality fusion data stream into the target format and encrypt it for transmission to the target system.

[0042] The specific steps include: The digital twin mapping engine calls the source system interface through the preset adapter module, parses the latest business data, and stores it in a temporary buffer. Align the buffer data with the source fields in the optimized mapping rule table. If the field is missing, it is marked as pending to ensure that the data matches the mapping rules.

[0043] The digital twin mapping engine starts the bidirectional dynamic mapping mechanism, first reads the mapping pairs in the mapping rule table, extracts the source system field values ​​from the buffer, extracts the corresponding attribute values ​​from the twin, and binds the field pairs through the matcher. The bidirectional dynamic mapping mechanism supports bidirectional synchronization, which is reflected in the synchronization of source system data to the twin when it is updated, and feedback to the source system when the twin data changes, thereby generating a fused data stream containing virtual and real associations and recording the binding relationship.

[0044] A comprehensive credibility score is generated based on historical mapping records, real-time semantic analysis, and business priorities (used to quantify the criticality of fields in the business, for example, the field Amount is more important than the DeviceID, and the business priority can be freely set according to business needs and personal needs). The comprehensive credibility score is based on weighted calculation, and the weight distribution can be 0.4 for history, 0.4 for real-time, and 0.2 for business. The specific weight setting can be customized according to the usage scenario and actual situation. The comprehensive credibility score is used to evaluate the reliability of each mapping rule.

[0045] According to the historical mapping records, a confidence threshold is set, for example, 0.7. If the comprehensive credibility score of a rule is lower than the confidence threshold (for example, due to an outlier causing a decrease in reliability or other reasons), it is marked as unreliable and sent back to the step of triggering the mapping rule for re-derivation; if it is higher than the confidence threshold, the mapping rule is retained. The adjustment process is recorded in the log to ensure traceability, so as to obtain the verified virtual-real fusion data flow.

[0046] The digital twin mapping engine loads the template of the target system and fills the virtual-reality fusion data into the target system template through the format converter. For example, if the target system template is XML, the virtual-reality fusion data is converted to XML format. After the conversion is completed, a checksum is generated and attached to the converted virtual-reality fusion data and transmitted to the target system.

[0047] Preferably, the traditional method is single-item data synchronization, or the use of mapping methods without verification methods, but the present invention significantly improves the integrity and reliability of virtual and real data fusion through bidirectional mapping and confidence evaluation, avoiding interaction failures caused by data deviations. In addition, the encrypted transmission and verification mechanism ensures data security, so that the fused data stream can efficiently adapt to the diverse needs of the target system, thereby improving the practicality and stability of the interaction process.

[0048] S5. The target system receives the data and returns feedback, and performs optimization based on the feedback results.

[0049] The specific steps include: After receiving the transmitted data, the target system first verifies the integrity of the data through the checksum, reads the transmitted data after confirming the integrity, extracts the field value and compares it with the expected format of the target system, and generates interaction feedback. The interaction feedback includes the interaction status (such as success or failure), processing delay and exception information, and then transmits the interaction feedback back.

[0050] S5.1. Adjust the mapping rule table according to the feedback from the target system (such as "Temperature is out of range"). For example, if the feedback indicates that Temperature is out of range, increase the Temperature weight, update the global mapping model synchronously, and output the updated mapping rule table and digital twin mapping engine.

[0051] Define a state set, such as a normal strategy and an abnormal recovery strategy. When the target system interaction feedback shows an abnormality, switch to the abnormal recovery strategy, give priority to transmitting key field data, and record state changes to optimize transmission efficiency.

[0052] This embodiment also provides a cross-platform and cross-system configurable data interaction system, including: The collection module accesses raw data from multiple heterogeneous data sources in real time, obtains metadata, traffic characteristics, and semantic analysis results of the raw data, and combines them into a comprehensive data set; The processing module processes the comprehensive dataset using the bimodal metadata inference model to obtain an initial mapping rule table; The optimization module establishes a global mapping model through federated learning, inputs the initial mapping rule table into the global mapping model, and obtains an optimized mapping rule table; The transmission module associates the target system source system data with the twin according to the optimized mapping rule table to form a virtual-real fusion data stream, and converts the virtual-real fusion data stream into the target format and encrypts it for transmission to the target system; In the optimization module, the target system receives data and returns feedback, and performs optimization based on the feedback results.

[0053] In summary, the present invention provides a preliminary rule framework for cross-system interaction by: utilizing a bimodal model to automatically derive field mapping relationships, and introducing a priority table to ensure that key fields are processed first when resources are limited. It greatly improves the efficiency and flexibility of rule generation through automated derivation and priority optimization, and can quickly adapt to changes in field semantics and load requirements in dynamic heterogeneous environments, thereby enhancing the real-time nature of data interaction. In addition, the source system data is integrated with the physical device status through digital twin technology, and the two-way mapping mechanism supports dynamic synchronization of data between virtual and real, providing a unified interactive data stream for the target system, achieving real-time consistency of virtual and real data, and effectively supporting the monitoring and decision-making of complex systems, thereby significantly improving the integrity and practical value of data interaction.

[0054] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A cross-platform and cross-system configurable data interaction method, characterized by: include, Access raw data from multiple heterogeneous data sources in real time, obtain metadata, traffic characteristics, and semantic analysis results of the raw data, and combine them into a comprehensive data set; The bimodal metadata inference model is used to process the comprehensive dataset to obtain the initial mapping rule table; Establish a global mapping model through federated learning, input the initial mapping rule table into the global mapping model, and obtain an optimized mapping rule table; Associate the source system data of the target system with the twin to form a virtual-real fusion data stream, convert the virtual-real fusion data stream into the target format and encrypt and transmit it to the target system; The target system receives data and returns feedback, and performs optimization based on the feedback results.

2. The cross-platform and cross-system configurable data interaction method according to claim 1, characterized in that: The semantic analysis result refers to the semantic analysis of the field name of the metadata through the BERT model to generate a similarity score; A comprehensive dataset in JSON format that integrates the metadata, traffic characteristics, and similarity scores of the original data.

3. The cross-platform and cross-system configurable data interaction method according to claim 2, characterized in that: The bimodal metadata inference model is used to process the comprehensive data set to obtain the initial mapping rule table, which specifically includes the following steps: The bimodal metadata inference model deeply analyzes the field semantics based on the metadata and similarity scores, generates preliminary mapping candidates, processes the traffic characteristics, determines the matching priority of the fields in the metadata, and generates a field priority table; The preliminary mapping candidates are merged with the field priority table to generate an initial mapping rule table.

4. The cross-platform and cross-system configurable data interaction method according to claim 3, characterized in that: A global mapping model is established through federated learning, and the initial mapping rule table is input into the global mapping model to obtain an optimized mapping rule table, which specifically includes the following steps: The federated learning framework distributes the initial mapping rule table to each training node, and each training node trains the MLP model based on local data; Upload the encrypted parameters to the asynchronous parameter server. After receiving the parameters of each training node, the server aggregates and generates a global mapping model. The initial mapping rule table is input into the global mapping model, and the optimized mapping rule table is output.

5. The cross-platform and cross-system configurable data interaction method according to claim 4, characterized in that: The forming of the virtual-real fusion data stream specifically includes the following steps: The optimized mapping rule table is loaded through the digital twin mapping engine, real-time data is extracted from the source system, associated with the twin data, and the fields are bound through a two-way dynamic mapping mechanism to form a virtual-real fusion data flow.

6. The cross-platform and cross-system configurable data interaction method according to claim 5, characterized in that: The virtual-reality fusion data stream is converted into a target format and encrypted and transmitted to the target system, specifically including the following steps: Use weighted decision-making technology to generate a comprehensive credibility score for virtual-real fusion data streams based on historical mapping records, real-time semantic analysis, and business priorities; A confidence threshold is set based on historical mapping records, and the comprehensive credibility score is compared with the confidence threshold. If it is lower than the confidence threshold, the process of re-deriving mapping rules is triggered; Convert the virtual-reality fusion data stream into a format that the target system can recognize, and attach a checksum to transmit it to the target system.

7. The cross-platform and cross-system configurable data interaction method according to claim 6, characterized in that: The target system receives data and returns feedback, and performs optimization based on the feedback results, which includes the following steps: The target system parses the received virtual-reality fusion data stream, generates interactive feedback, and transmits it back to the adaptive mapping engine through the data bus; The adaptive mapping engine adjusts the current mapping rule table according to the interactive feedback, updates the global mapping model through incremental training of federated learning, synchronously adjusts the digital twin state, and switches the strategy to optimize the transmission.

8. A cross-platform and cross-system configurable data interaction system, based on the cross-platform and cross-system configurable data interaction method according to any one of claims 1 to 7, characterized in that: include, The collection module accesses raw data from multiple heterogeneous data sources in real time, obtains metadata, traffic characteristics, and semantic analysis results of the raw data, and combines them into a comprehensive data set; The processing module processes the comprehensive dataset using the bimodal metadata inference model to obtain an initial mapping rule table; The optimization module establishes a global mapping model through federated learning, inputs the initial mapping rule table into the global mapping model, and obtains an optimized mapping rule table; The transmission module associates the target system source system data with the twin according to the optimized mapping rule table to form a virtual-real fusion data stream, and converts the virtual-real fusion data stream into the target format and encrypts it for transmission to the target system; In the optimization module, the target system receives data and returns feedback, and performs optimization based on the feedback results.

Images