Cloud service automatic migration platform and migration method

Abstract

The invention relates to the field of data migration and verification repair, and provides an automatic migration platform and migration method for a cloud service. The cloud service automatic migration platform comprises a data source adaptation layer, a dynamic feature extraction module, a joint check engine and an automatic repair module. The data source adaptation layer converts metadata of a source data platform to adapt to a target data platform, the dynamic feature extraction module extracts multi-feature data and dynamically adjusts features, the joint check engine compares and analyzes the multi-feature data by combining a check reference and machine learning and generates a report, and the automatic repair module repairs the data according to the report. The method can be compatible with various heterogeneous platforms, the comprehensiveness and the pertinence of feature extraction are improved, the accuracy and the intelligent level of inconsistency detection are improved, the repair operation can be automatically judged and executed, and the consistency and the availability of data after migration are effectively guaranteed.

Description

Technical Field

[0001] The present application relates to the technical field of data migration and verification and repair, and in particular to an automatic migration platform and migration method for cloud services. Background Art

[0002] With the rapid development of information technology, cloud computing has emerged and is widely used. Cloud services, with their efficiency, flexibility, and scalability, are playing an increasingly important role in enterprises' digital transformation and informatization efforts. To enhance their competitiveness, many companies are migrating their business systems to the cloud. This not only reduces hardware costs and maintenance burdens, but also enables rapid business deployment and iteration by leveraging the powerful resources and expertise of cloud service providers. Furthermore, cloud services provide enterprises with more convenient data storage and processing capabilities, helping them better tap into the value of their data and make more informed decisions.

[0003] During the cloud service migration process, related technologies primarily focus on several aspects. For data source adaptation, manual hard-coding is often used, converting and mapping the source data platform's metadata one by one to adapt it to the requirements of the target data platform. While this approach can address data format incompatibility to a certain extent, it consumes significant manpower and time, and is prone to human error. Furthermore, during data migration, data feature extraction is often relatively simplistic, focusing on only a few fixed feature indicators. This fails to fully reflect the true nature of the data, resulting in metadata conversion efficiency below 30% per hour and an inability to process unstructured data (such as DDL statements). For example, invention patent application publication number CN119011661A discloses a method, apparatus, electronic device, storage medium, and product for determining service migration information. The method includes receiving a distributed system's service migration information determination instruction, performing a database search based on the instruction to obtain initial feature data and reference feature data for the distributed system. The initial feature data is the distributed system's business information within the current data acquisition cycle, and the reference feature data is the distributed system's business information within a preset data acquisition cycle. This method only supports fixed feature extraction and cannot be dynamically adjusted. Furthermore, there is a lack of effective data verification mechanisms, which typically simply compare basic data attributes, making it difficult to detect deeper data inconsistencies. When data issues are discovered, there is no comprehensive automated repair method, and most data must be manually modified, which is inefficient.

[0004] However, related technologies have significant flaws. Manually adapting data sources is not only inefficient and error-prone, but also difficult to adapt to complex and changing data source environments. A single data feature extraction method cannot meet diverse data migration needs and cannot promptly identify potential risks during the data migration process. Simple data verification mechanisms cannot accurately identify data feature inconsistencies, while manual data repair prolongs the migration cycle, increases enterprise operating costs, and severely impacts the efficiency and quality of cloud service migration. Summary of the Invention

[0005] In order to solve the problems that related technologies are difficult to cope with complex and changeable data source environments, have a single data feature extraction method, and rely on manual intervention, this application provides a cloud service automatic migration platform and migration method.

[0006] On the one hand, this application provides a cloud service automatic migration platform, which adopts the following technical solutions: A cloud service automatic migration platform, including: The data source adaptation layer has a standardized metadata model and is used to transform the metadata of the source data platform to adapt it to the target data platform; a dynamic feature extraction module connected to the data source adaptation layer, the dynamic feature extraction module having multi-dimensional data features, and used to extract multi-feature data from the metadata of the source data platform, the intermediate data being migrated, and the migrated data of the target data platform during the data migration process, wherein the feature types and extraction granularity of the multi-dimensional data features are dynamically adjusted according to changes in the content of the metadata of the source data platform, the intermediate data, and the migrated data; a joint verification engine connected to the dynamic feature extraction module, the joint verification engine being capable of combining verification benchmarks with machine learning to perform comparative analysis on the multi-feature data and generating a feature inconsistency verification report when the features of the comparative analysis are inconsistent; An automatic repair module is connected to the joint verification engine and is used to determine whether there is information that can be automatically repaired based on the feature inconsistency verification report. When it is determined that the information that can be automatically repaired exists, an automatic repair instruction is generated. The target data platform can repair the migrated data according to the automatic repair instruction.

[0007] By adopting the above technical solution, the metadata of heterogeneous data sources is standardized and converted through the data source adaptation layer, which solves the problem of incompatibility of data structures and formats between different platforms and improves the adaptability of data migration; the dynamic feature extraction module extracts data features in real time and multi-dimensionally, and dynamically adjusts the extraction strategy according to the data content, ensuring the comprehensiveness and pertinence of feature extraction, laying the foundation for subsequent precise verification; the joint verification engine combines verification benchmarks with machine learning for comparative analysis, which improves the accuracy and intelligence level of data inconsistency detection; the automated repair module automatically judges and executes repair operations, significantly reducing manual intervention and improving the efficiency and quality of data migration, thereby realizing efficient, intelligent and reliable automated migration, verification and repair at the data level during the cloud service migration process, and ensuring the consistency and availability of the migrated data.

[0008] Optionally, the data source adaptation layer has multiple adapter plug-ins, each of which corresponds to the type of the source data platform. The adapter plug-in is used to obtain the original data of the source data platform, parse the original data of the source data platform to generate metadata of the source data platform, and map the metadata of the source data platform to the standardized metadata model.

[0009] By adopting the above technical solution, by equipping different types of source data platforms with special adapter plug-ins, flexible access and targeted processing of multiple heterogeneous data sources are achieved, enhancing the versatility and scalability of the platform; each plug-in is independently responsible for the acquisition, parsing and mapping of raw data to standardized metadata models, ensuring the accuracy and efficiency of metadata conversion, providing a standardized data foundation for subsequent unified data processing and verification processes, and further improving the platform's ability to handle diverse data sources and overall migration efficiency.

[0010] Optionally, when the original data of the source data platform is structured data, the adapter plug-in obtains the original data of the source data platform through the data interface of the source data platform, and extracts metadata information according to the data structure of the original data of the source data platform to generate metadata of the source data platform; When the original data of the source data platform is semi-structured data (such as text sentence data or other PDF semi-structured data), the adapter plug-in obtains the original data of the source data platform through a search method, decomposes the original data of the source data platform into structured intermediate representation data through syntax analysis and lexical analysis, and extracts metadata information based on the data structure of the structured intermediate representation data to generate metadata of the source data platform.

[0011] Using the above technical solution, different metadata acquisition and parsing strategies are designed for two types of raw data: structured data and semi-structured data (such as textual statements or other semi-structured PDF data). For structured data, metadata is extracted directly through the data interface and its inherent structure, which is a direct and efficient operation. For semi-structured data, metadata is acquired through search methods and converted into a structured intermediate representation using syntactic analysis and lexical analysis, and then extracted. This accurately processes the data structure information defined in the semi-structured data. This differentiated processing approach ensures that the data source adaptation layer can more accurately and efficiently extract metadata from different forms of raw data, further improving the accuracy of metadata conversion and the ability to adapt to various complex data source scenarios.

[0012] Optionally, the multi-dimensional data features include metadata dictionary information, the metadata dictionary information includes basic statistical features, field-level features, data distribution features and data quality features, and the multi-feature data includes basic statistical feature data, field-level feature data, data distribution feature data and data quality feature data; wherein, the dynamic feature extraction module extracts the basic statistical feature data from the metadata of the source data platform, the intermediate data in migration and the migrated data of the target data platform through big data calculation, extracts the field-level feature data through feature engineering, extracts the data distribution feature data through statistical analysis, and extracts the data quality feature data through the data audit platform.

[0013] By adopting this technical solution, the dynamic feature extraction module comprehensively and deeply captures key data attributes and potential issues by defining four multi-dimensional data features: basic statistics, field-level features, data distribution, and data quality. By employing targeted extraction methods such as big data computing, feature engineering, statistical analysis, and data audit platform processing, the module is able to comprehensively and deeply capture key data attributes and potential issues. This meticulous feature classification and extraction approach not only enhances the richness and representativeness of the extracted feature data but also provides a more refined and comprehensive data profile for the subsequent joint verification engine, enabling more accurate identification of various inconsistencies and quality issues during the data migration process, thereby enhancing the depth and breadth of verification.

[0014] Optionally, the joint verification engine compares and analyzes the multi-feature data with the verification benchmark, uses the multi-feature data and feature inconsistency verification report during the historical data migration process to train the machine learning model, and uses the machine learning model trained with historical data to compare and analyze the multi-feature data. When feature inconsistencies are identified in the multi-feature data based on the verification benchmark or the machine learning model, a feature inconsistency verification report is generated.

[0015] Using this technical solution, the joint verification engine implements a dual verification mechanism by combining traditional verification benchmark comparisons with intelligent analysis based on machine learning models. The verification benchmark ensures strict adherence to established rules, while the machine learning model learns from historical data and abnormal patterns to identify potential, poorly defined, or complex data inconsistencies. This hybrid verification strategy not only ensures verification accuracy and comprehensiveness, but also empowers the system with self-learning and intelligent evolution capabilities, enabling more effective identification of diverse data discrepancies. This significantly improves the intelligence, accuracy, and detection rate of data consistency verification.

[0016] Optionally, the automatic repair module is further used to determine whether there is information that cannot be automatically repaired based on the feature inconsistency verification report, and when it is determined that there is information that cannot be automatically repaired, generate a repair suggestion report including the information that cannot be automatically repaired and repair suggestions.

[0017] Using this technical solution, the automated repair module not only processes automatically repairable information but also increases its ability to identify and handle information that cannot be automatically repaired. By generating reports containing specific unrepairable information and repair suggestions, users are provided with clear problem diagnosis and follow-up guidance. This not only enables the platform to more comprehensively address various complex data inconsistency scenarios but also effectively assists with manual intervention, reducing user analysis and troubleshooting time, improving the efficiency and accuracy of complex problem handling, and further enhancing the platform's practicality and user experience.

[0018] On the other hand, the present application provides a method for automatically migrating cloud services, comprising the following steps: S1. Convert the metadata of the source data platform into a standardized metadata model of the data source adaptation layer, so that the source data platform is adapted to the target data platform; S2. During the data migration process, multi-feature data is extracted from the metadata of the source data platform, the intermediate data being migrated, and the migrated data of the target data platform using multi-dimensional data features, wherein the feature types and extraction granularity of the multi-dimensional data features are dynamically adjusted according to content changes of the metadata of the source data platform, the intermediate data, and the migrated data; S3. analyzing the multi-feature data, wherein the analysis mode includes combining a calibration benchmark with machine learning for comparison; S4. When the features of the comparative analysis in step S3 are inconsistent, a feature inconsistency verification report is generated; S5. Determine whether there is information that can be automatically repaired based on the feature inconsistency check report; S6. When it is determined in step S5 that there is information that can be automatically repaired, an automatic repair instruction is generated, and the target data platform can repair the migrated data according to the automatic repair instruction.

[0019] Using this technical solution, we implemented an automated verification and repair process for cloud service data migration through standardized steps. First, metadata standardization addressed heterogeneous platform adaptation issues. Next, dynamic, multi-dimensional feature extraction ensured the comprehensiveness and real-time nature of verification data. Finally, comparative analysis using joint verification benchmarks and machine learning improved the accuracy and intelligence of inconsistency detection. Finally, by determining and executing automated repair instructions, migration efficiency and data quality were enhanced. This approach systematically addressed the core pain points of data migration, achieving the technical effect of efficiently, intelligently, and reliably ensuring the consistency of migrated data.

[0020] Optionally, step S1 includes: S11. Obtain original data from the source data platform; S12. Parse the original data of the source data platform to generate metadata of the source data platform; S13. Mapping the metadata of the source data platform to a standardized metadata model; Wherein, in step S11, when it is determined that the original data of the source data platform is structured data, the original data of the source data platform is obtained through the data interface of the source data platform; in step S12, metadata information is extracted according to the data structure of the original data of the source data platform to generate metadata of the source data platform; In step S11, when it is determined that the original data of the source data platform is semi-structured data, the original data of the source data platform is obtained through search; in step S12, the original data of the source data platform is decomposed into structured intermediate representation data through grammatical analysis and lexical analysis, and metadata information is extracted according to the data structure of the structured intermediate representation data to generate metadata of the source data platform.

[0021] The above technical solution further refines the metadata conversion steps for the source data platform, employing differentiated acquisition and parsing methods for different types of raw data (structured and semi-structured). This refined processing flow ensures the ability to accurately extract metadata from various complex source data environments and successfully map it to standardized models. This improves the robustness of the entire data source adaptation process and its compatibility with different data source formats, laying a solid and accurate data foundation for subsequent data migration and verification.

[0022] Optionally, in step S2, the multi-dimensional data features include metadata dictionary information, the metadata dictionary information includes basic statistical features, field-level features, data distribution features and data quality features, and the multi-feature data includes basic statistical feature data, field-level feature data, data distribution feature data and data quality feature data, wherein the basic statistical feature data is extracted from the metadata of the source data platform, the intermediate data in migration and the migrated data of the target data platform through big data calculation, the field-level feature data is extracted through feature engineering, the data distribution feature data is extracted through statistical analysis, and the data quality feature data is extracted through a data audit platform; in step S3, the multi-feature data is compared and analyzed with the verification benchmark, the multi-feature data and feature inconsistency verification report in the historical data migration process are used to train the machine learning model, and the multi-feature data is compared and analyzed using the machine learning model trained with historical data; in step S4, when feature inconsistency is identified in the multi-feature data based on the verification benchmark or the machine learning model, a feature inconsistency verification report is generated.

[0023] The above technical solution specifies the specific feature types and extraction methods for dynamic feature extraction (step S2), as well as the specific analysis model for joint verification (steps S3 and S4). By clarifying the composition of multi-dimensional features and targeted extraction methods, the comprehensiveness and high quality of the verification input data are ensured. By combining verification benchmarks with machine learning models trained on historical data for dual comparative analysis, the accuracy, depth, and breadth of inconsistency identification are significantly enhanced, enabling the method to more effectively identify various data discrepancies during the migration process, thereby improving the reliability and intelligence of the entire migration verification process.

[0024] Optionally, the method for automatically migrating cloud services further includes: S7. Determine whether there is information that cannot be automatically repaired based on the feature inconsistency verification report; S8. When it is determined in step S7 that there is information that the system cannot be automatically repaired, a repair suggestion report including the information that the system cannot be automatically repaired and repair suggestions is generated.

[0025] The above technical solution complements the automated repair process with a mechanism for handling situations that cannot be automatically repaired. By adding steps to identify information that cannot be automatically repaired and generate a report containing specific information and remediation suggestions, the entire migration method more comprehensively covers various possible data inconsistency scenarios. This not only provides clear guidance for complex issues requiring manual intervention, reducing the difficulty of troubleshooting and resolving issues, but also improves the method's comprehensiveness, usability, and user-friendliness in complex migration environments.

[0026] In summary, this application includes at least one of the following beneficial technical effects: 1. The metadata of heterogeneous data sources is standardized and converted through the data source adaptation layer, which solves the problem of incompatibility of data structures and formats between different platforms and improves the adaptability of data migration. The dynamic feature extraction module extracts data features in real time and multi-dimensionally, and dynamically adjusts the extraction strategy according to the data content, ensuring the comprehensiveness and pertinence of feature extraction, laying the foundation for subsequent precise verification. The joint verification engine combines verification benchmarks with machine learning for comparative analysis, improving the accuracy and intelligence level of data inconsistency detection. The automated repair module automatically judges and executes repair operations, significantly reducing manual intervention and improving the efficiency and quality of data migration, thereby realizing efficient, intelligent and reliable automated migration, verification and repair at the data level during the cloud service migration process, and ensuring the consistency and availability of the migrated data.

[0027] 2. By equipping different types of source data platforms with specialized adapter plug-ins, flexible access and targeted processing of multiple heterogeneous data sources are achieved, enhancing the platform's versatility and scalability. Each plug-in is independently responsible for acquiring, parsing, and mapping raw data to standardized metadata models, ensuring the accuracy and efficiency of metadata conversion. This provides a standardized data foundation for subsequent unified data processing and verification processes, further improving the platform's ability to handle diverse data sources and overall migration efficiency.

[0028] 3. By defining four multi-dimensional data features—basic statistics, field-level features, data distribution, and data quality—and employing targeted extraction methods such as aggregation calculations, field traversal, statistical analysis, and data inspection, the dynamic feature extraction module is able to comprehensively and deeply capture key data attributes and potential issues. This meticulous feature classification and extraction approach not only enhances the richness and representativeness of the extracted feature data but also provides a more refined and comprehensive data portrait for the subsequent joint verification engine, enabling more accurate identification of various inconsistencies and quality issues during the data migration process, thereby enhancing the depth and breadth of verification.

[0029] 4. The joint verification engine implements a dual verification mechanism by combining traditional verification benchmark comparisons with intelligent analysis based on machine learning models. The verification benchmark ensures strict adherence to known rules, while the machine learning model learns from historical data and abnormal patterns to identify potential, undefined, or complex data inconsistencies. This hybrid verification strategy not only ensures verification accuracy and coverage, but also empowers the system with self-learning and intelligent evolution capabilities, enabling more effective identification of diverse data discrepancies. This significantly improves the intelligence, accuracy, and detection rate of data consistency verification.

[0030] 5. In addition to processing automatically repairable information, the automated repair module also adds the ability to identify and handle information that cannot be automatically repaired. By generating reports containing specific unrepairable information and repair suggestions, users are provided with clear problem diagnosis and clear follow-up guidance. This not only enables the platform to more comprehensively address various complex data inconsistency scenarios, but also effectively assists in the manual intervention process, reducing user analysis and troubleshooting time, improving the efficiency and accuracy of complex problem handling, and further enhancing the platform's practicality and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a module diagram of a cloud service automatic migration platform according to an embodiment of the present application; Figure 2 This is a module diagram of the data source adaptation layer in an embodiment of the present application; Figure 3 1 is a schematic diagram of a dynamic feature extraction module according to an embodiment of the present application; Figure 4 Schematic diagram of the module of the joint verification engine of the embodiment of the present application; Figure 5 1 is a schematic diagram of an automated repair module according to an embodiment of the present application; Figure 6 This is a flow chart of the main steps of a method for automatic migration of cloud services in an embodiment of the present application.

[0032] Explanation of the accompanying drawings: 10. Data source adaptation layer; 11. Connector management module; 12. Adapter plug-in module; 13. Standardized metadata model; 14. Unified data extraction interface; 20. Dynamic feature extraction module; 21. Data receiving module; 22. Feature data calculation engine; 23. Feature extraction logic module; 24. Feature data storage module; 30. Joint verification engine; 31. Feature data input module; 32. Benchmark engine; 33. Machine learning analysis module; 34. Comparison and verification logic module; 35. Report generation module; 40. Automatic repair module; 41. Inconsistency information input module; 42. Inconsistency analysis and diagnosis module; 43. Repair instruction generation module; 44. Repair execution module; 45. Repair result monitoring and verification trigger module; 46. Report and alarm module; 50. Source data platform; 60. Target data platform. DETAILED DESCRIPTION

[0033] The following is combined with Figure 1 —6 Further explain this application in detail.

[0034] An embodiment of the present application discloses an automatic migration platform for cloud services. Figure 1 This is a module diagram of a cloud service automatic migration platform in an embodiment of the present application. Figure 1 The cloud service automatic migration platform includes a data source adaptation layer (Data Source Adaptation Layer) 10, a dynamic feature extraction module (Dynamic Feature Extraction Module) 20, a joint validation engine (Joint Validation Engine) 30 and an automated repair module (Automated Repair Module) 40.

[0035] The data source adaptation layer 10 has a standardized metadata model (Standardized Metadata Model) 13. The data source adaptation layer 10 converts the metadata of the source data platform 50 through the standardized metadata model so that the source data platform 50 is adapted to the target data platform 60.

[0036] The dynamic feature extraction module 20 is connected to the data source adaptation layer 10. The dynamic feature extraction module 20 has multi-dimensional data features. The dynamic feature extraction module 20 extracts multi-feature data from the metadata of the source data platform 50, the intermediate data in migration, and the migrated data of the target data platform 60 during the data migration process based on the multi-dimensional data features. The source data platform 50 dynamically adjusts the feature types and extraction granularity of the multi-dimensional data features based on the content changes of the metadata, the intermediate data, and the migrated data.

[0037] The joint verification engine 30 is connected to the dynamic feature extraction module 20. The joint verification engine 30 can jointly verify the benchmark and the machine learning model, and perform comparative analysis on the multi-feature data through the verification benchmark and the machine learning model. When the features of the comparative analysis are inconsistent, a feature inconsistency verification report is generated.

[0038] The automated repair module 40 is connected to the joint verification engine 30. The automated repair module 40 determines whether there is information that can be automatically repaired or information that cannot be automatically repaired based on the feature inconsistency verification report. If it is determined that there is information that can be automatically repaired, the automated repair module 40 generates an automatic repair instruction, and the target data platform 60 can repair the migrated data according to the automatic repair instruction. If it is determined that there is information that cannot be automatically repaired, the automated repair module 40 generates a repair suggestion report containing the information that cannot be automatically repaired and repair suggestions, facilitating subsequent manual intervention based on the repair suggestion report.

[0039] The following further describes in detail the various modules of the cloud service automatic migration platform.

[0040] Figure 2This is a module diagram of the data source adaptation layer in the embodiment of the present application. Figure 2 The data source adaptation layer 10 includes a connector management module (Connector Management) 11, an adapter plug-in module (Adapter Plugins) 12, a standardized metadata model 13 and a unified data extraction interface (Unified Data Extraction Interface) 14.

[0041] The connector management module 11 is responsible for managing the process of establishing and maintaining connections with the source data platform 50. This module provides a unified connector interface that supports multiple connection methods (e.g., JDBC (Java Database Connectivity), ODBC (Open Database Connectivity), and platform-specific native APIs). This module 11 includes a connector interface that the adapter plug-in can call and internally encapsulates the specific connection logic for different databases or platforms.

[0042] The adapter plug-in module 12 includes a corresponding adapter plug-in (HDFS plug-in, Hive plug-in, MaxCompute plug-in, MySQL plug-in, etc.) for each supported source data platform 50 type (e.g., HDFS, Hive, MaxCompute, MySQL, etc.). Each adapter plug-in encapsulates all logic for interacting with the corresponding source data platform 50. The adapter plug-in establishes a connection with the source data platform 50 by invoking the unified connector interface provided by the connector management module 11. Optionally, the adapter plug-in module 12 incorporates a metadata-feature-driven adaptive adapter that uses machine learning to automatically identify unknown data source types (e.g., through data pattern matching or API detection), reducing manual plug-in development costs.

[0043] After establishing a connection with the source data platform 50, the adapter plug-in obtains the raw data from the source data platform 50, parses the raw data from the source data platform 50 to generate metadata for the source data platform 50, and maps the metadata of the source data platform 50 to the standardized metadata model 13. The adapter plug-in module 12 can implement a unified data extraction interface, read actual data from the source data platform 50, and mask differences in the underlying storage formats (e.g., Parquet, ORC) of different source data platforms 50.

[0044] When the original data of the source data platform 50 is structured data (such as JSON, XML, or platform-specific data objects), the adapter plug-in obtains the original data of the source data platform 50 through the API interface of the source data platform 50, and extracts metadata information based on the data structure of the original data of the source data platform 50 to generate metadata for the source data platform 50. When the original data of the source data platform 50 is semi-structured data (such as text statement data or other PDF semi-structured data), the adapter plug-in obtains the original data of the source data platform 50 (such as DDL statements) through a search method (such as a query statement), converts the original data of the source data platform 50 into structured intermediate representation data (such as an abstract syntax tree (AST) or other intermediate representation form) through syntax analysis (such as syntax parsing tools such as ANTLR) and lexical analysis, and extracts metadata information based on the data structure of the structured intermediate representation data to generate metadata for the source data platform 50.

[0045] The standardized metadata model 13 defines a standardized representation of common data types, table structures, partitioning methods, dependencies, and more, serving as a unified language for understanding and processing metadata within the system. The adapter plug-in maps the metadata of the source data platform 50 to the standardized metadata model 13. The dynamic feature extraction module 20 obtains the metadata of the source data platform 50 by accessing the standardized metadata model 13.

[0046] The unified data extraction interface 14 is used to provide a standardized interface so that the dynamic feature extraction module 20 can request and read data in a unified manner without having to worry about which specific source data platform 50 the data comes from or its storage format. Each of the adapter plug-ins is responsible for implementing the specific reading logic of the unified data extraction interface 14. When the dynamic feature extraction module 20 requests data through the unified data extraction interface 14, the request will be routed to the adapter plug-in currently in use, which will be responsible for reading the data from the actual data source and returning it.

[0047] The overall workflow of the data source adaptation layer 10 is as follows: when the system needs to interact with a source data platform 50, the corresponding adapter plug-in is selected and loaded through the data source adaptation layer 10; the adapter plug-in calls the connector management module 11 to establish a connection with the source data platform 50; the adapter plug-in communicates with the source data platform 50, obtains the original data and parses it to generate metadata; the adapter plug-in maps the parsed metadata to the standardized metadata model 13 for internal use in the system; when data needs to be read, the dynamic feature extraction module 20 calls the unified data extraction interface 14, and the unified data extraction interface 14 is directed to the running adapter plug-in, which is responsible for the actual data reading work, and provides the data stream or data block to the dynamic feature extraction module 20 through the unified data extraction interface 14.

[0048] The data source adaptation layer 10, based on the scalability and standardized interfaces and models of the adapter plug-ins, enables the system to flexibly integrate new source data platforms 50 without modifying the core logic. Each adapter plug-in implements the connection, metadata parsing / mapping, and data reading functions for a specific platform.

[0049] Figure 3 Schematic diagram of the dynamic feature extraction module of the embodiment of the present application. Figure 3 The dynamic feature extraction module 20 includes a data receiving module (Data Ingestion) 21, a feature data calculation engine (Feature Data Calculation Engine) 22, a feature extraction logic module (Feature Extraction Logic) 23 and a feature data storage module (Feature Data Storage Component) 24.

[0050] The data receiving module 21 is used to receive metadata from the source data platform 50 after conversion by the data source adaptation layer 10, intermediate data in the data migration channel, and migrated data that has been written to the target data platform 60. The data receiving module 21 can receive data continuously in real time or in batches.

[0051] The feature data calculation engine 22 is a core data processing unit, which stores multi-dimensional data features and is used to extract multi-feature data from the metadata of the source data platform 50, the intermediate data and the migrated data based on the multi-dimensional data features. The feature data calculation engine 22 can be implemented based on a stream processing (such as ApacheFlink, Spark Streaming) engine to support real-time, parallel and incremental computing, and achieve millisecond-level feature response. For multi-partitioned tables or parallel migration tasks, the feature data calculation engine 22 can start multiple computing instances in parallel to improve processing speed. In addition, the feature data calculation engine 22 can use stream processing or incremental computing technology to only update features for newly added or changed data parts to avoid repeated calculations. For ultra-large-scale data sets, the feature data calculation engine 22 can use approximate computing algorithms (such as HyperLogLog) to balance accuracy and efficiency.

[0052] The multi-dimensional data features include metadata dictionary information, such as basic statistical features, field-level features, data distribution features and data quality features, and the multi-feature data includes basic statistical feature data, field-level feature data, data distribution feature data and data quality feature data corresponding to the multi-dimensional data features.

[0053] The feature data calculation engine 22 applies different calculation and analysis methods to extract multi-feature data based on different multi-dimensional data features. For basic statistical features, it performs big data calculations on the received data, such as counting (number of records, number of files), summing, averaging, maximum and minimum values, and byte size statistics. For field-level features, it extracts them through feature engineering. For example, it traverses each field in the data record and counts the number of non-null values, the number of unique values, the length distribution of string strings, and the degree of conformance to specific formats (such as dates and email addresses). It then checks and calculates the value of each field. For data distribution features, it reflects the distribution of data values ​​through methods such as constructing histograms, calculating frequency distributions, and performing quantile analysis (such as quartiles). Field values ​​are then collected and grouped for statistics. For data quality features, it extracts them through the Data Quality Control (DQC) platform, which can apply predefined rules or patterns (such as regular expressions) to check whether the data complies with business rules or format requirements and calculate the proportion of illegal or erroneous data.

[0054] The feature extraction logic module 23 is configured to dynamically adjust the feature types and extraction granularity (i.e., the level of detail or precision used when extracting data features) of the multi-dimensional data features based on the received data. When the feature extraction logic module 23 detects a specific change in the data content, it automatically triggers a more detailed or in-depth extraction of features associated with the data. For example, when an unexpected change in the enumeration value of a field is detected, it automatically triggers a more detailed frequency distribution statistics for that field. This enables the feature data calculation engine 22 to intelligently determine which feature data to extract, at what granularity, and whether to perform a more in-depth analysis of certain feature data based on the actual data conditions and changes, thereby improving the relevance and efficiency of feature extraction. Optionally, a feedback-based control loop is introduced into the feature extraction logic module 23 to dynamically adjust the feature extraction intensity (e.g., fine-grained extraction under low load, coarse-grained extraction under high load) based on the resource utilization of the historical migration task, thereby avoiding excessive dynamic adjustment of feature extraction granularity (e.g., field-level traversal frequency) that may lead to wasted computing resources.

[0055] The feature extraction logic module 23 may be integrated into the feature data calculation engine 22 and be called and executed as a part of the feature data calculation engine 22 , or may be an independent calculation unit.

[0056] The feature data storage module 24 is used to receive and store the multi-feature data of each stage and batch calculated by the feature data calculation engine 22, and provides an interface for querying and obtaining the multi-feature data to the joint verification engine 30. The feature data storage module 24 typically uses a high-performance storage system (such as Redis or HBase) to enable fast access by the joint verification engine 30.

[0057] The overall workflow of the dynamic feature extraction module 20 is as follows: the data receiving module 21 receives the metadata of the source data platform 50 after conversion from the data source adaptation layer 10, the intermediate data in the data migration channel, and the migrated data that has been written to the target data platform 60; the data receiving module 21 sends the received data to the feature data calculation engine 22; the feature data calculation engine 22 performs real-time, parallel, and incremental calculations on the received data based on the multi-dimensional data features to extract multi-feature data. In this process, the feature types and extraction granularity of the multi-dimensional data features may be dynamically adjusted by the feature extraction logic module 23 according to changes in the content of the received data; the calculated multi-feature data is sent to the feature data storage module 24 for persistent storage; the joint verification engine 30 obtains the required multi-feature data by accessing the feature data storage module 24 for subsequent comparative analysis.

[0058] Figure 4 This is a module diagram of the joint verification engine of the embodiment of the present application. Figure 4 The joint verification engine 30 includes a feature data input module (Feature Data Input) 31, a standard engine (Standard Engine) 32, a machine learning analysis module (Machine Learning Analysis) 33, a comparison and verification logic module (Comparison and Validation Logic) 34 and a report generation module (Report Generation) 35.

[0059] The feature data input module 31 is connected to the feature data storage module 24 of the dynamic feature extraction module 20 and is used to read multiple feature data that need to be compared and analyzed.

[0060] The benchmark engine 32 is connected to the feature data input module 31 and is used to load and execute a preset verification benchmark to judge the multi-feature data and generate benchmark judgment information for violations of the verification benchmark. The verification benchmark is based on clearly defined judgment criteria such as thresholds, exact matching, and structural comparison.

[0061] The machine learning analysis module 33 is connected to the feature data input module 31, the benchmark engine 32, and the comparison and verification logic module 34, and includes a machine learning model. The machine learning model can use algorithms such as isolation forest and one-class support vector machines to train the model based on multiple feature data and inconsistency verification information during the historical data migration process.

[0062] The machine learning model is trained using multi-feature data and inconsistency verification information from the historical data migration process. The machine learning analysis module 33 uses the trained machine learning model and related analysis logic to detect anomalies in the input multi-feature data and generate anomaly analysis information. The comparison and verification logic module 34 receives the benchmark judgment information output by the benchmark engine 32 and the anomaly analysis information output by the machine learning analysis module 33. Combining built-in feature comparison algorithms (such as exact matching, fuzzy matching, statistical testing) and dependency verification algorithms, it performs a comprehensive comparison and consistency check to generate inconsistency verification information, which is then sent to the report generation module 35. Optionally, the comparison and verification logic module 34 uses an adaptive weighting algorithm to dynamically adjust the confidence weights of the verification benchmark and machine learning analysis based on data quality (e.g., reducing the benchmark weight in high-noise data scenarios). Compared to using only the verification benchmark for analysis and comparison, the false positive rate can be reduced by approximately 40%.

[0063] The report generation module 35 receives the inconsistency verification information and formats it to generate a detailed inconsistency verification report, which lists the difference items, the degree of difference, possible root causes, etc.

[0064] The overall workflow of the joint verification engine 30 is as follows: the feature data input module 31 obtains the multi-feature data that needs to be verified; the multi-feature data is sent to the benchmark engine 32 and the machine learning analysis module 33 for parallel processing; the benchmark engine 32 judges the multi-feature data based on the verification benchmark and generates benchmark judgment information that violates the verification benchmark; the machine learning analysis module 33 performs anomaly detection on the multi-feature data and generates anomaly analysis information; the comparison and verification logic module 34 receives the benchmark judgment information and anomaly analysis information, and combines its internal feature comparison algorithm and dependency verification algorithm to perform comprehensive verification and analysis on the multi-feature data to generate inconsistent verification information; the report generation module 35 generates a final inconsistent verification report based on the inconsistent verification information.

[0065] Figure 5 Schematic diagram of the module of the automatic repair module of the embodiment of the present application. Figure 5 The automated repair module 40 includes an inconsistency information input module 41, an inconsistency analysis and diagnosis module 42, a repair instruction generation module 43, a repair execution module 44, a repair result monitoring and validation trigger module 45, and a reporting and alerting module 46.

[0066] The inconsistency information input module 41 is connected to the report generation module 35 of the joint verification engine 30 and is used to receive the inconsistency verification report.

[0067] The inconsistency analysis and diagnosis module 42 is connected to the inconsistency information input module 41 and is used to conduct in-depth analysis of the received inconsistency verification report. The inconsistency analysis and diagnosis module 42 identifies the type, severity, and specific details of the discrepancy and uses automatic root cause diagnosis technology based on historical data and expert knowledge to determine the root cause of the discrepancy and whether there is any information that can be automatically repaired or not.

[0068] The repair instruction generation module 43 is connected to the inconsistency analysis and diagnosis module 42 and to a repair strategy knowledge base integrated within the automated repair module 40 or located externally. The repair instruction generation module 43 queries the repair strategy knowledge base for the automatically repairable information and generates automatic repair instructions. These automatic repair instructions include automatic repair scripts generated based on the discrepancy details or API call instructions for the target data platform 60. Optionally, the repair instruction generation module 43 includes an additional adaptation layer to automatically adapt the API / syntax specifications of the target data platform 60 before generating the repair instructions (e.g., converting general SQL to Snowflake). The automatic repair script can be a SQL DDL (Data Definition Language) or DML (Data Manipulation Language) script. For example, if a field is missing from the target table, the module can generate an ALTER TABLE ADD COLUMN DDL statement. If data format errors are detected, an UPDATE statement combined with a conversion function can be generated to correct the data value. Furthermore, a general Python script can be generated to execute more complex repair logic. Typically, template-based script generation capabilities are used to populate pre-set script templates based on specific difference parameters. However, for some target data platforms 60, data modification or reprocessing operations can be performed directly by calling their provided APIs, without the need to generate and execute scripts. For example, a cloud platform SDK can be used to adjust a table structure or re-import data from a partition.

[0069] The repair execution module 44 is connected to the repair instruction generation module 43 and the target data platform 60, and is used to send the automatic repair script to the target data platform 60 for execution, or call the target data platform 60 to perform the repair operation according to the API call instruction, and send the execution information of the automatic repair instruction (success, failure, error information) to the repair result monitoring and verification triggering module 45. The repair result monitoring and verification triggering module 45 is connected to the repair execution module 44 and the joint verification engine 30, and is used to monitor the execution information of the automatic repair instruction. When the automatic repair instruction is successfully executed, it automatically re-triggers the relevant verification items and notifies the joint verification engine 30 to re-verify the repaired data to confirm that the problem has been resolved.

[0070] The report and alarm module 46 is connected to the inconsistency analysis and diagnosis module 42, the repair execution module 44 and the repair result monitoring and verification trigger module 45, wherein the execution status and re-verification results of the automatic repair instruction are obtained from the repair execution module 44 and the repair result monitoring and verification trigger module 45, and the non-automatic repair information is obtained from the inconsistency analysis and diagnosis module 42. The report and alarm module 46 is used to summarize the status and results of the entire automated repair process, record the operation logs, results, problems encountered, etc. of the automatic repair, and generate a repair suggestion report containing non-automatic repair information and repair suggestions for situations where automatic repair information cannot be automatically repaired or automatic repair fails, and notify the user through a visual interface or an alarm system (email, SMS, etc.) to facilitate subsequent manual intervention by the user. The visual interface adopts the Vue interface design, and uses an interactive migration monitoring screen (which can display visual graphics such as abnormal data heat maps) to enhance the user experience.

[0071] The overall workflow of the automatic repair module 40 is as follows: the inconsistency information input module 41 receives the inconsistency verification report from the joint verification engine 30; the inconsistency analysis and diagnosis module 42 analyzes the inconsistency verification report to identify information that can be automatically repaired and information that cannot be automatically repaired; the information that can be automatically repaired is sent to the repair instruction generation module 43 to generate specific automatic repair instructions (script / API call); the automatic repair instructions are sent to the repair execution module 44 for actual operation; the execution result information of the automatic repair instructions is monitored by the repair result monitoring and verification triggering module 45, and triggers the joint verification engine 30 to perform re-verification; the status, diagnostic information, repair suggestions, execution results, etc. of the entire process are sent to the report and alarm module 46 for recording, display and notification; information that cannot be automatically repaired is also directly sent to the report and alarm module 46 to generate diagnostic reports and suggestions.

[0072] The implementation principle of an automatic migration platform for cloud services in an embodiment of the present application is as follows: metadata of heterogeneous data sources are standardized and converted through a data source adaptation layer 10, thereby solving the problem of incompatibility of data structures and formats between different platforms and improving the adaptability of data migration; data features are extracted in real time and in multiple dimensions through a dynamic feature extraction module 20, and the extraction strategy is dynamically adjusted according to the data content, thereby ensuring the comprehensiveness and pertinence of feature extraction and laying the foundation for subsequent precise verification; comparative analysis is performed through a joint verification engine 30 in combination with verification benchmarks and machine learning, thereby improving the accuracy and intelligence level of data inconsistency detection; automatic judgment and execution of repair operations by an automated repair module 40 significantly reduces manual intervention, improves the efficiency and quality of data migration, thereby realizing efficient, intelligent and reliable automated migration, verification and repair at the data level during the cloud service migration process, and ensuring the consistency and availability of the migrated data.

[0073] An embodiment of the present application also discloses a method for automatic migration of cloud services implemented by the cloud service automatic migration platform. Figure 6 This is a schematic diagram of the main steps of a method for automatic migration of cloud services in an embodiment of the present application. Figure 6 , the method comprises the following main steps: S1. Convert the metadata of the source data platform 50 into the standardized metadata model of the data source adaptation layer 10 , so that the source data platform 50 is adapted to the target data platform 60 .

[0074] S2. During the data migration process, multi-feature data is extracted from the metadata of the source data platform 50, the intermediate data being migrated, and the migrated data of the target data platform 60 using multi-dimensional data features, wherein the feature types and extraction granularity of the multi-dimensional data features are dynamically adjusted according to the content changes of the metadata of the source data platform 50, the intermediate data, and the migrated data.

[0075] S3. Analyze the multi-feature data, wherein the analysis mode includes combining a calibration benchmark with machine learning for comparison.

[0076] S4. When the features analyzed in step S3 are inconsistent, a feature inconsistency verification report is generated.

[0077] S5. Determine whether there is information that can be automatically repaired based on the feature inconsistency verification report.

[0078] S6. When it is determined in step S5 that there is information that can be automatically repaired, an automatic repair instruction is generated, and the target data platform 60 can repair the migrated data according to the automatic repair instruction.

[0079] S7. Determine whether there is information that cannot be automatically repaired based on the feature inconsistency verification report.

[0080] S8. When it is determined in step S7 that there is information that the system cannot be automatically repaired, a repair suggestion report including the information that the system cannot be automatically repaired and repair suggestions is generated.

[0081] The main step S1 is mainly implemented by the data source adaptation layer 10. Figure 2 The main step S1 includes the following subordinate steps: S11, obtaining the original data of the source data platform 50; S12, parsing the original data of the source data platform 50 to generate metadata of the source data platform 50; S13, mapping the metadata of the source data platform 50 to a standardized metadata model. In step S11, when it is determined that the original data of the source data platform 50 is structured data, the original data of the source data platform 50 is obtained through the data interface of the source data platform 50; in step S12, metadata information is extracted based on the data structure of the original data of the source data platform 50 to generate metadata of the source data platform 50; in step S11, when it is determined that the original data of the source data platform 50 is text statement data, the original data of the source data platform 50 is obtained through a query statement; in step S12, the original data of the source data platform 50 is decomposed into structured intermediate representation data through syntax analysis, and metadata information is extracted based on the data structure of the structured intermediate representation data to generate metadata of the source data platform 50.

[0082] The main step S2 is mainly implemented by the dynamic feature extraction module 20. Figure 3 In the main step S2, the multi-dimensional data features include basic statistical features, field-level features, data distribution features and data quality features, and the multi-feature data includes basic statistical feature data, field-level feature data, data distribution feature data and data quality feature data, wherein the basic statistical feature data is extracted from the metadata of the source data platform, the intermediate data in migration and the migrated data of the target data platform through aggregation calculation, the field-level feature data is extracted through field traversal, the data distribution feature data is extracted through statistical analysis, and the data quality feature data is extracted through data inspection.

[0083] The main steps S3 and S4 are mainly implemented by the joint verification engine 30. Figure 4 In the main step S3, the multi-feature data is compared and analyzed with the verification benchmark, the multi-feature data and the feature inconsistency verification report in the historical data migration process are used to train the machine learning model, and the multi-feature data is compared and analyzed using the machine learning model trained with historical data; in the main step S4, when feature inconsistencies are identified in the multi-feature data based on the verification benchmark or the machine learning model, a feature inconsistency verification report is generated.

[0084] The main steps S5 to S8 are mainly implemented by the automatic repair module 40. Figure 5 In main steps S5-S8, the presence of automatically repairable and non-automatically repairable information is determined based on the feature inconsistency check report. If it is determined that automatically repairable information exists, an automatic repair instruction is generated, and the target data platform 60 can repair the migrated data according to the automatic repair instruction. If it is determined that non-automatically repairable information exists, a repair suggestion report containing the non-automatically repairable information and repair suggestions is generated, facilitating subsequent manual intervention by the user based on the repair suggestion report.

[0085] The implementation principle of a method for automatic migration of cloud services in an embodiment of the present application is as follows: an automated verification and repair process for cloud service data migration is implemented through the cloud service automatic migration platform. First, the problem of heterogeneous platform adaptation is solved through metadata standardization; then, the comprehensiveness and real-time nature of the verification data are ensured through dynamic, multi-dimensional feature extraction; then, the accuracy and intelligence of inconsistency detection are improved by comparative analysis of joint verification benchmarks and machine learning; finally, by judging and executing automatic repair instructions, the migration efficiency and data quality are improved. This method systematically solves the core pain points in data migration and achieves the technical effect of efficiently, intelligently and reliably ensuring the consistency of migrated data.

[0086] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A cloud service automatic migration platform, characterized by: include: A data source adaptation layer (10) having a standardized metadata model (13) for converting metadata of a source data platform (50) to adapt to a target data platform (60); A dynamic feature extraction module (20) is connected to the data source adaptation layer (10), and the dynamic feature extraction module (20) has multi-dimensional data features and is used to extract multi-feature data from the metadata of the source data platform (50), the intermediate data being migrated, and the migrated data of the target data platform (60) during the data migration process, wherein the feature type and extraction granularity of the multi-dimensional data features are dynamically adjusted according to content changes of the metadata of the source data platform (50), the intermediate data, and the migrated data; A joint verification engine (30) is connected to the dynamic feature extraction module (20), and the joint verification engine (30) can combine a verification benchmark with machine learning to perform comparative analysis on the multi-feature data, and generate a feature inconsistency verification report when the features of the comparative analysis are inconsistent; An automatic repair module (40) is connected to the joint verification engine (30) and is used to determine whether there is information that can be automatically repaired based on the feature inconsistency verification report, and to generate an automatic repair instruction when it is determined that the information that can be automatically repaired exists. The target data platform (60) can repair the migrated data according to the automatic repair instruction.

2. The cloud service automatic migration platform according to claim 1, characterized in that: The data source adaptation layer (10) has a plurality of adapter plug-ins, each of which corresponds to the type of the source data platform (50). The adapter plug-in is used to obtain the original data of the source data platform (50), parse the original data of the source data platform (50) to generate metadata of the source data platform (50), and map the metadata of the source data platform (50) to the standardized metadata model (13).

3. The cloud service automatic migration platform according to claim 2, characterized in that: When the original data of the source data platform (50) is structured data, the adapter plug-in obtains the original data of the source data platform (50) through the data interface of the source data platform (50), and extracts metadata information according to the data structure of the original data of the source data platform (50) to generate metadata of the source data platform (50); When the original data of the source data platform (50) is semi-structured data, the adapter plug-in obtains the original data of the source data platform (50) through a search method, converts the original data of the source data platform (50) into structured intermediate representation data through syntax analysis and lexical analysis, and extracts metadata information based on the data structure of the structured intermediate representation data to generate metadata of the source data platform (50).

4. The cloud service automatic migration platform according to claim 1, characterized in that: The multi-dimensional data features include metadata dictionary information, the metadata dictionary information includes basic statistical features, field-level features, data distribution features and data quality features, and the multi-feature data includes basic statistical feature data, field-level feature data, data distribution feature data and data quality feature data; wherein the dynamic feature extraction module (20) extracts the basic statistical feature data from the metadata of the source data platform (50), the intermediate data in migration and the migrated data of the target data platform (60) through big data calculation, extracts the field-level feature data through feature engineering, extracts the data distribution feature data through statistical analysis, and extracts the data quality feature data through a data audit platform.

5. The cloud service automatic migration platform according to claim 1, characterized in that: The joint verification engine (30) compares and analyzes the multi-feature data with the verification benchmark, uses the multi-feature data and feature inconsistency verification report in the historical data migration process to train the machine learning model, and uses the machine learning model trained with the historical data to compare and analyze the multi-feature data. When it is identified that the multi-feature data has feature inconsistencies based on the verification benchmark or the machine learning model, a feature inconsistency verification report is generated.

6. The cloud service automatic migration platform according to claim 1, characterized in that: The automatic repair module (40) is further used to determine whether there is information that cannot be automatically repaired based on the feature inconsistency verification report, and when it is determined that there is information that cannot be automatically repaired, generate a repair suggestion report containing the information that cannot be automatically repaired and repair suggestions.

7. A method for automatic migration of cloud services, characterized in that: The following steps are involved: S1, converting metadata of a source data platform (50) into a standardized metadata model (13) of a data source adaptation layer (10), so that the source data platform (50) is adapted to a target data platform (60); S2. Extracting multi-feature data from the metadata of the source data platform (50), the intermediate data being migrated, and the migrated data of the target data platform (60) using multi-dimensional data features during the data migration process, wherein the feature types and extraction granularity of the multi-dimensional data features are dynamically adjusted according to content changes of the metadata of the source data platform (50), the intermediate data, and the migrated data; S3. analyzing the multi-feature data, wherein the analysis mode includes combining a calibration benchmark with machine learning for comparison; S4. When the features of the comparative analysis in step S3 are inconsistent, a feature inconsistency verification report is generated; S5. Determine whether there is information that can be automatically repaired based on the feature inconsistency check report; S6. When it is determined in step S5 that there is information that can be automatically repaired, an automatic repair instruction is generated, and the target data platform (60) can repair the migrated data according to the automatic repair instruction.

8. The method for automatic migration of cloud services according to claim 7, characterized in that: Step S1 includes: S11, obtaining original data from the source data platform (50); S12, parsing the original data of the source data platform (50) to generate metadata of the source data platform (50); S13, mapping the metadata of the source data platform (50) to the standardized metadata model (13); Wherein, in step S11, when it is determined that the original data of the source data platform (50) is structured data, the original data of the source data platform (50) is obtained through the data interface of the source data platform (50); in step S12, metadata information is extracted according to the data structure of the original data of the source data platform (50) to generate metadata of the source data platform (50); In step S11, when it is determined that the original data of the source data platform (50) is semi-structured data, the original data of the source data platform (50) is obtained by searching; in step S12, the original data of the source data platform (50) is decomposed into structured intermediate representation data through grammatical analysis and lexical analysis, and metadata information is extracted according to the data structure of the structured intermediate representation data to generate metadata of the source data platform (50).

9. The method for automatic migration of cloud services according to claim 7, characterized in that: In step S2, the multi-dimensional data feature metadata dictionary information, the metadata dictionary information includes basic statistical features, field-level features, data distribution features and data quality features, the multi-feature data includes basic statistical feature data, field-level feature data, data distribution feature data and data quality feature data, wherein the basic statistical feature data is extracted from the metadata of the source data platform (50), the intermediate data in migration and the migrated data of the target data platform (60) through big data calculation, the field-level feature data is extracted through feature engineering, the data distribution feature data is extracted through statistical analysis, and the data quality feature data is extracted through a data audit platform; in step S3, the multi-feature data is compared and analyzed with the verification benchmark, the multi-feature data and the feature inconsistency verification report in the historical data migration process are used to train the machine learning model, and the multi-feature data is compared and analyzed using the machine learning model trained with historical data; in step S4, when the multi-feature data is identified to have feature inconsistencies based on the verification benchmark or the machine learning model, a feature inconsistency verification report is generated.

10. The method for automatic migration of cloud services according to claim 7, wherein: Also includes: S7. Determine whether there is information that cannot be automatically repaired based on the feature inconsistency verification report; S8. When it is determined in step S7 that there is information that the system cannot be automatically repaired, a repair suggestion report including the information that the system cannot be automatically repaired and repair suggestions is generated.

Images