Cloud storage cross-domain data management method and system based on transfer learning

Abstract

This invention provides a cloud storage cross-domain data management method and system based on transfer learning, relating to the field of cloud storage technology. First, it takes the storage architecture characteristics of different cloud storage domains and the inherent attribute information of cross-domain data as input, and obtains a cloud storage cross-domain data migration adaptation benchmark through transfer learning preheating processing. Next, it extracts cross-domain migration knowledge from the transfer learning model; it constructs a cross-domain data flow channel by combining the cloud storage cross-domain data migration adaptation benchmark and cross-domain migration knowledge; it performs migration adaptation prediction on the cross-domain data flow channel through the transfer learning model; it dynamically adjusts the channel based on the prediction results, generates a dynamic cross-domain data management scheme, and executes the migration and storage control of cloud storage cross-domain data according to the dynamic cross-domain data management scheme. This achieves efficient and accurate migration and storage control of cloud storage cross-domain data, effectively improving the intelligence and flexibility of cross-domain data management, and reducing data management costs and risks.

Description

Technical Field

[0001] This invention relates to the field of cloud storage technology, and more specifically, to a cloud storage cross-domain data management method and system based on transfer learning. Background Technology

[0002] In today's era of digital information explosion, cloud storage technology, with its powerful data storage and management capabilities, has become an important choice for enterprises and individuals to store data. With the globalization and diversification of businesses, data often needs to be distributed across different cloud storage domains to achieve redundant backups, reduce storage costs, or meet the business needs of different regions. However, different cloud storage domains have their own unique storage architecture characteristics, such as differences in storage protocols, data formats, and access control mechanisms. Furthermore, cross-domain data also has its inherent attributes, such as data type, data volume, and data sensitivity.

[0003] Currently, traditional methods for cross-domain data management in cloud storage mainly rely on manual configuration and simple rule matching. Manual configuration is not only inefficient but also prone to errors due to human factors, making it unsuitable for large-scale, complex cross-domain data management needs. While simple rule matching can achieve data migration and storage to some extent, its lack of in-depth understanding and dynamic adaptability to the characteristics of different cloud storage domains and cross-domain data attributes often fails to guarantee the accuracy and efficiency of data migration, easily leading to data loss, data format incompatibility, and other problems, thus affecting the normal operation of business. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a cloud storage cross-domain data management method based on transfer learning, the method comprising: Using the storage architecture characteristics of different cloud storage domains and the inherent attribute information of cross-domain data as input, the adaptation benchmark elements for cross-domain data migration are formed after transfer learning preheating, thus obtaining the cloud storage cross-domain data migration adaptation benchmark. Cross-domain transfer knowledge is extracted from the transfer learning model. The cross-domain transfer knowledge is separated from the pre-trained weights of the transfer learning model. The cross-domain transfer knowledge includes data migration and adaptation experience between different cloud storage domains and data storage compatibility rules. A cross-domain data transfer channel is constructed by combining cloud storage cross-domain data migration adaptation benchmarks and cross-domain migration knowledge. The cross-domain data transfer channel includes the node association relationship and data transfer control logic for data transmission between different cloud storage domains. The transfer learning model is used to predict the migration adaptation of cross-domain data transfer channels. The transfer learning model calls cross-domain migration knowledge to analyze the degree of adaptation between cross-domain data transfer channels and different cloud storage domains, and generates cross-domain data migration adaptation prediction results. Based on the cross-domain data migration adaptation prediction results, the node association relationship and data flow control logic of the cross-domain data flow channel are dynamically adjusted to generate a cross-domain data dynamic management scheme. Based on the cross-domain data dynamic management scheme, the migration and storage control of cross-domain data in cloud storage are executed. The cross-domain data dynamic management scheme includes data migration timing arrangement and dynamic allocation rules for storage resources.

[0005] Furthermore, embodiments of the present invention also provide a cloud storage cross-domain data management system based on transfer learning, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned cloud storage cross-domain data management method based on transfer learning by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, the processor of the cloud storage cross-domain data management system based on transfer learning reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the cloud storage cross-domain data management system based on transfer learning to execute the above-described cloud storage cross-domain data management method based on transfer learning.

[0007] Based on the above, starting with the storage architecture characteristics of different cloud storage domains and the inherent attribute information of cross-domain data, a cross-domain data migration adaptation benchmark is formed through transfer learning preheating. This benchmark fully considers the characteristics of different cloud storage domains and data, extracting cross-domain migration knowledge from the transfer learning model. This cross-domain migration knowledge includes data migration adaptation experience and data storage compatibility rules between different cloud storage domains. The cross-domain data flow channel constructed by combining the adaptation benchmark and cross-domain migration knowledge can accurately describe the node association relationship and data flow control logic of data transmission between different cloud storage domains. Through the transfer learning model... By performing migration and adaptation predictions on cross-domain data transfer channels, the system can analyze the compatibility of channels with different cloud storage domains in advance, generate prediction results, and dynamically adjust cross-domain data transfer channels based on these results. This generates a dynamic cross-domain data management solution, which includes data migration timing arrangements and dynamic storage resource allocation rules. It can adjust data migration and storage strategies in real time according to actual conditions, achieving efficient and accurate migration and storage control of cross-domain data in cloud storage. This effectively improves the intelligence and flexibility of cross-domain data management, reduces data management costs and risks, and ensures the stable operation of business. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the execution flow of the cloud storage cross-domain data management method based on transfer learning provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of the cloud storage cross-domain data management system based on transfer learning provided in an embodiment of the present invention. Detailed Implementation

[0010] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a cloud storage cross-domain data management method based on transfer learning, provided in one embodiment of the present invention. The following is a detailed description of this cloud storage cross-domain data management method based on transfer learning.

[0011] Step S110: Using the storage architecture characteristics of different cloud storage domains and the inherent attribute information of cross-domain data as input, the adaptation benchmark elements for cross-domain data migration are formed after transfer learning preheating, thus obtaining the cloud storage cross-domain data migration adaptation benchmark.

[0012] This embodiment focuses on an enterprise cloud storage environment, assuming two different cloud storage domains, cloud storage domain A and cloud storage domain B, requiring cross-domain data migration management. First, step S110 is executed, which aims to process the storage architecture characteristics of different cloud storage domains and the inherent attribute information of cross-domain data.

[0013] Step S111: Collect storage architecture characteristic information and cross-domain data inherent attribute information for different cloud storage domains. The storage architecture characteristic information includes the connection architecture of storage nodes, storage media type, data processing flow, transmission interface specifications and resource scheduling mechanism. The cross-domain data inherent attribute information includes data structure type, access characteristics, storage requirements, update frequency and security level requirements.

[0014] Next, proceed to step S111. In this embodiment, information is collected for cloud storage domain A and cloud storage domain B. For cloud storage domain A, its storage node connection architecture adopts a star topology, with multiple storage nodes connected through a central switch; the storage medium type is mainly solid-state drives; the data processing flow is to first verify the data, then compress the data, and finally store it; the transmission interface specification is an Ethernet interface, supporting gigabit transmission rates; the resource scheduling mechanism is based on a priority queue, with high-priority tasks receiving priority resource allocation. Regarding the inherent attributes of cross-domain data, the data to be migrated includes structured data (such as customer information tables in relational databases), semi-structured data (such as order data in XML format), and unstructured data (such as image files of product design drawings). Access characteristics show that structured data is accessed more frequently, with multiple queries per hour on average, while unstructured data is accessed less frequently, but each access involves a larger amount of data. In terms of storage requirements, structured data requires high read / write performance to support frequent queries and updates, while unstructured data has greater storage space requirements. Regarding update frequency, structured data such as customer information tables are updated daily, order data is updated hourly, and unstructured data such as product design drawings is updated weekly. In terms of security level requirements, customer information and other privacy-related data have the highest security level, order data has a high security level, and product design drawings have a medium security level. For cloud storage domain B, its storage node connection architecture is a mesh topology, the storage media type uses a hybrid of solid-state drives and hard disk drives, the data processing flow involves data encryption followed by data sharding and storage, the transmission interface specification is a Fibre Channel interface, and the resource scheduling mechanism adopts load balancing scheduling.

[0015] Step S112: Input the collected storage architecture characteristic information and cross-domain data inherent attribute information into the preheating processing module of the transfer learning model for association mapping, and establish the corresponding association information between storage architecture characteristic information and cross-domain data inherent attribute information.

[0016] Then, step S112 is executed. After collecting the storage architecture characteristic information of cloud storage domain A and cloud storage domain B, as well as the inherent attribute information of cross-domain data, the above information is input into the preheating processing module of the transfer learning model. The function of this preheating processing module is to perform correlation mapping on the two types of input information and find the inherent relationship between them.

[0017] Step S1121: Perform feature structuring processing on the collected storage architecture characteristic information, and transform different types of data in the storage architecture characteristic information into a first structured feature vector in a unified format. The dimensions of the first structured feature vector correspond to different categories of storage architecture characteristics, and the feature values ​​of each dimension are transformed into values ​​in the range [0, 1] through normalization processing.

[0018] First, proceed to step S1121. For the storage architecture characteristics information of cloud storage domain A, the connection architecture of storage nodes, storage media type, data processing flow, transmission interface specifications, and resource scheduling mechanism are used as dimensions of the first structured feature vector. For example, solid-state drives (SSDs) can correspond to one dimension in storage media type. Through normalization, the feature values ​​are transformed to the [0, 1] interval based on factors such as the proportion of SSDs in this cloud storage domain. Similarly, the storage architecture characteristics information of cloud storage domain B is processed in a similar way to obtain the corresponding first structured feature vector.

[0019] Step S1122: Perform feature structuring processing on the inherent attribute information of cross-domain data. Using the same structuring processing standard as the storage architecture characteristic information, the inherent attribute information of cross-domain data is transformed into a second structured feature vector. The dimensions of the second structured feature vector correspond to the different categories of inherent attributes of cross-domain data.

[0020] Next, proceed to step S1122. For the inherent attribute information of cross-domain data, the dimensions of the second structured feature vector are defined by categories such as structure type, access characteristics, storage requirements, update frequency, and security level requirements. The same structured processing standard used for processing storage architecture characteristic information is applied. For example, for access characteristics, they are transformed into feature values ​​in the [0, 1] interval based on factors such as access frequency, thus obtaining the second structured feature vector.

[0021] Step S1123: Input the first structured feature vector and the second structured feature vector into the feature alignment unit of the preheating module of the transfer learning model. After adjusting the dimension order according to the preset feature dimension mapping table, perform association mining on the unified feature vector. The association mining is based on the semantic relevance and data dependency of each dimension feature in the feature vector. The mining process combines domain knowledge of cross-domain data transfer and uses cross-validation method to verify the preliminary results of association mining.

[0022] For example, step S11231: Construct a semantic relevance analysis model. The semantic relevance analysis model adopts a semantic similarity calculation algorithm based on word vectors to calculate the semantic similarity between features of different dimensions in the feature vector. The input of the semantic relevance analysis model is the descriptive text and representation meaning of the feature dimensions, and the output is the semantic similarity value.

[0023] When constructing the semantic relevance analysis model, a large amount of domain-specific vocabulary related to cross-domain data migration in cloud storage is first collected. This includes descriptive text for each feature dimension of storage architecture characteristics and inherent attributes of cross-domain data, such as "storage medium type" and "data access frequency." These vocabulary words are preprocessed by removing stop words and performing lexical normalization. Then, word vector training methods such as Word2Vec are used to transform the vocabulary into distributed word vectors. The model's input layer receives the descriptive text for each feature dimension and the corresponding word vectors representing the meaning. The hidden layer performs non-linear transformations on the word vectors through a multi-layer neural network, calculating the cosine similarity between word vectors of different feature dimensions. The output layer outputs this cosine similarity as a semantic similarity value, which ranges from 0 to 1. The closer the value is to 1, the higher the semantic relevance.

[0024] Step S11232: Input the unified storage architecture characteristic structured feature vector and the cross-domain data inherent attribute structured feature vector into the semantic relevance analysis model, and calculate the semantic similarity between each dimension feature in the storage architecture characteristic structured feature vector and each dimension feature in the cross-domain data inherent attribute structured feature vector.

[0025] The unified structured feature vectors of storage architecture characteristics (such as storage media type and transmission interface specifications) and the structured feature vectors of cross-domain data inherent attributes (such as data access frequency and storage format requirements) are input into the semantic relevance analysis model. The model pairs each feature dimension (e.g., "storage media type - solid-state drive") with each feature dimension (e.g., "data access frequency - high frequency") in the cross-domain data inherent attribute vector, calculating the semantic similarity value for each pair. For example, when calculating the semantic similarity between the "storage media read / write speed" dimension and the "data access frequency" dimension, the model compares the cosine similarity of their word vectors and outputs a specific semantic similarity value, reflecting the degree of semantic correlation between read / write speed and access frequency.

[0026] Step S11233: Record the semantic similarity values ​​output by the semantic relevance analysis model, organize the semantic similarity results according to the feature dimensions, and form a semantic similarity matrix. The semantic similarity matrix is ​​used to present the degree of semantic association between different feature dimension pairs.

[0027] The semantic similarity values ​​of all feature dimension pairs calculated by the semantic relevance analysis model are recorded and organized into a semantic similarity matrix, with storage architecture characteristics as rows and cross-domain data inherent attributes as columns. Each element in the matrix corresponds to the semantic similarity value of a feature dimension pair. For example, the element in the i-th row and j-th column of the matrix represents the semantic similarity between the i-th dimension of storage architecture characteristics and the j-th dimension of cross-domain data inherent attributes. This matrix provides a clear visual indication of which feature dimension pairs have a high degree of semantic correlation. For instance, higher matrix element values ​​for the "transmission interface transmission rate" dimension and the "data access frequency" dimension indicate a strong semantic relationship between them.

[0028] Step S11234: Construct a data dependency analysis model. The data dependency analysis model identifies the data dependency relationships between features of different dimensions in the feature vector by analyzing the generation logic, usage scenarios and interaction methods of feature data.

[0029] When constructing a data dependency analysis model, the first step is to define the types of data dependencies, including direct dependencies (e.g., the value of feature A directly affects the value of feature B), indirect dependencies (e.g., feature A affects feature B by influencing feature C), and co-dependent dependencies (e.g., features A and B jointly affect feature C). The model receives metadata information for each dimension of the feature vector, including the source of the feature data (e.g., which system or module generated it), the usage scenario (e.g., at which stage of data migration it is used), and the interaction method (e.g., the computational relationship with other features). Through rule-based reasoning and machine learning methods (e.g., dependency identification based on decision trees), the model analyzes the dependency paths and influence strength between features, thereby identifying data dependencies between features of different dimensions.

[0030] Step S11235: Input the unified feature vector into the data dependency analysis model. By analyzing the source of feature data generation, transmission path, processing flow and application scenario, identify the direct and indirect dependencies between feature dimensions.

[0031] The unified structured feature vectors of storage architecture characteristics and the structured feature vectors of cross-domain data inherent attributes are input into the data dependency analysis model. The model traces the generation source of data for each feature dimension; for example, the "storage node load status" feature is generated by the storage monitoring system, and the "data access frequency" feature is generated by the access log statistics module. It analyzes the data transmission path; for example, "storage media read / write rate" data is collected by sensors and transmitted to the data processing center. It also outlines the data processing flow; for example, the "data storage format" feature requires format parsing and compatibility verification. Finally, it considers the application scenario of the data; for example, the "transmission bandwidth" feature is used in data transmission scheduling scenarios, thus comprehensively identifying the dependencies between feature dimensions. For instance, the level of the "data access frequency" feature directly affects the value of the "storage node load status" feature, forming a direct dependency. The "transmission interface protocol" feature, by influencing the "data transmission rate" feature, in turn affects the "data access response time" feature, forming an indirect dependency.

[0032] Step S11236: Record the dependency types and dependency strengths identified by the data dependency analysis model. The dependency strength is determined based on the interaction frequency, influence degree and synergistic effect between feature data, forming a data dependency matrix.

[0033] After identifying the dependency types (direct dependency, indirect dependency, and co-dependency) between feature dimension pairs, the data dependency analysis model further calculates the dependency strength. The calculation of dependency strength comprehensively considers the interaction frequency between feature data (e.g., the number of times a change in feature data A triggers a change in feature data B per unit time), the degree of influence (e.g., the percentage change in feature data B resulting from a 10% change in feature data A), and the synergistic effect (e.g., the magnitude of the influence on feature C when features A and B work together). The dependency types and dependency strengths are then organized by feature dimension pair to form a data dependency matrix. Matrix elements contain dependency type identifiers and dependency strength values. For example, (direct dependency, 0.85) indicates that the feature dimension pair has a direct dependency relationship and a dependency strength of 0.85 (range 0 to 1, with larger values ​​indicating stronger dependencies).

[0034] Step S11237: Associate and fuse the corresponding elements in the semantic similarity matrix and the data dependency matrix to generate a feature dimension association matrix, which simultaneously records the semantic similarity and dependency of different feature dimension pairs.

[0035] The semantic similarity matrix and the data dependency matrix are correlated and fused. Using feature dimension pairs as indices, the element values ​​at corresponding positions in the two matrices are integrated to generate a feature dimension correlation matrix. For example, for the feature dimension pair "storage medium read / write rate - data access frequency", the semantic similarity value of 0.82 is extracted from the semantic similarity matrix, and the dependency type "direct dependency" and dependency strength of 0.78 are extracted from the data dependency matrix. This information is integrated and recorded in the corresponding positions of the feature dimension correlation matrix, so that the matrix simultaneously contains both semantic and dependency correlation information.

[0036] Step S11238: Set an association mining threshold based on the feature dimension association matrix. The association mining threshold includes a semantic similarity threshold and a dependency strength threshold.

[0037] Based on the actual needs of cross-domain data migration and historical experience, association mining thresholds are set. The semantic similarity threshold is determined according to the acceptable degree of semantic association, for example, set to 0.6, meaning feature pairs with a semantic similarity value greater than 0.6 are considered to have a strong semantic association. The dependency strength threshold is set according to the significance of the dependency relationship, for example, set to 0.5, meaning feature pairs with a dependency strength greater than 0.5 are considered to have a significant dependency relationship. The threshold settings need to be verified through multiple experiments to ensure that valuable associations are selected while avoiding the introduction of too many noisy associations.

[0038] Step S11239: Select feature dimension pairs whose semantic similarity and dependency strength both meet the association mining threshold as preliminary results of association mining. The preliminary results are sorted according to the level of semantic similarity and dependency strength.

[0039] Each element in the feature dimension association matrix is ​​traversed, and its semantic similarity value and dependency strength value are checked to see if they are both greater than or equal to the semantic similarity threshold and the dependency strength threshold, respectively. Feature dimension pairs that simultaneously meet both threshold conditions are selected as preliminary results for association mining. The feature dimension pairs in the preliminary results are sorted from high to low semantic similarity, and if the semantic similarity is the same, they are sorted from high to low dependency strength, forming an ordered preliminary result list for subsequent analysis and validation.

[0040] Step S112310: Analyze the preliminary results, combine the actual needs and adaptation goals of cross-domain data migration, use the expert knowledge base matching method to verify the rationality and practicality of the feature dimensions in the preliminary results for association, and conduct backtracking analysis based on historical data of association confidence to supplement potential associations missed in the association mining process, and complete the association mining.

[0041] For example, domain experts analyze the preliminary results of association mining, combining the actual needs of cross-domain data migration (such as improving migration efficiency and ensuring data security) and adaptation goals (such as precise matching of storage architecture and data attributes). They then verify the rationality and practicality of the feature dimensions in the associations using an expert knowledge base matching method. The expert knowledge base stores a large number of practically validated association rules, such as "data security level - transmission encryption protocol" being a key association pair. Simultaneously, backtracking analysis is conducted based on the association confidence of each feature dimension pair in historical data (such as the percentage of times the association pair has been proven effective in the past 100 migrations). If some feature dimension pairs do not meet the current threshold but have high historical association confidence, they are added as potential associations. Through expert validation and backtracking analysis, the preliminary results are corrected and supplemented, ultimately completing the association mining and obtaining accurate and practical feature dimension associations.

[0042] Step S1124: Identify feature dimension pairs in the structured feature vectors of storage architecture characteristics and the inherent attributes of cross-domain data that have semantic relationships. The semantic relationship is determined based on the degree of fit between the description content and the representation meaning of the feature dimension, covering direct semantic relationships and indirect semantic relationships.

[0043] Based on step S1123, proceed to step S1124. By analyzing the descriptive content and representational meaning of each dimension of the feature vector, identify feature dimension pairs with semantic relationships. Direct semantic relationships include storage medium type and data storage requirements; the high read / write speed of solid-state drives directly matches the high read / write performance storage requirements of structured data. Indirect semantic relationships include transmission interface specifications and data access characteristics; the transmission rate of the transmission interface affects the data access speed, which in turn indirectly relates to the data access frequency characteristics, i.e., data with high access frequency needs to be matched with a high transmission rate interface.

[0044] Step S1125: Calculate the association confidence of each feature dimension pair. The association confidence is determined based on the semantic matching degree between feature dimensions, the association frequency in historical data migration, and the association stability. The association confidence value accurately reflects the reliability of the association between feature dimension pairs.

[0045] Then, step S1125 is executed. For the identified feature dimension pairs, their association confidence is calculated. For example, the feature dimension pair of storage medium type and data storage requirements has a high degree of semantic matching. In historical cross-domain data migration cases, the association frequency between the two is also very high, and the association stability is good. That is, the association relationship between the two is relatively stable in different scenarios. Therefore, the association confidence of this feature dimension pair is high. On the other hand, the feature dimension pair of transmission interface specification and data update frequency has a relatively low degree of semantic matching, and the historical association frequency is also unstable. Therefore, its association confidence is low.

[0046] Step S1126: Filter feature dimension pairs that meet the preset conditions for association confidence. Use the filtered feature dimension pairs as the basis for association between storage architecture characteristic information and cross-domain data inherent attribute information. Construct an association mapping table based on the association basis. The association mapping table records the association relationship, association confidence, and association type of each feature dimension pair. The association type distinguishes between semantic association, data dependency association, and functional adaptation association, making the association relationship clearer and more identifiable.

[0047] Next, proceed to step S1126. A threshold for association confidence is preset, and feature dimension pairs with association confidence higher than this threshold are selected. For example, feature dimension pairs such as storage medium type and data storage requirements, and storage node data transmission interface characteristics and data access characteristics, whose association confidence meets preset conditions, are used as the basis for association. Then, an association mapping table is constructed, recording each feature dimension pair in the table. For example, (storage medium type - data storage requirements) has the association described as the read / write performance of the storage medium type needing to meet the data storage requirements, with an association confidence of 0.85 and an association type of semantic association; (transmission interface specifications - data access characteristics) has the association that the transmission rate of the transmission interface affects the data access speed, with an association confidence of 0.7 and an association type of functional adaptation association, etc.

[0048] Step S1127: Construct an association relationship model based on the association mapping table. The association relationship model transforms the static association information in the association mapping table into dynamic association logic, and finally establishes the corresponding association information between storage architecture characteristic information and cross-domain data inherent attribute information.

[0049] Finally, step S1127 is executed. Based on the contents of the association mapping table, an association model is constructed. This association model forms dynamic association logic by analyzing the association relationships, association confidence, and association types of each feature dimension pair. For example, when the data storage requirement is high read / write performance, the model will dynamically associate it with a solid-state drive (SSD) storage medium type according to the association mapping table; when the data access frequency is high, the model will associate it with an interface with a high transmission rate. Through the above dynamic association logic, the corresponding association information between storage architecture characteristic information and cross-domain data inherent attribute information is finally established.

[0050] Step S113: Based on the established corresponding association information, extract the key characteristic items that affect cross-domain data migration from the storage architecture characteristic information. The key characteristic items include the data transmission interface characteristics of the storage node, the read and write speed characteristics of the storage medium, the compatibility characteristics of the data processing flow, the response characteristics of the resource scheduling mechanism, and the protocol adaptation characteristics of the transmission interface.

[0051] After completing step S112 to establish the corresponding association information, proceed to step S113. Based on the corresponding association information, analyze the impact of each part of the storage architecture characteristic information on cross-domain data migration. The data transmission interface characteristics of the storage node directly determine the connection method and transmission rate of data during cross-domain transmission, which is crucial to data transmission efficiency; the read / write rate characteristics of the storage medium affect the storage and access speed of data in the target storage domain, which is essential for data migration with high read / write performance requirements; the compatibility characteristics of the data processing flow are related to whether data can be correctly processed and stored between different cloud storage domains. If the processing flow is incompatible, it may lead to data corruption or unusability; the responsiveness of the resource scheduling mechanism determines whether storage resources can respond to migration requirements in a timely manner during data migration, ensuring the smooth progress of the migration task; the protocol adaptation characteristics of the transmission interface ensure the consistency of the protocol during data transmission, avoiding transmission failure due to protocol incompatibility. Therefore, these key characteristic items are extracted.

[0052] Step S114: Extract key attribute items that affect cross-domain data migration from the inherent attribute information of cross-domain data. The key attribute items include the structural complexity of the data, the access frequency characteristics of the data, the storage format requirements of the data, the update frequency characteristics of the data, and the security level adaptation requirements of the data.

[0053] Next, proceed to step S114. From the inherent attribute information of the cross-domain data, extract key attribute items affecting cross-domain data migration based on their correlation with storage architecture characteristics. The structural complexity of the data affects the compatibility of the data processing flow; complex data requires the target storage domain's data processing flow to support its structural parsing during migration. The data access frequency characteristics are related to the data transmission interface characteristics of the storage nodes and the response characteristics of the resource scheduling mechanism; frequently accessed data requires a high-efficiency transmission interface and fast-response resource scheduling. The data storage format requirements determine whether the target storage domain's storage media and data processing flow are compatible with that format. The data update frequency characteristics affect the response strategy of the resource scheduling mechanism; frequently updated data requires the resource scheduling mechanism to quickly allocate resources to support update operations. The data security level adaptation requirements are related to the protocol adaptation characteristics of the transmission interface and security processing steps in the data processing flow, such as encryption and access control.

[0054] Step S115: Perform transfer learning preheating analysis on key feature items and key attribute items. The transfer learning preheating analysis determines the adaptation association strength by calculating the feature similarity and functional matching degree between key feature items and key attribute items, and generates the adaptation association strength distribution.

[0055] Then, step S115 is performed. The pre-warming module of the transfer learning model performs pre-warming analysis on the extracted key feature items and key attribute items. When calculating feature similarity, the similarity between key feature items and key attribute items in feature performance is compared. For example, the read / write rate characteristics of storage media and the access frequency characteristics of data have high similarity in feature performance, with high read / write rate and high access frequency showing high similarity. When calculating functional matching degree, it is analyzed whether the function of key feature items can meet the requirements of key attribute items, such as whether the protocol adaptation characteristics of the transmission interface can meet the encrypted transmission function in the data security level adaptation requirements. Through these two aspects of calculation, the adaptation association strength is determined, and an adaptation association strength distribution is formed. This adaptation association strength distribution reflects the adaptation association strength of different combinations of key feature items and key attribute items.

[0056] Step S116: Based on the distribution of adaptation association strength, select the combination of key characteristic items and key attribute items that meet the requirements of adaptation association strength, and use this combination as the core element of the adaptation benchmark for cross-domain data migration. The implementation of the core element of the adaptation benchmark directly determines the basic adaptation direction of cross-domain data migration.

[0057] After generating the adaptation association strength distribution in step S115, step S116 is executed. A threshold for the adaptation association strength is set, and combinations of key characteristic items and key attribute items that meet the strength requirements are selected. For example, combinations such as the read / write speed characteristics of the storage medium and the data access frequency characteristics, and the compatibility characteristics of the data storage format requirements and the data processing flow have high adaptation association strength and meet the requirements. These combinations are used as the core elements of the adaptation benchmark. These core elements directly determine the basic adaptation direction for cross-domain data migration, such as prioritizing the adaptation between the storage medium and the data access frequency, and the compatibility between the data processing flow and the data storage format.

[0058] Step S117: Supplement the auxiliary elements corresponding to the core elements of the adaptation benchmark. The auxiliary elements include the network transmission characteristics, storage resource occupancy status, network latency characteristics, bandwidth fluctuation characteristics, and load balancing characteristics of storage nodes in different cloud storage domains. The auxiliary elements and the core elements together constitute a complete adaptation evaluation index system.

[0059] Next, proceed to step S117. Supplement auxiliary elements to the core elements of the adaptation benchmark. Network transmission characteristics, including network topology and transmission protocols, affect the stability and efficiency of data transmission; storage resource occupancy status reflects the current storage space usage of the target storage domain. If the storage resource occupancy rate is too high, it may affect data storage and subsequent migration; network latency characteristics increase data transmission time, which has a significant impact on data migration with high real-time requirements; bandwidth fluctuation characteristics may lead to unstable data transmission rates, affecting the migration progress; the load balancing characteristics of storage nodes are related to the fairness and efficiency of resource scheduling, avoiding excessive load on some nodes that may affect the migration task. The above auxiliary elements, together with the core elements, comprehensively evaluate the adaptation of cross-domain data migration, forming a complete adaptation evaluation index system.

[0060] Step S118: Integrate the core elements and auxiliary elements of the adaptation benchmark to form an adaptation benchmark framework that includes element relationships, element adaptation standards, and element adjustment thresholds. The element relationships clarify the interaction between core elements and auxiliary elements, the element adaptation standards define the adaptation boundaries of each element, and the element adjustment thresholds specify the dynamic adjustment range of elements.

[0061] Then proceed to step S118. Integrate the core and auxiliary elements of the adaptation benchmark, clarifying the relationships between them. For example, the transmission interface specifications in the core elements interact with the network transmission characteristics in the auxiliary elements; the protocol of the transmission interface needs to match the network transmission protocol. The element adaptation standard specifies the adaptation boundaries of each element, such as the read / write rate of the storage medium needing to reach a certain range to meet the data access frequency requirements. The element adjustment threshold specifies the adjustment range of each element during dynamic changes; when network latency exceeds a certain threshold, the data transmission strategy or timing needs to be adjusted. This integration forms the adaptation benchmark framework.

[0062] Step S119: The adaptation benchmark framework is warmed up and strengthened through the transfer learning model. The transfer learning model uses some cross-domain data migration samples to calibrate the adaptation direction of the adaptation benchmark framework. The element weights in the adaptation benchmark framework are adjusted based on the sample migration effect feedback. The element adaptation standards and element adjustment thresholds in the adaptation benchmark framework are adjusted according to the warm-up and strengthening results. The adaptation details and adjustment range of each element are refined to form the final cloud storage cross-domain data migration adaptation benchmark.

[0063] Finally, step S119 is executed. The transfer learning model utilizes some historical cross-domain data migration samples, such as migration cases with similar architectures and data attributes to the current cloud storage domain A and cloud storage domain B, to warm up and strengthen the adaptation benchmark framework. Based on the migration effect feedback of the samples, the importance of each element in the migration process is analyzed, and the weights of core and auxiliary elements in the adaptation benchmark framework are adjusted. For example, if a protocol adaptation problem of the transmission interface causes migration failure in a certain sample, then the weight of the protocol adaptation feature of the transmission interface as a core element is increased in the adaptation benchmark framework. At the same time, based on the results of the warm-up and strengthening, the element adaptation standards and element adjustment thresholds are further adjusted, and the adaptation details of each element are refined, such as more accurately defining the adaptation range of storage medium read and write rates, adjusting the adjustment threshold of network latency, etc., ultimately forming the cloud storage cross-domain data migration adaptation benchmark.

[0064] Step S120: Extract cross-domain transfer knowledge from the transfer learning model. The cross-domain transfer knowledge is separated from the pre-trained weights of the transfer learning model. The cross-domain transfer knowledge includes data migration adaptation experience and data storage compatibility rules between different cloud storage domains.

[0065] After obtaining the cross-domain data migration adaptation benchmark for cloud storage, step S120 is executed. The purpose of this step is to extract cross-domain migration knowledge from the transfer learning model. This knowledge is an important basis for subsequently building cross-domain data flow channels and making migration adaptation predictions. It includes data migration adaptation experience and data storage compatibility rules between different cloud storage domains, which can guide the specific implementation of cross-domain data migration.

[0066] Step S121: Obtain the pre-trained weight set of the transfer learning model. The pre-trained weight set contains the weight parameters obtained by the transfer learning model under the data migration scenarios of different cloud storage domains. It covers the weight information of multiple network layers, including the input layer, feature extraction layer, association analysis layer, decision layer and output layer.

[0067] First, step S121 is executed. During the training process, the transfer learning model is pre-trained for different cloud storage domain data migration scenarios. These scenarios include cross-domain migration with different storage architectures and data attributes. The pre-trained weight set is the set of weight parameters obtained under these scenarios, covering multiple network layers such as the input layer, feature extraction layer, association analysis layer, decision layer, and output layer. The weight parameters of the input layer affect the initial processing of the storage architecture characteristics and inherent attribute information of the input data; the weight parameters of the feature extraction layer determine the ability to extract key features from the input information; the weight parameters of the association analysis layer are used to establish the association relationship between features; the weight parameters of the decision layer affect the final transfer adaptation decision; and the weight parameters of the output layer are related to the output of the transfer adaptation result.

[0068] Step S122: Perform weight hierarchical analysis on the pre-trained weight set. According to the network hierarchy of the transfer learning model, the pre-trained weight set is divided into input layer weights, feature extraction layer weights, association analysis layer weights, decision layer weights, and output layer weights. The weights of each layer are extracted separately and labeled with the corresponding network hierarchy identifier.

[0069] Next, proceed to step S122. Following the network hierarchy of the transfer learning model, the pre-trained weight set is analyzed hierarchically. Weight parameters belonging to the input layer are extracted and labeled as input layer weights; similarly, weight parameters for the feature extraction layer, association analysis layer, decision layer, and output layer are extracted and labeled with their corresponding hierarchical identifiers. This hierarchical processing makes the source and function of the weight parameters clearer, facilitating the subsequent identification of weight components related to cross-domain data transfer.

[0070] Step S123: Identify the weight components related to cross-domain data migration in each level of weight, and retain the weight components that have a consistent weight change trend under different scenarios and have a significant impact on the migration effect. The relevant weight components are determined by comparing the weight differences under different cloud storage domain pre-training scenarios.

[0071] Then proceed to step S123. Compare the differences in weights at each level under different cloud storage domain pre-training scenarios. For example, in the feature extraction layer, for the weight parameter of the storage medium type feature, in different pre-training scenarios, when the storage medium type is a solid-state drive and the data storage requirement is high read / write performance, the value of this weight parameter is higher, and the above trend is consistent in multiple scenarios. At the same time, the change of this weight parameter has a significant impact on the migration effect (such as data read / write speed and migration success rate), so it is identified as a weight component related to cross-domain data migration. Weight components with inconsistent weight change trends in different scenarios or with little impact on the migration effect are excluded.

[0072] Step S124: Aggregate the identified relevant weight components to form a cross-domain migration weight cluster. The cross-domain migration weight cluster contains all weight information related to cross-domain data migration adaptation and is classified and organized according to the corresponding network level and functional module.

[0073] After identifying the relevant weight components in step S123, step S124 is executed. The weight components are then categorized and organized according to their network layer (e.g., input layer, feature extraction layer) and functional module (e.g., feature extraction module, association analysis module). For example, weight components related to storage medium characteristics in the feature extraction layer are grouped together, and weight components related to storage architecture characteristics and data attributes in the association analysis layer are grouped together to form cross-domain migration weight clusters. This organization method makes the weight information more organized, facilitating subsequent knowledge decoding.

[0074] Step S125: Perform knowledge decoding on the cross-domain transfer weight cluster. Knowledge decoding transforms the weight information into interpretable cross-domain data transfer adaptation experience through the weight inversion structure of the transfer learning model. The weight inversion structure realizes the semantic transformation of weight information through the correspondence between weight parameters and transfer scenario features.

[0075] Next, step S125 is executed. The weight inversion structure of the transfer learning model stores the correspondence between weight parameters and migration scenario features. For example, a certain weight parameter corresponds to a migration scenario feature where the storage medium type is a solid-state drive and the data storage requirement is high read / write performance. Through the above correspondence, the weight information in the cross-domain migration weight cluster is semantically transformed, converting abstract weight parameters into specific, interpretable cross-domain data migration adaptation experience. For example, a high value of a certain weight component corresponds to the adaptation experience of "when migrating structured data with high access frequency, prioritize solid-state drive storage media with high read / write speed".

[0076] Step S1251: Construct the weight inversion structure of the transfer learning model. The weight inversion structure includes a weight parsing unit, a feature mapping unit, and an experience extraction unit. The weight parsing unit is responsible for the hierarchical parsing of weight parameters. The feature mapping unit realizes the mapping between weight parameters and transfer features. The experience extraction unit completes the extraction and summarization of adaptation experience.

[0077] First, proceed to step S1251. The constructed weight inversion structure comprises three key units. The weight parsing unit performs hierarchical parsing of the weight parameters in the cross-domain migration weight cluster, determining the network layer and functional module to which each weight parameter belongs, as well as its scope of influence during the migration process. The feature mapping unit maps the parsed weight parameters to specific migration scenario features, such as storage architecture characteristics and data attribute features, based on a pre-defined correspondence table between weight parameters and migration scenario features. The experience extraction unit analyzes and summarizes the mapped migration scenario features, extracting the cross-domain data migration adaptation experience contained within, such as adaptation rules for different storage media and data types.

[0078] Step S1252: Input the cross-domain transfer weight cluster into the weight parsing unit. The weight parsing unit performs hierarchical parsing of the weight parameters in the cross-domain transfer weight cluster, determines the network level, functional module and feature channel corresponding to each weight parameter, and marks the functional positioning and scope of each weight parameter.

[0079] Then, step S1252 is executed. The cross-domain migration weight cluster is input into the weight parsing unit, which performs detailed hierarchical parsing of the weight parameters. For example, a certain weight parameter is parsed as belonging to the storage medium feature extraction module of the feature extraction layer, and the corresponding data feature channel is the read / write rate feature of the storage medium. Its function is to measure the impact of the storage medium's read / write rate on data migration adaptation, and its scope covers the storage medium selection stage of data migration. Through the above parsing, the specific information of each weight parameter is clarified.

[0080] Step S1253: The parsed weight parameters are mapped to the corresponding cross-domain data migration feature space through the feature mapping unit. The feature mapping establishes the mapping relationship between the weight parameters and the migration features based on the degree of influence of the weight parameters on different migration features, and determines the migration feature type corresponding to each weight parameter.

[0081] Next, step S1253 is executed. The feature mapping unit maps the weight parameters according to their influence on different migration features. For example, if a certain weight parameter has a high influence on the storage medium read / write rate feature after parsing, then this weight parameter is mapped to the storage medium read / write rate feature type in the cross-domain data migration feature space. In this way, a clear mapping relationship between the weight parameters and migration features is established.

[0082] Step S1254: Identify the cross-domain data migration scenario features corresponding to the weight parameters. The scenario features include the architectural features, data attribute features, migration environment features, and migration target features of different cloud storage domains. The identification of scenario features is based on the mapping features of the weight parameters and pre-trained scenario information.

[0083] Based on step S1253, step S1254 is performed. Combining the mapping characteristics of the weight parameters and the pre-training scenario information, the corresponding cross-domain data migration scenario characteristics are identified. For example, the weight parameter mapped to the read / write rate characteristic type of the storage medium, combined with the storage architecture information of the cloud storage domain in the pre-training scenario, can be identified as "the storage medium of cloud storage domain A is a solid-state drive, and the data to be migrated is high-frequency structured data".

[0084] Step S1255: Analyze the variation pattern of weight parameters under different migration scenario features, sort out the fluctuation trend of weight parameters with scene features, summarize the correspondence between weight parameters and migration scenario features, and establish a weight parameter-scene feature correspondence model.

[0085] Then, step S1255 is executed. The changes in the value of this weight parameter under different migration scenario characteristics are analyzed. For example, the value is higher in scenarios where the storage medium is a solid-state drive and the data is high-frequency structured data, while the value is lower in scenarios where the storage medium is a hard disk drive and the data is low-frequency unstructured data. These fluctuation trends are then identified. Furthermore, the correspondence between the weight parameter and migration scenario characteristics is summarized: when the scenario characteristics are data with high read / write requirements combined with high read / write speed storage media, the weight parameter value is higher, and vice versa. Based on this, a weight parameter-scenario characteristic correspondence model is established.

[0086] Step S1256: Based on the correspondence between weight parameters and migration scenario features, extract data migration adaptation criteria under different migration scenarios. The data migration adaptation criteria include storage node selection criteria, data transmission adaptation criteria, storage format adaptation criteria, and resource scheduling adaptation criteria. Define specific condition judgment logic and execution parameter thresholds for each data migration adaptation criterion.

[0087] Execute step S1256. Based on the established weight parameter-scenario feature correspondence model, extract data migration adaptation criteria. For example, from the correspondence between the weight parameters and scenario features, extract the storage node selection criterion: "When the data to be migrated is high-frequency structured data, select a storage node with a solid-state drive as the storage medium," and define the condition judgment logic of this criterion as "determine whether the data access frequency is higher than a certain threshold and the data structure type is structured," with the execution parameter threshold being the specific threshold range of the access frequency.

[0088] Step S1257: Extract the points of convergence and difference between the data migration adaptation criteria and the actual operations in real cross-domain data migration cases, optimize the description and execution details of the data migration adaptation criteria based on case experience, and supplement each of the data migration adaptation criteria with specific execution parameters and judgment conditions.

[0089] Next, proceed to step S1257. Referring to actual cross-domain data migration cases, such as the previous case of migrating similar data between cloud storage domain A and cloud storage domain B, extract the points of convergence (e.g., the recommended storage node selection matches the actual operation) and the points of difference (e.g., the actual operation also considered the load of the storage nodes). Based on case experience, optimize the wording and execution details of the adaptation criteria, adding the execution parameter and judgment condition "simultaneously consider the current load rate of the storage node, which must be below a certain threshold" to the storage node selection criteria.

[0090] Step S1258: Summarize and integrate the verified data migration adaptation criteria, and divide them into adaptation criteria for the migration preparation stage, adaptation criteria for the migration execution stage, adaptation criteria for the storage adaptation stage, and adaptation criteria for the subsequent management and control stage according to the process stages of cross-domain data migration, forming a hierarchical cross-domain data migration adaptation experience.

[0091] Finally, step S1258 is executed. The validated and optimized data migration adaptation criteria are summarized and integrated according to the stages of the cross-domain data migration process. Adaptation criteria for the migration preparation stage include storage node selection and data format checking; adaptation criteria for the migration execution stage include data transmission protocol selection and transmission rate control; adaptation criteria for the storage adaptation stage include storage format conversion and storage resource allocation; and adaptation criteria for the subsequent management and control stage include data access permission configuration and storage status monitoring. Through the above summarization and integration, a hierarchical cross-domain data migration adaptation experience is formed.

[0092] Step S126: Extract data storage compatibility-related sample features from the pre-trained sample set of the transfer learning model. The sample features include the compatibility range of different cloud storage domains for data formats, data storage adaptation conditions, data access permission adaptation requirements, and data transmission encryption adaptation specifications.

[0093] While performing knowledge decoding in step S125, step S126 is executed. The pre-training sample set of the transfer learning model contains a large number of sample cases of data migration from different cloud storage domains. Sample features related to data storage compatibility are extracted from these sample sets, such as the compatibility range of cloud storage domain A for XML format data (supporting all versions of XML format), the adaptation condition of cloud storage domain B for data storage (data must undergo specific sharding processing), the adaptation requirements of different cloud storage domains for data access permissions (such as the need for specific user role permissions), and the encryption adaptation specifications for data transmission (such as the use of AES encryption algorithm), etc.

[0094] Step S127: Summarize data storage compatibility rules based on the extracted sample features. The data storage compatibility rules include data format conversion specifications, storage resource adaptation requirements, access permission adaptation specifications, transmission encryption adaptation standards, and data integrity assurance specifications. Each rule corresponds to an applicable cloud storage domain scenario and data type range.

[0095] Then, step S127 is executed. The extracted sample features are summarized to form data storage compatibility rules. For example, based on the sample features of the compatibility range of data formats for different cloud storage domains, data format conversion specifications are summarized: "When migrating XML format data from cloud storage domain A to cloud storage domain B, if cloud storage domain B does not support the XML version, it needs to be converted to a version supported by cloud storage domain B"; based on the sample features of data storage adaptation conditions, storage resource adaptation requirements are summarized: "For unstructured data, cloud storage domain B requires that the remaining storage space of the storage node be greater than 1.5 times the data size," etc. Each rule clearly defines the applicable cloud storage domain scenario (such as migration from cloud storage domain A to cloud storage domain B) and the data type range (such as unstructured data).

[0096] Step S128: Integrate cross-domain data migration adaptation experience and data storage compatibility rules to form an initial set of cross-domain migration knowledge. The initial set is divided into adaptation experience knowledge and compatibility rule knowledge according to knowledge type. Each type of knowledge is labeled with its corresponding applicable scenario and priority.

[0097] Execute step S128. Integrate the cross-domain data migration adaptation experience obtained in step S125 and the data storage compatibility rules summarized in step S127 to form an initial set of cross-domain migration knowledge. Divide the knowledge into adaptation experience knowledge (such as storage node selection experience, data transmission adaptation experience, etc.) and compatibility rule knowledge (such as data format conversion rules, access permission adaptation rules, etc.). Label each type of knowledge with applicable scenarios, such as "storage node selection experience is suitable for migration scenarios of high-frequency structured data," and label their priority, such as data security-related compatibility rules having higher priority than storage resource adaptation rules.

[0098] Step S129: Redundancy is removed from the initial set of cross-domain migration knowledge. After reorganizing the hierarchical structure of the knowledge set according to knowledge type and application scenario, the cross-domain migration knowledge after redundancy removal is reorganized according to the process steps of cross-domain data migration to form a cross-domain migration knowledge system covering the entire process of migration preparation, migration execution, storage adaptation and subsequent management, thus obtaining the final cross-domain migration knowledge.

[0099] Finally, step S129 is executed. The initial set of cross-domain migration knowledge is checked, and redundant knowledge content is removed, such as duplicate adaptation experience or compatibility rules for the same scenario, retaining only the most comprehensive and accurate information. Then, the hierarchical structure is reorganized according to knowledge type and application scenario, and further organized according to the cross-domain data migration process stages (migration preparation, migration execution, storage adaptation, and post-migration management). For example, the migration preparation stage includes experience in storage node selection and data format checking rules; the migration execution stage includes experience in data transmission adaptation and transmission encryption rules, forming a comprehensive cross-domain migration knowledge system covering the entire process, resulting in the final cross-domain migration knowledge.

[0100] Step S130: Construct a cross-domain data transfer channel by combining cloud storage cross-domain data migration adaptation benchmarks and cross-domain migration knowledge. The cross-domain data transfer channel includes node association relationships and data transfer control logic for data transmission between different cloud storage domains.

[0101] After extracting the cross-domain migration knowledge, step S130 is executed. This step aims to utilize the obtained cloud storage cross-domain data migration adaptation benchmark and cross-domain migration knowledge to build a cross-domain data flow channel, providing specific paths and control logic for data transmission between different cloud storage domains, and ensuring that data can be migrated safely and efficiently in the expected manner.

[0102] Step S131: Analyze the adaptation benchmark elements in the cloud storage cross-domain data migration adaptation benchmark, extract the storage node characteristics and data adaptation requirements from the adaptation benchmark elements, form the core information of the adaptation benchmark, and classify and organize the core information of the adaptation benchmark according to the storage node type and data type.

[0103] First, execute step S131. Analyze the cloud storage cross-domain data migration adaptation benchmark and extract its elements. For example, storage node characteristics include the transmission interface characteristics of storage nodes in cloud storage domain A and cloud storage domain B (such as the transmission rate and protocol type of Ethernet and Fibre Channel interfaces), the read / write speed characteristics of the storage medium, and the load balancing characteristics of the storage nodes; data adaptation requirements include storage format requirements, security level requirements, and access frequency adaptation requirements for different types of data (structured, semi-structured, and unstructured). Classify and organize the above storage node characteristics and data adaptation requirements according to storage node type (such as storage nodes in cloud storage domain A and storage nodes in cloud storage domain B) and data type to form the core information of the adaptation benchmark.

[0104] Step S132: Analyze the cross-domain data migration adaptation experience and data storage compatibility rules in the cross-domain migration knowledge, extract the node adaptation experience and flow control rules related to data flow, form the core information of cross-domain migration knowledge, and mark the corresponding adaptation priority and applicable conditions of the core information of cross-domain migration knowledge.

[0105] Next, proceed to step S132. Extract cross-domain data migration adaptation experience and data storage compatibility rules from the cross-domain migration knowledge, and extract content related to data flow. Node adaptation experience includes rules such as "high-frequency access structured data should be adapted to storage nodes with high read / write speeds," and flow control rules include rules such as "data transmission must be performed in descending order of security level." Assign adaptation priorities to the above node adaptation experience and flow control rules, with security-related rules having the highest priority and access frequency-related experience having the next highest priority; also, indicate applicable conditions, such as the condition "high-frequency access structured data" being applicable to specific node adaptation experiences.

[0106] Step S133: Associate and merge the core information of the adaptation benchmark and the core information of cross-domain migration knowledge to establish the correspondence between storage node characteristics and node adaptation experience, as well as the correspondence between data adaptation requirements and flow control rules, forming a fusion association table. The fusion association table is used to reflect the mapping logic and function of different information.

[0107] Then, proceed to step S133. Associate the storage node characteristics in the adaptation benchmark core information with the node adaptation experience in the cross-domain migration knowledge core information. For example, establish a correspondence between the high read / write speed storage node characteristics in cloud storage domain A and the node adaptation experience that "high-frequency access structured data should be adapted to high read / write speed storage nodes." Establish a correspondence between data adaptation requirements (such as the security level requirements for structured data) and flow control rules (such as "data with high security levels is transmitted first and encrypted"). Through the above association and fusion, a fusion association table is formed. This fusion association table reflects how storage node characteristics correspond to node adaptation experience, how data adaptation requirements correspond to flow control rules, and the mapping logic and operational methods between them.

[0108] Step S134: Based on the established fusion association table, select cloud storage domain storage nodes that meet the requirements. The selection criteria include the compatibility between storage node characteristics and node adaptation experience, the fit between data adaptation requirements and flow control rules, and the matching degree between storage node transmission capabilities and data transmission needs.

[0109] Execute step S134. Based on the correspondence in the fusion association table, formulate screening criteria. The compatibility between storage node characteristics and node adaptation experience is evaluated by comparing whether the actual characteristics of the storage node (e.g., read / write speed) match the characteristics required by node adaptation experience; the fit between data adaptation requirements and flow control rules is evaluated to determine whether the data adaptation requirements (e.g., security level) meet the conditions of the flow control rules; the matching degree between the storage node's transmission capabilities (e.g., transmission rate, bandwidth) and data transmission requirements (e.g., data volume, transmission time requirements) is considered, taking into account whether the storage node can complete the data transmission within the specified time. Based on these screening criteria, select suitable storage nodes from cloud storage domain A and cloud storage domain B, such as storage nodes in cloud storage domain A that meet high read / write speeds and low load rates, and storage nodes in cloud storage domain B that support specific data formats and have matching transmission interface protocols.

[0110] Step S135: Construct the association network of the filtered storage nodes. The association network includes the direct and indirect connection relationships between storage nodes, as well as the transmission characteristics corresponding to the connection relationships. The transmission characteristics include transmission rate, transmission delay, bandwidth usage, and transmission stability. The association network presents the connection status between nodes in the form of a topology.

[0111] After selecting the qualified storage nodes in step S134, proceed to step S135. Determine the connection relationships between these storage nodes, including direct connections (e.g., a storage node in cloud storage domain A is directly connected to a storage node in cloud storage domain B via a dedicated network line) and indirect connections (e.g., connections via intermediate network devices or other storage nodes). Simultaneously, record the transmission characteristics corresponding to the connection relationships. For example, direct connections have higher transmission rates, lower latency, stable bandwidth usage, and better transmission stability; indirect connections have relatively lower transmission rates, higher latency, bandwidth usage may be affected by intermediate nodes, and slightly lower transmission stability. Present the connection status between these storage nodes in a topology (e.g., star, mesh) to construct an association network.

[0112] Step S136: Construct data flow control logic based on the relationship network. The data flow control logic includes a node selection strategy for data transmission, a sequence control strategy for data transmission, and an exception handling strategy for data transmission. The node selection strategy clarifies the node selection logic under different data types and transmission scenarios. The sequence control strategy specifies the transmission order of different batches of data. The exception handling strategy defines the response methods and node switching mechanism when transmission anomalies occur.

[0113] Then, step S136 is executed. Based on the constructed relationship network, data flow control logic is built to ensure that data is transmitted in an orderly and stable manner between storage nodes.

[0114] Step S1361: Analyze the transmission and load characteristics of each storage node in the interconnected network. Transmission characteristics include transmission delay, transmission bandwidth, transmission stability, and transmission protocol compatibility. Load characteristics include current load status, load capacity limit, load growth trend, and load balancing capability. The analysis results are classified and recorded according to the storage node type.

[0115] First, proceed to step S1361. For each storage node in the interconnected network, analyze its transmission and load characteristics. Transmission latency characteristics include average transmission latency and latency fluctuation range; transmission bandwidth characteristics include maximum available bandwidth and currently used bandwidth; transmission stability characteristics are measured by the number of interruptions and data packet loss rate during historical transmission; transmission protocol compatibility characteristics refer to the types and versions of transmission protocols supported by the storage node. Regarding load characteristics, the current load status is reflected by indicators such as CPU utilization, memory utilization, and storage space utilization; the load capacity limit refers to the maximum load a storage node can withstand without significantly affecting performance; the load growth trend predicts future load changes based on recent load changes; load balancing capability refers to the storage node's ability to achieve load balancing through task scheduling and other methods when the load is high. Classify and record the above analysis results according to the storage node type (e.g., whether it belongs to cloud storage domain A or cloud storage domain B).

[0116] Step S1362: Construct a node selection strategy for data transmission based on the transmission and load characteristics of storage nodes. The node selection strategy includes node priority sorting rules and node dynamic switching rules. The node priority sorting rules determine the priority order of different nodes according to the transmission and load characteristics. The node dynamic switching rules clarify the switching conditions and switching methods when the node transmission capacity is insufficient or the load is too high.

[0117] Next, proceed to step S1362. Based on the transmission and load characteristics of each storage node, a node selection strategy is constructed. The node priority ranking rule comprehensively considers factors such as transmission latency, bandwidth, stability, and load status, assigning a priority score to each storage node. Nodes with low transmission latency, high bandwidth, good stability, and low load have higher priority scores, and the node priority order is determined from highest to lowest priority score. For example, in cloud storage domain A, a storage node with low transmission latency and current low load has a higher priority than a node with high transmission latency and high load. The node dynamic switching rule stipulates that node switching is triggered when the currently used storage node experiences insufficient transmission capacity (e.g., the actual transmission rate is lower than the data transmission requirement threshold) or excessive load (e.g., CPU utilization exceeds the load capacity limit threshold). Switching conditions include a transmission rate continuously below the threshold for a certain period of time and load indicators exceeding the limit; the switching method can be automatic switching to the second highest priority available node, or switching based on a preset list of backup nodes.

[0118] Step S1363: Extract the type and transmission requirements of cross-domain data. The transmission requirements of different types of data include transmission timeliness requirements, transmission reliability requirements, transmission security requirements, and transmission bandwidth requirements. Classify and archive cross-domain data according to the priority of data type and transmission requirements.

[0119] Then proceed to step S1363. Clearly define the types of data to be migrated across domains (structured, semi-structured, unstructured) and their transmission requirements. Transmission timeliness requirements refer to the need for data to be transmitted within a specified timeframe, such as order data needing to be transmitted within one hour. Transmission reliability requirements require minimizing data loss and corruption during transmission, such as customer information data requiring extremely high reliability. Transmission security requirements include encryption and access control requirements during data transmission, such as end-to-end encrypted transmission of privacy data. Transmission bandwidth requirements refer to the bandwidth requirements for data transmission, such as large unstructured data files requiring high bandwidth support. Classify and archive the cross-domain data according to data type and transmission requirement priority (e.g., security requirements have the highest priority, followed by timeliness requirements). For example, group high-security-level structured customer information data into one category and high-time-sensitivity-requirement semi-structured order data into another.

[0120] Step S1364: Construct a data transmission sequence control strategy based on the type of cross-domain data and transmission requirements. The sequence control strategy includes data transmission ordering rules and data fragment transmission order coordination rules. The ordering rules determine the transmission order based on the importance of the data and the timeliness requirements of transmission. The fragment transmission order coordination rules regulate the transmission order and splicing logic of different fragments of the same data.

[0121] Execute step S1364. Based on the data type and transmission requirements, construct a sequence control strategy. In the sequencing rules, data with high importance (e.g., security level) and strong timeliness requirements are transmitted first. For example, high-security customer information data is transmitted before unstructured data such as product design drawings, and order data with high timeliness requirements is transmitted before weekly updated product design drawings. For large data files that need to be transmitted in fragments (such as unstructured product design drawings), the fragment transmission sequence coordination rules specify the transmission order of fragments (e.g., transmission according to fragment number) and the splicing logic (e.g., the receiving end splices according to fragment number; if a fragment is lost, it requests retransmission of that fragment before continuing splicing).

[0122] Step S1365: Identify possible transmission anomaly scenarios in the network of relationships, perform detailed feature description and cause analysis for each transmission anomaly scenario, analyze the causes and impact range of different transmission anomaly scenarios, establish the correspondence between anomaly scenarios and impact range, and determine the degree of impact of different anomaly scenarios on data transmission progress, data integrity and transmission security.

[0123] Next, proceed to step S1365. Based on the structure of the interconnected network and the characteristics of the storage nodes, identify potential transmission anomaly scenarios, such as storage node connection interruptions, sudden drops in transmission rates, data transmission errors (checksum mismatches), and slow response due to sudden increases in node load. Describe the characteristics of each anomaly scenario; for example, a connection interruption manifests as a broken network connection, preventing data transmission and reception; a sudden drop in transmission rates manifests as an actual transmission rate significantly lower than normal. Cause analysis includes network failures, storage node hardware failures, software configuration errors, and network congestion. Analyze the impact range of each anomaly scenario; for example, a single storage node connection interruption only affects data transmission related to that node, while network congestion may affect data transmission across multiple nodes. Establish a correspondence between anomaly scenarios and their impact ranges, and assess the degree of impact on data transmission progress (e.g., increased transmission latency, task stagnation), data integrity (e.g., data loss, corruption), and transmission security (e.g., unauthorized access to data under abnormal conditions).

[0124] Step S1366: Construct an anomaly handling strategy for data transmission based on the abnormal scenario and scope of impact. The anomaly handling strategy includes anomaly detection rules, anomaly response mechanism, and anomaly recovery process. The anomaly detection rules clarify the identification indicators and detection methods of anomalies. The anomaly response mechanism defines the immediate response measures after an anomaly occurs. The anomaly recovery process specifies the steps and methods for restoring the normal state of data transmission.

[0125] Finally, step S1366 is executed. An anomaly handling strategy is constructed based on the identified transmission anomaly scenarios and their impact scope. Anomaly detection rules define the indicators and detection methods for various anomalies, such as periodically sending heartbeat packets to check the connection status of storage nodes; if a heartbeat packet is not received within a timeout period, it is considered a connection interruption; and real-time monitoring of the transmission rate, where a rate below a set threshold for a sustained period is considered a sudden drop in transmission rate. Anomaly response mechanisms include immediate measures such as stopping the current data transmission task, activating a backup transmission channel, and sending alarm information to the administrator. The anomaly recovery process standardizes the steps for restoring normal data transmission. For example, for connection interruption anomalies, the recovery process includes attempting to re-establish the connection; if this fails, switching to a backup node, re-initializing the transmission task, and continuing data transmission from the point of interruption; for data transmission errors, the recovery process includes requesting retransmission of erroneous data fragments and re-verifying the data.

[0126] Step S1367: Integrate the node selection strategy, sequence control strategy, and exception handling strategy; analyze the logical relationships and coordination between different strategies; when a strategy conflict is detected, resolve the conflict according to the strategy priority and execution sequence to form a preliminary framework for data flow control logic; strengthen the correlation of control strategies in the preliminary framework; determine the triggering conditions and execution priorities of different control strategies; construct the connection process and information interaction method between control strategies; and establish a strategy execution state machine to manage the activation, execution, and transition states of different strategies.

[0127] After completing steps S1361 to S1366, proceed to step S1367. Integrate the node selection strategy, sequence control strategy, and exception handling strategy. Analyze their logical relationships; for example, the node selection strategy provides transmission nodes for the sequence control strategy, the sequence control strategy determines the data transmission order on the selected nodes, and the exception handling strategy ensures the continuity of data flow when an exception occurs at a node or during transmission. The collaborative mechanism is as follows: when the exception handling strategy triggers a node switch, the node selection strategy reselects a node, and the sequence control strategy adjusts the data transmission order according to the new node. If a strategy conflict is detected, such as the sequence control strategy requiring priority transmission of certain data, but the optimal node corresponding to that data in the node selection strategy is currently unavailable, the conflict is resolved according to strategy priority (e.g., the exception handling strategy has higher priority than the sequence control strategy, and the sequence control strategy has higher priority than the node selection strategy) and execution timing. After establishing a preliminary framework for data flow control logic, the control strategies are reinforced for correlation. The triggering conditions (e.g., the node selection strategy triggers at the start of the data transmission task, and the exception handling strategy triggers when an exception is detected) and execution priorities of each strategy are determined. The connection process between strategies is constructed (e.g., after exception handling is completed, the sequential control strategy is notified to continue data transmission) and information exchange methods (e.g., the node selection strategy transmits the selected node information to the sequential control strategy). A strategy execution state machine is established to manage strategy activation (e.g., activating the strategy when the triggering conditions are met), execution (the strategy runs normally), and state transitions (e.g., after the exception handling strategy completes execution, the sequential control strategy transitions to continue execution).

[0128] Step S1368: Optimize the data flow control logic by simulating cross-domain data flow scenarios, adjust the rule parameters and execution flow in the control strategy, and establish a correspondence table between control logic parameters and network characteristic parameters based on the simulation results.

[0129] Finally, execute step S1368. Simulate cross-domain data flow scenarios, such as migrating a large amount of frequently accessed structured data from cloud storage domain A to cloud storage domain B. Run the data flow control logic in the simulation environment. Based on the problems encountered during the simulation (such as excessive transmission delay due to untimely node switching, and bandwidth waste due to unreasonable sequence control), adjust the rule parameters in the control strategy (such as the delay threshold in the node dynamic switching rule, and the priority score calculation weight in the sequence control strategy) and the execution process (such as optimizing the frequency of anomaly detection and adjusting the connection order between strategies). Based on the results of multiple simulations, analyze the optimal values ​​of the control logic parameters under different network characteristic parameters (such as network bandwidth, latency, and stability), and establish a correspondence table between the control logic parameters and network characteristic parameters. This allows for rapid adjustment of the control logic parameters according to the current network characteristics in practical applications, thereby optimizing the data flow effect.

[0130] Step S137: Integrate the node selection strategy, sequence control strategy, and exception handling strategy to form a flow control rule set. The flow control rule set clarifies the execution process, triggering conditions, and mutual cooperation methods of each strategy, and establishes a strategy execution state machine to manage the transition and coordination between strategies.

[0131] After constructing the data flow control logic in step S136, step S137 is executed. The node selection strategy, sequence control strategy, and exception handling strategy are further integrated to form a set of flow control rules. This set of flow control rules details the execution flow of each strategy (such as the specific steps of node selection, the execution stage of sequence control, and the operation flow of exception handling), triggering conditions (such as the node selection strategy being triggered at the start of the data transmission task, and the exception handling strategy being triggered when an exception is detected), and their coordination methods (such as how to notify the sequence control strategy to continue execution after the exception handling strategy has finished executing). By establishing a strategy execution state machine, the transitions and coordination between strategies are managed, such as transitioning from the node selection state to the sequence control state, transitioning from the sequence control state to the exception handling state when an exception occurs, and transitioning back to the sequence control state or node selection state after exception handling is completed.

[0132] Step S138: Integrate the relationship network and the set of flow control rules to form a preliminary architecture for cross-domain data flow channels. The preliminary architecture includes a node relationship topology and a set of flow control rules. The node relationship topology visually presents the connection relationship of storage nodes, and the set of flow control rules provides behavioral norms for data flow.

[0133] Then, step S138 is executed. The constructed relationship network (node ​​association topology) and the set of flow control rules are integrated. The node association topology graphically presents the connection relationships between storage nodes, including direct and indirect connections, as well as the transmission characteristics of the connections; the set of flow control rules provides behavioral specifications for the flow of data between these nodes, clarifying rules such as node selection, transmission order, and exception handling. The combination of the two forms the preliminary architecture of the cross-domain data flow channel. This preliminary architecture describes the storage node path that data passes through from the source cloud storage domain to the target cloud storage domain, as well as the flow control methods along the path.

[0134] Step S139: Adjust the node association relationships and flow control logic in the preliminary architecture based on the adaptation experience in cross-domain migration knowledge, optimize the node connection path based on the successful path patterns recorded in the adaptation experience, and refine the specific parameter settings of the control rules according to the compatibility rules.

[0135] Next, proceed to step S139. Referring to adaptation experience in cross-domain migration, such as the node connection path patterns used in previous successful cross-domain data migration cases, adjust the node relationships in the initial architecture. For example, if adaptation experience records that "indirect connections through intermediate forwarding nodes can improve transmission stability," then optimize the node connection path and add suitable intermediate forwarding nodes. Simultaneously, based on data storage compatibility rules, refine the specific parameter settings of the flow control rules. For instance, according to data format conversion specifications, set specific parameters for data format conversion in the flow control rules (such as the converted format version and encoding method).

[0136] Step S1310: Supplement the missing node association links and flow control details in the preliminary architecture, define specific input / output interfaces and execution logic for each functional module, and form a cross-domain data flow channel that includes the node association system and the flow control system.

[0137] Finally, execute step S1310. Check the preliminary architecture and supplement any missing node association links, such as necessary backup connections between storage nodes; supplement the flow control details, such as the verification frequency and logging requirements for different data types during transmission. Define specific input / output interfaces for each functional module in the cross-domain data flow channel (such as the node selection module, sequence control module, and exception handling module). The input interface specifies the data types and formats received by the module (e.g., the data types and transmission requirements received by the node selection module), and the output interface specifies the results output by the module (e.g., the node selection module outputs a list of selected nodes and their priorities). Define the execution logic for each functional module, such as the node selection module prioritizing nodes based on the input information and node selection strategy and outputting the results. Through these additions and definitions, a complete cross-domain data flow channel is formed, including a node association system (managing storage node connection relationships) and a flow control system (executing data flow control logic).

[0138] Step S140: The transfer learning model is used to predict the migration adaptation of the cross-domain data transfer channel. The transfer learning model calls cross-domain migration knowledge to analyze the degree of adaptation between the cross-domain data transfer channel and different cloud storage domains, and generates cross-domain data migration adaptation prediction results.

[0139] After the cross-domain data transfer channel is established, step S140 is executed. This step uses a transfer learning model and cross-domain migration knowledge to analyze and predict the compatibility of the cross-domain data transfer channel with different cloud storage domains (such as cloud storage domain A and cloud storage domain B), and generates cross-domain data migration compatibility prediction results.

[0140] Step S141: Transform the architecture information of the cross-domain data flow channel into input features that can be recognized by the transfer learning model. The input features include node association features, flow control features, and channel transmission features. The node association features reflect the connection method and association strength of the storage nodes. The flow control features embody the core logic of the flow control rules. The channel transmission features characterize the transmission capability and transmission characteristics of the channel.

[0141] First, execute step S141. Extract and transform features from the architecture information of the cross-domain data transfer channel. Node association features include the connection topology of storage nodes (e.g., star, mesh), connection type (direct connection, indirect connection), connection stability indicators, and association strength (measured comprehensively by connection bandwidth, latency, etc.). Transfer control features include the core logic of node selection strategies (e.g., priority ranking criteria), rules of sequence control strategies (e.g., conditions for determining transmission order), and response mechanisms of exception handling strategies. Channel transmission features include the channel's maximum transmission bandwidth, average transmission latency, transmission jitter range, data throughput, and supported transmission protocol types. These features are then transformed into vector forms recognizable by the transfer learning model and used as input features.

[0142] Step S142: Input the input features into the transfer learning model. The transfer learning model activates the cross-domain transfer knowledge invocation module to extract transfer adaptation experience and compatibility rules that match the input features from the cross-domain transfer knowledge. The extraction process achieves matching by calculating the cosine similarity between the input feature vector and the knowledge feature vector.

[0143] Next, proceed to step S142. The transformed input features are input into the transfer learning model, which activates the cross-domain transfer knowledge invocation module. This module extracts transfer adaptation experience and compatibility rules that match the input features from the cross-domain transfer knowledge. During extraction, the cosine similarity between the input feature vector and the knowledge feature vectors corresponding to each knowledge item in the cross-domain transfer knowledge is calculated. When the cosine similarity is higher than a set threshold, the knowledge item is considered to match the input feature, and it is extracted. For example, if the node association feature in the input features is "direct connection of high read / write speed storage nodes," and the cosine similarity is high with the knowledge feature vector of "high access frequency data adapts to direct connection of high read / write speed storage nodes" in the cross-domain transfer knowledge, then this transfer adaptation experience is extracted.

[0144] Step S143: Based on the extracted migration adaptation experience and compatibility rules, the transfer learning model analyzes the compatibility of the node association relationship of the cross-domain data flow channel with the storage architecture of different cloud storage domains. The analysis focuses on the compatibility between the node association method and the storage architecture, and the matching degree between the node transmission capacity and the carrying capacity of the storage architecture, generating node adaptation analysis results.

[0145] Then proceed to step S143. Utilizing the extracted migration adaptation experience and compatibility rules, analyze the compatibility of node associations with the storage architectures of cloud storage domains A and B. Node association methods (such as star connections and mesh connections) need to be compatible with the overall design of the storage architecture. For example, the mesh storage architecture of cloud storage domain B is highly compatible with the mesh node association method of cross-domain data transfer channels. Node transmission capabilities (such as the transmission rate of a single node and total bandwidth) need to match the carrying capacity of the storage architecture (such as the total data processing capacity of the storage architecture and network bandwidth resources) to avoid node transmission capabilities exceeding the carrying capacity of the storage architecture, leading to congestion, or falling far below the carrying capacity, resulting in resource waste. Generate node adaptation analysis results based on the analysis results, including compatibility assessment, matching score, and descriptions of existing adaptation problems.

[0146] Step S144: Analyze the compatibility between the cross-domain data transfer channel's transfer control logic and the data processing flow of different cloud storage domains, compare the execution order, triggering conditions, and operation specifications of the transfer control rules and the data processing flow, evaluate the degree of cooperation between the two, and generate transfer control compatibility analysis results.

[0147] Execute step S144. Compare and analyze the flow control logic of the cross-domain data transfer channel with the data processing flows of cloud storage domain A and cloud storage domain B. Verify whether the execution order of the flow control rules (e.g., data format conversion before transmission) matches the steps of the data processing flow (e.g., receiving data before format verification); whether the triggering conditions (e.g., transmission begins when data reaches a certain size) match the triggering events in the data processing flow (e.g., receiving a transmission request); and whether the operational specifications (e.g., data encryption standards) meet the security requirements of the data processing flow. Assess the degree of cooperation between the two, such as whether the flow control logic can be seamlessly integrated into the data processing flow, and whether adjustments are needed to avoid conflicts. Generate flow control compatibility analysis results, including compatibility assessment, cooperation issues, and improvement suggestions.

[0148] Step S145: Evaluate the compatibility between the transmission characteristics of the cross-domain data transfer channel and the network transmission characteristics of different cloud storage domains, analyze the adaptation of channel transmission rate and network bandwidth, the coordination of channel transmission delay and network delay, and the fit between channel transmission stability and network fluctuation characteristics, and generate transmission matching evaluation results.

[0149] Next, proceed to step S145. Evaluate the compatibility of the cross-domain data transfer channel's transmission characteristics with the network transmission characteristics of cloud storage domain A and cloud storage domain B. The channel transmission rate should be compatible with the network bandwidth to avoid data congestion caused by the channel transmission rate exceeding the network bandwidth, or insufficient utilization of bandwidth resources due to the channel transmission rate being too low. The channel transmission delay should be coordinated with the network delay; the channel transmission delay should not be significantly higher than the network delay, otherwise it will affect the overall transmission efficiency. The channel transmission stability should be consistent with the network fluctuation characteristics; in cloud storage domains with large network fluctuations, the channel should have strong anti-fluctuation capabilities. Based on the analysis, generate transmission matching evaluation results, including scores for each matching condition, existing mismatch problems, and optimization directions.

[0150] Step S146: Integrate the node adaptation analysis results, flow control compatibility analysis results, and transmission matching evaluation results, sort out the core adaptation information in each analysis result, determine the adaptation advantages and shortcomings in different dimensions, and form a comprehensive evaluation information on the adaptation degree of cross-domain data flow channels.

[0151] Then proceed to step S146. Integrate the node adaptation analysis results, flow control compatibility analysis results, and transmission matching evaluation results. Summarize the core adaptation information from each result, such as the high compatibility between node relationships and storage architecture, the coordination between flow control logic and data processing flow, and the good compatibility between channel transmission rate and network bandwidth. Determine the adaptation advantages (e.g., significant advantages in node adaptation) and shortcomings (e.g., coordination issues in flow control compatibility) across different dimensions (node ​​adaptation, flow control compatibility, and transmission matching), forming a comprehensive evaluation of the adaptation level of the cross-domain data flow channel.

[0152] For example, step S1461: Extract the core indicators from the node adaptation analysis results. The core indicators include node adaptation fit, node association rationality, node transmission capability matching degree, and node load adaptation degree. Each core indicator contains corresponding analysis data and characterization information.

[0153] Node fit and suitability measures the overall fit between node relationships and the storage architecture. The analyzed data includes fit scores, descriptions of fit points and differences. Node association rationality assesses the rationality of node connection methods and path selection. The analyzed data includes connection path efficiency scores and redundancy assessments. Node transmission capacity matching reflects the match between node transmission capacity and the storage architecture's carrying capacity. The analyzed data includes a matching percentage and quantitative values ​​of excess or insufficient transmission capacity. Node load fit assesses the fit between node load and the storage architecture's load balancing requirements. The analyzed data includes load fit scores and the proportion of nodes with excessively high or low loads. Each core indicator is accompanied by corresponding characterization information, explaining the meaning and evaluation basis of the indicator.

[0154] Step S1462: Extract the core indicators from the flow control compatibility analysis results. The core indicators include control logic compatibility, control process matching degree, strategy coordination adaptability, and trigger condition fit degree. Clarify the specific performance and influencing factors of each core indicator.

[0155] Control logic compatibility refers to the overall compatibility between the flow control logic and the data processing flow, specifically manifested in the number and severity of rule conflicts between the two. Control flow matching assesses the matching of execution order, specifically manifested in the overlap of process steps and the number of steps in reverse order. Strategy collaboration adaptability measures the effectiveness of collaborative work between strategies, specifically manifested in efficiency loss and error rate during collaborative execution. Trigger condition fit refers to the degree of matching between the trigger conditions of the flow control rules and the trigger events in the data processing flow, specifically manifested in the accuracy of condition matching and trigger latency. It is important to clarify the influencing factors of each core indicator, such as how control logic compatibility is affected by rule design philosophy and the complexity of the data processing flow.

[0156] Step S1463: Extract the core indicators from the transmission matching evaluation results. The core indicators include transmission characteristic matching degree, transmission efficiency fit degree, transmission stability adaptability and bandwidth adaptability. Summarize the evaluation data and conclusions for each core indicator.

[0157] The transmission characteristic matching degree comprehensively evaluates the overall matching between the channel's transmission characteristics and the network's transmission characteristics. Evaluation data includes a comprehensive matching score and the matching status of various transmission parameters. Transmission efficiency fit measures the degree of fit between the channel's transmission efficiency and the network's transmission efficiency. Evaluation data includes transmission time difference and throughput utilization. Transmission stability adaptability reflects the channel's stability performance under network fluctuations. Evaluation data includes the number of transmission interruptions and changes in packet loss rate during network fluctuations. Bandwidth adaptability evaluates the channel's utilization and adaptation to network bandwidth. Evaluation data includes bandwidth utilization rate and bandwidth waste or insufficiency rate. The evaluation conclusions for each core indicator are summarized, such as good transmission characteristic matching degree and bandwidth adaptability needing improvement.

[0158] Step S1464: Determine the weight of each core indicator. The weight is determined based on the degree of influence of each indicator on the adaptability of the cross-domain data flow channel. The higher the degree of influence, the greater the weight of the indicator.

[0159] For example, the matching degree of node transmission capabilities and the fit of transmission efficiency have a significant impact on the overall efficiency and success rate of data migration, and are therefore given higher weights; the rationality of node association and the fit of triggering conditions have a relatively smaller impact, and are therefore given lower weights. The determination of weights can be based on a comprehensive assessment combining expert experience, statistical analysis of historical migration cases, and adaptation experience from cross-domain migration knowledge.

[0160] Step S1465: Based on the weights of each core indicator, the node adaptation analysis results, flow control compatibility analysis results, and transmission matching evaluation results are weighted and integrated. The evaluation data of each core indicator are summarized to generate a comprehensive evaluation value. The comprehensive evaluation value intuitively reflects the overall adaptation level of the cross-domain data flow channel.

[0161] For example, the evaluation data for node transmission capacity matching is multiplied by its weight, and the evaluation data for node load adaptability is multiplied by its weight. Then, all the weighted evaluation data are summed to obtain a comprehensive evaluation value. The magnitude of the comprehensive evaluation value directly reflects the overall adaptability level of the cross-domain data flow channel; the higher the value, the better the adaptability.

[0162] Step S1466: Divide the fit level according to the distribution range of the comprehensive evaluation value, and each fit level corresponds to a specific range of comprehensive evaluation values.

[0163] For example, a comprehensive evaluation score in the range of 90-100 indicates an extremely high fit, 75-89 indicates a high fit, 60-74 indicates a moderate fit, 45-59 indicates a low fit, and below 45 indicates an extremely low fit. Each fit level corresponds to a specific range of comprehensive evaluation scores, making it easy to intuitively judge the fit level.

[0164] Step S1467: Extract the specific analysis details of each core indicator, and integrate the comprehensive evaluation value, the degree of fit, and the specific analysis details of each core indicator. Classify and organize them according to the fit dimension to form a comprehensive evaluation information of the degree of fit of the cross-domain data flow channel.

[0165] Extract detailed analysis of each core indicator, such as analysis data on node transmission capability matching and evaluation conclusions on transmission efficiency fit. Organize the comprehensive evaluation value, adaptation level, and these detailed analysis points according to adaptation dimensions (node ​​adaptation, flow control compatibility, and transmission matching) to form comprehensive evaluation information on the adaptation of cross-domain data flow channels, fully reflecting the channel's adaptation status across different dimensions.

[0166] Step S147: Determine the adaptation level of the cross-domain data transfer channel based on the comprehensive evaluation information, and generate a preliminary cross-domain data migration adaptation prediction result that includes the adaptation level, details of various analysis results, and adaptation optimization suggestions. The adaptation level is divided according to the preset range in which the comprehensive evaluation value is located. Different adaptation levels correspond to different adaptation optimization directions and adjustment strategies. The adaptation optimization suggestions propose specific adjustment directions and improvement measures for adaptation shortcomings, and clarify the key links and core elements of the adjustment.

[0167] The adaptation level is determined based on the comprehensive evaluation value in the comprehensive evaluation information. If the comprehensive evaluation value is in the range of 75-89, the adaptation level is high. Preliminary results of cross-domain data migration adaptation prediction are generated, including details of various analysis results such as adaptation level, node adaptation analysis results, flow control compatibility analysis results, and transmission matching evaluation results, as well as adaptation optimization suggestions. The adaptation optimization suggestions propose specific adjustment directions (such as optimizing the execution order of flow control rules) and improvement measures (such as redesigning triggering conditions) for adaptation shortcomings (such as coordination issues in the flow control compatibility dimension), clarifying the key aspects of adjustment (such as the flow control module) and core elements (such as triggering condition parameters).

[0168] Step S148: Perform transfer learning verification on the preliminary results of cross-domain data migration adaptation prediction. The transfer learning model re-invokes the cross-domain migration knowledge to verify the analysis logic and optimization suggestions in the preliminary results. When a conflict is found between the analysis logic and the empirical rules in the cross-domain migration knowledge, adjustments are made according to the preset conflict resolution strategy. Based on the verification results, the deviations in the preliminary results are corrected to obtain the final cross-domain data migration adaptation prediction results.

[0169] The transfer learning model re-invokes cross-domain transfer knowledge to verify the analytical logic (such as the reasoning process of node adaptation analysis) and optimization suggestions (such as suggestions to adjust node connection paths) in the preliminary results of cross-domain data migration adaptation prediction. If there is a conflict between the analytical logic and the empirical rules in the cross-domain transfer knowledge (such as a conflict between the node selection logic in the preliminary results and the logic of successful cases in adaptation experience), adjustments are made according to the preset conflict resolution strategy (such as taking the empirical rules in the cross-domain transfer knowledge as the standard, or making a comprehensive judgment based on the actual situation). Based on the verification results, deviations in the preliminary results are corrected, such as revising the adaptation level assessment and adjusting the adaptation optimization suggestions, to obtain the final cross-domain data migration adaptation prediction results.

[0170] Step S150: Based on the cross-domain data migration adaptation prediction results, dynamically adjust the node association relationship and data flow control logic of the cross-domain data flow channel to generate a cross-domain data dynamic management scheme. Based on the cross-domain data dynamic management scheme, execute the migration and storage control of cloud storage cross-domain data. The cross-domain data dynamic management scheme includes data migration timing arrangement and dynamic allocation rules for storage resources.

[0171] Based on the adaptation shortcomings and optimization suggestions identified in the forecast results, the node relationships (such as adjusting node connection paths and adding / removing nodes) and data flow control logic (such as modifying node selection strategies and optimizing sequence control rules) of the cross-domain data transfer channel are dynamically adjusted. Based on the adjusted cross-domain data transfer channel, a dynamic cross-domain data management scheme is generated. This scheme includes data migration timing arrangements, such as the start time, estimated completion time, and migration batch arrangements for different types of data; and dynamic storage resource allocation rules, such as dynamically adjusting the storage space allocation and computing resource allocation of each storage node based on the data migration progress and storage node load. The migration and storage management of cross-domain data in cloud storage are executed according to this scheme, ensuring that the data migration process is efficient, stable, and secure, and that storage resources are used rationally.

[0172] In one exemplary embodiment, a cloud storage cross-domain data management system based on transfer learning is provided. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2 As shown, this cloud storage cross-domain data management system based on transfer learning includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a cloud storage cross-domain data management method based on transfer learning. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the shell of the cloud storage cross-domain data management system based on transfer learning, or an external keyboard, touchpad, or mouse, etc.

[0173] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A cloud storage cross-domain data management method based on transfer learning, characterized in that, The method includes: Using the storage architecture characteristics of different cloud storage domains and the inherent attribute information of cross-domain data as input, the adaptation benchmark elements for cross-domain data migration are formed after transfer learning preheating, thus obtaining the cloud storage cross-domain data migration adaptation benchmark. Cross-domain transfer knowledge is extracted from the transfer learning model. The cross-domain transfer knowledge is separated from the pre-trained weights of the transfer learning model. The cross-domain transfer knowledge includes data migration and adaptation experience between different cloud storage domains and data storage compatibility rules. A cross-domain data transfer channel is constructed by combining cloud storage cross-domain data migration adaptation benchmarks and cross-domain migration knowledge. The cross-domain data transfer channel includes the node association relationship and data transfer control logic for data transmission between different cloud storage domains. The transfer learning model is used to predict the migration adaptation of cross-domain data transfer channels. The transfer learning model calls cross-domain migration knowledge to analyze the degree of adaptation between cross-domain data transfer channels and different cloud storage domains, and generates cross-domain data migration adaptation prediction results. Based on the cross-domain data migration adaptation prediction results, the node association relationship and data flow control logic of the cross-domain data flow channel are dynamically adjusted to generate a cross-domain data dynamic management scheme. Based on the cross-domain data dynamic management scheme, the migration and storage control of cross-domain data in cloud storage are executed. The cross-domain data dynamic management scheme includes data migration timing arrangement and dynamic allocation rules for storage resources.

2. The cloud storage cross-domain data management method based on transfer learning according to claim 1, characterized in that, The process involves taking storage architecture characteristics of different cloud storage domains and inherent attribute information of cross-domain data as input, and then performing transfer learning preheating to form adaptation benchmark elements for cross-domain data migration. This yields the cloud storage cross-domain data migration adaptation benchmark, which includes: Collect storage architecture characteristic information and cross-domain data inherent attribute information from different cloud storage domains. The storage architecture characteristic information includes the connection architecture of storage nodes, storage media type, data processing flow, transmission interface specifications and resource scheduling mechanism. The cross-domain data inherent attribute information includes data structure type, access characteristics, storage requirements, update frequency and security level requirements. The collected storage architecture characteristic information and cross-domain data inherent attribute information are input into the preheating processing module of the transfer learning model for association mapping, and the corresponding association information between storage architecture characteristic information and cross-domain data inherent attribute information is established. Based on the established corresponding association information, key characteristic items affecting cross-domain data migration are extracted from the storage architecture characteristic information. These key characteristic items include the data transmission interface characteristics of storage nodes, the read and write speed characteristics of storage media, the compatibility characteristics of data processing flow, the response characteristics of resource scheduling mechanism, and the protocol adaptation characteristics of transmission interface. Extract key attribute items that affect cross-domain data migration from the inherent attribute information of cross-domain data. The key attribute items include the structural complexity of the data, the access frequency characteristics of the data, the storage format requirements of the data, the update frequency characteristics of the data, and the security level adaptation requirements of the data. A transfer learning preheating analysis is performed on key feature items and key attribute items. The transfer learning preheating analysis determines the fit association strength by calculating the feature similarity and functional matching degree between key feature items and key attribute items, and generates the fit association strength distribution. Based on the distribution of adaptation association strength, a combination of key characteristic items and key attribute items that meet the requirements of adaptation association strength is selected. This combination is used as the core element of the adaptation benchmark for cross-domain data migration. The implementation of the core element of the adaptation benchmark directly determines the basic adaptation direction of cross-domain data migration. Supplementing the core elements of the adaptation benchmark with auxiliary elements, the auxiliary elements include the network transmission characteristics of different cloud storage domains, storage resource occupancy status, network latency characteristics, bandwidth fluctuation characteristics and load balancing characteristics of storage nodes. The auxiliary elements and the core elements together constitute a complete adaptation evaluation index system. The core and auxiliary elements of the adaptation benchmark are integrated to form an adaptation benchmark framework that includes element relationships, element adaptation standards, and element adjustment thresholds. The element relationships clarify the interaction between core and auxiliary elements, the element adaptation standards define the adaptation boundaries of each element, and the element adjustment thresholds specify the dynamic adjustment range of elements. The adaptation benchmark framework is preheated and strengthened by a transfer learning model. The transfer learning model uses some cross-domain data migration samples to calibrate the adaptation direction of the adaptation benchmark framework. Based on the feedback of the sample migration effect, the element weights in the adaptation benchmark framework are adjusted. According to the preheating and strengthening results, the element adaptation standards and element adjustment thresholds in the adaptation benchmark framework are adjusted, and the adaptation details and adjustment range of each element are refined to form the final cloud storage cross-domain data migration adaptation benchmark.

3. The cloud storage cross-domain data management method based on transfer learning according to claim 1, characterized in that, The extraction of cross-domain transfer knowledge from the transfer learning model includes: Obtain a pre-trained weight set for the transfer learning model. The pre-trained weight set contains weight parameters obtained by pre-training the transfer learning model under data migration scenarios in different cloud storage domains. It covers weight information of multiple network layers, including an input layer, a feature extraction layer, an association analysis layer, a decision layer, and an output layer. The pre-trained weight set is analyzed by weight hierarchy. According to the network hierarchy of the transfer learning model, the pre-trained weight set is divided into input layer weights, feature extraction layer weights, association analysis layer weights, decision layer weights and output layer weights. The weights of each layer are extracted separately and labeled with the corresponding network hierarchy identifier. Identify the weight components related to cross-domain data migration in each level of weight, and retain the weight components that have a consistent weight change trend and a significant impact on the migration effect under different scenarios. The relevant weight components are determined by comparing the weight differences under different cloud storage domain pre-training scenarios. The identified relevant weight components are aggregated to form a cross-domain migration weight cluster. The cross-domain migration weight cluster contains all weight information related to cross-domain data migration adaptation and is classified and organized according to the corresponding network level and functional module. Knowledge decoding is performed on the cross-domain transfer weight cluster. Knowledge decoding transforms weight information into interpretable cross-domain data transfer adaptation experience through the weight inversion structure of the transfer learning model. The weight inversion structure realizes the semantic transformation of weight information through the correspondence between weight parameters and transfer scenario features. Extract data storage compatibility-related sample features from the pre-trained sample set of the transfer learning model. These sample features include the compatibility range of different cloud storage domains with data formats, data storage adaptation conditions, data access permission adaptation requirements, and data transmission encryption adaptation specifications. Based on the extracted sample features, data storage compatibility rules are summarized. These rules include data format conversion specifications, storage resource adaptation requirements, access permission adaptation specifications, transmission encryption adaptation standards, and data integrity assurance specifications. Each rule corresponds to an applicable cloud storage domain scenario and data type range. Cross-domain data migration adaptation experience and data storage compatibility rules are integrated to form an initial set of cross-domain migration knowledge. The initial set is divided into adaptation experience knowledge and compatibility rule knowledge according to knowledge type, and each type of knowledge is labeled with the corresponding applicable scenario and priority. Redundancy is removed from the initial set of cross-domain migration knowledge, and the hierarchical structure of the knowledge set is reorganized according to knowledge type and application scenario. The cross-domain migration knowledge after redundancy removal is then reorganized according to the process steps of cross-domain data migration to form a cross-domain migration knowledge system covering the entire process of migration preparation, migration execution, storage adaptation and subsequent management, thus obtaining the final cross-domain migration knowledge.

4. The cloud storage cross-domain data management method based on transfer learning according to claim 1, characterized in that, The construction of a cross-domain data transfer channel by combining cloud storage cross-domain data migration adaptation benchmarks and cross-domain migration knowledge includes: The adaptation benchmark elements in the cloud storage cross-domain data migration adaptation benchmark are analyzed, and the storage node characteristics and data adaptation requirements in the adaptation benchmark elements are extracted to form the core information of the adaptation benchmark. The core information of the adaptation benchmark is classified and organized according to the storage node type and data type. This paper analyzes the cross-domain data migration adaptation experience and data storage compatibility rules in the cross-domain migration knowledge, extracts the node adaptation experience and flow control rules related to data flow, forms the core information of cross-domain migration knowledge, and marks the corresponding adaptation priority and applicable conditions of the core information of cross-domain migration knowledge. By associating and integrating the core information of the adaptation benchmark and the core information of cross-domain migration knowledge, a correspondence is established between the characteristics of storage nodes and the experience of node adaptation, as well as the correspondence between data adaptation requirements and flow control rules, forming a fusion association table. The fusion association table is used to reflect the mapping logic and the way of action between different information. Based on the established fusion association table, cloud storage domain storage nodes that meet the requirements are selected. The selection criteria include the compatibility between storage node characteristics and node adaptation experience, the fit between data adaptation requirements and flow control rules, and the matching degree between storage node transmission capabilities and data transmission needs. Construct a relational network of the filtered storage nodes. The relational network includes direct and indirect connections between storage nodes, as well as the transmission characteristics corresponding to the connections. The transmission characteristics include transmission rate, transmission delay, bandwidth usage, and transmission stability. The relational network presents the connection status between nodes in the form of a topology. The data flow control logic is constructed based on the relational network. The data flow control logic includes a node selection strategy for data transmission, a data transmission sequence control strategy, and a data transmission exception handling strategy. The node selection strategy clarifies the node selection logic under different data types and transmission scenarios. The sequence control strategy specifies the transmission order of different batches of data. The exception handling strategy defines the response methods and node switching mechanism when transmission exceptions occur. The node selection strategy, sequence control strategy, and exception handling strategy are integrated to form a flow control rule set. The flow control rule set clarifies the execution process, triggering conditions, and mutual cooperation methods of each strategy, and establishes a strategy execution state machine to manage the transition and coordination between strategies. By integrating the relationship network and the set of flow control rules, a preliminary architecture for cross-domain data flow channel is formed. The preliminary architecture includes a node relationship topology and a set of flow control rules. The node relationship topology is used to present the connection relationship of storage nodes, and the set of flow control rules provides behavioral norms for data flow. Based on the adaptation experience in cross-domain migration knowledge, the node association relationship and flow control logic in the initial architecture are adjusted, the node connection path is optimized based on the successful path pattern recorded in the adaptation experience, and the specific parameter settings of the control rules are refined according to the compatibility rules. The missing node association links and flow control details in the initial architecture are supplemented, and specific input / output interfaces and execution logic are defined for each functional module to form a cross-domain data flow channel that includes a node association system and a flow control system.

5. The cloud storage cross-domain data management method based on transfer learning according to claim 1, characterized in that, The method involves using a transfer learning model to predict the migration compatibility of cross-domain data transfer channels. This model utilizes cross-domain migration knowledge to analyze the compatibility between the cross-domain data transfer channels and different cloud storage domains, generating cross-domain data migration compatibility prediction results, including: The architecture information of the cross-domain data flow channel is transformed into input features that can be recognized by the transfer learning model. The input features include node association features, flow control features, and channel transmission features. The node association features reflect the connection method and association strength of the storage nodes, the flow control features embody the core logic of the flow control rules, and the channel transmission features characterize the transmission capability and transmission characteristics of the channel. The input features are fed into the transfer learning model, which activates the cross-domain transfer knowledge invocation module. The module extracts transfer adaptation experience and compatibility rules that match the input features from the cross-domain transfer knowledge. The extraction process achieves matching by calculating the cosine similarity between the input feature vector and the knowledge feature vector. Based on the extracted migration adaptation experience and compatibility rules, the transfer learning model analyzes the compatibility of node association relationships in cross-domain data flow channels with storage architectures of different cloud storage domains. It focuses on analyzing the compatibility between node association methods and storage architectures, the matching degree between node transmission capabilities and storage architecture carrying capacity, and generates node adaptation analysis results. Analyze the compatibility of the cross-domain data flow control logic with the data processing flow of different cloud storage domains, compare the execution order, triggering conditions and operation specifications of the flow control rules and the data processing flow, evaluate the degree of cooperation between the two, and generate flow control compatibility analysis results. The system assesses the compatibility of the transmission characteristics of cross-domain data transfer channels with the network transmission characteristics of different cloud storage domains, analyzes the adaptation of channel transmission rate and network bandwidth, the coordination of channel transmission delay and network delay, and the fit between channel transmission stability and network fluctuation characteristics, and generates transmission matching assessment results. By integrating node adaptation analysis results, flow control compatibility analysis results, and transmission matching evaluation results, the core adaptation information in each analysis result is sorted out, the adaptation advantages and shortcomings in different dimensions are determined, and a comprehensive evaluation information on the adaptation degree of cross-domain data flow channels is formed. Based on comprehensive evaluation information, the adaptation level of the cross-domain data transfer channel is determined, and a preliminary result of cross-domain data migration adaptation prediction is generated, which includes the adaptation level, details of various analysis results, and adaptation optimization suggestions. The adaptation level is divided according to the preset range in which the comprehensive evaluation value is located. Different adaptation levels correspond to different adaptation optimization directions and adjustment strategies. The adaptation optimization suggestions propose specific adjustment directions and improvement measures for adaptation shortcomings, and clarify the key links and core elements of the adjustment. The preliminary results of cross-domain data migration adaptation prediction are reviewed by transfer learning. The transfer learning model re-invokes cross-domain migration knowledge to verify the analysis logic and optimization suggestions in the preliminary results. When a conflict is found between the analysis logic and the empirical rules in the cross-domain migration knowledge, adjustments are made according to the preset conflict resolution strategy. Based on the review results, the deviations in the preliminary results are corrected to obtain the final cross-domain data migration adaptation prediction results.

6. The cloud storage cross-domain data management method based on transfer learning according to claim 2, characterized in that, The step of inputting the collected storage architecture characteristic information and cross-domain data inherent attribute information into the preheating processing module of the transfer learning model for association mapping, and establishing the corresponding association information between storage architecture characteristic information and cross-domain data inherent attribute information, includes: The collected storage architecture characteristic information is processed by feature structuring, and different types of data in the storage architecture characteristic information are transformed into a first structured feature vector in a unified format. The dimensions of the first structured feature vector correspond to different categories of storage architecture characteristics, and the feature values ​​of each dimension are transformed into values ​​in the range [0, 1] through normalization. The inherent attribute information of cross-domain data is processed by feature structuring. The same structuring standard as the storage architecture characteristic information is adopted to transform the inherent attribute information of cross-domain data into a second structured feature vector. The dimension of the second structured feature vector corresponds to the different categories of inherent attributes of cross-domain data. The first structured feature vector and the second structured feature vector are input into the feature alignment unit of the preheating module of the transfer learning model. After adjusting the dimension order according to the preset feature dimension mapping table, the unified feature vector is subjected to association mining. The association mining is based on the semantic relevance and data dependency of each dimension feature in the feature vector. The mining process combines domain knowledge of cross-domain data transfer and uses cross-validation method to verify the preliminary results of association mining. The structured feature vectors that identify storage architecture characteristics and the inherent attributes of cross-domain data contain semantically related feature dimension pairs. The semantic relationship is determined based on the degree of fit between the description content and the representation meaning of the feature dimension, covering both direct and indirect semantic relationships. Calculate the association confidence for each feature dimension pair. The association confidence is determined based on the semantic matching degree between feature dimensions, the association frequency in historical data migration, and the association stability. The association confidence value accurately reflects the reliability of the association between feature dimension pairs. Feature dimension pairs that meet the preset conditions are selected and used as the basis for associating storage architecture characteristic information with cross-domain data inherent attribute information. An association mapping table is constructed based on the association basis. The association mapping table records the association relationship, association confidence and association type of each feature dimension pair. The association type distinguishes semantic association, data dependency association and functional adaptation association, making the association relationship clearer and more identifiable. Based on the association mapping table, an association relationship model is constructed. The association relationship model transforms the static association information in the association mapping table into dynamic association logic, and finally establishes the corresponding association information between storage architecture characteristic information and cross-domain data inherent attribute information.

7. The cloud storage cross-domain data management method based on transfer learning according to claim 3, characterized in that, The knowledge decoding of the cross-domain transfer weight cluster, which transforms weight information into interpretable cross-domain data transfer adaptation experience through the weight inversion structure of the transfer learning model, includes: A weight inversion structure for a transfer learning model is constructed. The weight inversion structure includes a weight parsing unit, a feature mapping unit, and an experience extraction unit. The weight parsing unit is responsible for the hierarchical parsing of weight parameters, the feature mapping unit realizes the mapping between weight parameters and transfer features, and the experience extraction unit completes the extraction and summarization of adaptation experience. The cross-domain transfer weight cluster is input into the weight parsing unit. The weight parsing unit performs hierarchical parsing of the weight parameters in the cross-domain transfer weight cluster, determines the network layer, functional module and feature channel corresponding to each weight parameter, and marks the functional positioning and scope of each weight parameter. The parsed weight parameters are mapped to the corresponding cross-domain data migration feature space through the feature mapping unit. The feature mapping establishes the mapping relationship between the weight parameters and the migration features based on the degree of influence of the weight parameters on different migration features, and determines the migration feature type corresponding to each weight parameter. Identify the cross-domain data migration scenario features corresponding to the weight parameters. The scenario features include the architectural features, data attribute features, migration environment features, and migration target features of different cloud storage domains. The identification of scenario features is based on the mapping features of the weight parameters and pre-trained scenario information. Analyze the variation patterns of weight parameters under different migration scenario characteristics, sort out the fluctuation trend of weight parameters with scene characteristics, summarize the correspondence between weight parameters and migration scenario characteristics, and establish a weight parameter-scene feature correspondence model. Based on the correspondence between weight parameters and migration scenario characteristics, data migration adaptation criteria are extracted for different migration scenarios. The data migration adaptation criteria include storage node selection criteria, data transmission adaptation criteria, storage format adaptation criteria, and resource scheduling adaptation criteria. Specific condition judgment logic and execution parameter thresholds are defined for each data migration adaptation criterion. Extract the points of convergence and difference between the data migration adaptation criteria and the actual operations in real cross-domain data migration cases, optimize the description and execution details of the data migration adaptation criteria based on case experience, and supplement each of the data migration adaptation criteria with specific execution parameters and judgment conditions; The validated data migration adaptation criteria are summarized and integrated, and divided into adaptation criteria for the migration preparation stage, migration execution stage, storage adaptation stage, and subsequent management and control stage according to the process stages of cross-domain data migration, forming a hierarchical cross-domain data migration adaptation experience.

8. The cloud storage cross-domain data management method based on transfer learning according to claim 4, characterized in that, The data flow control logic based on the relational network includes: The transmission and load characteristics of each storage node in the interconnected network are analyzed. The transmission characteristics include transmission latency, transmission bandwidth, transmission stability, and transmission protocol compatibility. The load characteristics include current load status, load capacity limit, load growth trend, and load balancing capability. The analysis results are classified and recorded according to the storage node type. A node selection strategy for data transmission is constructed based on the transmission and load characteristics of storage nodes. The node selection strategy includes node priority ranking rules and node dynamic switching rules. The node priority ranking rules determine the priority order of different nodes according to transmission and load characteristics. The node dynamic switching rules clarify the switching conditions and switching methods when the node transmission capacity is insufficient or the load is too high. Extract the types and transmission requirements of cross-domain data. The transmission requirements of different types of data include transmission timeliness requirements, transmission reliability requirements, transmission security requirements, and transmission bandwidth requirements. Classify and archive cross-domain data according to the priority of data type and transmission requirements. Based on the type of cross-domain data and transmission requirements, a data transmission sequence control strategy is constructed. The sequence control strategy includes the data transmission order order rules and the data fragment transmission order coordination rules. The order order rules determine the transmission order based on the importance of the data and the transmission timeliness requirements. The fragment transmission order coordination rules regulate the transmission order and splicing logic of different fragments of the same data. Identify possible transmission anomaly scenarios in the network of relationships, perform detailed feature descriptions and cause analysis for each of the transmission anomaly scenarios, analyze the causes and impact range of different transmission anomaly scenarios, establish the correspondence between anomaly scenarios and impact ranges, and determine the degree of impact of different anomaly scenarios on data transmission progress, data integrity and transmission security. An anomaly handling strategy for data transmission is constructed based on abnormal scenarios and the scope of impact. The anomaly handling strategy includes anomaly detection rules, anomaly response mechanisms, and anomaly recovery procedures. Anomaly detection rules clarify the identification indicators and detection methods for anomalies. Anomaly response mechanisms define the immediate response measures after an anomaly occurs. Anomaly recovery procedures standardize the steps and methods for restoring data transmission to a normal state. The node selection strategy, sequence control strategy, and exception handling strategy are integrated. The logical relationship and coordination between different strategies are analyzed. When a strategy conflict is detected, the conflict is resolved according to the strategy priority and execution sequence relationship, forming a preliminary framework for data flow control logic. The correlation of the control strategies in the preliminary framework is strengthened, the triggering conditions and execution priorities of different control strategies are determined, the connection process and information interaction method between control strategies are constructed, and a strategy execution state machine is established to manage the activation, execution, and transition states of different strategies. The data flow control logic is optimized by simulating cross-domain data flow scenarios, adjusting the rule parameters and execution flow in the control strategy, and establishing a correspondence table between control logic parameters and network characteristic parameters based on the simulation results.

9. A cloud storage cross-domain data management system based on transfer learning, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the cloud storage cross-domain data management method based on transfer learning as described in any one of claims 1 to 8 by executing the machine-executable instructions.

10. A computer program product, characterized in that, The computer program product includes machine-executable instructions stored in a computer-readable storage medium. The processor of the cloud storage cross-domain data management system based on transfer learning reads the machine-executable instructions from the computer-readable storage medium and executes the machine-executable instructions, causing the cloud storage cross-domain data management system based on transfer learning to perform the cloud storage cross-domain data management method based on transfer learning as described in any one of claims 1 to 8.

Images