Artificial intelligence-based data archiving method and apparatus

Abstract

The application discloses an artificial intelligence-based data archiving method and device, and relates to the technical field of data management. The method comprises the following steps: extracting a multi-dimensional feature vector of to-be-archived data, and constructing a feature map of the to-be-archived data; inputting the multi-dimensional feature vector and the feature map into a collaborative analysis engine for multi-model collaborative analysis to obtain a collaborative analysis result; using a swarm intelligence optimization algorithm to perform global search according to the collaborative analysis result, and generating a candidate archiving strategy set; taking the candidate archiving strategy set as an exploration starting point of an archiving decision model, performing archiving decision according to the collaborative analysis result and a real-time environment state, and obtaining an optimal archiving strategy; executing the optimal archiving strategy on the to-be-archived data, and continuously learning the archiving decision model according to corresponding feedback information to obtain an updated archiving decision model. The problems of rigid archiving strategy, neglect of data correlation, insufficient optimization capability and incapability of self-evolution in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and in particular to a data archiving method and apparatus based on artificial intelligence. Background Technology

[0002] With the rapid development of information technology, the amount of data generated by enterprises and organizations is exploding. A large portion of this data is currently accessed infrequently but still has retention value and needs to be archived. Data archiving is the process of migrating this inactive data from expensive online storage devices to lower-cost offline or near-line storage devices. Its core objective is to minimize storage costs and optimize storage resource utilization while ensuring data accessibility.

[0003] Traditional data archiving methods primarily rely on manually set static strategies, such as "automatically migrating data that has not been accessed for more than 90 days to a tape library." These methods have the following significant drawbacks: 1) Rigid archiving strategy: Static rules cannot dynamically adapt to business changes and fluctuations in data value. For example, project data that has not been accessed for a long time may suddenly become important due to the launch of a new project, but a static strategy cannot predict this situation, which may lead to incorrect archiving decisions.

[0004] 2) Ignoring data relationships: Traditional methods often evaluate individual data in isolation, ignoring the complex business relationships between data. A seemingly unimportant data file may be the input of multiple key business processes, and its "relational value" far exceeds its intrinsic value.

[0005] 3) Insufficient optimization capabilities: Formulating an archiving strategy is a complex multi-objective optimization problem that requires simultaneously balancing factors such as storage costs, access latency, data security, and business value. Manual strategies struggle to find the globally optimal solution within the vast strategy space.

[0006] 4) Inability to self-evolve: Once the strategy is set, it remains fixed and cannot be adjusted or optimized based on the actual effects of archiving (such as user access delays and cost savings), resulting in stagnant system performance.

[0007] In recent years, although some machine learning-based archiving methods have emerged, most of them focus on single-dimensional prediction, such as predicting only access frequency. They fail to effectively integrate and coordinate temporal features, graph structure features, and real-time environmental states, and also lack the ability to perform efficient global search and dynamic decision-making in complex policy spaces.

[0008] Therefore, how to provide a data archiving method that can deeply integrate multi-dimensional data features, intelligently assess the comprehensive value of data, dynamically generate the optimal archiving strategy, and continuously self-optimize is a technical problem that urgently needs to be solved in the field of data management. Summary of the Invention

[0009] This invention provides a data archiving method and apparatus based on artificial intelligence, which solves the problems of rigid archiving strategies, neglect of data correlation, insufficient optimization capabilities, and inability to self-evolve in existing technologies.

[0010] In a first aspect, embodiments of the present invention provide a data archiving method based on artificial intelligence, the method comprising: Extract multi-dimensional feature vectors from the data to be archived, and construct a feature map of the data to be archived based on the business relationships between the data; Multi-dimensional feature vectors and feature maps are input into the collaborative analysis engine to perform multi-model collaborative analysis and obtain collaborative analysis results. Based on the collaborative analysis results, a swarm intelligence optimization algorithm is used to perform a global search in the preset archiving strategy space to generate a set of candidate archiving strategies. The candidate archiving strategy set is used as the starting point for exploring the archiving decision model. Based on the collaborative analysis results and the real-time environmental status, archiving decisions are made to obtain the optimal archiving strategy. The optimal archiving strategy is executed on the archived data, and the archiving decision model is continuously learned based on the corresponding feedback information to obtain an updated archiving decision model.

[0011] The technical solution provided in this application has at least the following beneficial effects: The collaborative analysis engine and archiving decision model built using artificial intelligence technology have enabled intelligent and automated data archiving, improving archiving efficiency and accuracy. Through the time-series access probability prediction model, collaborative prediction of data access probability, access intent, and business value has been achieved. When the business environment changes (such as the launch of a new project leading to an increase in the importance of old data), it can capture potential trends in access patterns and dynamically adjust the "hot" and "cold" assessment of data. This makes archiving decisions no longer based on rigid rules such as "no access for 90 days", but on intelligent prediction of future access possibilities, thereby making more dynamic and adaptive archiving decisions that meet current business needs and effectively avoiding erroneous archiving caused by rigid strategies. By using data feature graphs, data and its complex business relationships are structured. Based on this, by using a business relationship value and intent assessment model, information from neighboring nodes can be aggregated to extract "hub value" and "potential access intent" that are difficult to reflect from individual data in the graph structure. It can identify "high-relation value" data that has low access frequency but serves as key process input, thus avoiding the incorrect archiving of valuable business assets due to isolated assessments and realizing the depth and breadth of data value assessment. Employing the ACO-Levy swarm intelligence optimization algorithm, it performs efficient global search within a vast archiving strategy space. By integrating data value, cost, and latency through heuristic information, it effectively escapes local optima through the Levy flight mechanism and precisely guides the search direction through an adaptive pheromone update mechanism. Taking into account multiple dimensions such as storage cost, access latency, and data survival rate, it automatically generates a set of candidate archiving strategies that are close to the global optimum. Its optimization capabilities and strategy quality far exceed the levels that can be achieved by manual settings or traditional algorithms. It no longer pursues the local optimum of a single data point, but rather the global optimum of the entire data system in multiple dimensions such as cost, performance, and security, bringing tangible economic benefits and improved management efficiency to enterprises. By introducing a closed-loop reinforcement learning archiving decision model, the archiving strategy can be autonomously optimized based on actual results and continuously adapt to business changes. This solves the fundamental problems of difficult maintenance and poor adaptability of traditional strategies. The reward function of reinforcement learning comprehensively considers cost and performance, so that the system pursues long-term global optimum rather than local optimum of a single operation, thereby achieving a dual improvement in storage resource utilization efficiency and business assurance capability. A complete "execution-feedback-learning" closed loop has been established. By monitoring the actual feedback after the strategy is executed, the reward function is used to quantify the quality of the decision, and these experiences are stored in the experience replay pool to continuously train and optimize the archived decision model. This self-evolution mechanism enables the archived decision model to learn from the success and failure of each decision, just like an expert, and continuously adapt to new business models and environments, achieving continuous and automatic performance improvement, and completely solving the pain point of traditional systems being unable to optimize themselves.

[0012] In one alternative implementation, the collaborative analytics engine includes a parallel temporal access probability prediction model and a business association value and intent evaluation model. The temporal access probability prediction model is built based on the LSTM algorithm, and the business association value and intent evaluation model is built based on the GNN algorithm. The swarm intelligence optimization algorithm is constructed based on the ACO-Levy algorithm; The archiving decision model is built based on the DQN algorithm and includes an agent and an experience replay pool.

[0013] In one optional implementation, multi-dimensional feature vectors of the data to be archived are extracted, and a feature map of the data to be archived is constructed based on the business relationships between the data, including: Using a data server, a configurable connector is used to connect to several data sources within the enterprise, and a preset data scanning strategy is provided to confirm the data to be archived; Extract the basic features of the data to be archived, and concatenate the basic features from different dimensions to obtain a multi-dimensional feature vector; the basic features include data access features, data content features, data lifecycle features, and business association features; All data to be archived are used as nodes in the feature map, and multi-dimensional feature vectors are used as the corresponding node features. Based on the business association characteristics of the data to be archived, construct the edges of the feature graph and set the corresponding edge weights; Based on several nodes, several edges and their weights, construct a feature graph of the data to be archived.

[0014] In one optional implementation, multi-dimensional feature vectors and feature maps are input into a collaborative analysis engine for multi-model collaborative analysis, yielding collaborative analysis results, including: Multi-dimensional feature vectors and feature maps are input into the collaborative analysis engine, with the current data to be archived as the target data object, and features centered on the target data object are extracted from the feature map. N A degree relational network is obtained, resulting in a subgraph, where, N Preset relationship network parameters; The data access features in the multi-dimensional feature vector of the target data object and its historical data access feature sequence are input into the time-series access probability prediction model of the collaborative analysis engine, and the access probability prediction value is obtained based on the extracted time-series features. The subgraph of the target data object is input into the business association value and intent assessment model of the collaborative analysis engine, and predictions are made based on the extracted graph features to obtain the potential access intent assessment value and business value prediction value. By integrating the predicted access probability, potential access intent assessment, and business value prediction for each piece of data to be archived, corresponding collaborative analysis results are obtained.

[0015] In one alternative implementation, based on the collaborative analysis results, a swarm intelligence optimization algorithm is used to perform a global search within a predefined archiving strategy space to generate a candidate archiving strategy set, including: Based on the prediction indicators of the collaborative analysis results, the heuristic information calculation function and fitness function of the ACO-Levy algorithm are defined. The data to be archived is mapped to ACO individuals of the ACO-Levy algorithm, the actions of ACO individuals are encoded as archiving measures in the archiving strategy, and the walking path of ACO individuals is encoded as an archiving strategy including several archiving measures. Initialize the ACO population in the preset archiving strategy space, set the maximum number of iterations, and initialize the pheromone matrix using the Tent chaotic mapping sequence; Based on the heuristic information generated by the heuristic information computation function, the Levy flight mechanism is introduced to construct a path for each ACO individual and obtain the corresponding complete path; Based on the collaborative analysis results of each ACO individual, the fitness function is used to calculate the fitness value of each complete path, and the ACO individual with the best fitness value is regarded as the excellent individual. An adaptive convergence factor is introduced to induce pheromone evaporation and release pheromones along the complete path corresponding to outstanding individuals. Repeat the global search steps described above until the maximum number of iterations is reached or the fitness value of the best individuals meets the requirements. Then terminate the iteration and use the complete paths of several ACO individuals with excellent fitness values ​​as a candidate archiving strategy set.

[0016] In one alternative implementation, the formula for the heuristic information computation function is: In the formula, For path Heuristic information; i Indicator of the amount of data to be archived; j For action indication quantity; For the first i Predicted business value from the collaborative analysis results of data to be archived; For the first i The potential access intent assessment value in the collaborative analysis results of the data to be archived; To execute the first j Estimated storage cost of actions; To execute the first j Estimated access delays after the action; For heuristic information coefficients; The formula for the fitness function is: In the formula, ACO individual X The fitness value of the path; For the first i Predicted access probability of data to be archived; The fitness coefficient; M This represents the total number of data items to be archived. To execute the first j The action corresponds to the data survival rate score of the storage layer (for example, the survival rate of tape library may be lower than that of cloud storage).

[0017] In one optional implementation, the candidate archiving strategy set is used as the starting point for exploring the archiving decision model. Based on the collaborative analysis results and the real-time environmental state, an archiving decision is made to obtain the optimal archiving strategy, including: Extract at least one candidate archiving strategy from the candidate archiving strategy set that corresponds to the current data to be archived, and write all candidate archiving strategies into the action space of the agent in the archiving decision model; Collect the real-time environmental status of the database and write the collaborative analysis results corresponding to the real-time environmental status and the current data to be archived into the state space of the agent of the archiving decision model. Based on the state space, the agent's DQN network using the archiving decision model selects the action with the highest Q value in the action space to obtain the optimal archiving policy.

[0018] In one optional implementation, an optimal archiving strategy is executed on the data to be archived, and the archiving decision model is continuously learned based on the corresponding feedback information to obtain an updated archiving decision model, including: The optimal archiving strategy is executed on the data to be archived, the data to be archived is stored in the database, it is transformed into archived data, and the access logs of the archived data are monitored in real time. Based on the access logs, the corresponding comprehensive reward value, i.e., feedback information, is obtained using the reward function of the archived decision model. Based on the feedback information, corresponding real-time experiences are generated and stored in the experience regression pool of the archived decision model. Periodically, a number of historical experiences are randomly extracted from the experience regression pool, and the archived decision model is continuously learned based on a number of real-time experiences and a number of historical experiences to obtain an updated archived decision model.

[0019] In one alternative implementation, the reward function of the archiving decision model is formulated as follows: In the formula, This is the total reward value; The difference in storage costs before and after implementing the optimal archiving strategy; The latency that occurs when a user actually accesses the service; The reward for a successful pre-received item; This is the reward coefficient.

[0020] Secondly, embodiments of the present invention provide an artificial intelligence-based data archiving apparatus for implementing a data archiving method. The apparatus includes: The feature extraction unit is used to extract multi-dimensional feature vectors from the data to be archived, and to construct a feature map of the data to be archived based on the business relationships between the data. The multi-model collaborative analysis unit is used to input multi-dimensional feature vectors and feature maps into the collaborative analysis engine to perform multi-model collaborative analysis and obtain collaborative analysis results. The global search unit is used to perform a global search in the preset archiving strategy space based on the collaborative analysis results and using a swarm intelligence optimization algorithm to generate a set of candidate archiving strategies. The archiving decision unit is used to take the candidate archiving strategy set as the starting point for exploring the archiving decision model, and make archiving decisions based on the collaborative analysis results and real-time environmental status to obtain the optimal archiving strategy. The continuous learning unit is used to execute the optimal archiving strategy on the data to be archived, and to continuously learn the archiving decision model based on the corresponding feedback information to obtain an updated archiving decision model.

[0021] A third aspect of this invention provides an electronic device, which includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, such that the at least one processor can perform the method proposed in the first aspect of the present invention.

[0022] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the steps of an artificial intelligence-based data archiving method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the functional units of a data archiving device based on artificial intelligence provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] The present invention will be further described below with reference to the accompanying drawings.

[0026] Reference Figure 1 , Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0027] like Figure 1 As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0028] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0029] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and an electronic program for an artificial intelligence-based data archiving device.

[0030] exist Figure 1 In the electronic device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the electronic program of the artificial intelligence-based data archiving device stored in the memory 1005 through the processor 1001 and executes the artificial intelligence-based data archiving method provided in the embodiment of the present invention.

[0031] Reference Figure 2 The present invention provides an artificial intelligence-based data archiving method, the method comprising: S201: Extract multi-dimensional feature vectors from the data to be archived, and construct a feature map of the data to be archived based on the business relationships between the data; S202: Input multi-dimensional feature vectors and feature maps into the collaborative analysis engine to perform multi-model collaborative analysis and obtain collaborative analysis results; S203: Based on the collaborative analysis results, a swarm intelligence optimization algorithm is used to perform a global search in the preset archiving strategy space to generate a set of candidate archiving strategies; S204: Using the candidate archiving strategy set as the starting point for exploring the archiving decision model, archiving decisions are made based on the collaborative analysis results and real-time environmental conditions to obtain the optimal archiving strategy; S205: Execute the optimal archiving strategy for the data to be archived, and continuously learn the archiving decision model based on the corresponding feedback information to obtain an updated archiving decision model.

[0032] The technical solution provided in this application has at least the following beneficial effects: The collaborative analysis engine and archiving decision model built using artificial intelligence technology have enabled intelligent and automated data archiving, improving archiving efficiency and accuracy. Through the time-series access probability prediction model, collaborative prediction of data access probability, access intent, and business value has been achieved. When the business environment changes (such as the launch of a new project leading to an increase in the importance of old data), it can capture potential trends in access patterns and dynamically adjust the "hot" and "cold" assessment of data. This makes archiving decisions no longer based on rigid rules such as "no access for 90 days", but on intelligent prediction of future access possibilities, thereby making more dynamic and adaptive archiving decisions that meet current business needs and effectively avoiding erroneous archiving caused by rigid strategies. By using data feature graphs, data and its complex business relationships are structured. Based on this, by using a business relationship value and intent assessment model, information from neighboring nodes can be aggregated to extract "hub value" and "potential access intent" that are difficult to reflect from individual data in the graph structure. It can identify "high-relation value" data that has low access frequency but serves as key process input, thus avoiding the incorrect archiving of valuable business assets due to isolated assessments and realizing the depth and breadth of data value assessment. Employing the Ant Colony Optimization (ACO)-Levy swarm intelligence optimization algorithm, it performs efficient global search within a vast archiving strategy space. By integrating data value, cost, and latency through heuristic information, it effectively escapes local optima through the Levy flight mechanism and precisely guides the search direction through an adaptive pheromone update mechanism. Taking into account multiple dimensions such as storage cost, access latency, and data survival rate, it automatically generates a set of candidate archiving strategies that are close to the global optimum. Its optimization capabilities and strategy quality far exceed the levels that can be achieved by manual settings or traditional algorithms. It no longer pursues the local optimum of a single data point, but rather the global optimum of the entire data system in multiple dimensions such as cost, performance, and security, bringing tangible economic benefits and improved management efficiency to enterprises. By introducing a closed-loop reinforcement learning archiving decision model, the archiving strategy can be autonomously optimized based on actual results and continuously adapt to business changes. This solves the fundamental problems of difficult maintenance and poor adaptability of traditional strategies. The reward function of reinforcement learning comprehensively considers cost and performance, so that the system pursues long-term global optimum rather than local optimum of a single operation, thereby achieving a dual improvement in storage resource utilization efficiency and business assurance capability. A complete "execution-feedback-learning" closed loop has been established. By monitoring the actual feedback after the strategy is executed, the reward function is used to quantify the quality of the decision, and these experiences are stored in the experience replay pool to continuously train and optimize the archived decision model. This self-evolution mechanism enables the archived decision model to learn from the success and failure of each decision, just like an expert, and continuously adapt to new business models and environments, achieving continuous and automatic performance improvement, and completely solving the pain point of traditional systems being unable to optimize themselves.

[0033] In one alternative implementation, the collaborative analytics engine includes a parallel temporal access probability prediction model and a business association value and intent evaluation model. The temporal access probability prediction model is built based on the Long Short-Term Memory (LSTM) algorithm, and the business association value and intent evaluation model is built based on the Graph Neural Network (GNN) algorithm. In this embodiment, the time-series access probability prediction model focuses on predicting the likelihood of a single piece of data being accessed in the future: Model selection: Long Short-Term Memory network is adopted because it can effectively handle the long-term dependency problem in time series data and overcome the gradient vanishing / exploding problem of traditional RNNs; Input data: For the target data object, extract the "data access feature" part from its multi-dimensional feature vector, especially its historical access timestamp sequence (e.g., the number of daily accesses in the past two years). This sequence is preprocessed into a fixed-length sliding window and used as the input to the time series access probability prediction model. Prediction process: The LSTM network of the time-series access probability prediction model learns complex patterns in the access sequence, such as periodicity (e.g., weekday effect, end-of-month effect), trend (e.g., gradual decrease in access volume during the project lifecycle) and suddenness (e.g., occasional audit access), through its gating mechanism (input gate, forget gate, output gate). Output: The model output is a floating-point number between 0 and 1, which is the access probability prediction value. It represents the probability that the data will be accessed within a preset time window (such as the next 30 days). For example, an access probability prediction value of 0.85 means that the data is very likely to be accessed. The business relevance and intent assessment model focuses on evaluating the "hub" value and "potential" value of data within the business network: Model selection: GNN network is adopted, specifically graph convolutional network or graph attention network. GNN can learn directly on graph structure and update the representation of the center node by aggregating the information of neighboring nodes. Input data: For the target data object, first extract the data centered on it from the feature map. N A degree-relationship network is formed, creating a subgraph. N It is a configurable hyperparameter, typically set to 2 or 3, to balance computational complexity and information integrity. This subgraph contains the target data object and its direct and indirect related data. Prediction process: The GNN model performs multiple rounds of message passing on the subgraph. In each round, each node aggregates the feature vectors of its neighboring nodes (combined with edge weights) and updates its own feature vector. After multiple layers of propagation, the final feature vector of each node incorporates the structured information in its neighborhood. Based on this final vector rich in contextual information, the model makes predictions through one or more fully connected layers. Output results: The model outputs two key metrics: Business value prediction: A normalized score (e.g., 0-100) that assesses the contribution of data to the overall business process. Data that is relied upon by many key nodes will have a higher business value prediction. Potential access intent assessment value: A normalized score (e.g., 0-1) that assesses the strength of intent that other high-value data might bring when accessing the target data object. For example, if a core report is frequently accessed after accessing a raw data table, then the potential access intent assessment value of that raw data table will be high, implying that it has the potential to be "accessed as a side effect". The swarm intelligence optimization algorithm is constructed based on the ACO-Levy algorithm; In this embodiment, the ACO-Levy algorithm aims to solve the problem of finding high-quality solutions in a large policy space. The archiving policy space is defined as the Cartesian product of all data to be archived to all available archiving measures, and its size grows exponentially with the amount of data. This embodiment adopts an improved ACO algorithm that integrates chaos theory, Levy flight mechanism and adaptive mechanism to perform efficient global search. The archiving decision model is built on the Deep Q-Network (DQN) algorithm and includes an agent and an experience replay pool.

[0034] In one optional implementation, multi-dimensional feature vectors of the data to be archived are extracted, and a feature map of the data to be archived is constructed based on the business relationships between the data, including: S2011: Using a data server, a configurable connector is used to connect to several data sources within the enterprise, and a preset data scanning strategy is provided to confirm the data to be archived; In this embodiment, the data source includes, but is not limited to: File systems, such as network file systems, general internet file systems, and Hadoop Distributed File System, are scanned through file system application programming interfaces (APIs) or inotify mechanisms. Databases such as MySQL, PostgreSQL, and Oracle obtain table and record access information through database logs or periodic queries; Application systems, such as enterprise resource planning systems, customer relationship management systems, Git, and Jira, obtain business metadata such as project, task, and code commit information through API interfaces; Administrators can preset flexible data scanning strategies, which can include various rules, such as: Time rule: The last modification time is more than 180 days and the last access time is more than 90 days; Size rule: Large files exceeding 1GB in size and accessed infrequently; Tagging rules: Data that is marked as "cold data" or "project completed" by the business system; Combination rules: Data that simultaneously meets multiple of the above conditions; based on these strategies, the data source is scanned periodically or on demand, automatically identifying and generating a list of data to be archived; S2012: Extract the basic features of the data to be archived, and concatenate the basic features of different dimensions to obtain a multi-dimensional feature vector; the basic features include data access features, data content features, data lifecycle features, and business association features; The basic features include: Data access characteristics: These are dynamic characteristics used to capture data usage patterns, including: historical access frequency (such as daily / weekly access counts), access timestamp sequences (for time series analysis), distribution of accessing users / roles, access type (read / write), access source IP, etc. These characteristics are usually obtained by parsing from database audit logs, file system logs, or application logs. Data content characteristics: These are the inherent attributes of the data, including: data size, file type / extension, MIME type, data creation time, data owner, data digest (such as MD5, SHA-256, used for deduplication and integrity verification), keywords / tags (extracted from document titles or content through natural language processing techniques), and data schema (for structured data). Data lifecycle characteristics: These are characteristics related to data governance and compliance, including: data classification level (e.g., public, internal, confidential), compliant retention period (e.g., retention period required under GDPR, SOX Act), business line or project to which the data belongs, and data lifecycle stage (e.g., creation, active, archive candidate, destruction). Business association characteristics: This is one of the key features of the present invention, used to describe the location of data in the business process, including: parent / child data relationship (such as which data tables a report depends on), process dependency relationship (such as which workflow the data is the input or output of), common access pattern (datasets that are frequently accessed by the same user in the same session). These characteristics can be discovered by parsing ETL scripts, analyzing database foreign keys, or through log mining. S2013: Treat all data to be archived as nodes in the feature map, and use multi-dimensional feature vectors as the corresponding node features; S2014: Based on the business association characteristics of the data to be archived, construct the edges of the feature graph and set the corresponding edge weights; In this embodiment, the types of edges can be diverse, such as "belonging to the same project", "is the input of", "shared access", etc.; edge weights are used to quantify the strength of the association. Weights can be calculated based on various factors, such as the frequency of shared access, the strength of dependency in the data flow, and the tightness of association defined by business rules. The higher the weight, the more inseparable the two pieces of data are in terms of business. S2015: Construct a feature map of the data to be archived based on several nodes, several edges and their weights.

[0035] In one optional implementation, multi-dimensional feature vectors and feature maps are input into a collaborative analysis engine for multi-model collaborative analysis, yielding collaborative analysis results, including: S2021: Input multi-dimensional feature vectors and feature maps into the collaborative analysis engine, use the current data to be archived as the target data object, and extract the data centered on the target data object from the feature map. N A degree relational network is obtained, resulting in a subgraph, where, N Preset relationship network parameters; S2022: Input the data access features and their historical data access feature sequences from the multi-dimensional feature vector of the target data object into the time-series access probability prediction model of the collaborative analysis engine, and make predictions based on the extracted time-series features to obtain the access probability prediction value. S2023: Input the subgraph of the target data object into the business association value and intent assessment model of the collaborative analysis engine, and make predictions based on the extracted graph features to obtain the potential access intent assessment value and business value prediction value. S2024: Integrates the predicted access probability, potential access intent assessment, and business value prediction of each piece of data to be archived to obtain corresponding collaborative analysis results, which comprehensively and three-dimensionally depict the "hotness" of the data, its "business importance" and "potential value", providing high-quality input for subsequent strategy optimization and decision-making.

[0036] In one alternative implementation, based on the collaborative analysis results, a swarm intelligence optimization algorithm is used to perform a global search within a predefined archiving strategy space to generate a candidate archiving strategy set, including: S2031: Based on the prediction indicators of the collaborative analysis results, define the heuristic information calculation function and fitness function of the ACO-Levy algorithm; The formula for the heuristic information computation function is: In the formula, For path Heuristic information; i Indicator of the amount of data to be archived; j For action indication quantity; For the first i Predicted business value from the collaborative analysis results of data to be archived; For the first i The potential access intent assessment value in the collaborative analysis results of the data to be archived; To execute the first j Estimated storage cost of actions; To execute the first j Estimated access delays after the action; For heuristic information coefficients; this function defines the ACO individual's information coefficient for the first... i Select the data to be archived j The "prior preference" during an action does not depend on pheromones, but is based on the results of co-analysis and the attributes of the action itself. Its design aims to guide ACO individuals toward "high-value, low-cost" choices. The formula for the fitness function is: In the formula, ACO individual X The fitness value of the path; For the first i Predicted access probability values ​​for data to be archived; The fitness coefficient; M This represents the total number of data items to be archived. To execute the first j The action corresponds to a data survival rate score for the storage layer (e.g., the survival rate of a tape library may be lower than that of cloud storage); this function is the ultimate criterion for evaluating the merits of an archiving strategy (the complete path X of an ACO individual), and is a multi-objective optimization function that aims to maximize overall fitness. It encourages the selection of low-latency, low-cost, and high-survival-rate storage media for data with high access probability; S2032: Map the data to be archived to ACO individuals of the ACO-Levy algorithm, encode the actions of ACO individuals into archiving measures in the archiving strategy, and encode the walking path of ACO individuals into an archiving strategy that includes several archiving measures. In this embodiment, the archiving strategy optimization problem is mapped to the path finding problem of the ACO algorithm. Each piece of data to be archived corresponds to a decision point on the path. Each optional archiving measure (such as A1: retain SSD, A2: migrate HDD, A3: migrate tape) corresponds to an optional path starting from that point. An ACO individual starts from the starting point and selects an action for each piece of data in sequence. The complete path it has traversed represents a complete archiving strategy. S2033: Initialize the ACO population in the preset archiving strategy space, set the maximum number of iterations, and initialize the pheromone matrix using the Tent chaotic mapping sequence; The formula is: In the formula, For path The initial pheromone concentration; For the first k One chaotic variable; These are the upper and lower bounds of the pheromone; For chaotic variable indicators; In the formula, For the first n, n+ One chaotic variable; the Tent chaotic mapping sequence makes the pheromone distribution in the entire solution space present a pseudo-random, ergodic chaotic state, providing richer initial exploration clues for ACO individuals and helping to discover more potential search areas; S2034: Based on the heuristic information generated by the heuristic information calculation function, the Levy flight mechanism is introduced to construct a path for each ACO individual and obtain the corresponding complete path; Each ACO individual starts with an empty solution, sequentially selects an action for each piece of data to be archived, and uses the Levy flight threshold. (Select mode within [0.7, 1], such as 0.85): Randomly generate values ​​between [0, 1]. ,if < The selection probability is determined by both pheromone concentration and heuristic information, as shown in the formula: In the formula, For the first k ACO individuals for paths The probability of choosing; Number of iterations t path pheromone concentration; Number of iterations t path Heuristic information; These are the pheromone importance factor and the heuristic information importance factor; l Indicator of the amount of data to be archived; t This represents the current iteration number; this ensures the convergence of the algorithm and allows it to utilize existing high-quality information. if ≥ There is a small probability of triggering Levy flight. In this case, the ACO individual does not choose according to the above probability, but directly jumps to a "distant" path where the current pheromone is not the highest. The target path of this jump is determined by the Levy distribution. The formula for the step size of Levy flight is: In the formula, The stride length for Levy's flight; For random numbers that follow a standard normal distribution, It is a constant (usually taken as 1.5); the target path for the jump. By step size The algorithm is determined by mapping to the action index space; the Lévy flight mechanism is introduced to allow ants to make long-distance jumps in a probabilistic manner, thereby enhancing the algorithm's ability to escape local optima. S2035: Based on the collaborative analysis results of each ACO individual, use the fitness function to calculate the fitness value of each complete path, and take the ACO individual with the best fitness value as the excellent individual; S2036: Introduce an adaptive convergence factor to volatilize pheromones and release pheromones along the complete path corresponding to excellent individuals; The formula for pheromone evaporation is: In the formula, Number of iterations t+ Path 1 The concentration of pheromones after pheromone volatilization; Based on the volatile content; The convergence factor; In the formula, These are the maximum and minimum values ​​of the convergence factor; , To adjust the parameters; It is the hyperbolic tangent function; T This represents the maximum number of iterations. To guide the search more precisely, only the best individuals in this iteration (or the top 10% of ACO individuals in terms of fitness) are allowed to release pheromones. The amount released is proportional to the fitness value of the path. This elite strategy strengthens the high-quality paths while avoiding interference from the poorer paths. The formula for pheromone release is: In the formula, For the first k ACO individuals for paths The amount of pheromones released; It is a constant. For the first k The fitness value of an ACO individual for a given path; in the early stages of iteration, The pheromone levels are relatively high, meaning even poor solutions can release a certain amount of pheromone, maintaining population diversity; in the later stages of iteration, The smaller the value, the more significant the pheromone release; only high-quality solutions can release significant pheromones, thus accurately guiding the search direction. S2037: Repeat the above global search steps until the number of iterations reaches the maximum number of iterations or the fitness value of the excellent individuals meets the requirements, then terminate the iteration and use the complete paths of several ACO individuals with excellent fitness values ​​as a candidate archiving strategy set.

[0037] In one optional implementation, the candidate archiving strategy set is used as the starting point for exploring the archiving decision model. Based on the collaborative analysis results and the real-time environmental state, an archiving decision is made to obtain the optimal archiving strategy, including: S2041: Extract at least one candidate archiving strategy from the candidate archiving strategy set that corresponds to the current data to be archived, and write all candidate archiving strategies into the action space of the agent of the archiving decision model. S2042: Collect the real-time environmental status of the database (e.g., real-time available capacity of each storage layer (SSD, HDD, Tape), current network I / O bandwidth, CPU load, queue length, etc., these indicators are normalized and then concatenated), and write the real-time environmental status and the collaborative analysis results corresponding to the current data to be archived into the state space of the agent of the archiving decision model. S2043: Based on the state space, the agent's DQN network of the archiving decision model selects the action with the highest Q value in the action space to obtain the optimal archiving policy.

[0038] In one optional implementation, an optimal archiving strategy is executed on the data to be archived, and the archiving decision model is continuously learned based on the corresponding feedback information to obtain an updated archiving decision model, including: S2051: Execute the optimal archiving strategy for the data to be archived, store the data to be archived in the database, transform it into archived data, and monitor the access logs of the archived data in real time; S2052: Based on the access logs, use the reward function of the archived decision model to obtain the corresponding comprehensive reward value, i.e., feedback information; In this embodiment, after the optimal archiving strategy is executed, a continuous monitoring phase begins. Access logs are collected through log brokers, database triggers, etc., to track subsequent access behavior to the archived data in real-time or near real-time and record the following: Access events: whether it was accessed, access time, and accessing user; Performance metrics: actual data recovery time, access latency; Cost changes: The change in storage costs for this data before and after the strategy is implemented; The formula for the reward function of the archiving decision model is: In the formula, This is the total reward value; The difference in storage costs before and after implementing the optimal archiving strategy; The latency that occurs when a user actually accesses the service; The reward for a successful pre-received item; This is the reward coefficient; S2053: Based on the feedback information, generate the corresponding real-time experience and store the real-time experience in the experience regression pool of the archived decision model; S2054: Periodically extract a number of historical experiences from the experience regression pool, and continuously learn the archived decision model based on a number of real-time experiences and a number of historical experiences to obtain an updated archived decision model.

[0039] This invention also provides an artificial intelligence-based data archiving device, referring to... Figure 3 The diagram shows a functional unit diagram of an artificial intelligence-based data archiving device 300 according to the present invention. The device may include the following units: The feature extraction unit 301 is used to extract multi-dimensional feature vectors of the data to be archived, and to construct a feature map of the data to be archived based on the business relationship between the data. The multi-model collaborative analysis unit 302 is used to input multi-dimensional feature vectors and feature maps into the collaborative analysis engine to perform multi-model collaborative analysis and obtain collaborative analysis results. The global search unit 303 is used to perform a global search in the preset archiving strategy space based on the collaborative analysis results and using a swarm intelligence optimization algorithm to generate a set of candidate archiving strategies. The archiving decision unit 304 is used to take the candidate archiving strategy set as the starting point for exploring the archiving decision model, and make archiving decisions based on the collaborative analysis results and real-time environmental status to obtain the optimal archiving strategy. The continuous learning unit 305 is used to execute the optimal archiving strategy on the data to be archived, and to continuously learn the archiving decision model based on the corresponding feedback information to obtain an updated archiving decision model.

[0040] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the artificial intelligence-based data archiving method of the present invention.

[0041] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0042] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0043] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the artificial intelligence-based data archiving method of the present invention.

[0044] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable hardware devices (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0045] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0046] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0047] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0048] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "And / or" indicates that either one or both can be chosen. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0049] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A data archiving method based on artificial intelligence, characterized in that, The method includes: Extract multi-dimensional feature vectors from the data to be archived, and construct a feature map of the data to be archived based on the business relationships between the data; The multi-dimensional feature vectors and feature maps are input into the collaborative analysis engine for multi-model collaborative analysis, yielding collaborative analysis results, including: The multi-dimensional feature vectors and feature maps are input into the collaborative analysis engine. The current data to be archived is used as the target data object, and features centered on the target data object are extracted from the feature map. N A degree relational network is obtained, resulting in a subgraph, where, N The parameters are preset for the relationship network; the collaborative analysis engine includes a parallel time-series access probability prediction model and a business association value and intent evaluation model. The time-series access probability prediction model is constructed based on the LSTM algorithm, and the business association value and intent evaluation model is constructed based on the GNN algorithm. The data access features in the multi-dimensional feature vector of the target data object and its historical data access feature sequence are input into the time-series access probability prediction model of the collaborative analysis engine, and the access probability prediction value is obtained based on the extracted time-series features. The subgraph of the target data object is input into the business association value and intent assessment model of the collaborative analysis engine, and predictions are made based on the extracted graph features to obtain the potential access intent assessment value and business value prediction value. By integrating the predicted access probability, potential access intent assessment, and business value prediction of each piece of data to be archived, the corresponding collaborative analysis results are obtained. Based on the collaborative analysis results, a swarm intelligence optimization algorithm is used to perform a global search in the preset archiving strategy space to generate a candidate archiving strategy set; the swarm intelligence optimization algorithm is constructed using the ACO-Levy algorithm. Using the candidate archiving strategy set as the starting point for exploring the archiving decision model, archiving decisions are made based on the collaborative analysis results and real-time environmental conditions to obtain the optimal archiving strategy, including: Extract at least one candidate archiving strategy from the candidate archiving strategy set that corresponds to the current data to be archived, and write all candidate archiving strategies into the action space of the agent in the archiving decision model; the archiving decision model is constructed based on the DQN algorithm and includes an agent and an experience replay pool; Collect the real-time environmental status of the database and write the collaborative analysis results corresponding to the real-time environmental status and the current data to be archived into the state space of the agent of the archiving decision model. Based on the state space, the agent's DQN network using the archiving decision model selects the action with the highest Q value in the action space to obtain the optimal archiving policy. The optimal archiving strategy is executed on the archived data, and the archiving decision model is continuously learned based on the corresponding feedback information to obtain an updated archiving decision model.

2. The data archiving method based on artificial intelligence according to claim 1, characterized in that, Extract multi-dimensional feature vectors from the data to be archived, and construct a feature map of the data to be archived based on the business relationships between the data, including: Using a data server, a configurable connector is used to connect to several data sources within the enterprise, and a preset data scanning strategy is provided to confirm the data to be archived; Extract the basic features of the data to be archived, and concatenate the basic features of different dimensions to obtain a multi-dimensional feature vector; the basic features include data access features, data content features, data lifecycle features, and business association features; All data to be archived are used as nodes in the feature map, and multi-dimensional feature vectors are used as the corresponding node features. Based on the business association characteristics of the data to be archived, construct the edges of the feature graph and set the corresponding edge weights; Based on several nodes, several edges and their weights, construct a feature graph of the data to be archived.

3. The data archiving method based on artificial intelligence according to claim 2, characterized in that, Based on the collaborative analysis results, a swarm intelligence optimization algorithm is used to perform a global search within a predefined archiving strategy space to generate a candidate archiving strategy set, including: Based on the prediction metrics of the collaborative analysis results, the heuristic information computation function and fitness function of the ACO-Levy algorithm are defined. The data to be archived is mapped to ACO individuals of the ACO-Levy algorithm, the actions of ACO individuals are encoded as archiving measures in the archiving strategy, and the walking path of ACO individuals is encoded as an archiving strategy including several archiving measures. Initialize the ACO population in the preset archiving strategy space, set the maximum number of iterations, and initialize the pheromone matrix using the Tent chaotic mapping sequence; Based on the heuristic information generated by the heuristic information computation function, the Levy flight mechanism is introduced to construct a path for each ACO individual and obtain the corresponding complete path; Based on the collaborative analysis results of each ACO individual, the fitness function is used to calculate the fitness value of each complete path, and the ACO individual with the best fitness value is regarded as the excellent individual. An adaptive convergence factor is introduced to induce pheromone evaporation and release pheromones along the complete path corresponding to outstanding individuals. Repeat the global search steps described above until the maximum number of iterations is reached or the fitness value of the best individuals meets the requirements. Then terminate the iteration and use the complete paths of several ACO individuals with excellent fitness values ​​as a candidate archiving strategy set.

4. The data archiving method based on artificial intelligence according to claim 3, characterized in that, The formula for the heuristic information computation function is: In the formula, For path Heuristic information; i Indicator of the amount of data to be archived; j For action indication quantity; For the first i Predicted business value from the collaborative analysis results of data to be archived; For the first i The potential access intent assessment value in the collaborative analysis results of the data to be archived; To execute the first j Estimated storage cost of actions; To execute the first j Estimated access delays after the action; For heuristic information coefficients; The formula for the fitness function is: In the formula, ACO individual X The fitness value of the path; For the first i Predicted access probability values ​​for data to be archived; The fitness coefficient; M This represents the total number of data items to be archived. To execute the first j The action corresponds to the data survival rate score of the storage layer.

5. The data archiving method based on artificial intelligence according to claim 4, characterized in that, The optimal archiving strategy is executed on the data to be archived, and the archiving decision model is continuously learned based on the corresponding feedback information to obtain an updated archiving decision model, including: The optimal archiving strategy is executed on the data to be archived, the data to be archived is stored in the database, it is transformed into archived data, and the access logs of the archived data are monitored in real time. Based on the access logs, the corresponding comprehensive reward value, i.e., feedback information, is obtained using the reward function of the archived decision model. Based on the feedback information, corresponding real-time experience is generated and stored in the experience regression pool of the archived decision model; Periodically, a number of historical experiences are randomly extracted from the experience regression pool, and the archived decision model is continuously learned based on a number of real-time experiences and a number of historical experiences to obtain an updated archived decision model.

6. The data archiving method based on artificial intelligence according to claim 5, characterized in that, The formula for the reward function of the archiving decision model is: In the formula, This is the total reward value; The difference in storage costs before and after implementing the optimal archiving strategy; The latency that occurs when a user actually accesses the service; The reward for a successful pre-received item; This is the reward coefficient.

7. A data archiving device based on artificial intelligence, used to implement the data archiving method as described in any one of claims 1-6, characterized in that, The device includes: The feature extraction unit is used to extract multi-dimensional feature vectors from the data to be archived, and to construct a feature map of the data to be archived based on the business relationships between the data. The multi-model collaborative analysis unit is used to input the multi-dimensional feature vectors and feature maps into the collaborative analysis engine to perform multi-model collaborative analysis and obtain collaborative analysis results. This includes: inputting the multi-dimensional feature vectors and feature maps into the collaborative analysis engine, using the current data to be archived as the target data object, and extracting data centered on the target data object from the feature map. N A degree relational network is obtained, resulting in a subgraph, where, N The system uses preset relational network parameters. The collaborative analysis engine includes a parallel temporal access probability prediction model and a business association value and intent evaluation model. The temporal access probability prediction model is built based on the LSTM algorithm, and the business association value and intent evaluation model is built based on the GNN algorithm. Data access features and their historical data access feature sequences from the multi-dimensional feature vector of the target data object are input into the temporal access probability prediction model of the collaborative analysis engine, and predictions are made based on the extracted temporal features to obtain the access probability prediction value. The subgraph of the target data object is input into the business association value and intent evaluation model of the collaborative analysis engine, and predictions are made based on the extracted graph features to obtain the potential access intent evaluation value and business value prediction value. The access probability prediction value, potential access intent evaluation value, and business value prediction value of each piece of data to be archived are integrated to obtain the corresponding collaborative analysis result. The global search unit is used to perform a global search in a preset archiving strategy space based on the collaborative analysis results and using a swarm intelligence optimization algorithm to generate a candidate archiving strategy set; the swarm intelligence optimization algorithm is constructed using the ACO-Levy algorithm. An archiving decision unit is used to take the set of candidate archiving strategies as the starting point for exploring the archiving decision model. Based on the collaborative analysis results and the real-time environmental state, it makes archiving decisions to obtain the optimal archiving strategy. This includes: extracting at least one candidate archiving strategy from the set of candidate archiving strategies that corresponds to the current data to be archived, and writing all candidate archiving strategies into the action space of the agent in the archiving decision model; the archiving decision model is constructed based on the DQN algorithm and includes an agent and an experience replay pool; collecting the real-time environmental state of the database and writing the real-time environmental state and the collaborative analysis results corresponding to the current data to be archived into the state space of the agent in the archiving decision model; and, based on the state space, using the DQN network of the agent in the archiving decision model, selecting the action with the highest Q value in the action space to obtain the optimal archiving strategy. The continuous learning unit is used to execute the optimal archiving strategy on the data to be archived, and to continuously learn the archiving decision model based on the corresponding feedback information to obtain an updated archiving decision model.

Images