Intelligent enterprise data modeling management system based on dynamic data sources

Abstract

This invention discloses an intelligent enterprise data modeling and management system based on dynamic data sources, comprising: a data acquisition module for acquiring real-time data from dynamic data sources via blockchain technology, ensuring data consistency and preventing tampering; a data processing module for cleaning and formatting real-time data, and dynamically adjusting data fields; a data modeling module for modeling data entities and relationships using knowledge graphs, and extracting features through graph convolutional networks and graph aggregation mechanisms; an access control module for controlling data access permissions through smart contracts and recording operation logs; a dynamic update module for dynamically updating the knowledge graph based on the access control results using an improved A3C algorithm; and a decision support module for generating traceable decision reports to support efficient decision-making and resource allocation. This invention improves data consistency, security, and decision support capabilities through blockchain, knowledge graphs, and reinforcement learning.

Description

Technical Field

[0001] This invention relates to the field of big data technology, and in particular to an intelligent enterprise data modeling and management system based on dynamic data sources. Background Technology

[0002] With the advent of the big data era, enterprises have accumulated a large number of dynamic data sources in their daily operations. This data is diverse and frequently updated, and traditional data management and modeling methods are unable to efficiently address these challenges. Existing enterprise data management methods often rely on fixed data structures and static storage methods, which makes it difficult to flexibly respond to changes in data sources, especially in the dynamic changes of data types, field structures, and data relationships. Furthermore, existing technologies often neglect the security of data access and the guarantee of data consistency, resulting in insufficient assurance of the authenticity, integrity, and security of data in a distributed environment.

[0003] Currently, many enterprises employ knowledge graph technology in their data modeling processes, aiming to represent complex relationships between entities through graph structures. However, existing knowledge graph technologies based on graph databases typically use static data modeling methods and cannot effectively update dynamically in real time when data relationships change. These methods fail to meet the needs of dynamic data sources, lacking flexible automatic update mechanisms and intelligent decision support capabilities, making it difficult for data modeling to align with the real-time business needs of enterprises.

[0004] On the other hand, as enterprise data security issues become increasingly serious, existing data access control technologies are mostly focused on traditional role-based access management, lacking the ability to dynamically control and fine-grained manage permissions. This approach cannot respond to data changes in real time and lacks access control schemes that are closely integrated with business scenarios, thus failing to effectively prevent data leakage and abuse.

[0005] Therefore, how to provide an intelligent enterprise data modeling and management system based on dynamic data sources is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent enterprise data modeling and management system based on dynamic data sources. This invention makes full use of blockchain technology, knowledge graphs, and reinforcement learning algorithms, and describes in detail how to achieve intelligent modeling and management of enterprise data through processes such as real-time data acquisition based on dynamic data sources, data cleaning and format standardization, data modeling, access control, and dynamic updates.

[0007] The intelligent enterprise data modeling and management system based on dynamic data sources according to embodiments of the present invention includes: The data acquisition module is used to collect real-time data from multiple dynamic data sources using blockchain technology. It categorizes and encrypts the real-time data according to timestamps and sources, and uses consensus algorithms to ensure data consistency and prevent tampering. The data processing module is used to clean and standardize the format of the collected real-time data, form a set of data entities, and dynamically adjust the data fields according to changes in the data source. The data modeling module is used to model data entities and their relationships using knowledge graphs. Each data entity is mapped to a graph node, and graph convolutional networks are used to perform deep modeling of the relationships between nodes. Local and global features are extracted using graph aggregation mechanisms. The access control module is used to control data access permissions through smart contracts on the blockchain based on the node characteristics in the knowledge graph, automatically execute data queries and exchanges according to preset rules, and record all operation logs. The dynamic update module is used to dynamically update the knowledge graph based on the results of access control using the improved A3C algorithm, and adaptively adjust the data entity and relation model by learning new data patterns. The decision support module generates traceable decision management reports by using updated knowledge graphs and operation logs recorded on the blockchain, enabling enterprises to make efficient decisions and allocate resources based on real-time data changes.

[0008] Optionally, the data acquisition module specifically includes: By connecting to the API interfaces of various data sources or the sensors of IoT devices, data is acquired in real time from various data sources using a custom data acquisition protocol; Each piece of data is timestamped locally during collection and is marked according to the source information to ensure the accuracy and reliability of the data source. The data is encrypted locally using the AES-256 encryption algorithm to generate an encrypted data packet. The data packet, along with the data source information and timestamp, is packaged into a block, the hash value of the block is calculated using a hash algorithm, and then added to the blockchain head. Blockchain uses a chain of hashes between blocks to connect data structures, ensuring the order of blockchain data. Each time a new block is generated, network nodes participate in consensus calculation through proof-of-stake or proof-of-work mechanisms to verify the legality of the new block, thereby ensuring data consistency and preventing tampering. The hash value of the current data block and the hash value of the previous block are recorded in the block header of each block to ensure that the data is immutable and traceable. During the storage process, distributed storage technology is used to distribute block data across multiple blockchain nodes to prevent data loss due to single point of failure.

[0009] Optionally, the data processing module specifically includes: Receive real-time data transmitted from various data sources, which typically contains inconsistent data formats, missing values, noisy data, and format errors; To perform data cleaning, predefined rules are applied to each data field for data validation to check whether each piece of data meets the expected range. The cleaning process specifically includes: If data is missing, depending on the characteristics of the field, algorithms such as mean filling, previous value filling, or interpolation based on adjacent data are used to handle missing items; For duplicate data, a deduplication mechanism based on hash function is used. The hash value of the data item is compared with the hash value of the processed data. If the hash values ​​are the same, the data is considered to be duplicated and is automatically removed. For fields with incorrect formatting, regular expressions are used to correct the field formatting to ensure that all fields conform to the preset standard format. After data cleaning, the data is converted into a unified standard format through regular expression mapping and type conversion mechanisms. All timestamps will be converted into the unified ISO 8601 format, numerical data will be standardized into a unified unit, and all field names will be standardized according to the specification to ensure data consistency. During the standardization process, data fields are transformed into fixed data structures: JSON or XML format, to facilitate subsequent storage and use; As the data source changes, new fields or attributes will be dynamically added to the data structure. Based on the real-time monitoring mechanism for changes in the data source, new fields will be automatically mapped to the existing data structure. If the data source deletes some fields, the relevant data will be deleted and removed from the database.

[0010] Optionally, the data modeling module specifically includes: Based on enterprise needs and entity characteristics provided by data sources, a knowledge graph is used to build relationships between entity types and entities. Data entities are mapped to graph nodes through predefined entity types. Each node contains multiple attribute fields, which are provided by the data source and have been standardized. The attribute values ​​of each node are transformed into fixed-dimensional vector representations through the embedding layer. The pre-trained word vector model Word2Vec is used to transform textual data into vectors and to standardize numerical data. Establish relationships between nodes by constructing edges between nodes through manually designed rules or based on context information provided by the data source. These edges represent direct associations or indirect connections between nodes. A graph convolutional network is used to process each node, passing information from neighboring nodes to the current node to form a new node representation; The representation of each node is updated iteratively multiple times through a multi-layer graph convolutional network. In each convolutional layer, the node aggregates information by weighted average of the features of its neighboring nodes. These weighting coefficients are determined by the adjacency relationship of the nodes. The attention weight of each pair of adjacent nodes is calculated. The calculation of the attention weight is carried out through a self-attention mechanism, which dynamically adjusts the intensity of information transmission based on node characteristics and edge relationships, thereby avoiding too much or too little information transmission. Through multi-layered graph convolution and attention mechanisms, the node representation gradually integrates features from the neighborhood and the global network, ultimately obtaining a comprehensive representation of each node, which includes local information and global structural features of the node. By using max pooling or average pooling methods, the features of all nodes in the graph are aggregated to extract global features of the graph for subsequent decision analysis and prediction tasks.

[0011] Optionally, the access control module specifically includes: During the data access process, based on the user's identity and role type, the smart contract will query the Access Control List (ACL) to see if the user has access rights to the target data. The access control information of each node is stored as an encrypted data structure on the blockchain. The smart contract verifies the user's identity through a public-private key mechanism. After successful verification, the contract will automatically obtain the node's permission data. In a knowledge graph, a node's access permissions are determined by its attribute fields and relationship types. Each node's access rules are defined by the node's tags, weights, and relationships with other nodes in the knowledge graph. The contract will match the rules according to the node's permission level. If the match is successful, the user is allowed to access the relevant data. Data access requests are specified down to the level of specific fields. For each requested field, the smart contract determines whether the field can be queried, updated, or deleted according to the rules. If the field meets the preset rules, the contract will perform data operations: querying specific field values ​​or exchanging data, and return the results through an encrypted channel. Each operation, whether successfully executed or rejected, will be automatically logged by a smart contract. The log content includes the executor's identity, operation time, operation content, operation data, and permission verification results. All logs are generated into immutable records using hash values ​​and written to the blockchain's log chain. All log entries are timestamped and the storage is encrypted using the AES-256 encryption algorithm to prevent data leakage and unauthorized access.

[0012] Optionally, the dynamic update module specifically includes: Based on the access control results, a state space and action space are defined using a reinforcement learning framework, where: The state space represents the attributes of each node and edge in the knowledge graph, including the node's category, attribute values, and the connections between nodes; The action space is defined as the operations performed on nodes or edges, including: adding a new node, deleting a node, modifying node attributes, and adding or deleting edges; Each action is generated in parallel by multiple independent Actor networks, and the value of all actions in the current state is evaluated through a shared Critic network. The Actor network predicts the optimal action based on the current state and historical data, and is trained using the policy gradient method to update the network weights. The Critic network estimates the value of each action and reduces variance through an advantage function, thereby optimizing the stability of policy learning. In each update, the A3C algorithm calculates the advantage function of each action based on the learned state-action value function Q, and adjusts the parameters of Actor and Critic through gradient update algorithm to make the structure and node information of the knowledge graph more accurate. During the update process, the A3C algorithm improves its technology by introducing a multi-threaded parallel training mechanism, allowing multiple Actor networks to explore different policy paths simultaneously, and by sharing a Critic network for global evaluation. Each update action adjusts the relationships between data entities and nodes in the knowledge graph by weighted summation of node features and relationships, ensuring that each node update operation takes into account changes in the global graph structure and avoids local optima.

[0013] The beneficial effects of this invention are: First, this invention uses blockchain technology to ensure the security and consistency of data collection, achieving the immutability and traceability of data in a distributed environment, and preventing potential risks during data storage and transmission.

[0014] Secondly, by introducing knowledge graphs and the improved A3C algorithm, this invention can dynamically adjust the data model according to changes in real-time data sources, thereby realizing flexible modeling and automatic updating of data entities and their relationships, avoiding the problem that traditional static modeling methods cannot adapt to complex and ever-changing business needs and data patterns.

[0015] Finally, this invention achieves fine-grained access control for data access through smart contracts, automatically performing permission verification and data operations based on user identity and role. This not only ensures data security but also improves the efficiency and intelligence of data access, effectively reducing the risk of human error and data leakage. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a module structure diagram of the intelligent enterprise data modeling and management system based on dynamic data sources proposed in this invention; Figure 2 This is a flowchart of the data acquisition module method for the intelligent enterprise data modeling and management system based on dynamic data sources proposed in this invention; Figure 3 This is a flowchart of the data processing module method of the intelligent enterprise data modeling and management system based on dynamic data sources proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figure 1 An intelligent enterprise data modeling and management system based on dynamic data sources includes: The data acquisition module is used to collect real-time data from multiple dynamic data sources using blockchain technology. It categorizes and encrypts the real-time data according to timestamps and sources, and uses consensus algorithms to ensure data consistency and prevent tampering. The data processing module is used to clean and standardize the format of the collected real-time data, form a set of data entities, and dynamically adjust the data fields according to changes in the data source. The data modeling module is used to model data entities and their relationships using knowledge graphs. Each data entity is mapped to a graph node, and graph convolutional networks are used to perform deep modeling of the relationships between nodes. Local and global features are extracted using graph aggregation mechanisms. The access control module is used to control data access permissions through smart contracts on the blockchain based on the node characteristics in the knowledge graph, automatically execute data queries and exchanges according to preset rules, and record all operation logs. The dynamic update module is used to dynamically update the knowledge graph based on the results of access control using the improved A3C algorithm, and adaptively adjust the data entity and relation model by learning new data patterns. The decision support module generates traceable decision management reports by using updated knowledge graphs and operation logs recorded on the blockchain, enabling enterprises to make efficient decisions and allocate resources based on real-time data changes.

[0019] refer to Figures 2-3 In this embodiment, the data acquisition module specifically includes: By connecting to the API interfaces of various data sources or the sensors of IoT devices, data is acquired in real time from various data sources using a custom data acquisition protocol; Each piece of data is timestamped locally during collection and is marked according to the source information to ensure the accuracy and reliability of the data source. The data is encrypted locally using the AES-256 encryption algorithm to generate an encrypted data packet. The data packet, along with the data source information and timestamp, is packaged into a block. The hash value of the block is calculated using a hash algorithm and added to the blockchain head. Blockchain uses a chain of hashes between blocks to connect data structures, ensuring the order of blockchain data. Each time a new block is generated, network nodes participate in consensus calculation through proof-of-stake or proof-of-work mechanisms to verify the legality of the new block, thereby ensuring data consistency and preventing tampering. The hash value of the current data block and the hash value of the previous block are recorded in the block header of each block to ensure that the data is immutable and traceable. During the storage process, distributed storage technology is used to distribute block data across multiple blockchain nodes to prevent data loss due to single point of failure.

[0020] In this embodiment, the data processing module specifically includes: Receive real-time data from various data sources. Real-time data often contains inconsistent data formats, missing values, noisy data, and format errors. To perform data cleaning, predefined rules are applied to each data field for data validation to check whether each piece of data meets the expected range. The cleaning process specifically includes: If data is missing, depending on the characteristics of the field, algorithms such as mean filling, previous value filling, or interpolation based on adjacent data are used to handle missing items; For duplicate data, a deduplication mechanism based on hash function is used. The hash value of the data item is compared with the hash value of the processed data. If the hash values ​​are the same, the data is considered to be duplicated and is automatically removed. For fields with incorrect formatting, regular expressions are used to correct the field formatting to ensure that all fields conform to the preset standard format. After data cleaning, the data is converted into a unified standard format through regular expression mapping and type conversion mechanisms. All timestamps will be converted into the unified ISO 8601 format, numerical data will be standardized into a unified unit, and all field names will be standardized according to the specification to ensure data consistency. During the standardization process, data fields are transformed into fixed data structures: JSON or XML format, to facilitate subsequent storage and use; As the data source changes, new fields or attributes will be dynamically added to the data structure. Based on the real-time monitoring mechanism for changes in the data source, new fields will be automatically mapped to the existing data structure. If the data source deletes some fields, the relevant data will be deleted and removed from the database.

[0021] In this embodiment, the data modeling module specifically includes: Based on enterprise needs and entity characteristics provided by data sources, a knowledge graph is used to build relationships between entity types and entities. Data entities are mapped to graph nodes through predefined entity types. Each node contains multiple attribute fields, which are provided by the data source and have been standardized. The attribute values ​​of each node are transformed into fixed-dimensional vector representations through the embedding layer. The pre-trained word vector model Word2Vec is used to transform textual data into vectors and to standardize numerical data. Establish relationships between nodes by constructing edges between nodes through manually designed rules or based on context information provided by the data source. These edges represent direct associations or indirect connections between nodes. A graph convolutional network is used to process each node, passing information from neighboring nodes to the current node to form a new node representation; The representation of each node is updated iteratively multiple times through a multi-layer graph convolutional network. In each convolutional layer, the node aggregates information by weighted average of the features of its neighboring nodes. These weighting coefficients are determined by the adjacency relationship of the nodes. The attention weight of each pair of adjacent nodes is calculated. The calculation of the attention weight is carried out through a self-attention mechanism, which dynamically adjusts the intensity of information transmission based on node characteristics and edge relationships, thereby avoiding too much or too little information transmission. Through multi-layered graph convolution and attention mechanisms, the node representation gradually integrates features from the neighborhood and the global network, ultimately obtaining a comprehensive representation of each node, which includes local information and global structural features of the node. By using max pooling or average pooling methods, the features of all nodes in the graph are aggregated to extract global features of the graph for subsequent decision analysis and prediction tasks.

[0022] In this embodiment, the process of forming the new node representation specifically includes: Graph convolutional networks initialize the feature vector of each node and use the adjacency matrix of the graph to represent the structural relationships between nodes; Each layer of graph convolution operation propagates information by weighted summation of the features of adjacent nodes. The weight coefficients are determined by the degree of the node and the weight of the edge. For each graph convolution layer, the weighted sum of the current node and its neighboring nodes is calculated, and the result is passed to the activation function for nonlinear transformation to generate a new node representation; In each iteration, the graph convolutional layer achieves information fusion by weighted averaging of the features of neighboring nodes, and updates the feature representation of each node; The weighted summation process can be achieved by adjusting the weighting coefficients of the features of each neighboring node. These coefficients are determined by the similarity between the node and its neighbors, the node's attributes, and the type of the edges. Graph convolution operations aggregate broader neighborhood information in each layer through multiple stacks, enabling each node to obtain information from more distant nodes and relationships, thus continuously enriching the node representation. During graph convolution computation, the Adam optimizer is used to adjust the weights of the graph convolutional layers through backpropagation, ensuring the formation of new node representations that contain multi-level information from a wider neighborhood.

[0023] In this embodiment, the access control module specifically includes: During the data access process, based on the user's identity and role type, the smart contract will query the Access Control List (ACL) to see if the user has access rights to the target data. The access control information of each node is stored as an encrypted data structure on the blockchain. The smart contract verifies the user's identity through a public-private key mechanism. After successful verification, the contract will automatically obtain the node's permission data. In a knowledge graph, a node's access permissions are determined by its attribute fields and relationship types. Each node's access rules are defined by the node's tags, weights, and relationships with other nodes in the knowledge graph. The contract will match the rules according to the node's permission level. If the match is successful, the user is allowed to access the relevant data. Data access requests are specified down to the level of specific fields. For each requested field, the smart contract determines whether the field can be queried, updated, or deleted according to the rules. If the field meets the preset rules, the contract will perform data operations: querying specific field values ​​or exchanging data, and return the results through an encrypted channel. Each operation, whether successfully executed or rejected, will be automatically logged by a smart contract. The log content includes the executor's identity, operation time, operation content, operation data, and permission verification results. All logs are generated into immutable records using hash values ​​and written to the blockchain's log chain. All log entries are timestamped and the storage is encrypted using the AES-256 encryption algorithm to prevent data leakage and unauthorized access.

[0024] In this embodiment, the node access rules specifically include: Each node in the knowledge graph forms access control information based on its attribute fields and its relationship with other nodes. The access control information is stored in the blockchain in the form of hash values ​​using the AES-256 encryption algorithm. Each node's permission control data includes access level, validity period, field-level operation permissions, and upstream and downstream relationships of the data. Based on the fields in each request, the contract will match rules according to the node's permission level, specifically including: The access control list (ACL) is matched based on the user's identity and role information to determine the scope of the user's access permissions. The contract calculates the permission verification result based on the node's tag information and the relationship structure in the knowledge graph, combined with the node's weight and the relationship type with related nodes; The access permissions for each field are calculated according to specific rules, and the smart contract dynamically determines whether the access conditions are met. For data query, update, and delete operations, the smart contract will verify whether the user has read, write, and modify permissions for the corresponding fields according to preset permission rules, including: If the permission conditions are met, the operation is executed and the result is returned through an encrypted channel; If the permission conditions are not met, the operation will be rejected and a corresponding permission deficiency message will be returned.

[0025] In this embodiment, the dynamic update module specifically includes: Based on the access control results, a state space and action space are defined using a reinforcement learning framework, where: The state space represents the attributes of each node and edge in the knowledge graph, including the node's category, attribute values, and the connections between nodes; Furthermore, graph structure embedding vectors are introduced as state representations, and the local neighborhood features of nodes are jointly encoded with global topological features. This distinguishes it from the state representation methods in existing technologies that are based only on original attributes or low-dimensional features, thereby improving the semantic integrity and discriminative ability of the state representation. The action space is defined as the operations performed on nodes or edges, including: adding a new node, deleting a node, modifying node attributes, and adding or deleting edges; Furthermore, a constraint filtering mechanism is introduced into the action space. Candidate actions are pre-screened using semantic rules and access control results in the knowledge graph to eliminate operations that do not conform to business logic or access constraints. Compared with the traditional unconstrained action space, this effectively reduces invalid exploration and improves convergence efficiency. Each action is generated in parallel by multiple independent Actor networks, and the value of all actions in the current state is evaluated through a shared Critic network. Among them, each Actor network adopts a heterogeneous structure design, and different Actor networks learn policies for different types of nodes or relationships, thereby distinguishing them from homogeneous Actor networks in existing technologies and improving their adaptability to complex graph structures. The Actor network predicts the optimal action based on the current state and historical data, and is trained using the policy gradient method to update the network weights. The Critic network estimates the value of each action and reduces variance through an advantage function, thereby optimizing the stability of policy learning. Furthermore, the advantage function introduces time difference error and structural consistency constraints, which makes the value assessment consider not only the return but also the rationality of graph structure changes, thus distinguishing it from the traditional assessment method based solely on the return. In each update, the A3C algorithm calculates the advantage function of each action based on the learned state-action value function Q, and adjusts the parameters of Actor and Critic through gradient update algorithm to make the structure and node information of the knowledge graph more accurate. During the update process, the A3C algorithm improves its technology by introducing a multi-threaded parallel training mechanism, allowing multiple Actor networks to explore different policy paths simultaneously, and by sharing a Critic network for global evaluation. In addition, a parameter synchronization delay control mechanism is introduced during multi-threaded training to constrain the gradient updates of different Actors within a time window, thereby avoiding policy oscillations caused by asynchronous updates and improving training stability. Each update action adjusts the relationships between data entities and nodes in the knowledge graph by weighted summation of node features and relationships, ensuring that each node update operation takes into account changes in the global graph structure and avoids local optima.

[0026] In this embodiment, the update process of the A3C algorithm specifically includes: The A3C algorithm calculates the Q-value of each action in the current state through environmental feedback. The Q-value is represented by the state-action value function, which reflects the long-term reward of each action in the current state. The Actor network calculates the optimal action policy based on the Q-value and the current state using the policy gradient method, and updates the network weights through backpropagation so that future policies can obtain higher rewards. Furthermore, an entropy regularization term is introduced into the policy gradient to enhance the policy exploration capability and avoid premature convergence to a suboptimal solution. The Critic network evaluates the actual value of each action chosen by the Actor network, calculates the difference between the actual reward and the expected reward, i.e., the advantage function. The advantage function is used to reduce the variance caused by the state-action value function during training and guide the Actor network to optimize the decision path. Each action selection and weight update is optimized through multiple backpropagations, and the propagation of global information is ensured by sharing the Critic network, reducing policy differences between different Actor networks; Furthermore, by introducing a cross-Actor experience sharing buffer mechanism, experience reuse among different Actors can be achieved, thereby improving sample utilization efficiency; In each update cycle, the A3C algorithm adjusts the attributes of nodes and the connections of edges in the knowledge graph through a gradient update strategy to ensure that the data entity relationships in the knowledge graph continuously approach the optimal structure. All hyperparameters are dynamically adjusted based on actual training results, making the knowledge graph more consistent with real-world data patterns and adaptable to dynamic changes, including: Specifically, an adaptive learning rate adjustment mechanism and a weight decay mechanism based on performance feedback are adopted to dynamically adjust the learning rates of the Actor network and the Critic network, thereby differentiating the fixed parameter setting method and improving the robustness and generalization ability of the A3C algorithm in dynamic data environments.

[0027] Example 1: To verify the feasibility of this invention in practice, it was applied to a production management scenario in a manufacturing industry. The production process of this manufacturing enterprise involves multiple dynamic data sources, including sensor data from production equipment, employee operation records, order information, and external market demand. As the production environment and market conditions change, the format, content, and relationships of these data sources also change. Traditional data management methods cannot respond to these changes in real time, resulting in delayed data updates, low model accuracy, and untimely decision support, severely impacting the enterprise's production efficiency and decision-making quality.

[0028] In this scenario, data acquisition leverages blockchain technology to ensure data consistency and tamper-proofness. Real-time data is collected from various production equipment, sensors, and external market data sources via blockchain. The system categorizes data by timestamp and source and stores it encrypted. Every change to the data is confirmed through a consensus algorithm, ensuring consistency and immutability in a distributed environment. In this way, enterprises can ensure the authenticity of their production data, and once data enters the system, it cannot be tampered with or deleted, thus providing solid data support for subsequent decision-making.

[0029] Next, in the data cleaning and format standardization stage, all collected raw data undergoes preprocessing to remove irregular, missing, and redundant data, and is then standardized in format. This ensures that all data is stored in the database according to a unified standard and format, facilitating subsequent analysis and modeling. Different data sources in the enterprise's production process, such as equipment operating parameters, work order information, and quality inspection results, can be integrated and processed according to a unified standard, significantly reducing data format inconsistencies and improving data processing efficiency.

[0030] In the data modeling phase, enterprises utilize knowledge graphs to model data entities and their relationships based on specific production needs. By constructing relationships between nodes and edges, smart contracts ensure data access control. Each data entity is mapped to a graph node based on its characteristics, and the relationships between nodes are defined through contextual information from the data source. Building upon this, deep modeling of the relationships between nodes is performed using graph convolutional networks and self-attention mechanisms to continuously adjust and optimize node characteristics, ensuring the dynamic updating and accuracy of the data model.

[0031] In practical applications, this invention enables dynamic updates to the knowledge graph through an improved A3C algorithm, automatically adjusting the data model based on changes in real-time data sources to respond to data changes in enterprise production. For example, when a device sensor detects a performance degradation in a particular device, the A3C algorithm automatically identifies and updates the nodes related to that device, adjusts the relationship between the device and other nodes, and optimizes production scheduling strategies. Through this method, enterprises can react to production anomalies in a timely manner, avoiding delays or errors in production planning.

[0032] By using smart contracts to control access management, enterprises can ensure that users with different roles and identities can only access data within their authorized scope. Data access relies not only on user authentication but also on dynamic permission matching based on node relationships and attributes in a knowledge graph. This mechanism effectively prevents unauthorized access and ensures data security.

[0033] To verify the effectiveness of this invention, we conducted a comparative analysis of data before and after implementation. The relevant data is as follows: Table 1 Comparison of Production Management Data As shown in Table 1, this invention demonstrated significant improvements in several key performance indicators after implementation. Data acquisition latency decreased from 30 minutes to 5 minutes, production scheduling response time decreased from 60 minutes to 15 minutes, and production efficiency increased by 50%. These improvements not only enhanced the company's production efficiency and decision-making speed but also significantly strengthened the security and consistency of data management. Furthermore, through smart contract-based access control, the number of unauthorized access violations decreased from 5 to 0, greatly ensuring the security of the company's data.

[0034] In summary, the application of this invention in enterprise data management significantly improves performance in areas such as data acquisition, model updates, production scheduling, data consistency, and security. It addresses the shortcomings of traditional data management methods when dealing with dynamic data sources, achieving the beneficial effects of optimizing production efficiency and ensuring data security.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent enterprise data modeling and management system based on dynamic data sources, characterized in that: include: The data acquisition module is used to collect real-time data from multiple dynamic data sources using blockchain technology. It categorizes and encrypts the real-time data according to timestamps and sources, and uses consensus algorithms to ensure data consistency and prevent tampering. The data processing module is used to clean and standardize the format of the collected real-time data, form a set of data entities, and dynamically adjust the data fields according to changes in the data source. The data modeling module is used to model data entities and their relationships using knowledge graphs. Each data entity is mapped to a graph node, and graph convolutional networks are used to perform deep modeling of the relationships between nodes. Local and global features are extracted using graph aggregation mechanisms. The access control module is used to control data access permissions through smart contracts on the blockchain based on the node characteristics in the knowledge graph, automatically execute data queries and exchanges according to preset rules, and record all operation logs. The dynamic update module is used to dynamically update the knowledge graph based on the results of access control using the improved A3C algorithm, and adaptively adjust the data entity and relation model by learning new data patterns. The decision support module generates traceable decision management reports by using updated knowledge graphs and operation logs recorded on the blockchain, enabling enterprises to make efficient decisions and allocate resources based on real-time data changes.

2. The intelligent enterprise data modeling and management system based on dynamic data sources according to claim 1, characterized in that, The data acquisition module specifically includes: By connecting to the API interfaces of various data sources or the sensors of IoT devices, data is acquired in real time from various data sources using a custom data acquisition protocol; Each piece of data is timestamped locally during collection and is marked according to the source information to ensure the accuracy and reliability of the data source. The data is encrypted locally using the AES-256 encryption algorithm to generate an encrypted data packet. The data packet, along with the data source information and timestamp, is packaged into a block, the hash value of the block is calculated using a hash algorithm, and then added to the blockchain head. Blockchain uses a chain of hashes between blocks to connect data structures, ensuring the order of blockchain data. Each time a new block is generated, network nodes participate in consensus calculations through proof-of-stake or proof-of-work mechanisms to verify the legitimacy of the new block; The block header of each block records the hash value of the current data block and the hash value of the previous block; During the storage process, distributed storage technology is used to disperse and store block data across multiple blockchain nodes.

3. The intelligent enterprise data modeling and management system based on dynamic data sources according to claim 1, characterized in that, The data processing module specifically includes: Receive real-time data transmitted from various data sources, which typically contains inconsistent data formats, missing values, noisy data, and format errors; To perform data cleaning, predefined rules are applied to each data field for data validation to check whether each piece of data meets the expected range. The cleaning process specifically includes: If data is missing, depending on the characteristics of the field, algorithms such as mean filling, previous value filling, or interpolation based on adjacent data are used to handle missing items. For duplicate data, a deduplication mechanism based on hash function is used. The hash value of the data item is compared with the hash value of the processed data. If the hash values ​​are the same, the data is considered to be duplicated and is automatically removed. For fields with incorrect formatting, regular expressions are used to correct the field formatting to ensure that all fields conform to the preset standard format. After data cleaning, the data is converted into a unified standard format through regular expression mapping and type conversion mechanisms. All timestamps will be converted into the unified ISO 8601 format, numerical data will be standardized into a unified unit, and all field names will be standardized according to the specification to ensure data consistency. During the standardization process, data fields are transformed into fixed data structures: JSON or XML format, to facilitate subsequent storage and use; As the data source changes, new fields or attributes will be dynamically added to the data structure. Based on the real-time monitoring mechanism for changes in the data source, new fields will be automatically mapped to the existing data structure. If the data source deletes some fields, the relevant data will be deleted and removed from the database.

4. The intelligent enterprise data modeling and management system based on dynamic data sources according to claim 1, characterized in that, The data modeling module specifically includes: Based on enterprise needs and entity characteristics provided by data sources, a knowledge graph is used to build relationships between entity types and entities. Data entities are mapped to graph nodes through predefined entity types. Each node contains multiple attribute fields, which are provided by the data source and have been standardized. The attribute values ​​of each node are transformed into fixed-dimensional vector representations through the embedding layer. The pre-trained word vector model Word2Vec is used to transform textual data into vectors and to standardize numerical data. Establish relationships between nodes by constructing edges between nodes through manually designed rules or based on context information provided by the data source. These edges represent direct associations or indirect connections between nodes. A graph convolutional network is used to process each node, passing information from neighboring nodes to the current node to form a new node representation; The representation of each node is updated iteratively multiple times through a multi-layer graph convolutional network. In each convolutional layer, the node aggregates information by weighted average of the features of its neighboring nodes. These weighting coefficients are determined by the adjacency relationship of the nodes. The attention weight of each pair of adjacent nodes is calculated. The calculation of the attention weight is carried out through a self-attention mechanism, which dynamically adjusts the intensity of information transmission based on node characteristics and edge relationships, thereby avoiding too much or too little information transmission. Through multi-layered graph convolution and attention mechanisms, the node representation gradually integrates features from the neighborhood and the global network, ultimately obtaining a comprehensive representation of each node, which includes local information and global structural features of the node. By using max pooling or average pooling methods, the features of all nodes in the graph are aggregated to extract global features of the graph for subsequent decision analysis and prediction tasks.

5. The intelligent enterprise data modeling and management system based on dynamic data sources according to claim 1, characterized in that, The access control module specifically includes: During the data access process, based on the user's identity and role type, the smart contract will query the Access Control List (ACL) to see if the user has access rights to the target data. The access control information of each node is stored as an encrypted data structure on the blockchain. The smart contract verifies the user's identity through a public-private key mechanism. After successful verification, the contract will automatically obtain the node's permission data. In a knowledge graph, a node's access permissions are determined by its attribute fields and relationship types. Each node's access rules are defined by the node's tags, weights, and relationships with other nodes in the knowledge graph. The contract will match the rules according to the node's permission level. If the match is successful, the user is allowed to access the relevant data. Data access requests are specified down to the level of specific fields. For each requested field, the smart contract determines whether the field can be queried, updated, or deleted according to the rules. If the field meets the preset rules, the contract will perform data operations: querying specific field values ​​or exchanging data, and return the results through an encrypted channel. Each operation, whether successfully executed or rejected, will be automatically logged by a smart contract. The log content includes the executor's identity, operation time, operation content, operation data, and permission verification results. All logs are generated into immutable records using hash values ​​and written to the blockchain's log chain. All log entries are timestamped and the storage is encrypted using the AES-256 encryption algorithm to prevent data leakage and unauthorized access.

6. The intelligent enterprise data modeling and management system based on dynamic data sources according to claim 1, characterized in that, The dynamic update module specifically includes: Based on the access control results, a state space and action space are defined using a reinforcement learning framework, where: The state space represents the attributes of each node and edge in the knowledge graph, including the node's category, attribute values, and the connections between nodes; The action space is defined as the operations performed on nodes or edges, including: adding a new node, deleting a node, modifying node attributes, and adding or deleting edges; Each action is generated in parallel by multiple independent Actor networks, and the value of all actions in the current state is evaluated through a shared Critic network. The Actor network predicts the optimal action based on the current state and historical data, and is trained using the policy gradient method to update the network weights. The Critic network estimates the value of each action and reduces variance through an advantage function, thereby optimizing the stability of policy learning. In each update, the A3C algorithm calculates the advantage function of each action based on the learned state-action value function Q, and adjusts the parameters of Actor and Critic through gradient update algorithm to make the structure and node information of the knowledge graph more accurate. During the update process, the A3C algorithm introduces a multi-threaded parallel training mechanism, allowing multiple Actor networks to explore different policy paths simultaneously and perform global evaluation by sharing a Critic network. Each update action adjusts the relationships between data entities and nodes in the knowledge graph by weighted summation of node features and relationships, ensuring that each node update operation takes into account changes in the global graph structure and avoids local optima.

Images