Industrial chain atlas modeling method and system based on deep learning

Abstract

The invention provides an industrial chain atlas modeling method and system based on deep learning, and relates to the technical field of industrial chain analysis and modeling, and the method comprises the steps: collecting multi-source heterogeneous data in an industrial chain, including demand information, resource states, process features and operation and maintenance data, and preprocessing the multi-source heterogeneous data to form a standardized data set; and performing semantic vectorization processing on the text type data in the standardized data set, and performing node embedding on the multi-relation graph to generate entity low-dimensional representation fusing text and structural features. According to the method, by integrating multi-source heterogeneous data, fusing text and structural features, constructing an industrial chain knowledge graph and performing dynamic optimization, accurate adjustment of production line configuration and task allocation strategies is realized, meanwhile, the industrial trend is effectively predicted, risk paths are analyzed and resource bottlenecks are identified in combination with a risk propagation model, a scientific basis is provided for industrial chain decision-making, and the risk propagation model has a wide application prospect. And the operation efficiency and stability of an industrial chain are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial chain analysis and modeling, and in particular to an industrial chain graph modeling method and system based on deep learning. Background Art

[0002] Most traditional industrial chain map modeling methods are highly dependent on expert experience and manual analysis, which leads to strong subjectivity in the model construction process and makes it difficult to fully and objectively reflect the actual situation of the industrial chain.

[0003] For example, in the automotive manufacturing industry chain, traditional methods may only rely on industry experts' intuitive judgment and experience summary of parts suppliers, vehicle manufacturers, sellers, etc. to build an industry chain map. However, this expert experience-based method may ignore some emerging companies or potential important suppliers, resulting in an incomplete or biased map.

[0004] Secondly, with the rapid development of information technology, the amount of data related to the industrial chain has exploded. Traditional industrial chain graph modeling methods are inefficient in processing large-scale data and are difficult to adapt to the rapidly changing market environment. In the e-commerce industry chain, massive amounts of transaction data, user behavior data, logistics data, etc. are involved. Traditional methods may use manual sorting or simple data analysis tools to process these data, resulting in slow processing speed and low accuracy. For example, during shopping festivals such as "Double Eleven", the transaction volume surges, and traditional methods find it difficult to update the industrial chain graph in real time to reflect the latest market conditions, which may cause companies to miss market opportunities or make wrong decisions.

[0005] In addition, in the intelligent manufacturing industry chain, technologies such as equipment networking and real-time data collection have led to a sharp increase in the amount of data. Traditional methods may face problems such as storage difficulties and insufficient computing resources when processing these data. For example, a smart manufacturing company hopes to build a comprehensive industry chain map including equipment status, production process, supply chain information, etc., but traditional methods are difficult to efficiently process such a large amount of data, resulting in a long map construction cycle and untimely updates, which affects the company's real-time monitoring of the production process and its ability to optimize decision-making. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide an industrial chain graph modeling method and system based on deep learning to improve the intelligence and automation level of industrial chain management.

[0007] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0008] In a first aspect, a method for modeling an industrial chain graph based on deep learning is provided, the method comprising:

[0009] Collect multi-source heterogeneous data in the industrial chain, including demand information, resource status, process characteristics and operation and maintenance data, and pre-process the multi-source heterogeneous data to form a standardized data set;

[0010] Perform semantic vectorization on text data in standardized datasets and embed nodes in multi-relational graphs to generate low-dimensional representations of entities that integrate text and structural features.

[0011] According to the low-dimensional representation of entities, the industrial chain ontology is constructed, including structured data, semi-structured text and unstructured building information model, and the industrial chain knowledge graph is constructed according to the industrial chain ontology;

[0012] According to the industrial chain knowledge graph, the two-stage deep Q network is used to dynamically optimize the production line configuration and task allocation strategy, and the optimization results are updated to the node attributes and relationship weights of the knowledge graph to generate an optimized dynamic industrial chain knowledge graph;

[0013] Based on the optimized dynamic industrial chain knowledge graph, the node importance weights are calculated through the graph attention network, and combined with the risk propagation model to perform industrial trend prediction, risk propagation path analysis and resource bottleneck identification, and finally generate decision recommendations.

[0014] Furthermore, the text data in the standardized data set is semantically vectorized, and the nodes of the multi-relation graph are embedded to generate a low-dimensional representation of the entity that integrates text and structural features, including:

[0015] Perform word segmentation and part-of-speech tagging on the text data in the standardized data set, and extract the preliminary semantic features of the text, including word frequency vectors and co-occurrence matrices; extract entities from the standardized data set, including enterprises, equipment, processes, and associations, including supply and demand relationships, and collaborative relationships, and construct the initial structure of the multi-relationship graph;

[0016] Map the word frequency vectors and co-occurrence matrix to the corresponding nodes of the multi-relation graph, and add preliminary semantic representation to each node;

[0017] Calculate the structural features of each node in the multi-relational graph, including degree centrality and local clustering coefficient features, and combine the structural features with the semantic features to form a composite feature vector containing text and structural information;

[0018] A linear projection operation is performed on the composite feature vector, and a nonlinear transformation is performed on the projected features to generate a low-dimensional representation vector of the entity.

[0019] Furthermore, based on the low-dimensional representation of entities, the industrial chain ontology is constructed, including structured data, semi-structured text and unstructured building information model, and the industrial chain knowledge graph is constructed based on the industrial chain ontology, including:

[0020] Based on the low-dimensional representation of entities, define the ontology model of the industrial chain, including structured data, semi-structured text and unstructured building information model;

[0021] Map the entities, relationships and attributes in the ontology model to the nodes, edges and attribute fields of the graph database to generate an industrial chain knowledge graph containing node, edge and relationship weights.

[0022] Furthermore, in the industrial chain knowledge graph, nodes represent entities in the industrial chain, including enterprises, equipment, and processes, and edges represent relationships between entities, including supply and demand, collaboration, and dependency.

[0023] Furthermore, according to the industrial chain knowledge graph, the two-stage deep Q network is used to dynamically optimize the production line configuration and task allocation strategy, and the optimization results are updated to the node attributes and relationship weights of the knowledge graph to generate an optimized dynamic industrial chain knowledge graph, including:

[0024] According to the current state of the industry chain knowledge graph, a two-stage deep Q network is used to simulate the combination of production line configuration adjustment and task allocation strategy;

[0025] The two-stage deep Q network is used to dynamically adjust and optimize the production line configuration and task allocation strategy in the knowledge graph environment. The target Q value is calculated and the parameters are updated during the iterative learning process to determine the final combination of production line configuration and task allocation strategy.

[0026] Update the attributes of the equipment nodes in the knowledge graph based on the final combination of production line configuration and task allocation strategy;

[0027] After the attributes of the device nodes are updated, the relationship weights between the nodes in the knowledge graph are re-evaluated and adjusted according to the new task allocation strategy;

[0028] Integrate all updated node attributes and relationship weights to generate an optimized dynamic industrial chain knowledge graph.

[0029] Furthermore, a two-stage deep Q network is used to dynamically adjust and optimize the production line configuration and task allocation strategy in the knowledge graph environment. The target Q value is calculated and the parameters are updated during the iterative learning process to determine the final combination of production line configuration and task allocation strategy, including:

[0030] Calculate the difference ratio of the efficiency between the new state and the old state, the cost difference between the old state and the new state, and the difference ratio of the resource utilization between the new state and the old state, and get an immediate reward;

[0031] Predict the next state based on the current state, and select the action that maximizes the Q value predicted by the current Q network as the final action under the corresponding predicted state;

[0032] Input the future state and final action into the target Q network to obtain the Q value prediction of the future state;

[0033] The immediate reward and the Q-value prediction of the future state are combined to obtain the target Q-value.

[0034] Furthermore, based on the optimized dynamic industry chain knowledge graph, the node importance weights are calculated through the graph attention network, and combined with the risk propagation model to predict industry trends, analyze risk propagation paths, and identify resource bottlenecks, and finally generate decision recommendations, including:

[0035] Extract node features, edge features and topological structure information from the optimized dynamic industry chain knowledge graph;

[0036] The graph attention network is used to process node features, edge features and topological structure information, automatically learn the association strength between nodes, and assign an importance weight to each node;

[0037] Combine the node importance weights with the predefined risk propagation model to evaluate the risk propagation path and potential impact in the industrial chain and obtain the risk propagation analysis results;

[0038] According to the results of risk propagation analysis, the future development trend of the industrial chain is predicted, including identifying key nodes and potential risk points in the industrial chain to obtain the industrial trend prediction results;

[0039] Analyze the node importance weights, risk propagation models and industry trend forecast results, identify key risk propagation nodes and paths, as well as resource bottleneck nodes and potential resource shortages in the industrial chain, and generate decision-making recommendations, including adjustments to production line configurations, optimization of task allocation, formulation of risk prevention and control measures, and planning of resource allocation and scheduling.

[0040] The second aspect is an industrial chain graph modeling system based on deep learning, including:

[0041] The data processing module is used to collect multi-source heterogeneous data from the industrial chain, including demand information, resource status, process characteristics and operation and maintenance data, and pre-process the multi-source heterogeneous data to form a standardized data set;

[0042] The knowledge fusion module is used to perform semantic vectorization on text data in the standardized data set and embed nodes in the multi-relation graph to generate a low-dimensional representation of entities that integrates text and structural features;

[0043] The graph construction module is used to construct the industry chain knowledge graph using the low-dimensional representation of entities that integrate text and structural features, where nodes represent different entities in the industry chain and edges represent the relationships between entities;

[0044] Reinforcement learning module, which is used to dynamically optimize production line configuration and task allocation strategy through a two-stage deep Q network to generate an optimized dynamic industry chain knowledge graph;

[0045] The decision support module is used to forecast industry trends, analyze risk propagation paths and identify resource bottlenecks based on the optimized industry chain knowledge graph, and ultimately generate decision recommendations.

[0046] According to a third aspect, a computing device includes:

[0047] one or more processors;

[0048] The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method described.

[0049] In a fourth aspect, a computer-readable storage medium stores a program, and when the program is executed by a processor, the method described is implemented.

[0050] The above solution of the present invention includes at least the following beneficial effects:

[0051] By collecting multi-source heterogeneous data such as demand information, resource status, process characteristics, and operation and maintenance data in the industrial chain and preprocessing them to form a standardized data set, the operating status and characteristics of the industrial chain can be fully reflected. This helps to eliminate data silos and improve data availability and consistency. The text data in the standardized data set is semantically vectorized, and the nodes of the multi-relationship graph are embedded to generate a low-dimensional representation of the entity that integrates text and structural features. This method can make full use of the information in text data and structural data, capture the complex relationships between entities, and improve the accuracy and richness of entity representation.

[0052] The ontology of the industrial chain is constructed based on the low-dimensional representation of entities, including structured data, semi-structured text and unstructured building information models, and the knowledge graph of the industrial chain is further constructed. This helps to formalize the entities, relationships and attributes in the industrial chain, form a computable and reasonable knowledge system, and provide strong support for the analysis and optimization of the industrial chain. The production line configuration and task allocation strategy are dynamically optimized through a two-stage deep Q network, and the optimization results are updated to the node attributes and relationship weights of the knowledge graph to generate an optimized dynamic industrial chain knowledge graph. This method can respond to changes in the industrial chain in real time, improve production efficiency and resource utilization, reduce production costs, and enhance the competitiveness of the industrial chain.

[0053] According to the optimized dynamic industry chain knowledge graph, the node importance weights are calculated through the graph attention network, and the industry trend forecast, risk propagation path analysis and resource bottleneck identification are carried out in combination with the risk propagation model. This helps to discover potential problems and risks in the industry chain in advance, provide support for decision-making, and avoid or reduce losses. The final decision-making recommendations are based on comprehensive data analysis and in-depth model reasoning, and are scientific and targeted. These recommendations can guide enterprises in the industry chain to carry out reasonable production line configuration, task allocation, risk prevention and control, and resource scheduling, and improve the overall operating efficiency and stability of the industry chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of an industrial chain graph modeling method based on deep learning provided in an embodiment of the present invention.

[0055] Figure 2 It is a schematic diagram of an industrial chain graph modeling system based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0057] like Figure 1 As shown, an embodiment of the present invention proposes a method for industrial chain graph modeling based on deep learning, and the method comprises the following steps:

[0058] Step 1: Collect multi-source heterogeneous data in the industrial chain, including demand information, resource status, process characteristics, and operation and maintenance data, and pre-process the multi-source heterogeneous data to form a standardized data set;

[0059] Step 2: semantically vectorize the text data in the standardized data set and embed nodes in the multi-relation graph to generate a low-dimensional representation of the entity that integrates text and structural features;

[0060] Step 3: Based on the low-dimensional representation of entities, the industrial chain ontology is constructed, including structured data, semi-structured text, and unstructured building information model, and the industrial chain knowledge graph is constructed based on the industrial chain ontology;

[0061] Step 4: According to the industrial chain knowledge graph, the production line configuration and task allocation strategy are dynamically optimized through a two-stage deep Q network, and the optimization results are updated to the node attributes and relationship weights of the knowledge graph to generate an optimized dynamic industrial chain knowledge graph;

[0062] Step 5: Based on the optimized dynamic industrial chain knowledge graph, the node importance weights are calculated through the graph attention network, and the risk propagation model is combined to perform industrial trend prediction, risk propagation path analysis, and resource bottleneck identification, and finally generate decision recommendations.

[0063] In the embodiment of the present invention, by collecting multi-source heterogeneous data such as demand information, resource status, process characteristics and operation and maintenance data in the industrial chain, the key information of each link of the industrial chain can be fully covered. Preprocessing the multi-source heterogeneous data to form a standardized data set helps to eliminate format differences and noise between data and improve data quality and consistency.

[0064] Step 2: semantic vectorization of text data in the standardized data set can convert text information into a vector form that can be understood by computers and capture the semantic information in the text. Node embedding is performed on the multi-relational graph to generate a low-dimensional representation of the entity that integrates text and structural features. It can simultaneously consider the text description of the entity and its structural position in the graph, improving the accuracy and comprehensiveness of the entity representation.

[0065] Step 3: Construct the industrial chain ontology based on the low-dimensional representation of entities, including structured data, semi-structured text and unstructured building information models, which can clearly define the entities, attributes and relationships in the industrial chain, and provide a clear framework and semantic basis for the construction of the knowledge graph. Constructing the industrial chain knowledge graph based on the industrial chain ontology can present the entities and relationships in the industrial chain in a graphical way, and intuitively display the structure and association of the industrial chain.

[0066] Step 4: Dynamically optimize the production line configuration and task allocation strategy through the two-stage deep Q network, which can adjust the production line configuration and task allocation in real time according to the current production status and demand, and improve production efficiency and resource utilization. Update the optimization results to the node attributes and relationship weights of the knowledge graph to generate an optimized dynamic industry chain knowledge graph, which can reflect the latest status and changes of the industry chain in real time.

[0067] Step 5, by calculating the importance weight of the nodes through the graph attention network, the importance of each node in the industrial chain can be accurately evaluated, and key nodes and potential risk points can be identified. Combining the risk propagation model to predict industrial trends, analyze risk propagation paths, and identify resource bottlenecks, it is possible to discover potential risks and bottlenecks in the industrial chain in advance, provide scientific decision-making suggestions for decision makers, and help enterprises formulate effective risk prevention and control and resource allocation strategies. The decision-making suggestions finally generated can provide enterprises with specific action guidelines, help them optimize the industrial chain structure, improve production efficiency, reduce risks, and achieve sustainable development.

[0068] In a preferred embodiment of the present invention, the above step 1, collecting multi-source heterogeneous data in the industrial chain, including demand information, resource status, process characteristics and operation and maintenance data, and preprocessing the multi-source heterogeneous data to form a standardized data set, may include:

[0069] In the embodiment of the present invention, demand information of customers, suppliers, partners, etc. is collected through questionnaires, interviews, etc. Historical sales data is analyzed to predict future demand trends, and macro demand information is obtained by referring to industry research reports, market analysis articles, etc. Resource status data is collected in real time through sensors, RFID and other devices, and resource status information is extracted from the ERP system. For data that cannot be collected automatically, it is entered into the system manually. Operation and maintenance data is collected in real time through the equipment monitoring system, historical maintenance records are analyzed, key operation and maintenance information is extracted, and user feedback on equipment usage is collected as a supplement to operation and maintenance data.

[0070] Use hashing algorithms, sorting and comparison methods to remove duplicate data, and select filling, deletion or interpolation methods to handle missing values ​​according to data characteristics. Detect and handle outliers through statistical methods, machine learning algorithms, etc. Convert data in different formats (such as text, pictures, videos, etc.) into a unified format. Convert character encoding, date format, etc. into standard encoding and format, and unify data in different units into the same unit.

[0071] Scale the data to the interval [0, 1] or [-1, 1]. Convert the data to a standard normal distribution with a mean of 0 and a variance of 1. Integrate the cleaned, converted, and normalized / standardized data into a unified data set, and add labels, annotations, and other information to the data set.

[0072] In a preferred embodiment of the present invention, the above step 2, performing semantic vectorization processing on the text data in the standardized data set, and embedding nodes in the multi-relation graph to generate a low-dimensional representation of the entity that integrates text and structural features, may include:

[0073] Step 210, perform word segmentation and part-of-speech tagging on the text data in the standardized data set, extract preliminary semantic features of the text, including word frequency vectors and co-occurrence matrices; extract entities from the standardized data set, including enterprises, equipment, processes, and associations, including supply and demand relationships, and collaborative relationships, and construct the initial structure of a multi-relationship graph;

[0074] Step 211, mapping the word frequency vector and the co-occurrence matrix to the corresponding nodes of the multi-relation graph, and adding a preliminary semantic representation to each node;

[0075] Step 212, calculating the structural features of each node in the multi-relation graph, including degree centrality and local clustering coefficient features, and combining the structural features with the semantic features to form a composite feature vector containing text and structural information;

[0076] Step 213, performing a linear projection operation on the composite feature vector and performing a nonlinear transformation on the projected features to generate a low-dimensional representation vector of the entity.

[0077] In an embodiment of the present invention, a word segmentation tool (such as Jieba, NLTK, etc.) is used to segment text data in a standardized data set, and the text is divided into individual words. The words after word segmentation are tagged with parts of speech (such as nouns, verbs, adjectives, etc.) to identify the grammatical role of the words in the sentence. The frequency of each word in the text is counted to form a word frequency vector to reflect the importance of the word in the text, and the number of co-occurrences between words is calculated to form a co-occurrence matrix to reflect the association between words.

[0078] Entity extraction: Extract entities such as enterprises, equipment, processes, etc. from standardized data sets.

[0079] Relationship extraction: Identify the relationships between entities, such as supply and demand relationships, collaboration relationships, etc.

[0080] Build the initial structure: Use entities as nodes and relationships as edges to build the initial structure of the multi-relationship graph.

[0081] Step 211, for each node in the multi-relation graph, the corresponding word frequency vector and co-occurrence matrix information are mapped to the node according to the appearance of the entity it represents in the text. The values ​​in the word frequency vector and co-occurrence matrix are used as attributes of the node, or the information in multiple texts is integrated into a single node in some way (such as weighted average). After mapping, each node contains the text semantic information related to it, forming the initial semantic representation of the node.

[0082] Step 212, calculate the degree of each node in the multi-relational graph (i.e., the number of edges connected to the node), reflecting the importance of the node in the graph. Calculate the local clustering coefficient of each node, reflecting the closeness between the node and its neighboring nodes. Concatenate the structural features (degree centrality, local clustering coefficient) of the node with the semantic features (word frequency vector, information after co-occurrence matrix mapping) to form a composite feature vector containing text and structural information.

[0083] Step 213, linearly project the composite feature vector to map it from a high-dimensional space to a low-dimensional space, which can be achieved through matrix multiplication, where the projection matrix can be obtained through training. The projected features are subjected to nonlinear transformation (such as using an activation function) to increase the expressive power of the model. After linear projection and nonlinear transformation, a low-dimensional representation vector of the entity is obtained, which combines text and structural features.

[0084] Suppose there is a dataset about the automobile manufacturing industry chain, which contains text data (such as corporate news, process descriptions, etc.) and structured data (such as supply and demand relationships and collaborative relationships between companies).

[0085] Perform word segmentation and part-of-speech tagging on the news text, extract word frequency vectors and co-occurrence matrices. For example, for the news "A certain automobile company reached a cooperation agreement with its supplier", after word segmentation, it becomes "A certain automobile company reached a cooperation agreement with its supplier", and calculate word frequency and co-occurrence relationships.

[0086] Extract entities: automobile companies, suppliers.

[0087] Extraction relationship: supply and demand relationship (automotive companies-suppliers);

[0088] Construct the initial graph structure.

[0089] Map the word frequency and co-occurrence information in the news to the "automobile enterprise" and "supplier" nodes in the graph. Calculate the degree centrality and local clustering coefficient of the node, and concatenate the structural features with the semantic features to form a composite feature vector. Perform linear projection and nonlinear transformation on the composite feature vector to obtain the low-dimensional representation vector of "automobile enterprise" and "supplier".

[0090] By fusing text and structural features, the generated low-dimensional representation vector can more comprehensively reflect the characteristics and associations of entities and improve the expressiveness of the model. Low-dimensional representation vectors reduce the dimension of data, reduce the risk of overfitting, and enhance the generalization ability of the model. Low-dimensional representation vectors facilitate subsequent machine learning tasks such as similarity calculation, cluster analysis, and classification, providing strong support for the analysis and decision-making of the industrial chain. Through dimensionality reduction processing, the storage and computing costs of data are reduced, and the analysis efficiency is improved. The fusion of text and structural features helps to discover the potential associations between entities and provides new ideas for the optimization and upgrading of the industrial chain.

[0091] In a preferred embodiment of the present invention, the above step 3, constructing an industrial chain ontology according to the low-dimensional representation of the entity, including structured data, semi-structured text and unstructured building information model, and constructing an industrial chain knowledge graph according to the industrial chain ontology, may include:

[0092] Step 310, defining an ontology model of the industrial chain based on the low-dimensional representation of the entity, including structured data, semi-structured text, and unstructured building information model;

[0093] Step 311, map the entities, relationships and attributes in the ontology model to the nodes, edges and attribute fields of the graph database to generate an industrial chain knowledge graph containing node, edge and relationship weights.

[0094] In an embodiment of the present invention, the low-dimensional representation vectors of the entities generated in the previous steps are read and analyzed. These vectors are obtained by semantically vectorizing text data and embedding nodes in multi-relationship graphs, integrating text and structural features. According to the entity types and their associations revealed by the low-dimensional representation of the entities, the ontology model framework of the industrial chain begins to be constructed. This framework needs to cover structured data (such as corporate information, transaction records, product specifications, etc.), semi-structured text (such as process descriptions, technical document summaries, market analysis reports, etc.) and unstructured building information models (such as 3D factory layouts, equipment models, process flow charts, etc.).

[0095] Within the framework of the ontology model, clearly define various entities (such as enterprises, equipment, processes, raw materials, etc.), the relationships between them (such as supply and demand relationships, production collaboration relationships, technical dependencies, etc.), and the attribute fields of each entity and relationship (such as enterprise name, equipment model, process parameters, transaction amount, etc.). By comparing the low-dimensional representation of the entity with the definition of the ontology model, the accuracy and completeness of the model are verified. If inconsistencies or omissions are found, the machine will automatically adjust the definition of the ontology model to ensure that the model can accurately reflect the actual structure and characteristics of the industrial chain.

[0096] Step 311, select a suitable graph database (such as Neo4j, ArangoDB, JanusGraph, etc.) to store and query the industrial chain knowledge graph. The graph database can efficiently process the relationship between nodes and edges, and is suitable for storing and querying complex knowledge graphs. Map each entity in the ontology model to a node in the graph database. Each node contains all the attribute fields of the entity and is used to store detailed information of the entity. Map each relationship in the ontology model to an edge in the graph database. Each edge connects two nodes and represents the relationship between them. At the same time, the edge also contains relationship type and weight information to describe the nature and closeness of the relationship. The relationship weight can be set according to the association strength or similarity information in the low-dimensional representation of the entity. After mapping entities to nodes, relationships to edges, the machine generates an industrial chain knowledge graph containing nodes, edges and relationship weights. This graph graphically displays the entities and relationships in the industrial chain.

[0097] Suppose there is a task to build a knowledge graph about the automobile manufacturing industry chain.

[0098] Based on the low-dimensional representation of entities, an ontology model containing entities such as automobile manufacturers, parts suppliers, production equipment, and process flows is defined. At the same time, relationship types such as supply and demand relationships, production collaboration relationships, and technical dependencies are defined, as well as attribute fields such as company name, equipment model, process parameters, and transaction amount. Neo4j is selected as the graph database, and the entities in the ontology model are mapped to nodes and the relationships are mapped to edges. For example, "a certain automobile manufacturer" is mapped to a node, including attributes such as company name, address, and contact information; "a certain parts supplier" is mapped to another node, including attributes such as supplier name, product specifications, and price; "the supply and demand relationship between a certain automobile manufacturer and a certain parts supplier" is mapped to an edge, including relationship type and weight information (such as transaction frequency, transaction amount, etc.). After mapping, a knowledge graph of the automobile manufacturing industry chain containing multiple nodes and edges is generated. This graph graphically displays the entities and relationships in the automobile manufacturing industry chain.

[0099] By constructing the industrial chain ontology and knowledge graph, it is possible to integrate and manage various types of data such as structured data, semi-structured text and unstructured building information models, and realize comprehensive integration and unified management of data. The industrial chain knowledge graph graphically displays the entities and relationships in the industrial chain, making the data more intuitive and easy to understand. At the same time, the efficient query and traversal functions of the graph database can quickly locate and analyze the key entities and relationships in the industrial chain, improving the efficiency of data analysis. The industrial chain knowledge graph can reveal the potential associations and laws between entities, and provide new ideas and directions for the optimization and upgrading of the industrial chain. For example, by analyzing the supply and demand relationship and the production collaboration relationship, the bottleneck links and potential risk points in the industrial chain can be found. The analysis results based on the industrial chain knowledge graph can provide scientific decision-making suggestions and support for decision makers. At the same time, through the mining and analysis of historical data, the future development trend and potential risks of the industrial chain can also be predicted, providing strong support for the strategic planning and risk management of enterprises. The construction of the industrial chain knowledge graph helps to promote the collaboration and cooperation between upstream and downstream enterprises in the industrial chain, and promote technological innovation and industrial upgrading. For example, by sharing information and resources in the knowledge graph, enterprises can carry out R&D cooperation and market expansion activities more efficiently.

[0100] In another preferred embodiment of the present invention, in the industrial chain knowledge graph, nodes represent entities in the industrial chain, including enterprises, equipment, and processes, and edges represent relationships between entities, including supply and demand, collaboration, and dependency, which may include:

[0101] In the embodiment of the present invention, the nodes in the industrial chain knowledge graph represent entities in the industrial chain, including enterprises, equipment, processes, etc. Each node contains detailed attribute information, such as enterprise name, address, contact information, etc.; equipment model, specifications, manufacturer, etc.; process name, parameters, steps, etc.

[0102] Edge representation: Edges represent the relationship between entities, including supply and demand relationships (such as the relationship between an enterprise and its suppliers), collaborative relationships (such as cooperation or alliances between enterprises), dependency relationships (such as the dependency of a process on equipment or raw materials), etc. Each edge contains relationship type and weight information to describe the nature and closeness of the relationship.

[0103] In a preferred embodiment of the present invention, the above step 4, based on the industrial chain knowledge graph, dynamically optimizes the production line configuration and task allocation strategy through a two-stage deep Q network, and updates the optimization results to the node attributes and relationship weights of the knowledge graph to generate an optimized dynamic industrial chain knowledge graph, which may include:

[0104] Step 410, based on the current state of the industrial chain knowledge graph, a two-stage deep Q network is used to simulate the combination of production line configuration adjustment and task allocation strategy;

[0105] Step 411, using a two-stage deep Q network to dynamically adjust and optimize the production line configuration and task allocation strategy in the knowledge graph environment, calculating the target Q value and updating the parameters in the iterative learning process to determine the final combination of production line configuration and task allocation strategy;

[0106] Step 412, updating the attributes of the equipment nodes in the knowledge graph according to the final production line configuration and task allocation strategy combination;

[0107] Step 413, after the attributes of the device node are updated, the relationship weights between the nodes in the knowledge graph are re-evaluated and adjusted according to the new task allocation strategy;

[0108] Step 414, integrate all updated node attributes and relationship weights to generate an optimized dynamic industry chain knowledge graph.

[0109] In an embodiment of the present invention, the current state is read from the industrial chain knowledge graph, including the attributes of the equipment nodes (such as equipment type, production capacity, maintenance status, etc.), the attributes of the task nodes (such as task type, task volume, priority, etc.), and the relationship weights between nodes (such as supply and demand relationship, collaboration relationship, etc.). Initialize the model parameters of the double-stage deep Q network (Double DQN), including the weights of the Q network and the target Q network. The double-stage deep Q network reduces the problem of overestimation of the Q value and improves the stability of learning by introducing two Q networks. Using the double-stage deep Q network, different combinations of production line configuration adjustments and task allocation strategies are simulated according to the current state of the knowledge graph. Each strategy combination corresponds to a Q value, which represents the expected return of the strategy combination.

[0110] Step 411, execute the current strategy combination in the knowledge graph environment, observe the environmental feedback (such as production line production efficiency, task completion time, etc.), and calculate the immediate reward based on the feedback. Select the next action (i.e., production line configuration adjustment or task allocation strategy) according to the ε-greedy strategy to balance exploration and exploitation.

[0111] The target Q network is used to calculate the target Q value of the next state, and the parameters of the Q network are updated by the gradient descent method based on the difference between the current Q value and the target Q value. During the iterative learning process, the machine continuously repeats the above steps to gradually optimize the production line configuration and task allocation strategy. After multiple iterations, the final combination of production line configuration and task allocation strategy is determined, which has the highest expected return.

[0112] Step 412, parse the final combination of production line configuration and task allocation strategy, extract the configuration information related to the equipment node (such as equipment enable / disable status, equipment production parameter adjustment, etc.). According to the parsing results, update the attributes of the equipment node in the knowledge graph. For example, update the production status of a certain equipment from "idle" to "enabled", and adjust its production parameters to adapt to the new task requirements.

[0113] Step 413, analyze the new task allocation strategy and identify the affected node pairs (such as device and task nodes, device and device nodes, etc.). Re-evaluate the relationship weights between the affected nodes according to the new task allocation strategy. For example, if a device is assigned more tasks, the supply and demand relationship weights between it and the relevant task nodes may increase. According to the re-evaluation results, adjust the relationship weights between the nodes in the knowledge graph to reflect the impact of the new task allocation strategy on the industrial chain structure.

[0114] Step 414: All updated device node attributes, task node attributes, and relationship weights between nodes are integrated to form a complete knowledge graph data. The integrated data is used to generate an optimized dynamic industry chain knowledge graph. The graph reflects the latest production line configuration and task allocation strategies, as well as their impact on the industry chain structure.

[0115] Suppose there is an optimization task about an intelligent manufacturing production line.

[0116] The current status is read from the industry chain knowledge graph, including the production capacity, maintenance status, and current task allocation of each device. A two-stage deep Q network is used to simulate different combinations of production line configuration adjustments and task allocation strategies, and the strategy combination is optimized through iterative learning. For example, it may be found that enabling an idle device and assigning it to a high-priority task can improve overall production efficiency. According to the optimized strategy combination, the attributes of the device nodes in the knowledge graph are updated. For example, the status of a device is updated from "idle" to "enabled", and its production parameters are adjusted. The new task allocation strategy is analyzed and the relationship weights between the affected nodes are re-evaluated. For example, if a device is assigned more tasks, the weight of the supply and demand relationship between it and the relevant task nodes increases. All updated node attributes and relationship weights are integrated to generate an optimized dynamic industry chain knowledge graph. The graph reflects the latest production line configuration and task allocation strategies, as well as their impact on the structure of smart manufacturing production lines.

[0117] The dynamic optimization of production line configuration and task allocation strategy through the two-stage deep Q network can significantly improve the production efficiency and task completion speed of the production line. The optimized industrial chain knowledge graph reflects the latest production line configuration and task allocation strategy, which helps to enhance the collaboration and flexibility between upstream and downstream enterprises in the industrial chain and improve the overall response speed. Based on the optimized knowledge graph, scientific decision-making suggestions and support can be provided to decision makers. At the same time, through the mining and analysis of historical data, the machine can also predict the future development trend and potential risks of the industrial chain. The construction of the dynamic industrial chain knowledge graph provides strong support for the continuous innovation and optimization of the industrial chain. Enterprises can continuously adjust and optimize the production line configuration and task allocation strategy based on the information and suggestions in the knowledge graph to adapt to market changes and customer needs. The optimized knowledge graph graphically displays the entities and relationships in the industrial chain, making the data more intuitive and easy to understand. At the same time, the efficient query and traversal functions of the graph database can quickly locate and analyze the key entities and relationships in the industrial chain, improving the efficiency of data analysis.

[0118] In another preferred embodiment of the present invention, the above step 411 uses a two-stage deep Q network to dynamically adjust and optimize the production line configuration and task allocation strategy in the knowledge graph environment, calculates the target Q value and updates the parameters in the iterative learning process to determine the final combination of production line configuration and task allocation strategy, including:

[0119] Calculate the difference ratio of the efficiency between the new state and the old state, the cost difference between the old state and the new state, and the difference ratio of the resource utilization between the new state and the old state, and get an immediate reward;

[0120] Predict the next state based on the current state, and select the action that maximizes the Q value predicted by the current Q network as the final action under the corresponding predicted state;

[0121] Input the future state and final action into the target Q network to obtain the Q value prediction of the future state;

[0122] The immediate reward and the Q-value prediction of the future state are combined to obtain the target Q-value.

[0123] In the embodiment of the present invention, the parameters of the initialization Q network and the target Q network are and , yes A delayed copy of . Set the discount factor =0.9, indicating the current value of future rewards, sets the weight coefficient =1.0, =1.5, =1.2, which respectively indicate the importance of efficiency, cost and resource utilization in reward calculation. For each state transition, calculate the immediate reward ;in, and are the efficiencies in the new state and the old state, respectively. and are the costs in the new state and the old state respectively, and The resource utilization in the new state and the old state respectively. Use the current Q network to predict the next state The Q value of , and select the action that maximizes the Q value, that is . The future state and the selected final action Input into the target Q network to get the Q value prediction of the future state ,in, is in a given state and actions Next, the current Q network (whose parameters are ) predicted Q value.

[0124] The target Q value is obtained by combining the immediate reward and the Q value prediction of the future state, that is, the target Q value = .

[0125] Use the error between the target Q value and the current Q network's prediction to update the parameters of the Q network , using the gradient descent method. Every certain number of steps, the parameters of the Q network are Parameters copied to the target Q network In order to maintain the stability of the target Q network. By calculating immediate rewards and future rewards in real time, the two-stage deep Q network can dynamically adjust the production line configuration and task allocation strategy to adapt to the ever-changing production environment. By optimizing the reward calculation of efficiency, cost and resource utilization, the system can tend to choose actions that can improve production efficiency and resource utilization, thereby reducing production costs. The two-stage deep Q network reduces the deviation of Q value estimation by introducing the target Q network, thereby improving the scientificity and accuracy of decision-making. At the same time, the use of discount factors enables the system to balance the importance of immediate rewards and future rewards and make more long-term decisions. Through continuous iterative learning and updating of parameters, it can gradually adapt to different production environments and task requirements, and improve the adaptability and robustness of the system.

[0126] In a preferred embodiment of the present invention, the above step 5, based on the optimized dynamic industrial chain knowledge graph, calculates the node importance weights through the graph attention network, and combines the risk propagation model to perform industrial trend prediction, risk propagation path analysis and resource bottleneck identification, and finally generates decision recommendations, which may include:

[0127] Step 510, extracting node features, edge features and topological structure information from the optimized dynamic industrial chain knowledge graph;

[0128] Step 511, using a graph attention network to process node features, edge features, and topological structure information, automatically learning the association strength between nodes, and assigning an importance weight to each node;

[0129] Step 512, combining the node importance weight with a predefined risk propagation model to evaluate the risk propagation path and potential impact in the industrial chain, and obtaining a risk propagation analysis result;

[0130] Step 513, predicting the future development trend of the industrial chain based on the risk propagation analysis results, including identifying key nodes and potential risk points in the industrial chain, and obtaining industrial trend prediction results;

[0131] Step 514, analyze the node importance weights, risk propagation models and industry trend forecast results, identify key risk propagation nodes and paths, as well as resource bottleneck nodes and potential resource shortage problems in the industrial chain, and generate decision-making recommendations, including adjustments to production line configurations, optimization of task allocation, formulation of risk prevention and control measures, and planning of resource allocation and scheduling.

[0132] In the embodiment of the present invention, each node in the knowledge graph is traversed to extract attribute information related to the node, such as the type of node (enterprise, product, technology, etc.), the size of the node (such as the number of employees and output value of the enterprise), and the state of the node (such as the operating status of the enterprise, the market demand for the product, etc.). For each edge in the knowledge graph, the type of edge (such as supply relationship, cooperative relationship, etc.), the weight of the edge (indicating the strength or frequency of the relationship) and other information are extracted, the connection relationship between the nodes is recorded, and an adjacency matrix is ​​constructed to represent the topological structure of the knowledge graph.

[0133] Step 511, construct a graph attention network (GAT) model, taking node features and edge features as input. Through the self-attention mechanism of the GAT model, the association strength between nodes is automatically learned, that is, the importance contribution of each node to its neighboring nodes. According to the learned association strength, an importance weight is assigned to each node to indicate the relative importance of the node in the knowledge graph.

[0134] Step 512, define a risk propagation model, such as a variant based on the SIR (Susceptible-Infected-Recovered) model or the SEIR (Susceptible-Exposed-Infected-Recovered) model, and consider the impact of the importance weight of the node on the risk propagation. Input the node importance weight into the risk propagation model, simulate the risk propagation process in the industrial chain, and record the infection status and propagation path of each node. Based on the simulation results, evaluate the potential impact of the risk on each node in the industrial chain, such as production interruption, demand decline, etc.

[0135] Step 513, based on the risk propagation analysis results, identify the key nodes that have the greatest impact on the overall stability of the industrial chain. Analyze the risk propagation path and potential impact, and identify potential risk points in the industrial chain, such as supply chain disruptions, changes in market demand, etc. Combined with the analysis of key nodes and potential risk points, predict the future development trend of the industrial chain, such as the market demand for certain products will increase, and certain technologies will be gradually eliminated.

[0136] Step 514, analyze the risk propagation analysis results, identify key risk propagation nodes and paths, and identify resource bottleneck nodes in the industrial chain, i.e., those nodes that have the greatest impact on the overall operating efficiency of the industrial chain, by combining the node importance weights and industry trend forecast results. Analyze the resource demand and supply of resource bottleneck nodes to identify potential resource shortages. Based on the analysis of risk propagation nodes and paths, resource bottleneck nodes, and potential resource shortages, generate decision recommendations, such as adjusting production line configuration, optimizing task allocation, formulating risk prevention and control measures, and planning resource allocation and scheduling.

[0137] Suppose there is a dynamic knowledge graph about the automotive industry chain, which contains nodes such as automakers, parts suppliers, and dealers, as well as edges such as supply relationships and cooperation relationships between them. Extract the characteristics of each node (such as enterprise size, product type, etc.), edge characteristics (such as the strength of the supply relationship), and topological structure information (such as the adjacency matrix). Use the graph attention network to process this information and assign an importance weight to each node. For example, large automakers may have higher importance weights.

[0138] Combine the node importance weights with the risk propagation model to simulate the propagation path and potential impact of risks (such as supply chain disruptions) in the industry chain. For example, if a key component supplier has a problem, it may cause production disruptions for multiple automakers. Based on the results of the risk propagation analysis, predict the future development trend of the automotive industry chain. For example, it is predicted that the market demand for some new energy vehicles will increase, while the market demand for some traditional fuel vehicles will decrease. Analyze the node importance weights, risk propagation models, and industry trend forecast results to identify key risk propagation nodes and paths (such as some key component suppliers), resource bottleneck nodes (such as some manufacturers with limited production capacity), and potential resource shortages (such as insufficient supply of some key components). Then generate decision recommendations, such as increasing the inventory of key components, optimizing production line configuration to increase production capacity, etc.

[0139] By combining the graph attention network and the risk propagation model, the importance of nodes and risk propagation paths can be more accurately evaluated, thereby generating more scientific decision-making recommendations. By identifying key risk propagation nodes and paths, risk prevention and control measures can be formulated in advance to reduce the impact of risks on the industrial chain. By identifying resource bottleneck nodes and potential resource shortages, resource allocation and scheduling can be planned more reasonably to improve resource utilization efficiency. By predicting industry trends and identifying key nodes, enterprises in the industrial chain can be guided to strengthen cooperation, achieve coordinated development, and improve the competitiveness of the entire industrial chain.

[0140] like Figure 2 As shown, an embodiment of the present invention further provides an industrial chain graph modeling system based on deep learning, including:

[0141] The data processing module is used to collect multi-source heterogeneous data from the industrial chain, including demand information, resource status, process characteristics and operation and maintenance data, and pre-process the multi-source heterogeneous data to form a standardized data set;

[0142] The knowledge fusion module is used to perform semantic vectorization on text data in the standardized data set and embed nodes in the multi-relation graph to generate a low-dimensional representation of entities that integrates text and structural features;

[0143] The graph construction module is used to construct the industry chain knowledge graph using the low-dimensional representation of entities that integrate text and structural features, where nodes represent different entities in the industry chain and edges represent the relationships between entities;

[0144] Reinforcement learning module, which is used to dynamically optimize production line configuration and task allocation strategy through a two-stage deep Q network to generate an optimized dynamic industry chain knowledge graph;

[0145] The decision support module is used to forecast industry trends, analyze risk propagation paths and identify resource bottlenecks based on the optimized industry chain knowledge graph, and ultimately generate decision recommendations.

[0146] It should be noted that the system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0147] The embodiment of the present invention further provides a computing device, comprising: a processor, a memory storing a computer program, wherein when the computer program is executed by the processor, the method described above is executed. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0148] The embodiment of the present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the method described above. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0149] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for modeling an industrial chain graph based on deep learning, characterized in that: The method comprises: Collect multi-source heterogeneous data in the industrial chain, including demand information, resource status, process characteristics and operation and maintenance data, and pre-process the multi-source heterogeneous data to form a standardized data set; Perform semantic vectorization on text data in standardized datasets and embed nodes in multi-relational graphs to generate low-dimensional representations of entities that integrate text and structural features. According to the low-dimensional representation of entities, the industrial chain ontology is constructed, including structured data, semi-structured text and unstructured building information model, and the industrial chain knowledge graph is constructed according to the industrial chain ontology; According to the industrial chain knowledge graph, the two-stage deep Q network is used to dynamically optimize the production line configuration and task allocation strategy, and the optimization results are updated to the node attributes and relationship weights of the knowledge graph to generate an optimized dynamic industrial chain knowledge graph; Based on the optimized dynamic industrial chain knowledge graph, the node importance weights are calculated through the graph attention network, and combined with the risk propagation model to perform industrial trend prediction, risk propagation path analysis and resource bottleneck identification, and finally generate decision recommendations.

2. The industrial chain graph modeling method based on deep learning according to claim 1 is characterized in that: Semantic vectorization is performed on text data in the standardized dataset, and nodes are embedded in the multi-relational graph to generate low-dimensional representations of entities that integrate text and structural features, including: Perform word segmentation and part-of-speech tagging on the text data in the standardized data set, and extract the preliminary semantic features of the text, including word frequency vectors and co-occurrence matrices; extract entities from the standardized data set, including enterprises, equipment, processes, and associations, including supply and demand relationships, and collaborative relationships, and construct the initial structure of the multi-relationship graph; Map the word frequency vectors and co-occurrence matrix to the corresponding nodes of the multi-relation graph, and add preliminary semantic representation to each node; Calculate the structural features of each node in the multi-relational graph, including degree centrality and local clustering coefficient features, and combine the structural features with the semantic features to form a composite feature vector containing text and structural information; A linear projection operation is performed on the composite feature vector, and a nonlinear transformation is performed on the projected features to generate a low-dimensional representation vector of the entity.

3. The industrial chain graph modeling method based on deep learning according to claim 2 is characterized in that: According to the low-dimensional representation of entities, the industrial chain ontology is constructed, including structured data, semi-structured text and unstructured building information model, and the industrial chain knowledge graph is constructed according to the industrial chain ontology, including: Based on the low-dimensional representation of entities, define the ontology model of the industrial chain, including structured data, semi-structured text and unstructured building information model; Map the entities, relationships and attributes in the ontology model to the nodes, edges and attribute fields of the graph database to generate an industrial chain knowledge graph containing node, edge and relationship weights.

4. The industrial chain graph modeling method based on deep learning according to claim 3 is characterized in that: In the industrial chain knowledge graph, nodes represent entities in the industrial chain, including enterprises, equipment, and processes, and edges represent relationships between entities, including supply and demand, collaboration, and dependency.

5. The industrial chain graph modeling method based on deep learning according to claim 4 is characterized in that: According to the industrial chain knowledge graph, the two-stage deep Q network is used to dynamically optimize the production line configuration and task allocation strategy, and the optimization results are updated to the node attributes and relationship weights of the knowledge graph to generate an optimized dynamic industrial chain knowledge graph, including: According to the current state of the industry chain knowledge graph, a two-stage deep Q network is used to simulate the combination of production line configuration adjustment and task allocation strategy; The two-stage deep Q network is used to dynamically adjust and optimize the production line configuration and task allocation strategy in the knowledge graph environment. The target Q value is calculated and the parameters are updated during the iterative learning process to determine the final combination of production line configuration and task allocation strategy. Update the attributes of the equipment nodes in the knowledge graph based on the final combination of production line configuration and task allocation strategy; After the attributes of the device nodes are updated, the relationship weights between the nodes in the knowledge graph are re-evaluated and adjusted according to the new task allocation strategy; Integrate all updated node attributes and relationship weights to generate an optimized dynamic industrial chain knowledge graph.

6. The method for industrial chain graph modeling based on deep learning according to claim 5 is characterized in that: The two-stage deep Q network is used to dynamically adjust and optimize the production line configuration and task allocation strategy in the knowledge graph environment. The target Q value is calculated and the parameters are updated during the iterative learning process to determine the final combination of production line configuration and task allocation strategy, including: Calculate the difference ratio of the efficiency between the new state and the old state, the cost difference between the old state and the new state, and the difference ratio of the resource utilization between the new state and the old state, and get an immediate reward; Predict the next state based on the current state, and select the action that maximizes the Q value predicted by the current Q network as the final action under the corresponding predicted state; Input the future state and final action into the target Q network to obtain the Q value prediction of the future state; The immediate reward and the Q-value prediction of the future state are combined to obtain the target Q-value.

7. The industrial chain graph modeling method based on deep learning according to claim 6 is characterized in that: According to the optimized dynamic industry chain knowledge graph, the node importance weights are calculated through the graph attention network, and the risk propagation model is combined to predict industry trends, analyze risk propagation paths, and identify resource bottlenecks, and finally generate decision recommendations, including: Extract node features, edge features and topological structure information from the optimized dynamic industry chain knowledge graph; The graph attention network is used to process node features, edge features and topological structure information, automatically learn the association strength between nodes, and assign an importance weight to each node; Combine the node importance weights with the predefined risk propagation model to evaluate the risk propagation path and potential impact in the industrial chain and obtain the risk propagation analysis results; According to the results of risk propagation analysis, the future development trend of the industrial chain is predicted, including identifying key nodes and potential risk points in the industrial chain to obtain the industrial trend prediction results; Analyze the node importance weights, risk propagation models and industry trend forecast results, identify key risk propagation nodes and paths, as well as resource bottleneck nodes and potential resource shortages in the industrial chain, and generate decision-making recommendations, including adjustments to production line configurations, optimization of task allocation, formulation of risk prevention and control measures, and planning of resource allocation and scheduling.

8. An industrial chain graph modeling system based on deep learning, the system implements the method according to any one of claims 1 to 7, characterized in that: include: The data processing module is used to collect multi-source heterogeneous data from the industrial chain, including demand information, resource status, process characteristics and operation and maintenance data, and pre-process the multi-source heterogeneous data to form a standardized data set; The knowledge fusion module is used to perform semantic vectorization on text data in the standardized data set and embed nodes in the multi-relation graph to generate a low-dimensional representation of entities that integrates text and structural features; The graph construction module is used to construct the knowledge graph of the industrial chain by using the low-dimensional representation of entities that integrate text and structural features. The nodes represent different entities in the industrial chain, and the edges represent the relationships between entities. Reinforcement learning module, which is used to dynamically optimize production line configuration and task allocation strategies through a two-stage deep Q network to generate an optimized dynamic industry chain knowledge graph; The decision support module is used to forecast industry trends, analyze risk propagation paths and identify resource bottlenecks based on the optimized industry chain knowledge graph, and ultimately generate decision recommendations.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images