Chemical industrial park industrial chain data knowledge graph construction method and system based on big data

Abstract

The invention relates to the technical field of knowledge graph construction, in particular to a chemical industry park industrial chain data knowledge graph construction method and system based on big data. The method comprises the following steps: acquiring industrial chain data of a chemical industry park; nodes and edges of the knowledge graph are defined, an evolution graph convolutional network model is used for predicting the number change relation of adjacent nodes and placing the number change relation in the edges, and a product reserve relation knowledge graph is constructed; introducing timestamps to update node definitions, converting the product reserve relationship knowledge graph into a product time relationship knowledge graph by using a dynamic time sequence graph network algorithm, and combining repeated routes and products of the chemical industry park to obtain a final product time relationship knowledge graph; and displaying the product reserve relationship knowledge graph and the product time relationship knowledge graph by using a visual interface. According to the chemical industry park industrial chain data knowledge graph construction method and system based on big data, time sequence management and dynamic optimization of a product reserve relationship are realized by constructing two knowledge graphs.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge graph construction, and more specifically, to a method and system for constructing a knowledge graph of chemical industrial park industrial chain data based on big data. Background Art

[0002] The method and system for constructing a knowledge graph of chemical industrial park industrial chain data based on big data aims to improve the data processing efficiency of the chemical industrial park industrial chain and optimize the resource scheduling and decision-making support of the industrial chain. By using an evolutionary graph convolutional network model to calculate and analyze the change in production capacity relationship caused by the change in the number of adjacent chemical products with production up and down relationships, a knowledge graph of product reserve relationships with the function of reflecting the change in the number relationship of adjacent chemical products and real-time update is constructed, and a time stamp is introduced on the basis of the knowledge graph of product reserve relationships to construct a knowledge graph of product time relationships that can reflect the production sequence and interdependence between different chemical products. The two knowledge graphs are stored independently and cooperate with each other to realize intelligent storage management of chemical industrial park industrial chain data, dynamic update of product reserves, and on-demand query of product resources.

[0003] Existing knowledge graph construction methods and systems are usually static knowledge graphs, which can only simply store industrial chain data, and cannot handle complex industrial chain relationships that change over time and changes in product relationships caused by dynamic updates of product quantities. Moreover, due to the strong temporal and interdependent nature of chemical industrial park industrial chain data, ordinary knowledge graphs cannot update product quantities in real time, predict relationship changes from data changes, and reflect the upstream and downstream relationships between products, which will lead to the problem that when multiple employees query at the same time, the required product status cannot be queried in real time, resulting in product resource shortage conflicts. Therefore, a method and system for constructing a knowledge graph of chemical industrial park industrial chain data based on big data are provided. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for constructing a knowledge graph of chemical industrial park industrial chain data based on big data, so as to solve the problem proposed in the above background art that due to the strong temporal and interdependent nature of chemical industrial park industrial chain data, ordinary knowledge graphs cannot update product quantities in real time, predict relationship changes from data changes, and reflect the upstream and downstream relationships between products, which will lead to the problem that when multiple employees query at the same time, the required product status cannot be queried in real time, resulting in product resource shortage conflicts.

[0005] To achieve the above object, the present invention aims to provide a method for constructing a knowledge graph of chemical industrial park industrial chain data based on big data, including:

[0006] S1. Obtain chemical industrial park industrial chain data from the industrial database, and use natural language processing technology to extract industrial chain product entity information and product relationship information;

[0007] S2. Define the nodes and edges of the knowledge graph based on the product entity information and product relationship information, use the evolving graph convolutional network model to predict the change relationship of the number of adjacent nodes and place it in the edge, and construct the product reserve relationship knowledge graph;

[0008] The evolving graph convolutional network model is implemented based on the graph convolutional network integrated with the multi-layer perceptron model, and is used to consider the changes of nodes and edges over time to handle the situation where the graph structure changes over time, and to predict the impact of the change in the number of nodes on the edges;

[0009] S3. Based on the product reserve relationship knowledge graph, introduce timestamps to update the node definition, use the dynamic time-series graph network algorithm to transform the product reserve relationship knowledge graph into a product time relationship knowledge graph, and merge the duplicate routes and products in the chemical industrial park to obtain the final product time relationship knowledge graph;

[0010] S4. Use the visualization interface to display the product reserve relationship knowledge graph and the product time relationship knowledge graph for users to perform query operations according to their needs.

[0011] As a further improvement of this technical solution, in S1, the product entity information includes the Chinese and English names, quantity, molecular weight, molecular formula, structural formula, production capacity, appearance, physical properties, specifications, process information, market price, and suppliers of each product;

[0012] The product relationship information includes raw material supply relationship, production relationship, consumption relationship, substitution relationship, and transportation relationship.

[0013] As a further improvement of this technical solution, in S2, based on the product entity information and product relationship information, define the nodes and edges of the knowledge graph, use the evolving graph convolutional network model to predict the change relationship of the number of adjacent nodes and place it in the edge, and construct the product reserve relationship knowledge graph. The specific method steps are as follows:

[0014] S2.1. Define product entity nodes based on product entity information;

[0015] S2.2. Define product relationship edges based on product relationship information;

[0016] S2.3. Use the evolving graph convolutional network model to predict the production capacity relationship caused by the change in the number of adjacent chemical products with production up and down relationships and place it in the corresponding product relationship edge;

[0017] S2.4. Based on the product entity nodes and product relationship edges, construct the product reserve relationship knowledge graph.

[0018] As a further improvement of this technical solution, in S2.1, define product entity nodes based on product entity information. The specific method is as follows:

[0019] N i = (P i , Q i , X i );

[0020] Among them, i is the product entity index; N i is the i-th product entity node; P i is the name of the i-th product entity; Q i is the quantity of the i-th product entity; X i are the features of the i-th product entity other than the name and quantity.

[0021] In S2.2, define the product relationship edge based on the product relationship information. The specific method is as follows:

[0022] E ij = (N i , N j , R ij , w ij , ΔQ ij );

[0023] Among them, j is the product entity index; N j is the j-th product entity node; E ij is the product relationship edge between the i-th product entity node and the j-th product entity node; R ij is the product relationship information; w ij is the product relationship edge weight; ΔQ ij is the product relationship change information.

[0024] As a further improvement of this technical solution, in S2.3, use the evolutionary graph convolutional network model to predict the production capacity relationship caused by the change in the quantity of adjacent chemical products with production up and down relationships and place it in the corresponding product relationship edge. The specific method steps are as follows:

[0025] S2.3.1. Set the time step as t, and each product entity node has a corresponding quantity at each time step t

[0026] S2.3.2. Based on the graph convolutional network model, aggregate the feature information of neighbor nodes and calculate the new feature representation of each product entity node:

[0027]

[0028] Among them, X i (t+1) are the features of the i-th product entity node at time step t + 1 other than the name and quantity; μ is the index of the neighbor node; is the set of neighbor nodes of the i-th product entity node; Aij is the adjacency matrix; d i is the degree of the i-th product entity node; X μ (t) are the features of neighbor node μ except for name and quantity; W is the weight matrix; σ(·) is the Sigmoid non-linear activation function;

[0029] S2.3.3. Based on the features of the i-th product entity node except for name and quantity at time step t + 1, fuse the multi-layer perceptron model to predict the quantity change of the product entity at time step t + 1:

[0030]

[0031] where, is the product quantity change of the i-th product entity node at time step t + 1; X i (t+1) are the features of the i-th product entity node except for name and quantity at time step t + 1; X j (t+1) are the features of the j-th product entity node except for name and quantity at time step t + 1; A' ij is the node connection strength weight; MLP([·]) is the multi-layer perceptron model operation;

[0032] And similarly, the product quantity change based on the j-th product entity node at time step t + 1 can be obtained

[0033] S2.3.4. Based on the product quantity changes of the i-th product entity node and the j-th product entity node at time step t + 1 and Analyze the change in the production capacity relationship between the i-th chemical product and the adjacent j-th chemical product, and place it in the product relationship edge;

[0034] As a further improvement of this technical solution, in S2.3.4, based on the product quantity changes of the i-th product entity node and the j-th product entity node at time step t + 1 and Analyze the change in the production capacity relationship between the i-th chemical product and the adjacent j-th chemical product, and place it in the product relationship edge. The specific method is as follows:

[0035] S2.3.4.1. Based on the product relationship information R ij Combine the product quantity changes of the i-th product entity node and the j-th product entity node at time step t + 1 and Calculate the product relationship information at time step t + 1:

[0036]

[0037] Among them, is the product relationship information at time step t, which is equivalent to the product relationship information R ij ; The change in product relationship information at time step t + 1; γ1 is the weight coefficient of the i-th product entity; γ2 is the weight coefficient of the j-th product entity; γ3 is the product relationship weight coefficient; ∈ is the smoothing term;

[0038] S2.3.4.2. Update the product relationship edge weight of the product relationship edge with the change in product relationship information at time step t + 1:

[0039]

[0040] Among them, is the product relationship edge weight at time step t, which is equivalent to the product relationship edge weight w ij ; is the product relationship edge weight at time step t + 1; α is the weight coefficient of the product relationship edge weight; β is the weight coefficient of the change in product relationship information;

[0041] Based on the evolutionary graph convolutional network model, calculate and analyze the change in production capacity relationship caused by the change in the number of adjacent chemical products with production up and down relationships, and use it in the product reserve relationship knowledge graph to query in real time whether the production capacity relationship between adjacent chemical products can meet the current production demand, and also use it to update the quantity and production capacity change of each chemical product in real time;

[0042] In S2.4, based on the product entity node and the product relationship edge, construct a product reserve relationship knowledge graph, specifically as follows:

[0043] G1 = ({N1, N2, …, N i , …, N n}, {E' ij ∣1 ≤ i ≠ j ≤ n});

[0044]

[0045] Among them, G1 is the product reserve relationship knowledge graph; n is the total number of product entity nodes; E' ij is the product relationship edge between the i-th product entity node and the j-th product entity node after adding prediction data.

[0046] As a further improvement of this technical solution, in S3, based on the product reserve relationship knowledge graph, introduce a timestamp to update the node definition, use the dynamic time series graph network algorithm to transform the product reserve relationship knowledge graph into a product time relationship knowledge graph, and merge the duplicate routes and products in the chemical industrial park to obtain the final product time relationship knowledge graph. The specific method steps are as follows:

[0047] S3.1. Introduce timestamps to update the defined product entity nodes and product relationship edges:

[0048]

[0049] E” ij =(N i , N j , R ij , w ij );

[0050] Among them, is the i-th product entity node including the timestamp after update; E” ij is the product relationship edge between the i-th product entity node and the j-th product entity node after update; is the node timestamp;

[0051] S3.2. Use the dynamic time series graph network algorithm to transform the product reserve relationship knowledge graph into a product time relationship knowledge graph;

[0052] S3.3. Use the incremental hierarchical time series memory model to adaptively merge the duplicate routes and product entity nodes in the chemical industrial park to obtain the final product time relationship knowledge graph.

[0053] As a further improvement of this technical solution, in the above S3.2, using the dynamic time series graph network algorithm to transform the product reserve relationship knowledge graph into a product time relationship knowledge graph, the specific method steps are as follows:

[0054] S3.2.1. Use the graph neural network to perform time series embedding on each product entity node and product relationship edge:

[0055]

[0056] Among them, is the time series embedding of the i-th product entity node including the timestamp after update ; is the set of neighbor nodes of the i-th product entity node;

[0057]

[0058] Among them, is the time series embedding of the product relationship edge between the i-th product entity node and the j-th product entity node; is the edge timestamp;

[0059] S3.2.2. Combine the time series information of the product entity nodes and product relationship edges to obtain the product time relationship knowledge graph:

[0060]

[0061] Among them, G2 is the product time relationship knowledge graph; is the nth product entity node with a timestamp after update of the temporal embedding.

[0062] As a further improvement of this technical solution, in S3.3, an incremental hierarchical temporal memory model is used to adaptively merge the duplicate routes and product entity nodes in the chemical industrial park to obtain the final product time relationship knowledge graph. The specific method steps are as follows:

[0063] S3.3.1. At each time step, based on the temporal embeddings of each product entity node, calculate the cosine similarity between each pair of product entity nodes, and set a node similarity threshold δ to determine whether they can be merged:

[0064]

[0065] If then the ith product entity node and the jth product entity node are similar and can be merged, otherwise they cannot be merged;

[0066] Among them, is the cosine similarity between the ith product entity node and the jth product entity node; δ is the node similarity threshold;

[0067] S3.3.2. At each time step, based on the temporal embeddings of each product relationship edge, calculate the cosine similarity between each pair of product relationship edges, and set an edge similarity threshold τ to determine whether they can be merged:

[0068]

[0069] If then the product relationship edge and the product relationship edge are similar and can be merged, otherwise they cannot be merged;

[0070] Among them, is the cosine similarity between the product relationship edge and the product relationship edge ; τ is the edge similarity threshold; s is the product entity node index; l is the product entity node index; is the temporal embedding of the product relationship edge between the sth product entity node and the lth product entity node;

[0071] S3.3.3. Merge the similar product entity nodes and product relationship edges to obtain the final product time relationship knowledge graph.

[0072] On the other hand, the present invention provides a system for constructing a knowledge graph of chemical industrial park industrial chain data based on big data, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the method for constructing a knowledge graph of chemical industrial park industrial chain data based on big data according to any one of the above.

[0073] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0074] 1. In the method and system for constructing a knowledge graph of chemical industrial park industrial chain data based on big data, based on the evolutionary graph convolutional network model, it is possible to predict the quantity change of each product entity in the chemical industrial park industrial chain over time, dynamically capture the relationship evolution between product entity nodes in the knowledge graph, and optimize the resource allocation of the industrial chain according to this change, reducing the risks of redundant production and supply chain breakage.

[0075] 2. In the method and system for constructing a knowledge graph of chemical industrial park industrial chain data based on big data, by constructing a knowledge graph of product reserve relationships, the problem that the relationship changes with the output change between products cannot be presented in the static knowledge graph is solved. By constructing a knowledge graph of product time relationships, the problem that the upstream and downstream relationships of products cannot be expressed in the knowledge graph of product reserve relationships is made up for. The two knowledge graphs complement each other in terms of functional requirements, realizing the time sequence management and dynamic optimization of product reserve relationships, thereby improving the resource scheduling accuracy in the chemical industrial park industrial chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is the overall method flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0077] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0078] Embodiment 1: Please refer to Figure 1 As shown, this embodiment provides a method for constructing a knowledge graph of chemical industrial park industrial chain data based on big data, including the following steps:

[0079] S1. Obtain chemical industrial park industrial chain data from the industrial database, and use natural language processing technology to extract industrial chain product entity information and product relationship information;

[0080] In S1, the product entity information includes the Chinese and English names, quantity, molecular weight, molecular formula, structural formula, production capacity, appearance, physical properties, specifications, process information, market price, and suppliers of each product;

[0081] The product relationship information includes raw material supply relationship, production relationship, consumption relationship, substitution relationship, and transportation relationship.

[0082] In this embodiment, relevant product entity information and product relationship information are obtained from the industrial database of the chemical industrial park and preprocessed, including removing unnecessary symbols, HTML tags, whitespace characters, and other irrelevant information; splitting long texts into meaningful words; filtering out common but meaningless words; for English texts, using lemmatization technology to restore words to their basic forms; using named entity recognition technology to extract the Chinese and English names, quantity, molecular weight, molecular formula, structural formula, production capacity, appearance, physical properties, specifications, process information, market price, and suppliers of each product from the text;

[0083] Based on the extracted entities, identify and extract the relationships between entities. For example, raw material B is required for the production of product A, and enterprise X has a supply relationship with product Y; use a relationship extraction algorithm to automatically extract relationships from the text, and by analyzing the dependency relationships of sentences, identify the syntactic relationships between entities and infer the logical relationships between entities;

[0084] The product relationship information includes raw material supply relationship, production relationship, consumption relationship, substitution relationship, and transportation relationship, which are specifically as follows:

[0085] The raw material supply relationship means that upstream products provide raw materials for downstream products; the production relationship is that a product is produced from one or more upstream raw materials or intermediate products through a certain production process; the consumption relationship is that downstream products or applications consume upstream products; the substitution relationship is that one product can replace the function or use of another product; the transportation relationship is the transportation path and method of raw materials and products between different enterprises or production stages.

[0086] S2. Define the nodes and edges of the knowledge graph based on the product entity information and product relationship information, use the evolving graph convolutional network model to predict the change relationship of the number of adjacent nodes and place it in the edge, and construct the product reserve relationship knowledge graph;

[0087] The evolving graph convolutional network model is implemented based on the graph convolutional network integrated with the multi-layer perceptron model, and is used to consider the changes of nodes and edges over time to handle the situation where the graph structure changes over time, and to predict the impact of the change in the number of nodes on the edges;

[0088] In S2, nodes and edges of the knowledge graph are defined based on product entity information and product relationship information. An evolving graph convolutional network model is used to predict the change relationship of the number of adjacent nodes and place it in the edge, constructing a knowledge graph of product reserve relationships. The specific method steps are as follows:

[0089] S2.1. Define product entity nodes based on product entity information;

[0090] S2.2. Define product relationship edges based on product relationship information;

[0091] S2.3. Use an evolving graph convolutional network model to predict the production capacity relationship caused by the change in the number of adjacent chemical products with production up and down relationships and place it in the corresponding product relationship edge;

[0092] S2.4. Based on product entity nodes and product relationship edges, construct a knowledge graph of product reserve relationships. In S2.1, product entity nodes are defined based on product entity information. The specific method is as follows:

[0093] N i =(P i ,Q i ,X i );

[0094] Among them, i is the product entity index; N i is the i-th product entity node; P i is the name of the i-th product entity; Q i is the quantity of the i-th product entity; X i is the feature of the i-th product entity other than name and quantity;

[0095] In S2.2, product relationship edges are defined based on product relationship information. The specific method is as follows:

[0096] E ij =(N i ,N j ,R ij ,w ij ,ΔQ ij );

[0097] Among them, j is the product entity index; N j is the j-th product entity node; E ij is the product relationship edge between the i-th product entity node and the j-th product entity node; R ij is the product relationship information; w ij is the product relationship edge weight; ΔQ ij is the product relationship change information.

[0098] In this embodiment, X iThe features of the i-th product entity other than name and quantity, i.e., the molecular weight, molecular formula, structural formula, production capacity, appearance, physical properties, specifications, process information, market price, and supplier of the product in the product entity information. This X i This data can be queried when the user needs to conduct a detailed query; in the product reserve relationship knowledge graph, it does not exist explicitly at the product entity node, which not only simplifies the graph structure but also satisfies the detailed recording of the chemical industrial park industrial chain data in the knowledge graph;

[0099] w ij The weight of the product relationship edge, which represents the importance of the i-th product entity node and the j-th product entity node for the integrated knowledge graph, and can better reflect the indispensable importance of the two product entities in the entire chemical industrial park industrial chain, and provides a basis for the subsequent transformation of the knowledge graph.

[0100] In S2.3, the evolutionary graph convolutional network model is used to predict the production capacity relationship caused by the change in the quantity of adjacent chemical products with production up and down relationships and place it in the corresponding product relationship edge. The specific method steps are as follows:

[0101] S2.3.1: Set the time step as t, and each product entity node has a corresponding quantity at each time step t

[0102] As the time step progresses, the quantity of products in the graph will change, and the graph convolutional network will update the feature information of the nodes according to the change in the quantity of each node. A new graph structure needs to be established for each time step, where the quantity of the nodes will be updated according to the change in the previous step to form a new feature representation X i (t) ;

[0103] S2.3.2: Based on the graph convolutional network model, aggregate the feature information of neighbor nodes and calculate the new feature representation of each product entity node:

[0104]

[0105] Among them, X i (t+1) is the feature of the i-th product entity node at time step t + 1 other than name and quantity; μ is the index of the neighbor node; is the set of neighbor nodes of the i-th product entity node; A ij is the adjacency matrix; d i is the degree of the i-th product entity node; X μ (t) is the feature of the neighbor node μ other than name and quantity; W is the weight matrix; σ(·) is the Sigmoid non-linear activation function;

[0106] S2.3.3. Based on the features of the i-th product entity node at time step t+1 other than name and quantity, a multi-layer perceptron model is integrated to predict the quantity change of the product entity at time step t+1:

[0107]

[0108] in, is the product quantity change of the i-th product entity node at time step t+1; X i (t+1) is the feature of the ith product entity node at time step t+1 except for the name and quantity; X j (t+1) is the feature of the j-th product entity node at time step t+1 except for the name and quantity; A' ij is the node connection strength weight; MLP([·]) is the multi-layer perceptron model operation;

[0109] And based on S2.3.1-S2.3.3, the product quantity change based on the j-th product entity node at time step t+1 can be obtained by the same logic

[0110] S2.3.4 Product quantity changes based on the i-th product entity node and the j-th product entity node at time step t+1 and Analyze the changes in the capacity relationship between the i-th chemical product and the adjacent j-th chemical product, and place them in the product relationship edge.

[0111] In this embodiment, the evolutionary graph convolutional network model is implemented based on the graph convolutional network fused with a multi-layer perceptron model. Node representation is learned through graph convolution operations, and the evolutionary changes of node features are processed using MLP, so that the time dependency between the graph structure and the attributes of product entity nodes can be captured, the changes in the number of products at product entity nodes can be predicted, and the changes in the relationship caused by the changes in the number of products can be analyzed. The time step t is introduced into the evolutionary graph convolutional network model. This model only considers the changes in the number of products at each product entity node over time, and does not consider the time changes of the entire knowledge graph.

[0112] In S2.3.4, the product quantity change between the i-th product entity node and the j-th product entity node based on time step t+1 and Analyze the change in the capacity relationship between the i-th chemical product and the adjacent j-th chemical product, and place it in the product relationship edge. The specific method is as follows:

[0113] S2.3.4.1. Based on product relationship information R ijThe change in the quantity of the i-th product entity node and the j-th product entity node at time step t+1 and Calculate the product relationship information at time step t+1:

[0114]

[0115] where is the product relationship information at time step t, which is equivalent to the product relationship information R ij ; The change in the product relationship information at time step t+1; γ1 is the weight coefficient of the i-th product entity; γ2 is the weight coefficient of the j-th product entity; γ3 is the product relationship weight coefficient; ∈ is the smoothing term;

[0116] S2.3.4.2. Update the product relationship edge weight of the product relationship edge with the change in the product relationship information at time step t+1:

[0117]

[0118] where is the product relationship edge weight at time step t, which is equivalent to the product relationship edge weight w ij ; is the product relationship edge weight at time step t+1; α is the weight coefficient of the product relationship edge weight; β is the weight coefficient of the change in the product relationship information;

[0119] Based on the evolutionary graph convolutional network model, calculate and analyze the change in the production capacity relationship caused by the change in the quantity of adjacent chemical products with a production up-down relationship, which is used to query in real time whether the production capacity relationship between adjacent chemical products can meet the current production demand in the product reserve relationship knowledge graph, and is also used to update the quantity and production capacity change of each chemical product in real time;

[0120] In this embodiment, the evolving graph convolutional network model is used to update in real time the quantity and production capacity changes of each chemical product in the product reserve relationship knowledge graph, ensuring that the product reserve relationship knowledge graph is updated in real time as the product data of the chemical industrial park changes, thus guaranteeing the real-time nature of the product reserve relationship knowledge graph. At the same time, the evolving graph convolutional network model is used to calculate and analyze the production capacity relationship changes caused by the quantity changes of adjacent chemical products with an upstream-downstream production relationship. For example, for products polyethylene (PE) and polypropylene (PP), they have an upstream-downstream relationship in the production process. The production process of polyethylene requires the use of by-products of polypropylene, while the production of polypropylene depends on ethylene as a raw material, and ethylene is produced through steam cracking. Due to the mutual dependence in the production process, the production of polyethylene and polypropylene needs to be coordinated and adjusted according to market demand and reserve volume. In this case, the evolving graph convolutional network model operates as follows: Assume that at time step t+1, the demand for polyethylene increases by 10%, while the production volume of polypropylene decreases by 5%. Through the evolving graph convolutional network model, the quantity changes between polyethylene and polypropylene can be captured, and the changes in the production capacity relationship between polyethylene and polypropylene can be calculated. The increase in the production capacity of polyethylene may require an increase in the supply of ethylene, while the decrease in the production capacity of polypropylene may lead to adjustments in the ethylene supply chain. According to the dynamic changes in the production chain, the evolving graph convolutional network model will update in real time the changes in the production capacity relationship between polyethylene and polypropylene, specifically including: whether the production line of polyethylene has sufficient capacity to expand production under the current production demand; whether the production line of polypropylene will be affected by ethylene supply restrictions, thereby affecting the by-product supply capacity of polyethylene; whether the supply of ethylene can keep up with the production demands of polypropylene and polyethylene. The evolving graph convolutional network model will, based on the above changes, update the product reserve relationship knowledge graph in real time and conduct queries: By querying the edge weights of the relationship between polyethylene and polypropylene in the graph, confirm whether the production capacities of these two products are coordinated and can meet the current production demands; for example, if the production volume of polypropylene decreases, whether it will lead to insufficient ethylene supply, thereby affecting the production of polyethylene and causing bottlenecks in the production capacity chain between polyethylene and ethylene; in the product reserve relationship knowledge graph, be used to query in real time whether the production capacity relationship between adjacent chemical products can meet the current production demands.

[0121] In S2.4, based on the product entity nodes and product relationship edges, a product reserve relationship knowledge graph is constructed as follows:

[0122] G1 = ({N1, N2, …, N i , …, N n , {E' ij ∣1 ≤ i ≠ j ≤ n});

[0123]

[0124] Among them, G1 is the knowledge graph of product reserve relationships; n is the total number of product entity nodes; E' ij is the product relationship edge between the i-th product entity node and the j-th product entity node after adding prediction data.

[0125] In this embodiment, we infer the change in the product relationship between them by calculating the change in the quantity of the i-th and j-th product entity nodes, and further update the weight of the product relationship edge:

[0126] and corresponding to the change ratio of the quantities of the i-th and j-th product entity nodes, an ∈ smoothing term is added to avoid a zero denominator. then takes into account the historical information of the product relationship, which can help us capture the stability and continuity of the existing relationship. If the relationship between the i-th and j-th product entity nodes was strong before, it may affect future changes;

[0127] After calculating the change in the product relationship information, the next step is to update the weight of the product relationship edge according to this change. The weight of the relationship edge represents the importance or strength of the relationship between products, and is usually used to measure the connection degree or interaction frequency between products;

[0128] In the chemical industrial park industrial chain, the relationships between products are not static, but change over time and demand. As the production volume of a certain product changes, it may affect its supply relationship, dependence relationship, or production capacity scheduling with other products. A static product relationship graph cannot accurately reflect the actual dynamic interactions between products in the industrial chain. By using an evolutionary graph convolutional network model to dynamically update the product relationship graph, considering the direct impact of the change in product quantity on the product relationship, the product relationship can be updated in a timely manner according to factors such as demand and production capacity fluctuations, thus reflecting the actual interactions between products in the industrial chain; the impact of product quantity changes on the relationships between products is very complex. For example, when the output of a product increases, it may cause changes in the demand for upstream or downstream products, but this impact is usually non-linear and affected by multiple factors. Traditional knowledge graph models may only provide static relationships between products, and cannot provide a dynamic view of how to adjust relationships over time and trends, making it difficult to predict future changes in the industrial chain, especially when facing demand fluctuations or production bottlenecks. By considering the impact of time step changes and product quantity changes on the relationship, this method provides the ability to predict the dynamic changes in the industrial chain. The update of the edge weight and product relationship not only enables the graph to reflect the changes between products in real time, but also provides useful trend predictions for decision-makers.

[0129] S3. Introduce a timestamp to update the node definition based on the product reserve relationship knowledge graph, use the dynamic time-series graph network algorithm to transform the product reserve relationship knowledge graph into a product time relationship knowledge graph, and merge the duplicate routes and products in the chemical industrial park to obtain the final product time relationship knowledge graph;

[0130] In S3, introduce a timestamp to update the node definition based on the product reserve relationship knowledge graph, use the dynamic time-series graph network algorithm to transform the product reserve relationship knowledge graph into a product time relationship knowledge graph, and merge the duplicate routes and products in the chemical industrial park to obtain the final product time relationship knowledge graph. The specific method steps are as follows:

[0131] S3.1. Introduce a timestamp to update the definition of product entity nodes and product relationship edges:

[0132]

[0133] E” ij =(N i ,N j ,R ij ,w ij );

[0134] Among them, is the i-th product entity node including the timestamp after update; E” ij is the product relationship edge between the i-th product entity node and the j-th product entity node after update; is the node timestamp;

[0135] In this embodiment, the i-th product entity node including the timestamp after update additionally includes the timestamp for sorting out the time sequence between product entity nodes; the product relationship edge E” between the i-th product entity node and the j-th product entity node after update ij removes the predicted data and because this product time relationship knowledge graph does not need to consider the changes between product entities, but only represents the broad relationships between products, so there is no need to consider the predicted data in the product reserve relationship knowledge graph;

[0136] S3.2. Use the dynamic time-series graph network algorithm to transform the product reserve relationship knowledge graph into a product time relationship knowledge graph;

[0137] In this embodiment, there are often strict dependency relationships and time constraints among multiple production links and products. For example, the production of ethylene needs to be carried out after the production of polyethylene, and the production of polyethylene depends on naphtha as a raw material. However, the product reserve relationship knowledge graph cannot well reflect the time dependency relationships between these products and cannot reflect the dynamic changes in the production process. Especially when it comes to complex dependency relationships involving multiple products, introducing timestamps in the nodes and updating, transforming the product reserve relationship in the knowledge graph into a product time relationship knowledge graph to reflect the production sequence and mutual dependency relationships between different products;

[0138] S3.3. Use the incremental hierarchical temporal memory model to adaptively merge the repeated routes and product entity nodes in the chemical industrial park to obtain the final product time relationship knowledge graph.

[0139] In S3.2, the dynamic temporal graph network algorithm is used to transform the product reserve relationship knowledge graph into a product time relationship knowledge graph. The specific method steps are as follows:

[0140] S3.2.1. Use a graph neural network to perform temporal embedding on each product entity node and product relationship edge:

[0141]

[0142] where, is the temporal embedding of the i-th product entity node with a timestamp after update ; is the set of neighbor nodes of the i-th product entity node;

[0143]

[0144] where, is the temporal embedding of the product relationship edge between the i-th product entity node and the j-th product entity node; is the edge timestamp;

[0145] S3.2.2. Combine the temporal information of the product entity nodes and product relationship edges to obtain the product time relationship knowledge graph:

[0146]

[0147] where, G2 is the product time relationship knowledge graph; is the temporal embedding of the n-th product entity node with a timestamp after update ;

[0148] In S3.3, the incremental hierarchical temporal memory model is used to adaptively merge the repeated routes and product entity nodes in the chemical industrial park to obtain the final product time relationship knowledge graph. The specific method steps are as follows:

[0149] S3.3.1. At each time step, based on the temporal embeddings of each product entity node, calculate the cosine similarity between each pair of product entity nodes, and set a node similarity threshold δ to determine whether they can be merged:

[0150]

[0151] If then the i-th product entity node and the j-th product entity node are similar and can be merged; otherwise, they cannot be merged.

[0152] Among them, is the cosine similarity between the i-th product entity node and the j-th product entity node; δ is the node similarity threshold;

[0153] S3.3.2. At each time step, based on the temporal embeddings of each product relationship edge, calculate the cosine similarity between each pair of product relationship edges, and set an edge similarity threshold τ to determine whether they can be merged:

[0154]

[0155] If then the product relationship edge and the product relationship edge are similar and can be merged; otherwise, they cannot be merged.

[0156] Among them, is the cosine similarity between the product relationship edge and the product relationship edge ; τ is the edge similarity threshold; s is the index of the product entity node; l is the index of the product entity node; is the temporal embedding of the product relationship edge between the s-th product entity node and the l-th product entity node;

[0157] S3.3.3. Merge the similar product entity nodes and product relationship edges to obtain the final product time relationship knowledge graph.

[0158] In this embodiment, the product time relationship knowledge graph is constructed based on the temporal embeddings of the product entity nodes and product relationship edges of the product reserve relationship knowledge graph, and then based on this temporal embedding information; by using graph neural networks and temporal embedding methods, we can capture the dynamic characteristics of product entities and relationship edges over time, which enables the product time relationship knowledge graph to reflect the impact of time evolution on product relationships; compared with the product reserve relationship knowledge graph, the product time relationship knowledge graph can consider the time-varying relationships between products.

[0159] S4. Use a visualization interface to display the product reserve relationship knowledge graph and the product time relationship knowledge graph for users to perform query operations according to their needs.

[0160] Embodiment 2:

[0161] This embodiment provides a system for constructing a knowledge graph of chemical industrial park industrial chain data based on big data, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the method for constructing a knowledge graph of chemical industrial park industrial chain data based on big data described in any one of the above.

[0162] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and all these changes and improvements fall within the scope of the present invention claimed.

Claims

1. A method for constructing a chemical park industry chain data knowledge graph based on big data, characterized in that: The following steps are involved: S1. Obtain the industrial chain data of the chemical park from the industrial database, and use natural language processing technology to extract the product entity information and product relationship information of the industrial chain; S2. Define the nodes and edges of the knowledge graph based on product entity information and product relationship information, use the evolutionary graph convolutional network model to predict the relationship between the number of adjacent nodes and place them in the edges to build a product reserve relationship knowledge graph; The evolutionary graph convolutional network model is implemented based on the graph convolutional network fused with a multi-layer perceptron model, and is used to consider the changes of nodes and edges over time to handle the situation where the graph structure changes over time, and to predict the impact of changes in the number of nodes on the edges; S3. Based on the product reserve relationship knowledge graph, the timestamp update node definition is introduced, and the product reserve relationship knowledge graph is converted into the product time relationship knowledge graph using the dynamic time series graph network algorithm, and the repeated routes and products of the chemical park are merged to obtain the final product time relationship knowledge graph; S4. Use a visual interface to display the product reserve relationship knowledge graph and the product time relationship knowledge graph, so that users can perform query operations according to their needs.

2. The method for constructing a chemical park industry chain data knowledge graph based on big data according to claim 1, characterized in that: In S1, the product entity information includes the Chinese and English names, quantity, molecular weight, molecular formula, structural formula, production capacity, appearance, physical properties, specifications, process information, market price and supplier of each product; Product relationship information includes raw material supply relationships, production relationships, consumption relationships, substitution relationships, and transportation relationships.

3. The method for constructing a chemical park industry chain data knowledge graph based on big data according to claim 2 is characterized in that: In S2, the nodes and edges of the knowledge graph are defined based on the product entity information and product relationship information, and the evolutionary graph convolutional network model is used to predict the relationship between the number of adjacent nodes and place them in the edges to construct the product reserve relationship knowledge graph. The specific method steps are as follows: S2.

1. Define product entity nodes based on product entity information; S2.2, define product relationship edges based on product relationship information; S2.3, using the evolutionary graph convolutional network model to predict the capacity relationship caused by the change in the quantity of adjacent chemical products with production up-down relationship and place it in the corresponding product relationship edge; S2.

4. Based on product entity nodes and product relationship edges, construct a product reserve relationship knowledge graph.

4. The method for constructing a chemical park industry chain data knowledge graph based on big data according to claim 3 is characterized in that: In S2.1, a product entity node is defined based on product entity information. The specific method is as follows: N i =(P i ,Q i ,X i ); Where i is the product entity index; N i is the i-th product entity node; P i is the name of the i-th product entity; Q i is the number of the i-th product entity; X i is the characteristics of the i-th product entity except name and quantity; In S2.2, product relationship edges are defined based on product relationship information. The specific method is as follows: E ij =(N i ,N j ,R ij ,w ij ,ΔQ ij ); Where j is the product entity index; N j is the jth product entity node; E ij is the product relationship edge between the i-th product entity node and the j-th product entity node; R ij is product relationship information; ij is the product relationship edge weight; ΔQ ij Product relationship change information.

5. The method for constructing a chemical park industry chain data knowledge graph based on big data according to claim 4 is characterized in that: In S2.3, the evolution graph convolutional network model is used to predict the capacity relationship caused by the change in the quantity of adjacent chemical products with production up-down relationship and place it in the corresponding product relationship edge. The specific method steps are as follows: S2.3.

1. Set the time step to t. Each product entity node has a corresponding number at each time step t. S2.3.

2. Based on the graph convolutional network model, aggregate the feature information of neighbor nodes and calculate the new feature representation of each product entity node: Among them, X i (t+1) is the feature of the ith product entity node at time step t+1 except for the name and quantity; μ is the index of the neighbor node; is the neighbor node set of the i-th product entity node; A ij is the adjacency matrix; d i is the degree of the ith product entity node; X μ (t) are the features of the neighbor node μ except the name and quantity; W is the weight matrix; σ(·) is the Sigmoid nonlinear activation function; S2.3.

3. Based on the features of the i-th product entity node at time step t+1 other than name and quantity, a multi-layer perceptron model is integrated to predict the quantity change of the product entity at time step t+1: in, is the product quantity change of the i-th product entity node at time step t+1; X i (t+1) is the feature of the ith product entity node at time step t+1 except for the name and quantity; X j (t+1) is the feature of the j-th product entity node at time step t+1 except for the name and quantity; A' ij is the node connection strength weight; MLP([·]) is the multi-layer perceptron model operation; Similarly, the product quantity change based on the j-th product entity node at time step t+1 can be obtained S2.3.

4. Product quantity changes based on the i-th product entity node and the j-th product entity node at time step t+1 and Analyze the changes in the capacity relationship between the i-th chemical product and the adjacent j-th chemical product, and place them in the product relationship edge.

6. The method for constructing a chemical park industry chain data knowledge graph based on big data according to claim 5 is characterized in that: In S2.3.4, the product quantity change between the i-th product entity node and the j-th product entity node based on time step t+1 and Analyze the change in the capacity relationship between the i-th chemical product and the adjacent j-th chemical product, and place it in the product relationship edge. The specific method is as follows: S2.3.4.

1. Based on product relationship information R ij Combine the product quantity changes of the i-th product entity node and the j-th product entity node at time step t+1 and Calculate product relationship information at time step t+1: in, is the product relationship information at time step t, which is equivalent to the product relationship information R ij ; The change of product relationship information at time step t+1; γ1 is the weight coefficient of the i-th product entity; γ2 is the weight coefficient of the j-th product entity; γ3 is the product relationship weight coefficient; ∈ is a smoothing term; S2.3.4.

2. Update the product relationship edge weight of the product relationship edge based on the product relationship information change at time step t+1: in, is the product relationship edge weight at time step t, which is equivalent to the product relationship edge weight w ij ; is the product relationship edge weight at time step t+1; α is the weight coefficient of the product relationship edge weight; β is the weight coefficient of the product relationship information change; In S2.4, a product reserve relationship knowledge graph is constructed based on product entity nodes and product relationship edges, as follows: G1({N1,N2,...,N i ,...,N n },{E′ ij ∣1≤i≠j≤n}) Among them, G1 is the product reserve relationship knowledge graph; n is the total number of product entity nodes; E' ij It is the product relationship edge between the i-th product entity node and the j-th product entity node after adding the prediction data.

7. The method for constructing a chemical park industry chain data knowledge graph based on big data according to claim 6 is characterized in that: In S3, based on the product reserve relationship knowledge graph, the timestamp update node definition is introduced, the product reserve relationship knowledge graph is converted into the product time relationship knowledge graph using the dynamic time sequence graph network algorithm, and the repeated routes and products of the chemical park are merged to obtain the final product time relationship knowledge graph. The specific method steps are as follows: S3.

1. Introduce timestamps to update and define product entity nodes and product relationship edges: E″ ij =(N i ,N j ,R ij ,w ij ); in, is the ith product entity node containing the timestamp after update; e″ ij is the product relationship edge between the i-th product entity node and the j-th product entity node after update; is the node timestamp; S3.

2. Use the dynamic temporal graph network algorithm to transform the product reserve relationship knowledge graph into the product time relationship knowledge graph; S3.

3. Use the incremental hierarchical temporal memory model to adaptively merge repeated routes and product entity nodes in the chemical park to obtain the final product time relationship knowledge graph.

8. The method for constructing a chemical park industry chain data knowledge graph based on big data according to claim 7, characterized in that: In S3.2, a dynamic time-series graph network algorithm is used to transform the product reserve relationship knowledge graph into a product time relationship knowledge graph. The specific method steps are as follows: S3.2.

1. Use graph neural network to perform temporal embedding for each product entity node and product relationship edge: in, The i-th product entity node containing the timestamp after the update Temporal embedding of is the neighbor node set of the i-th product entity node; in, is the temporal embedding of the product relationship edge between the i-th product entity node and the j-th product entity node; is the edge timestamp; S3.2.

2. Combine the time series information of product entity nodes and product relationship edges to obtain the product time relationship knowledge graph: Among them, G2 is the product time relationship knowledge graph; The nth product entity node containing the timestamp after the update The temporal embedding of .

9. The method for constructing a chemical park industry chain data knowledge graph based on big data according to claim 8, characterized in that: In S3.3, the incremental hierarchical temporal memory model is used to adaptively merge repeated routes and product entity nodes in the chemical park to obtain the final product time relationship knowledge graph. The specific method steps are as follows: S3.3.

1. At each time step, based on the temporal embedding of each product entity node, the cosine similarity between each product entity node is calculated, and the node similarity threshold δ is set to determine whether it can be merged: like If the i-th product entity node is similar to the j-th product entity node, they can be merged; otherwise, they cannot be merged; in, is the cosine similarity between the i-th product entity node and the j-th product entity node; δ is the node similarity threshold; S3.3.

2. At each time step, based on the temporal embedding of each product relationship edge, calculate the cosine similarity of each product relationship edge, and set the edge similarity threshold τ to determine whether it can be merged: like The product relationship edge Product relationship Similar ones can be merged, otherwise they cannot be merged; in, Product relationship edge Product relationship The cosine similarity of ; τ is the edge similarity threshold; s is the product entity node index; l is the product entity node index; is the temporal embedding of the product relationship edge between the sth product entity node and the lth product entity node; S3.3.

3. Merge similar product entity nodes and product relationship edges to obtain the final product time relationship knowledge graph.

10. A chemical park industry chain data knowledge graph construction system based on big data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the method for constructing a chemical park industrial chain data knowledge graph based on big data as described in any one of claims 1 to 9.

Images