A graphene technology route prediction method based on big model and knowledge graph

Abstract

The present invention discloses a graphene technology route prediction method based on a large model and a knowledge graph, including S1, constructing a graphene interdisciplinary data set; S2, using text mining technology to extract key information related to the development of graphene technology from the collected graphene interdisciplinary data set; S3, identifying the potential direction of graphene technology development; S4, constructing a knowledge graph; S5, combining the knowledge graph in step S4 with the pattern recognition and trend analysis results of step S3, combining convolutional neural networks and graph neural networks to learn and predict the future development route of graphene technology; S6, dynamically updating the knowledge graph and the adaptive graph convolutional network model according to the prediction results; S7, outputting a prediction report. The present invention can generate a list of market application fields according to the predicted technical direction, help enterprises and R&D teams better plan market strategies and R&D investment, and ensure the close integration of technological development and market demand.

Description

Technical Field

[0001] The present invention relates to the field of motion recognition technology, and in particular to a graphene technology route prediction method based on a large model and a knowledge graph. Background Art

[0002] Graphene is a two-dimensional material composed of a single layer of carbon atoms with excellent physical, chemical and electrical properties. Its high conductivity, high strength and high transparency make it have broad application prospects in electronic devices, energy storage, sensors and biomedicine. However, the research and development of graphene technology involves a large amount of interdisciplinary data, including scientific literature, patent information, experimental data and market analysis. These data come from various sources and formats, and traditional methods are often inefficient and error-prone when integrating these data. At the same time, the rapid development of graphene technology and its application fields requires predictive models to be able to flexibly adapt to new research results and changes in market demand. However, traditional predictive models are often static and difficult to adapt to new situations once established.

[0003] In the prior art, a single data mining method or a simple statistical analysis method is usually used to predict the development of graphene technology. However, these methods have the following defects:

[0004] First, existing data integration methods often lack systematicity and comprehensiveness when dealing with interdisciplinary data. Since graphene technology involves multiple disciplines, the formats, structures and sources of relevant data are significantly different. Traditional data mining methods are difficult to effectively integrate and process these heterogeneous data, resulting in low data utilization and affecting the accuracy of prediction results. Secondly, most of the existing prediction models are statistical analysis methods based on static data, which cannot dynamically update and adapt to new research results and market changes. With the continuous development of graphene technology, new research results and market demands continue to emerge. Static prediction models are difficult to reflect these changes in a timely manner, resulting in lagging prediction results and unable to provide effective decision-making support for R&D teams and enterprises. Thirdly, the existing technology route prediction methods lack in-depth analysis of key information such as technology nodes, research directions and material properties. Traditional methods often make predictions based on simple indicators such as the number of literature or patents, ignoring the complexity and diversity of technological development, and it is difficult to accurately identify the potential directions and key nodes of technological development.

[0005] In addition, existing methods lack the comprehensive application of knowledge graphs and deep learning technologies in the prediction process. Knowledge graphs can effectively represent the structured information of domain knowledge and support complex queries and analysis, but traditional prediction methods often ignore this. Deep learning technology can automatically extract features and patterns from large amounts of data to improve the accuracy and efficiency of predictions, but existing methods often lack targeted optimization and design when applying deep learning technology.

[0006] Therefore, how to provide a graphene technology route prediction method based on a large model and knowledge graph is an urgent problem that technicians in this field need to solve. Summary of the invention

[0007] One purpose of the present invention is to propose a graphene technology route prediction method based on a large model and a knowledge graph. By identifying and predicting potential market application fields, the present invention not only provides a predicted direction of technology development and a time series of key technology nodes, but also can generate a list of market application fields according to the predicted technology direction, helping enterprises and R&D teams to better plan market strategies and R&D investment, and ensure the close integration of technology development and market demand.

[0008] A graphene technology route prediction method based on a large model and a knowledge graph according to an embodiment of the present invention comprises the following steps:

[0009] S1. Collect scientific literature, patent information, market analysis, scientific research team cooperation network information and international research funding trends in the field of graphene to build a graphene interdisciplinary dataset;

[0010] S2. Use text mining technology to extract key information related to the development of graphene technology from the collected graphene interdisciplinary data set, the key information including technology nodes, research directions and material properties;

[0011] S3. Use the adaptive graph convolutional network model to perform pattern recognition and trend analysis on the extracted key information to identify the potential direction of graphene technology development;

[0012] S4. Build a knowledge graph, which includes the entities of graphene technology and the relationships between them, including the dependencies between technologies, the use of materials, and the innovations of technologies;

[0013] S5, combining the knowledge graph in step S4 with the pattern recognition and trend analysis results in step S3, and combining convolutional neural networks and graph neural networks to learn and predict the future development path of graphene technology;

[0014] S6. Dynamically update the knowledge graph and adaptive graph convolutional network model based on the prediction results to adapt to new research results and market demand changes;

[0015] S7. Output a forecast report, which includes the forecast direction of technology development, the time series of key technology nodes, and potential market application areas.

[0016] Optionally, the S1 includes the following steps:

[0017] S11. Collect scientific literature in the field of graphene D lit , using automated crawler technology to obtain literature from scientific databases and extract technology nodes, research directions, and material property data:

[0018]

[0019] Where N is the number of scientific papers, M is the number of technology nodes in the papers, and α ij is the weight parameter, is the feature extraction function, δ ij (t) represents the time weight function, crawl is the automated crawler technology function, is the data of the jth technology node in the i-th document at time t;

[0020] S12. Collect patent information in the field of graphene D pat , by accessing the patent database and using natural language processing technology to parse patent documents, obtain data on technical innovation points, patent holders and technical application fields, the steps of parsing patent documents include:

[0021]

[0022] Among them, L represents the number of patent documents, K represents the number of information points in the documents, and β km , γ km and δ km is the weight parameter of different information points, NLP innovation 、NLP holder and NLP application is a natural language processing technology function, which is used to extract data on technological innovations, patent holders, and technical application fields. is the mth information point in the kth patent document;

[0023] S13. Collect market analysis reports D mar , obtain market demand, market growth trends and major market participants data from market research institutions, and use data mining technology to structure unstructured market analysis text:

[0024]

[0025] Where P represents the number of market analysis reports, Q represents the number of market participants in the reports, and γno is the weight parameter, extract and transform are data mining technology functions, is the rth characteristic of the oth market participant in the nth report, U is the number of market characteristics, κ nor and λ nor is the weight parameter of the feature;

[0026] S14. Collect information on the scientific research team's collaborative network team , obtain the data of scientific research team members, cooperation relationships and research projects from the scientific research cooperation platform, and use social network analysis technology to construct the scientific research team cooperation network diagram:

[0027]

[0028] Among them, V represents the set of scientific research team members and research project nodes, E represents the set of cooperation relationship edges between members, and each edge e j ∈E represents the cooperative relationship between two members, and the weight ω(e j ) represents the cooperation intensity, and the constructed cooperation network graph is represented as a weighted undirected graph;

[0029] S15. Collect international research funding trends fund , obtain the funding projects, funding amounts and research content data from the funding agency website, and use data crawling technology to store the information in a structured manner:

[0030]

[0031] Among them, U represents the number of funded projects, V represents the number of funded contents, tv is the weight parameter, scrape is the data crawling technology function, is the wth feature of the vth funding content of the tth funding project, X is the number of funding features, μ tvw is the weight parameter of the feature;

[0032] S16. Perform data cleaning, data deduplication, and data standardization on all types of collected data:

[0033]

[0034] Among them, X represents the amount of original data, Y represents the number of fields to be processed, and η wx is the weight parameter, m is the preprocessing function, is the wth original data field and the xth processed field;

[0035] S17, build a graphene interdisciplinary dataset, and combine the D lit , Dpat , D mar , D team , D fund Data integration, using a unified data format and data model to represent:

[0036]

[0037] Among them, Z represents the number of data sources, AA represents the amount of data in each data source, and θ yz is the weight parameter, n is the data fusion function, The y-th data source and the z-th data.

[0038] Optionally, S2 includes the following steps:

[0039] S21. Preprocess the graphene interdisciplinary dataset Dgra using natural language processing technology, including word segmentation, part-of-speech tagging, and named entity recognition:

[0040]

[0041] Where N is the number of datasets, M is the number of documents in each dataset, and ω ij is the weight parameter, NER is the named entity recognition function, is the part-of-speech tagging function, ψ is the word segmentation function, is the jth document in the i-th dataset, D pre is the preprocessed data set;

[0042] S22. Extracting technology node information T from the preprocessed data set node , using an adaptive graph convolutional network model to identify technology nodes:

[0043]

[0044] Among them, P represents the number of preprocessed data sets, Q represents the number of technology nodes in each data set, and α uv is the weight parameter, σ is the activation function, H is the number of graph convolutional layers, G is the number of nodes in the graph convolutional layer, γ kl is the adaptive weight parameter, ReLU is the activation function, W kl is the weight matrix of the graph convolutional network, A uv is the adjacency matrix, is the feature vector of the first node in the kth layer, b kl is the bias term;

[0045] S23. Extract research direction information R from the preprocessed data set dir , use topic models to classify and identify research directions:

[0046]

[0047] Among them, L represents the number of preprocessed data sets, K represents the number of research directions in each data set, and β mn is the weight parameter, θ m is the topic distribution matrix, V is the number of topic models, η zn is the adaptive weight parameter, Dirichlet is the Dirichlet distribution function, is the feature extraction function, is the weight matrix of the topic model;

[0048] S24, extracting material property information M from the preprocessed data set prop , using feature extraction algorithms to parse and identify material properties:

[0049]

[0050] Where S represents the number of preprocessed data sets, T represents the number of material properties in each data set, and δ ab is the weight parameter, W is the number of feature extraction layers, ζ cw is the adaptive weight parameter, tanh is the hyperbolic tangent activation function, X is the number of nodes in the feature extraction layer, Y is the number of features in the feature extraction layer, κ de is the adaptive weight parameter, and ψ are feature extraction and processing functions, λ de is the feature extraction weight matrix, ∈ de is the bias term;

[0051] S25. Structural storage of the extracted technology nodes, research directions and material property information to form a structured data set D struct :

[0052] D struct ={T node , R dir ,M prop};

[0053] Among them, D struct is a structured data set, including technology nodes T node 、Research Direction dir and material properties M prop .

[0054] Optionally, S3 includes the following steps:

[0055] S31. Build an adaptive graph convolutional network model GCN adapt , initialize the model parameters Where L is the number of network layers, W (l) is the weight matrix of the first layer, b (l) is the bias term of the first layer;

[0056] S32, extract the structured data set D struct Convert to graph structure data G = (V, E), where V is the node set and E is the edge set. Define the node feature matrix X and adjacency matrix A:

[0057] X = [x1, x2, ..., x N ] T ;

[0058] A=[a ij ] N×N ;

[0059] Among them, x i is the feature vector of the i-th node, a ij is the edge weight between the i-th and j-th nodes, and N is the number of nodes;

[0060] S33, use graph convolution operation to perform convolution operation on node features and calculate the node embedding representation H of each layer (l) , and its recursive relationship is:

[0061] H (l) =σ(D -1 / 2 AD -1 / 2 H (l-1) W (l) +b (l) );

[0062] Where D is the degree matrix, H (0) =X, σ is the activation function;

[0063] S34, update the node embedding representation through the adaptive graph convolutional network, combined with the edge weight α ij and the adaptive coefficient γ ij , calculate the final node representation:

[0064] H * =∑ (i，j)∈E α ij γ ij ·σ(D -1 / 2 AD -1 / 2 H (L) W * +b * );

[0065] Among them, W * and b * is the weight matrix and bias term of the final layer;

[0066] S35, for the final node representation H * Conduct clustering and classification analysis to identify potential directions for the development of graphene technology dir , and its calculation formula is:

[0067]

[0068] Among them, K is the number of clusters, βk is the weight parameter of each cluster, Cluster is the clustering function, P is the number of nodes, Q is the number of edges, and are the parameters used in the clustering process;

[0069] S36. Based on the cluster analysis results, conduct trend analysis, predict the development path and potential application areas of graphene technology, and generate a technology development trend report R trend :

[0070]

[0071] Among them, D is the time dimension, C is the technology dimension, θ dc is the weight parameter, F is the number of potential application areas, is the domain weight parameter, T dir is the technical direction, and Proj is the projection function of time and technology.

[0072] Optionally, S4 includes the following steps:

[0073] S41. Define the entity set E in the field of graphene technology, including technology, tools, materials, research institutions, researchers, patents, and literature entity types:

[0074] E = {e1, e2, ..., e n};

[0075] Where n represents the number of entities, e i represents the i-th entity;

[0076] S42. Define the relationship set R between entities, including technical dependency, material usage, cooperation, and reference relationships:

[0077] R = {r1, r2, ..., r m};

[0078] Where m represents the number of relations, r j represents the jth relationship;

[0079] S43. Use natural language processing technology and data mining technology to extract entities and their relationships from the graphene interdisciplinary dataset and construct the initial knowledge graph KG init :

[0080] KG init =(E, R);

[0081] S44. Initial knowledge graph KG init Perform semantic analysis and annotation of entities and relationships to generate a knowledge graph KG with weights and attributes weighted :

[0082] KG weighted =(E, R, W E , W R );

[0083] Among them, W E is the weight set of the entity, W R is the weight set of the relationship;

[0084] S45. Combine the embedding algorithm to vectorize the knowledge graph and generate the vector representation KG of the knowledge graph. embed , and its calculation formula is:

[0085]

[0086] Among them, ω ij is the weight parameter, NER is the named entity recognition function, embed(e i , r j ) is the embedding function, is the i-th entity and j-th relationship in the knowledge graph;

[0087] S46. Combine adaptive graph convolutional networks to perform pattern recognition and trend analysis on knowledge graphs and identify potential directions T dir , identify potential directions, and thus form the final knowledge graph KG final .

[0088] Optionally, S5 includes the following steps:

[0089] S51, the knowledge graph KG constructed in step S4 final The node embedding representation H generated in step S3 * Combine to form a comprehensive data set D combined :

[0090] D combined ={KG weighted , H *};

[0091] S52. Define the convolutional neural network model CNN graph and graph neural network model GNN graph , initialize the model parameters:

[0092]

[0093]

[0094] Among them, L CNN and L GNN Represents the number of network layers of CNN and GNN respectively;

[0095] S53. Use the convolutional neural network model to analyze the comprehensive dataset D combined Perform feature extraction and generate convolutional feature representation F CNN :

[0096]

[0097] Among them, σ is the activation function, K is the number of data blocks, conv is the convolution operation function, η m is the adaptive weight parameter, is the feature extraction function;

[0098] S54. Using the graph neural network model to analyze the comprehensive dataset D combined Perform feature extraction and generate graph convolution feature representation F GNN :

[0099]

[0100] Where N is the number of nodes, a ij is the weight in the adjacency matrix, is the feature representation of the i-th node in the l-1th layer, γ p is the adaptive weight parameter, ReLU is the activation function, ψ p is the feature extraction function;

[0101] S55, represent the convolution feature F CNN With graph convolution feature representation F GNN Fusion is performed to generate the final feature representation F final :

[0102]

[0103] Among them, Q is the number of feature blocks, λ q is the fusion weight coefficient, δ r is the adaptive weight parameter, softmax is the normalization function, is the feature fusion function;

[0104] S56. Using the fused features to represent F final Predict the future development path and generate prediction results P route :

[0105]

[0106] Among them, S is the number of prediction layers, θ s is the prediction layer weight parameter, Predict is the prediction function, μ t is the adaptive weight parameter, Attention is the attention mechanism function, v t is the attention weight parameter;

[0107] S57, according to the prediction result P route Generate a future development roadmap for graphene technology to identify technology development directions and potential application areas.

[0108] Optionally, the S7 includes the following steps:

[0109] S71. According to the prediction result P route , extract the predicted direction of technology development;

[0110] S72. Construct a time series of key technology nodes, including predicting the occurrence time and importance score of each key technology node;

[0111] S73. Identify potential market application areas and generate a list of market application areas based on the predicted technical direction and time series;

[0112] S74. Integrate forecast directions, time series of key technology nodes and potential market application areas to generate a comprehensive forecast report;

[0113] S75. Display the comprehensive forecast report in a visual form, including line graphs, bar graphs and pie charts, and output the final version of the comprehensive forecast report for reference by R&D teams, enterprises and policymakers for making R&D or investment decisions.

[0114] The beneficial effects of the present invention are:

[0115] (1) The present invention collects and integrates interdisciplinary data such as scientific literature, patent information, and market analysis in the field of graphene, and uses text mining technology to extract key information such as technical nodes, research directions, and material properties to construct a comprehensive interdisciplinary dataset for graphene technology. By combining an adaptive graph convolutional network with a convolutional neural network, efficient data integration and analysis are achieved, overcoming the problem of inefficient processing of heterogeneous data in traditional methods.

[0116] (2) The method of the present invention combines the adaptive graph convolutional network model and the convolutional neural network with the graph neural network, and can dynamically update the prediction model to adapt to new research results and changes in market demand. Through continuous learning and updating, the timeliness and accuracy of the prediction results are maintained, providing reliable decision-making support for R&D teams, enterprises and policy makers.

[0117] (3) The present invention constructs a knowledge graph and uses an embedding algorithm and an adaptive graph convolutional network to perform pattern recognition and trend analysis, deeply explores the key nodes and research directions of graphene technology development, and identifies potential technology development paths and important technology nodes. This process improves the ability to capture complex technology relationships and development dynamics, and makes up for the shortcomings of traditional methods that only make simple predictions based on the number of documents or patents.

[0118] (4) By identifying and predicting potential market application areas, the present invention not only provides the predicted direction of technology development and the time series of key technology nodes, but also generates a list of market application areas based on the predicted technology direction, helping enterprises and R&D teams to better plan market strategies and R&D investment, and ensure the close integration of technology development and market demand. BRIEF DESCRIPTION OF THE DRAWINGS

[0119] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0120] Figure 1 A flowchart of a graphene technology route prediction method based on a large model and knowledge graph proposed in the present invention. DETAILED DESCRIPTION

[0121] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0122] refer to Figure 1 , a graphene technology route prediction method based on a large model and knowledge graph, including the following steps:

[0123] S1. Collect scientific literature, patent information, market analysis, scientific research team cooperation network information and international research funding trends in the field of graphene to build a graphene interdisciplinary dataset;

[0124] In this implementation, S1 includes the following steps:

[0125] S11. Collect scientific literature in the field of graphene D lit, using automated crawler technology to obtain literature from scientific databases and extract technology nodes, research directions, and material property data:

[0126]

[0127] Where N is the number of scientific papers, M is the number of technology nodes in the papers, and α ij is the weight parameter, is the feature extraction function, δ ij (t) represents the time weight function, crawl is the automated crawler technology function, is the data of the jth technology node in the i-th document at time t;

[0128] S12. Collect patent information in the field of graphene D pat , by accessing the patent database and using natural language processing technology to parse patent documents, obtain data on technological innovations, patent holders and technological application fields. The steps for parsing patent documents include:

[0129]

[0130] Among them, L represents the number of patent documents, K represents the number of information points in the documents, and β km , γ km and δ km is the weight parameter of different information points, NLP innovation 、NLP holder and NLP application is a natural language processing technology function, which is used to extract data on technological innovations, patent holders, and technical application fields. is the mth information point in the kth patent document;

[0131] S13. Collect market analysis reports D mar , obtain market demand, market growth trends and major market participants data from market research institutions, and use data mining technology to structure unstructured market analysis text:

[0132]

[0133] Where P represents the number of market analysis reports, Q represents the number of market participants in the reports, and γ no is the weight parameter, extract and transform are data mining technology functions, is the rth characteristic of the oth market participant in the nth report, U is the number of market characteristics, κ nor and λ nor is the weight parameter of the feature;

[0134] S14. Collect information on the scientific research team's collaborative network team , obtain the data of scientific research team members, cooperation relationships and research projects from the scientific research cooperation platform, and use social network analysis technology to construct the scientific research team cooperation network diagram:

[0135]

[0136] Among them, V represents the set of scientific research team members and research project nodes, E represents the set of cooperation relationship edges between members, and each edge e j ∈E represents the cooperative relationship between two members, and the weight ω(e j ) represents the cooperation intensity, and the constructed cooperation network graph is represented as a weighted undirected graph;

[0137] S15. Collect international research funding trends fund , obtain the funding projects, funding amounts and research content data from the funding agency website, and use data crawling technology to store the information in a structured manner:

[0138]

[0139] Among them, U represents the number of funded projects, V represents the number of funded contents, tv is the weight parameter, scrape is the data crawling technology function, is the wth feature of the vth funding content of the tth funding project, X is the number of funding features, μtv w is the weight parameter of the feature;

[0140] S16. Perform data cleaning, data deduplication, and data standardization on all types of collected data:

[0141]

[0142] Among them, X represents the amount of original data, Y represents the number of fields to be processed, and η wx is the weight parameter, m is the preprocessing function, is the wth original data field and the xth processed field;

[0143] S17, build a graphene interdisciplinary dataset, and combine the D lit , D pat , D mar , D team , D fund Data integration, using a unified data format and data model to represent:

[0144]

[0145] Among them, Z represents the number of data sources, AA represents the number of data in each data source, θyz is the weight parameter, and n is the data fusion function. The y-th data source and the z-th data.

[0146] S2. Use text mining technology to extract key information related to the development of graphene technology from the collected graphene interdisciplinary datasets. The key information includes technology nodes, research directions and material properties.

[0147] In this implementation, S2 includes the following steps:

[0148] S21. Using natural language processing technology to analyze the graphene interdisciplinary dataset D gra Perform preprocessing, including word segmentation, part-of-speech tagging, and named entity recognition:

[0149]

[0150] Where N is the number of datasets, M is the number of documents in each dataset, and ω ij is the weight parameter, NER is the named entity recognition function, is the part-of-speech tagging function, ψ is the word segmentation function, is the jth document in the i-th dataset, D pre is the preprocessed data set;

[0151] S22. Extracting technology node information T from the preprocessed data set node , using an adaptive graph convolutional network model to identify technology nodes:

[0152]

[0153] Among them, P represents the number of preprocessed data sets, Q represents the number of technology nodes in each data set, and α uv is the weight parameter, σ is the activation function, H is the number of graph convolutional layers, G is the number of nodes in the graph convolutional layer, γ kl is the adaptive weight parameter, ReLU is the activation function, W kl is the weight matrix of the graph convolutional network, A uv is the adjacency matrix, is the feature vector of the first node in the kth layer, b kl is the bias term;

[0154] S23. Extract research direction information R from the preprocessed data set dir , use topic models to classify and identify research directions:

[0155]

[0156] Among them, L represents the number of preprocessed data sets, K represents the number of research directions in each data set, and β mn is the weight parameter, θ m is the topic distribution matrix, V is the number of topic models, η zn is the adaptive weight parameter, Dirichlet is the Dirichlet distribution function, is the feature extraction function, is the weight matrix of the topic model;

[0157] S24, extracting material property information M from the preprocessed data set prop , using feature extraction algorithms to parse and identify material properties:

[0158]

[0159] Where S represents the number of preprocessed data sets, T represents the number of material properties in each data set, and δ ab is the weight parameter, W is the number of feature extraction layers, ζ cw is the adaptive weight parameter, tanh is the hyperbolic tangent activation function, X is the number of nodes in the feature extraction layer, Y is the number of features in the feature extraction layer, κ de is the adaptive weight parameter, and ψ are feature extraction and processing functions, λ de is the feature extraction weight matrix, ∈ de is the bias term;

[0160] S25. Structural storage of the extracted technology nodes, research directions and material property information to form a structured data set D struct :

[0161] D struct ={T node , R dir ,M prop};

[0162] Among them, D struct is a structured data set, including technology nodes T node 、Research Direction dir and material properties M prop .

[0163] S3. Use the adaptive graph convolutional network model to perform pattern recognition and trend analysis on the extracted key information to identify the potential direction of graphene technology development;

[0164] In this implementation, S3 includes the following steps:

[0165] S31. Build an adaptive graph convolutional network model GCN adapt, initialize the model parameters Where L is the number of network layers, W (l) is the weight matrix of the lth layer, b (l) is the bias term of the first layer;

[0166] S32, extract the structured data set D struct Convert to graph structure data G = (V, E), where V is the node set and E is the edge set. Define the node feature matrix X and adjacency matrix A:

[0167] x=[x1,x2,...,x N ] T ;

[0168] A=[a ij ] N×N ;

[0169] Among them, x i is the feature vector of the i-th node, a ij is the edge weight between the i-th and j-th nodes, and N is the number of nodes;

[0170] S33, use graph convolution operation to perform convolution operation on node features and calculate the node embedding representation H of each layer (l) , and its recursive relationship is:

[0171] H (l) =σ(D -1 / 2 AD -1 / 2 H (l-1) W (l) +b (l) );

[0172] Where D is the degree matrix, H (0) =X, σ is the activation function;

[0173] S34, update the node embedding representation through the adaptive graph convolutional network, combined with the edge weight α ij and the adaptive coefficient γ ij , calculate the final node representation:

[0174] H * =∑ (i，j)∈E α ij γ ij ·σ(D -1 / 2 AD -1 / 2 H (L) W * +b * );

[0175] Among them, W * and b *is the weight matrix and bias term of the final layer;

[0176] S35, for the final node representation H * Conduct clustering and classification analysis to identify potential directions for the development of graphene technology dir , and its calculation formula is:

[0177]

[0178] Among them, K is the number of clusters, βk is the weight parameter of each cluster, Cluster is the clustering function, P is the number of nodes, Q is the number of edges, and are the parameters used in the clustering process;

[0179] S36. Based on the cluster analysis results, conduct trend analysis, predict the development path and potential application areas of graphene technology, and generate a technology development trend report R trend :

[0180]

[0181] Among them, D is the time dimension, C is the technology dimension, θ dc is the weight parameter, F is the number of potential application areas, is the domain weight parameter, T dir is the technical direction, and Proj is the projection function of time and technology.

[0182] S4. Build a knowledge graph, which includes the entities of graphene technology and the relationships between them, including the dependencies between technologies, the use of materials, and the innovations of technologies;

[0183] In this implementation, S4 includes the following steps:

[0184] S41. Define the entity set E in the field of graphene technology, including technology, tools, materials, research institutions, researchers, patents, and literature entity types:

[0185] E = {e1, e2, ..., e n};

[0186] Where n represents the number of entities, e i represents the i-th entity;

[0187] S42. Define the relationship set R between entities, including technical dependency, material usage, cooperation, and reference relationships:

[0188] R = {r1, r2, ..., r m};

[0189] Where m represents the number of relations, r j represents the jth relationship;

[0190] S43. Use natural language processing technology and data mining technology to extract entities and their relationships from the graphene interdisciplinary dataset and construct the initial knowledge graph KG init :

[0191] KG init =(E, R);

[0192] S44. Initial knowledge graph KG init Perform semantic analysis and annotation of entities and relationships to generate a knowledge graph KG with weights and attributes weighted :

[0193] KG weighted =(E, R, W E , W R );

[0194] Among them, W E is the weight set of the entity, W R is the weight set of the relationship;

[0195] S45. Combine the embedding algorithm to vectorize the knowledge graph and generate the vector representation KG of the knowledge graph. embed , and its calculation formula is:

[0196]

[0197] Among them, ω ij is the weight parameter, NER is the named entity recognition function, embed(e i , r j ) is the embedding function, is the i-th entity and j-th relationship in the knowledge graph;

[0198] S46. Combine adaptive graph convolutional networks to perform pattern recognition and trend analysis on knowledge graphs and identify potential directions T dir , identify potential directions, and thus form the final knowledge graph KG final .

[0199] S5, combining the knowledge graph in step S4 with the pattern recognition and trend analysis results in step S3, and combining convolutional neural networks and graph neural networks to learn and predict the future development path of graphene technology;

[0200] In this implementation, S5 includes the following steps:

[0201] S51, the knowledge graph KG constructed in step S4 finalThe node embedding representation H generated in step S3 * Combine to form a comprehensive data set D combined :

[0202] D combined ={KG weighted , H *};

[0203] S52. Define the convolutional neural network model CNN graph and graph neural network model GNN graph , initialize the model parameters:

[0204]

[0205]

[0206] Among them, L CNN and L GNN Represents the number of network layers of CNN and GNN respectively;

[0207] S53. Use the convolutional neural network model to analyze the comprehensive dataset D combined Perform feature extraction and generate convolutional feature representation F CNN :

[0208]

[0209] Among them, σ is the activation function, K is the number of data blocks, conv is the convolution operation function, η m is the adaptive weight parameter, is the feature extraction function;

[0210] S54. Using the graph neural network model to analyze the comprehensive dataset D combined Perform feature extraction and generate graph convolution feature representation F GNN :

[0211]

[0212]

[0213] Where N is the number of nodes, a ij is the weight in the adjacency matrix, is the feature representation of the i-th node in the l-1th layer, γ p is the adaptive weight parameter, ReLU is the activation function, ψ p is the feature extraction function;

[0214] S55, represent the convolution feature F CNN With graph convolution feature representation F GNNFusion is performed to generate the final feature representation F final :

[0215]

[0216] Among them, Q is the number of feature blocks, λ q is the fusion weight coefficient, δ r is the adaptive weight parameter, softmax is the normalization function, is the feature fusion function;

[0217] S56. Using the fused features to represent F final Predict the future development path and generate prediction results P route :

[0218]

[0219] Among them, S is the number of prediction layers, θ s is the prediction layer weight parameter, Predict is the prediction function, μ t is the adaptive weight parameter, Attention is the attention mechanism function, v t is the attention weight parameter;

[0220] S57, according to the prediction result P route Generate a future development roadmap for graphene technology to identify technology development directions and potential application areas.

[0221] S6. Dynamically update the knowledge graph and adaptive graph convolutional network model based on the prediction results to adapt to new research results and market demand changes;

[0222] S7. Output a forecast report, which includes the forecast direction of technology development, the time series of key technology nodes, and potential market application areas.

[0223] In this implementation, S7 includes the following steps:

[0224] S71. According to the prediction result P route , extract the predicted direction of technology development;

[0225] S72. Construct a time series of key technology nodes, including predicting the occurrence time and importance score of each key technology node;

[0226] S73. Identify potential market application areas and generate a list of market application areas based on the predicted technical direction and time series;

[0227] S74. Integrate forecast directions, time series of key technology nodes and potential market application areas to generate a comprehensive forecast report;

[0228] S75. Display the comprehensive forecast report in a visual form, including line graphs, bar graphs and pie charts, and output the final version of the comprehensive forecast report for reference by R&D teams, enterprises and policymakers for making R&D or investment decisions.

[0229] Embodiment 1:

[0230] We applied the technology of the present invention to a high-tech enterprise that focuses on the research and development of graphene technology to test the feasibility and effectiveness of the present invention. The high-tech enterprise expects to improve its competitiveness and technological innovation capabilities in the market through forward-looking technology route predictions in the next five years. The R&D projects involved in the enterprise include highly conductive graphene composite materials, graphene-enhanced composite materials, graphene-based sensors and graphene films.

[0231] The company's R&D team first collected a large amount of scientific literature, patent information, market analysis reports and other data related to the field of graphene to build an interdisciplinary data set. Next, text mining technology was used to extract key information such as technology nodes, research directions and material properties, and pattern recognition and trend analysis were performed through an adaptive graph convolutional network model.

[0232] By constructing a knowledge graph, the extracted information is structured to form the relationships between the various entities of graphene technology. Subsequently, the knowledge graph is deeply learned by combining convolutional neural networks and graph neural networks to predict the future development path of graphene technology. Finally, a comprehensive forecast report is generated, which details the forecast direction of technology development, the time series of key technology nodes, and potential market application areas.

[0233] After applying this method, the company conducted data analysis and technical forecasts from January 2023 to April 2024. The specific data are as follows:

[0234] In January 2023, the company began to collect data in the field of graphene, involving 2,000 scientific papers, 1,500 patent information, 300 market analysis reports and 50 international research funding dynamics. Through text mining technology, 3,000 technology nodes, 500 research directions and 2,000 material property data were extracted.

[0235] In March 2023, an initial knowledge graph was constructed, containing 10,000 entities and 25,000 relationships. Adaptive graph convolutional networks were used for pattern recognition to identify 500 potential technology development directions.

[0236] In April 2023, we combined convolutional neural networks and graph neural networks to generate a comprehensive data set and conduct deep learning to predict the future development path of graphene technology. The accuracy of the prediction model was verified by comparing historical data and market trends.

[0237] In June 2023, based on the forecast results, the company successfully achieved laboratory-scale preparation technology for highly conductive graphene composites, marking the first key node in this technology direction.

[0238] In August 2023, the company further realized the industrial production process of graphene-enhanced composite materials, proving the effectiveness of the predictive model in industrial applications.

[0239] In December 2023, a breakthrough was achieved in the commercial application of graphene-based sensors, which were widely used in medical diagnosis and environmental monitoring.

[0240] In February 2024, the company achieved a breakthrough in the technology of mass production of graphene films, which significantly reduced production costs and promoted the development of flexible electronic products.

[0241] In April 2024, the widespread application of graphene in the new generation of flexible electronic devices achieved the long-term goal in the forecast.

[0242] Forecast report:

[0243] 1. Predicted direction of technological development

[0244] Through the analysis and trend forecast of graphene technology, the following main technology development directions are identified:

[0245] Highly conductive graphene composites: for advanced electronic devices and battery technology.

[0246] Graphene enhanced composites: applications in aerospace and automotive industries.

[0247] Graphene-based sensors: for medical diagnostics and environmental monitoring.

[0248] Graphene films: used in flexible electronics and transparent conductive films.

[0249] 2. Time series of key technology nodes

[0250] The predicted key technology nodes and their time series are as follows:

[0251] 2023: Laboratory-scale preparation technology for highly conductive graphene composites.

[0252] 2023: Industrial production process of graphene-enhanced composites.

[0253] 2024: Commercial application of graphene-based sensors.

[0254] 2024: Breakthrough technology for mass production of graphene films.

[0255] 2024: Graphene is widely used in the new generation of flexible electronic devices.

[0256] 3. Potential market application areas

[0257] Based on the forecast direction and time series of technology development, the following potential market application areas are identified:

[0258] Electronic device manufacturers: Especially those requiring highly conductive materials such as smartphones, batteries, and flexible electronics.

[0259] Aerospace and automotive manufacturing: Application of graphene-enhanced composites in the demand for lightweight and high-strength materials.

[0260] Medical equipment and environmental monitoring industries: The high sensitivity and accuracy of graphene-based sensors are suitable for medical diagnosis and environmental pollution detection.

[0261] Display technology and solar cell industry: Graphene films have broad application prospects in transparent conductive films and flexible displays.

[0262] Visual display:

[0263] Line chart of technology development direction: shows the research progress and time nodes of each technology direction.

[0264] Time series bar chart of key technical nodes: displays the time and importance score of each key node.

[0265] Pie chart of potential market application areas: showing the application proportion and prospects of different market areas.

[0266] The present invention collects and integrates interdisciplinary data such as scientific literature, patent information, market analysis, etc. in the field of graphene, uses text mining technology to extract key information such as technical nodes, research directions and material properties, and constructs a comprehensive interdisciplinary data set for graphene technology. Through the combination of adaptive graph convolutional networks and convolutional neural networks, efficient data integration and analysis are achieved, overcoming the problem of inefficient processing of heterogeneous data in traditional methods.

[0267] The method of the present invention combines the adaptive graph convolutional network model and the convolutional neural network with the graph neural network, and can dynamically update the prediction model to adapt to new research results and changes in market demand. Through continuous learning and updating, the timeliness and accuracy of the prediction results are maintained, providing reliable decision-making support for R&D teams, enterprises and policy makers.

[0268] This invention builds a knowledge graph and uses an embedding algorithm and an adaptive graph convolutional network to perform pattern recognition and trend analysis, deeply explores the key nodes and research directions of graphene technology development, and identifies potential technology development paths and important technology nodes. This process improves the ability to capture complex technology relationships and development dynamics, and makes up for the shortcomings of traditional methods that only make simple predictions based on the number of documents or patents.

[0269] By identifying and predicting potential market application areas, the present invention not only provides the predicted direction of technology development and the time series of key technology nodes, but also can generate a list of market application areas according to the predicted technology direction, helping enterprises and R&D teams to better plan market strategies and R&D investment, and ensure the close integration of technology development and market demand.

[0270] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A graphene technology route prediction method based on a large model and knowledge graph, characterized in that: The steps include: S1. Collect scientific literature, patent information, market analysis, scientific research team cooperation network information and international research funding trends in the field of graphene to build a graphene interdisciplinary dataset; S2. Use text mining technology to extract key information related to the development of graphene technology from the collected graphene interdisciplinary data set, the key information including technology nodes, research directions and material properties; S3. Use the adaptive graph convolutional network model to perform pattern recognition and trend analysis on the extracted key information to identify the potential direction of graphene technology development; S4. Build a knowledge graph, which includes the entities of graphene technology and the relationships between them, including the dependencies between technologies, the use of materials, and the innovations of technologies; S5, combining the knowledge graph in step S4 with the pattern recognition and trend analysis results in step S3, and combining convolutional neural networks and graph neural networks to learn and predict the future development path of graphene technology; S6. Dynamically update the knowledge graph and adaptive graph convolutional network model based on the prediction results to adapt to new research results and market demand changes; S7. Output forecast report, which includes the forecast direction of technology development, the time series of key technology nodes and potential market application areas; The S5 comprises the following steps: S51, the knowledge graph KG constructed in step S4 final The node embedding representation H generated in step S3 * Combine to form a comprehensive data set D combined : D combined ={KG final ,H * }; S52. Define the convolutional neural network model CNN graph and graph neural network model GNN graph , initialize the model parameters: Among them, L CNN and L GNN Represents the number of network layers of CNN and GNN respectively; S53. Use the convolutional neural network model to analyze the comprehensive dataset D combined Perform feature extraction and generate convolutional feature representation F CNN : Among them, σ is the activation function, K is the number of data blocks, conv is the convolution operation function, η m is the adaptive weight parameter, is the feature extraction function; S54. Using the graph neural network model to analyze the comprehensive dataset D combined Perform feature extraction and generate graph convolution feature representation F GNN : Where N is the number of nodes, a ij is the weight in the adjacency matrix, is the feature representation of the i-th node in the l-1th layer, γ p is the adaptive weight parameter, ReLU is the activation function, ψ p is the feature extraction function; S55, represent the convolution feature F CNN With graph convolution feature representation F GNN Fusion is performed to generate the final feature representation F final : Among them, Q is the number of feature blocks, λ q is the fusion weight coefficient, δ r is the adaptive weight parameter, softmax is the normalization function, is the feature fusion function; S56. Using the fused features to represent F final Predict the future development path and generate prediction results P route : Among them, S is the number of prediction layers, θ s is the prediction layer weight parameter, Predict is the prediction function, μ t is the adaptive weight parameter, Attention is the attention mechanism function, ν t is the attention weight parameter; S57, according to the prediction result P route Generate a future development roadmap for graphene technology to identify technology development directions and potential application areas.

2. A graphene technology route prediction method based on a large model and knowledge graph according to claim 1, characterized in that: The S2 comprises the following steps: S21. Using natural language processing technology to analyze the graphene interdisciplinary dataset D gra Perform preprocessing, including word segmentation, part-of-speech tagging, and named entity recognition: Where N is the number of datasets, M is the number of documents in each dataset, and ω ij is the weight parameter, NER is the named entity recognition function, is the part-of-speech tagging function, ψ is the word segmentation function, is the jth document in the i-th dataset, D pre is the preprocessed data set; S22. Extracting technology node information T from the preprocessed data set node , using an adaptive graph convolutional network model to identify technology nodes: Among them, P represents the number of preprocessed data sets, Q represents the number of technology nodes in each data set, and α uv is the weight parameter, σ is the activation function, H is the number of graph convolutional layers, G is the number of nodes in the graph convolutional layer, γ kl is the adaptive weight parameter, ReLU is the activation function, W kl is the weight matrix of the graph convolutional network, A uv is the adjacency matrix, is the feature vector of the lth node in the kth layer, b kl is the bias term; S23. Extract research direction information R from the preprocessed data set dir , use topic models to classify and identify research directions: Among them, L represents the number of preprocessed data sets, K represents the number of research directions in each data set, and β mn is the weight parameter, θ m is the topic distribution matrix, V is the number of topic models, η zn is the adaptive weight parameter, Dirichlet is the Dirichlet distribution function, is the feature extraction function, is the weight matrix of the topic model; S24, extracting material property information M from the preprocessed data set prop , using feature extraction algorithms to parse and identify material properties: Where S represents the number of preprocessed data sets, T represents the number of material properties in each data set, and δ ab is the weight parameter, W is the number of feature extraction layers, ζ cw is the adaptive weight parameter, tanh is the hyperbolic tangent activation function, X is the number of nodes in the feature extraction layer, Y is the number of features in the feature extraction layer, κ de is the adaptive weight parameter, and ψ are feature extraction and processing functions, λ de is the feature extraction weight matrix, ∈ de is the bias term; S25. Structural storage of the extracted technology nodes, research directions and material property information to form a structured data set D struct : D struct ={T node ,R dir ,M prop }; Among them, D struct is a structured data set, including technology nodes T node 、Research Direction dir and material properties M prop .

3. A graphene technology route prediction method based on a large model and knowledge graph according to claim 2, characterized in that: The S3 comprises the following steps: S31. Build an adaptive graph convolutional network model GCN adapt , initialize the model parameters Where L is the number of network layers, W (l) is the weight matrix of the lth layer, b (l) is the bias term of the lth layer; S32, extract the structured data set D struct Convert to graph structure data G = (V, E), where V is the node set and E is the edge set. Define the node feature matrix X and adjacency matrix A: X=[x1,x2,…,x N ] T ; A=[a ij ] N×N ; Among them, x i is the feature vector of the i-th node, a ij is the edge weight between the i-th and j-th nodes, and N is the number of nodes; S33, use graph convolution operation to perform convolution operation on node features and calculate the node embedding representation H of each layer (l) , and its recursive relationship is: H (l) =σ(D -1 / 2 AD -1 / 2 H (l-1) W (l) +b (l) ); Where D is the degree matrix, H (0) =X, σ is the activation function; S34, update the node embedding representation through the adaptive graph convolutional network, combined with the edge weight α ij and the adaptive coefficient γ ij , calculate the final node representation: H * =∑ (i,j)∈E α ij ·γ ij ·σ(D -1 / 2 AD -1 / 2 H (L) W * +b * ); Among them, W * and b * is the weight matrix and bias term of the final layer; S35, for the final node representation H * Conduct clustering and classification analysis to identify potential directions for the development of graphene technology dir , and its calculation formula is: Among them, K is the number of clusters, β k is the weight parameter of each cluster, Cluster is the clustering function, P is the number of nodes, Q is the number of edges, and are the parameters used in the clustering process; S36. Based on the cluster analysis results, conduct trend analysis, predict the development path and potential application areas of graphene technology, and generate a technology development trend report R trend : Among them, D is the time dimension, C is the technology dimension, θ dc is the weight parameter, F is the number of potential application areas, is the domain weight parameter, T dir is the technical direction, and Proj is the projection function of time and technology.

4. A graphene technology route prediction method based on a large model and knowledge graph according to claim 3, characterized in that: The S4 comprises the following steps: S41. Define the entity set E in the field of graphene technology, including technology, tools, materials, research institutions, researchers, patents, and literature entity types: E={e1,e2,…,e n }; Where n represents the number of entities, e i represents the i-th entity; S42. Define the relationship set R between entities, including technical dependency, material usage, cooperation, and reference: R＝{r1,r2,…,r m}； Where m represents the number of relations, r j represents the jth relationship; S43. Use natural language processing and data mining techniques to extract entities and their relationships from the graphene interdisciplinary dataset and construct an initial knowledge graph KG init : KG init =(E,R); S44. Initial knowledge graph KG init Perform semantic analysis and annotation of entities and relationships to generate a knowledge graph KG with weights and attributes weighted : KG weighted =(E,R,W E ,W R ); Among them, W E is the weight set of the entity, W R is the weight set of the relationship; S45. Combine the embedding algorithm to vectorize the knowledge graph and generate the vector representation KG of the knowledge graph. embed , and its calculation formula is: Among them, ω ij is the weight parameter, NER is the named entity recognition function, embed(e i ,r j ) is the embedding function, is the i-th entity and j-th relationship in the knowledge graph; S46. Combine adaptive graph convolutional networks to perform pattern recognition and trend analysis on knowledge graphs and identify potential directions T dir , identify potential directions, and thus form the final knowledge graph KG final .

5. A graphene technology route prediction method based on a large model and knowledge graph according to claim 4, characterized in that: The S7 comprises the following steps: S71. According to the prediction result P route , extract the predicted direction of technology development; S72. Construct a time series of key technology nodes, including predicting the occurrence time and importance score of each key technology node; S73. Identify potential market application areas and generate a list of market application areas based on the predicted technical direction and time series; S74. Integrate forecast directions, time series of key technology nodes and potential market application areas to generate a comprehensive forecast report; S75. Display the comprehensive forecast report in a visual form, including line graphs, bar graphs and pie charts, and output the final version of the comprehensive forecast report for reference by R&D teams, enterprises and policymakers for making R&D or investment decisions.

Images