Industrial internet information model modeling optimization method based on natural language processing

Abstract

The invention provides an industrial internet information model modeling optimization method based on natural language processing, and relates to the technical field of computers. The method comprises the following steps: inputting a preset data source and model prompt information into a preset natural language processing model, and extracting to obtain modeling elements; the model prompt information comprises first-stage prompt information and second-stage prompt information; dividing the modeling elements into a corresponding number of modeling element clusters based on the target cluster number; constructing an information model based on the modeling element cluster; the information model is used for indicating the relationship among the modeling elements; the information model comprises a tree structure layer or a combination of the tree structure layer and a net structure layer; wherein the tree structure layer comprises an object class layer corresponding to the information model, a cluster class layer corresponding to each modeling element cluster and a representative element layer corresponding to each modeling element cluster; the problem that an information model is poor in effect and reliability can be solved. And the effect and reliability of the information model are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an industrial Internet information modeling optimization method based on natural language processing. Background Art

[0002] Information model is an abstract and standardized description of real-world information and data, expressing the structure, relationship and flow of information through unified standards. In industrial production, information model is used to describe the relationship between elements such as production process, equipment, technology and materials, ensuring that different systems and equipment can exchange and share information efficiently and accurately, and supporting production decision-making and control. As the basis for standardized expression and flow of production information, information model is an effective solution to improve the integration and interoperability of production information.

[0003] The traditional natural language processing-based industrial Internet information modeling optimization method includes: manually collecting and processing data and building information models.

[0004] However, due to the difficulty in obtaining data and difficulty in standardization, traditional modeling methods rely on manual collection and processing of data and lack sufficient automated tools and technologies, which is time-consuming and labor-intensive, resulting in incomplete and inaccurate model construction, poor data quality and integrity, and thus poor effectiveness and reliability of information models. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides an industrial Internet information model modeling optimization method based on natural language processing to eliminate or improve one or more defects existing in the prior art, and can solve the problem of poor effect and reliability of the information model.

[0006] One aspect of the present invention provides an industrial Internet information model modeling optimization method based on natural language processing, the method comprising the following steps:

[0007] Input the preset data source and model prompt information into the preset natural language processing model to extract modeling elements; the model prompt information includes first-stage prompt information and second-stage prompt information; the first-stage prompt information is used to prompt the preset natural language processing model to extract entity types from the preset data source according to the preset entity type list; the second-stage prompt information is used to prompt the preset natural language processing model to extract modeling elements based on the extracted entity types;

[0008] Divide the modeling elements into a corresponding number of modeling element clusters based on the number of target clusters;

[0009] An information model is constructed based on modeling element clusters; the information model is used to indicate the relationship between each modeling element, including the internal relationship between modeling elements within each modeling element cluster and the cross-cluster relationship between modeling elements between different modeling element clusters. The information model includes a tree structure layer, or a combination of a tree structure layer and a mesh structure layer.

[0010] In some embodiments of the present invention, before dividing the modeling elements into a corresponding number of modeling element clusters based on the target cluster number, the method further includes:

[0011] Perform cluster analysis on modeling elements based on a preset cluster number set, and calculate the silhouette coefficients corresponding to clustering of modeling elements under different cluster numbers;

[0012] The target number of clusters is determined from a preset number of clusters based on the silhouette coefficient.

[0013] In some embodiments of the present invention, before dividing the modeling elements into a corresponding number of modeling element clusters based on the target cluster number, the method further includes:

[0014] Determine the target semantic similarity threshold corresponding to each modeling element cluster;

[0015] The modeling elements in each modeling element cluster are compared pairwise to determine the semantic similarity corresponding to each pair of modeling elements;

[0016] In each modeling element cluster, modeling element pairs whose semantic similarity is greater than or equal to the corresponding target semantic similarity threshold are clustered.

[0017] In some embodiments of the present invention, determining a target semantic similarity threshold corresponding to each modeling element cluster includes:

[0018] Within a preset semantic similarity threshold range, analyzing and obtaining a first relationship between the silhouette coefficient corresponding to each modeling element cluster and the semantic similarity, and a second relationship between the number of modeling elements corresponding to each modeling element cluster and the semantic similarity;

[0019] Based on the first relationship and the second relationship, a target semantic similarity threshold corresponding to each modeling element cluster is determined within a preset semantic similarity threshold range.

[0020] In some embodiments of the present invention, before determining the target semantic similarity threshold corresponding to each modeling element cluster, the method further includes:

[0021] determining the number of modeling elements in each modeling element cluster;

[0022] For modeling element clusters with a number of modeling elements greater than 2, a step of determining a target semantic similarity threshold corresponding to each modeling element cluster is performed.

[0023] In some embodiments of the present invention, after inputting the preset data source and model prompt information into the preset natural language processing model and extracting the modeling elements, it also includes: converting the modeling elements into corresponding word vector forms.

[0024] In some embodiments of the present invention, the tree structure layer includes an object class layer corresponding to the information model, a cluster category layer corresponding to each modeling element cluster, and a representative element layer corresponding to each modeling element cluster; the object class layer is used to indicate the modeling object corresponding to the information model; the cluster category layer is used to indicate the category corresponding to each modeling element cluster; and the representative element layer is used to indicate the representative element corresponding to each modeling element cluster.

[0025] In some embodiments of the present invention, constructing an information model based on a modeling element cluster includes:

[0026] Based on the modeling element clusters and association relationships, a network structure layer is constructed;

[0027] Determine the average word vector of the modeling element word vectors in each modeling element cluster respectively, and compare the semantic similarity with the preset modeling element cluster category to obtain a comparison result; the comparison result is used to indicate the similarity between each modeling element cluster and each preset modeling element cluster category;

[0028] Based on the comparison results, determine the category corresponding to each modeling element cluster and construct the cluster category layer;

[0029] Based on the internal relationships, the representative elements corresponding to each modeling element cluster are determined, and the representative element layer is constructed.

[0030] In some embodiments of the present invention, constructing an information model based on a modeling element cluster further includes:

[0031] At least two information sub-models are constructed based on the cluster of modeling elements;

[0032] An information model is obtained by combining at least two information sub-models.

[0033] In some embodiments of the present invention, the preset data source includes a text data source and / or a knowledge base corresponding to a preset natural language processing model; when the preset data source includes a text data source, the model prompt information is input into the preset natural language processing model, and before the modeling elements are extracted, it also includes: inputting the text data source as an attachment into the preset natural language processing model.

[0034] The industrial Internet information model modeling optimization method based on natural language processing of the present invention can solve the problem of poor effect and reliability of information models; by presetting the natural language processing model, according to the preset entity type list indicated by the first-stage model prompt information, the entity type is automatically obtained from the preset data source, and then the entity information is automatically extracted from the preset data source through the second-stage model prompt information and entity type, without relying on manual collection and processing, avoiding the problem of poor data quality and integrity, thereby improving the effect and reliability of the information model; at the same time, using a unified preset entity type list to obtain entity types can overcome the compatibility barriers between different standards and improve the compatibility of information models.

[0035] In addition, after determining the target semantic similarity threshold, the modeling elements in each modeling element cluster are optimized, that is, according to the target semantic similarity threshold, it is determined whether the modeling elements can be aggregated to reduce the number of modeling elements, improve model accuracy, and thus optimize model performance.

[0036] Additional advantages, purposes, and features of the present invention will be described in part in the following description, and will become apparent to those skilled in the art after studying the following, or may be learned from the practice of the present invention. The purposes and other advantages of the present invention may be achieved and obtained by the structures specifically indicated in the specification and the accompanying drawings.

[0037] Those skilled in the art will appreciate that the objectives and advantages that can be achieved with the present invention are not limited to the above specific description, and the above and other objectives that can be achieved by the present invention will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings:

[0039] Figure 1 A flowchart of an industrial Internet information modeling optimization method based on natural language processing provided by an embodiment of the present invention;

[0040] Figure 2 A schematic diagram of text-driven prompt learning provided by an embodiment of the present invention;

[0041] Figure 3 A schematic diagram of knowledge-driven prompt learning provided by an embodiment of the present invention;

[0042] Figure 4 A graph showing the relationship between the silhouette coefficient and the number of clusters provided in one embodiment of the present invention;

[0043] Figure 5A schematic diagram of the division of modeling element clusters provided by an embodiment of the present invention;

[0044] Figure 6 A schematic diagram of a first relationship and a second relationship provided in one embodiment of the present invention;

[0045] Figure 7 Another schematic diagram of a first relationship and a second relationship provided for an embodiment of the present invention;

[0046] Figure 8 Another schematic diagram of a first relationship and a second relationship provided for an embodiment of the present invention;

[0047] Fig. 9 Another schematic diagram of a first relationship and a second relationship provided for an embodiment of the present invention;

[0048] Fig.10 Another schematic diagram of a first relationship and a second relationship provided for an embodiment of the present invention;

[0049] Fig.11 A semantic similarity heat map provided by an embodiment of the present invention;

[0050] Fig.12 Another semantic similarity heat map provided by an embodiment of the present invention;

[0051] Fig.13 Another semantic similarity heat map provided by an embodiment of the present invention;

[0052] Fig.14 Another semantic similarity heat map provided by an embodiment of the present invention;

[0053] Fig.15 Another semantic similarity heat map provided by an embodiment of the present invention;

[0054] Fig.16 A schematic diagram of a framework of an information model provided by an embodiment of the present invention;

[0055] Fig.17 A schematic diagram of another information model framework provided by an embodiment of the present invention;

[0056] Fig.18 A hierarchical clustering dendrogram provided by an embodiment of the present invention;

[0057] Fig.19 A schematic diagram of another information model framework provided by an embodiment of the present invention;

[0058] Fig. 20 A radar chart showing the category similarity of a cluster of modeling elements provided by an embodiment of the present invention;

[0059] Fig.21 A schematic diagram of another information model framework provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0061] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.

[0062] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0063] It should also be noted that, unless otherwise specified, the term “connection” herein may refer not only to a direct connection but also to an indirect connection involving an intermediate.

[0064] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0065] The following is a detailed introduction to the industrial Internet information modeling optimization method based on natural language processing provided in this application.

[0066] like Figure 1 As shown, the embodiment of the present application provides an industrial Internet information model modeling optimization method based on natural language processing. The implementation of the method can rely on a computer program, which can be run on a computer device such as a smart phone, a tablet computer, a personal computer, or a server. This embodiment does not limit the operating subject of the method. The method includes at least the following steps:

[0067] Step S101: input model prompt information into a preset natural language processing model to extract modeling elements.

[0068] In some embodiments of the present invention, extracting modeling elements refers to the process of extracting retrievable and editable entity information related to the modeling purpose of the modeling object from a preset data source through a preset natural language processing model. The extracted modeling elements are used to construct an information model.

[0069] Among them, the preset natural language processing model refers to a pre-trained language model, including but not limited to Chat Generative Pre-trained Transformer (ChatGPT) or Language Model for Dialogue Applications (LaMDA); entity information refers to specific data describing actual objects, concepts or events, including their attributes, characteristics, status, etc. For example, taking the industrial field as an example, entity information refers to independently existing things or concepts that are identified and defined in the industrial field, such as manufacturer, equipment name, production date, etc.

[0070] Before extracting modeling elements, you also need to clarify the modeling purpose of the modeling object. For example, if the modeling purpose is to manage equipment operation, the extraction of modeling elements should focus on the equipment operation status; if the modeling purpose is to assemble manufacturing equipment, the modeling elements include information on equipment functions and appearance.

[0071] In some embodiments of the present invention, the preset data source includes a text data source and / or a knowledge base corresponding to a preset natural language processing model; wherein the text data source includes technical documents, operation manuals, white papers and other unstructured texts related to the modeling object; the knowledge base corresponding to the preset natural language processing model includes the language knowledge and background knowledge learned during the model training process, as well as the knowledge acquired from the network through a search engine.

[0072] When the preset data source is a text data source, the model prompt information is input into the preset natural language processing model, and before the modeling elements are extracted, the method further includes: inputting the text data source as an attachment into the preset natural language processing model.

[0073] Upload the text data source to the preset natural language processing model in the form of a neighborhood, extract information from the preset data source using a text-driven prompt learning method, combine the preset data source and model prompt information as the input of the preset natural language processing model, use the preset natural language processing model to obtain more complete and accurate modeling elements, and filter out irrelevant elements in the text data source to reduce the search space and computational complexity. Through the rich semantic information accumulated by the preset natural language processing model after training on large-scale text data, the subtle semantic relationship between sentences or paragraphs is captured to understand the complex context. Extracting precise modeling elements from the preset data source makes the results more relevant and accurate.

[0074] In some embodiments of the present invention, the process of obtaining modeling elements from a preset data source is designed as multiple rounds of question and answer executed sequentially, including two stages, each of which includes multiple rounds of question and answer. Accordingly, the model prompt information includes first-stage prompt information and second-stage prompt information; wherein the first-stage prompt information is used to prompt the preset natural language processing model to extract entity types from the preset data source according to the preset entity type list; the second-stage prompt information is used to prompt the preset natural language processing model to extract modeling elements based on the extracted entity types.

[0075] The entity type list refers to a list of categories or types used to classify and identify different entities in the text, including but not limited to identifiers, attributes, events, services, parameters, relationships, classes, etc.

[0076] In actual implementation, in each round of dialogue, prompts can be constructed based on the previously extracted information as input to the preset natural language processing model; at the same time, in the second stage, according to the entity types extracted in the first stage, combined with specific constraints such as modeling purpose and key modeling elements, relevant information is further extracted and classified. Finally, the information extracted from each round of dialogue is combined into the final structured data, that is, the required modeling elements.

[0077] For example: Reference Figure 2 , taking the construction of an information model in the industrial field, the preset data source is a text-based instruction manual, white paper, technical documentation and other materials related to the modeling object, and the preset data source is uploaded as an attachment; in the first stage, the first stage prompt information is used to instruct the preset natural language processing model to extract entity types (such as identifiers, attributes) from the preset data source according to a given entity type list; in the second stage, the second stage prompt information is used to instruct the preset natural language processing model to extract entities from the preset data source according to the entity type as modeling elements (such as equipment number, part number, cutting speed, feed speed).

[0078] In actual implementation, the access to text data sources is limited; at the same time, there is a problem that the text quality of the acquired text data source is low, resulting in a problem that the extracted modeling elements have poor effects. Therefore, this embodiment also proposes a method of knowledge-driven prompt learning to extract modeling elements, and the preset data source also includes a knowledge base corresponding to the preset natural language processing model.

[0079] Compared with the text-driven prompt learning method, the knowledge-driven prompt learning method does not require any additional input, but directly generates relevant entities based on the knowledge base of the preset natural language processing model by designing targeted prompts. It can enhance the versatility and adaptability of the information model, and can effectively extract entity information and maintain high accuracy when the data source information is missing or incomplete.

[0080] For example: Reference Figure 3 , in order to build an information model for the industrial field; in the first stage, the first stage prompt information is used to instruct the preset natural language processing model to extract entity types (such as identifiers, events) from the knowledge base of the preset natural language processing model according to a given entity type list; in the second stage, the second stage prompt information is used to instruct the preset natural language processing model to extract entities from the knowledge base of the preset natural language processing model according to the entity type as modeling elements (such as version numbers, artifact loads, etc.).

[0081] Since text-driven hint learning methods may be limited by text quality and coverage, and knowledge-driven hint learning methods may be limited by known information in the knowledge base, in actual implementation, the two can be combined to complement each other's shortcomings, maximize coverage of information in related fields, reduce omissions, and improve the comprehensiveness and accuracy of information extraction.

[0082] Take the automatic extraction of modeling elements of CNC machine tools as an example; combining the text-driven method and the knowledge-driven method, in the text-driven method, CNC equipment series standards, CNC machine tool operation manuals and CNC machine tool technical documents are selected as preset data sources to input the preset natural language processing model, and the entity type and entity information are extracted based on the seven key elements of the 3IM framework (Industrial Internet Information Model, 3IM) (identifier, attribute, event, service, parameter, relationship, class). In the first stage, the prompt information of the first stage is: "Given the text is the content of the attachment, given the entity type list: [identifier, attribute, event, service, parameter, relationship, class], what entity types may be included in this text?"; in the second stage, the prompt information of the second stage is "Based on the above text, please output the entities with the entity type of "identifier" in the form of a list, similar to: ["entity type", "entity name"], ["entity type", "entity name 2"]...", and "Among the entities with the entity type of "identifier" just given, which entities are related to "CNC machine tools"? ", etc. In the knowledge-driven method, entity types and entity information are extracted from the knowledge base of the preset natural language processing model itself. In the first stage, the prompt information is "Given a list of entity types: [identifier, attribute, event, service, parameter, relationship], what entity types may the modeling object "CNC machine tool" contain?" In the second stage, the prompt information is "Please output entities with entity type "identifier" in the form of a list, similar to: ["entity type", "entity name"], ["entity type", "entity name 2"]...".

[0083] Step S102: dividing the modeling elements into a corresponding number of modeling element clusters based on the target number of clusters.

[0084] In some embodiments of the present invention, in order to reasonably classify modeling elements so that the constructed information model is clearer, more orderly, and easier to maintain and expand, it is necessary to divide the modeling elements according to the number of target clusters to obtain a corresponding number of modeling element clusters.

[0085] The target number of clusters is obtained by globally optimizing the modeling elements, performing cluster analysis on the modeling elements according to a preset number of clusters, and determining the number of clusters from the preset number of clusters based on the obtained analysis results.

[0086] Specifically, first, a preset cluster number set is set, which includes several cluster numbers; then, under each cluster number, the modeling elements are clustered to obtain modeling element clusters, and the corresponding silhouette coefficients are calculated. After that, by comparing the silhouette coefficients under different cluster numbers, the compactness of the modeling elements in the modeling element cluster and the separation degree of the elements between clusters are comprehensively evaluated, so as to determine the target cluster number, and divide the modeling elements into the corresponding number of clusters according to the target cluster number.

[0087] Specifically, before dividing the modeling elements into a corresponding number of modeling element clusters based on the target number of clusters, it also includes: performing cluster analysis on the modeling elements based on a preset number of clusters set, calculating the silhouette coefficients corresponding to the clustering of the modeling elements under different numbers of clusters; and determining the target number of clusters in the preset number of clusters set based on the silhouette coefficients.

[0088] In some embodiments of the present invention, the target number of clusters refers to the number of clusters corresponding to the largest silhouette coefficient among the silhouette coefficients.

[0089] refer to Figure 4 , take the automatic extraction of modeling elements of CNC machine tools as an example to illustrate, set the preset cluster number set to [2,10], then perform hierarchical clustering on all modeling elements within the range of this set, calculate the corresponding silhouette coefficient under each cluster number, and draw a relationship between the cluster number and the silhouette coefficient, as shown in Figure 4 As shown. Figure 4From the analysis, we can see that in the initial stage, as the number of clusters increases, the number of modeling elements in each cluster decreases, the compactness within the cluster increases, and the difference between clusters increases, so the silhouette coefficient increases. When the number of clusters continues to increase to a certain extent, the number of modeling elements contained in each cluster is too small, the stability and representativeness of the cluster deteriorate, and the silhouette coefficient deteriorates. Although continuing to increase the number of clusters may find a better clustering solution again, so that the silhouette coefficient has recovered, it has not reached the highest point in the initial stage. Therefore, when the silhouette coefficient is the largest when the number of clusters is 6, it means that the compactness within the cluster and the separation between clusters are the best at this time, and the comprehensive effect is good. Therefore, the target number of clusters is determined to be 6.

[0090] After determining the target number of clusters, the modeling elements of the CNC machine tool are divided into a corresponding number of modeling element clusters. In order to provide an intuitive reference for further analysis, all modeling elements are clustered and the corresponding cluster scatter plots are drawn, as shown in Figure 2. Figure 5 shown.

[0091] Figure 5 There are six clusters displayed, named Cluster0, Cluster1, Cluster2, Cluster3, Cluster4 and Cluster5. Among them, there are 5 modeling elements in Cluster0, Cluster0 = {cutting speed, spindle speed, feed adjustment rate, feed override, feed speed}; there are 4 modeling elements in Cluster1, Cluster1 = {operating status, working status, alarm status, operating time}; there are 5 modeling elements in Cluster2, Cluster2 = {equipment type, factory batch, manufacturer, serial number, production date}; there are 8 modeling elements in Cluster3, Cluster3 = {spindle servo system, spindle drive system, main transmission system, servo system, position follow-up system, feed servo system, feed transmission system, feed drive system}; there are 12 modeling elements in Cluster4, Cluster4 = {CNC system, programmable controller, auxiliary system, sensor, cooling system, lubrication system, protection system, hydraulic system, oil pressure system, fixture system, chip removal system, clamping system}; there are 2 modeling elements in Cluster5, Cluster5 = {material inbound, material outbound}.

[0092] However, although a limited number of modeling elements are obtained, these modeling elements may contain unnecessary redundant information, which makes the information model relationship construction complex, increases the system development and maintenance costs, and reduces the response speed. At the same time, if too few modeling elements are retained, the information model may lose necessary details and key semantic information, resulting in information loss and insufficient model accuracy. Therefore, in order to further screen and optimize the modeling elements, a balance is achieved between model complexity and information integrity.

[0093] In some embodiments of the present invention, the modeling elements in each modeling element cluster obtained by division are locally optimized according to the target semantic similarity threshold to reduce the number of modeling elements. The target semantic similarity threshold is an indicator for judging whether the modeling elements can be aggregated, and is used to evaluate the similarity of the modeling elements in the modeling element cluster. In the same modeling element cluster, if the semantic similarity of two modeling elements exceeds the target semantic similarity threshold, one of the modeling elements is selected to be retained and the other modeling element is deleted until the similarity of the modeling elements in the modeling element cluster is lower than the target semantic similarity threshold.

[0094] Specifically, after dividing the modeling elements into a corresponding number of modeling element clusters based on the target number of clusters, it also includes: determining a target semantic similarity threshold corresponding to each modeling element cluster; comparing the modeling elements in each modeling element cluster pairwise to determine the semantic similarity corresponding to each pair of modeling elements; and in each modeling element cluster, aggregating the modeling element pairs whose semantic similarity is greater than or equal to the corresponding target semantic similarity threshold.

[0095] Specifically, within the preset semantic similarity threshold range, the changes between the silhouette coefficient of each modeling element cluster and the number of retained modeling elements are analyzed, and the target semantic similarity threshold corresponding to each modeling element cluster is selected according to the analysis results; in this process, by analyzing the silhouette coefficients corresponding to different semantic similarity thresholds, it is determined which semantic similarity threshold makes the clustering effect better; at the same time, by observing the number of retained modeling elements, the modeling elements can be locally optimized as much as possible while ensuring the clustering effect, balancing the clustering quality and data volume, while improving the integrity and practicality of the model.

[0096] Specifically, determining a target semantic similarity threshold corresponding to each modeling element cluster includes: within a preset semantic similarity threshold range, analyzing to obtain a first relationship between a silhouette coefficient corresponding to each modeling element cluster and the semantic similarity, and a second relationship between the number of modeling elements corresponding to each modeling element cluster and the semantic similarity; based on the first relationship and the second relationship, determining a target semantic similarity threshold corresponding to each modeling element cluster within a preset semantic similarity threshold range.

[0097] Take the automatic extraction of modeling elements of CNC machine tools as an example to illustrate. Figure 5When the modeling element clusters (Cluster0, Cluster1, Cluster2, Cluster3 and Cluster4) in are locally optimized, the semantic similarity threshold range is set to [0.75, 1.05] with a step size of 0.05. Too low a threshold may remove a large number of modeling elements, resulting in incomplete information expression. Therefore, setting the lower threshold to 0.75 can strike a balance between data integrity and simplicity and ensure data quality. At the same time, although the theoretical maximum value of semantic similarity is 1, the upper threshold reaches 1.05, which is to ensure that the algorithm can still perform well when approaching the limit value, and to ensure its reliability in practical applications. In actual implementation, the semantic similarity threshold range can be adjusted according to actual conditions. This embodiment does not limit the value and step size of the semantic similarity threshold range.

[0098] In order to intuitively demonstrate the process of selecting the semantic similarity threshold, Cluster0 to Cluster4 are analyzed in detail. Figures 6 to 10 The results are presented in . Specifically, each modeling element cluster has two corresponding relationship subgraphs, including the first relationship subgraph corresponding to the semantic similarity threshold-silhouette coefficient and the second relationship subgraph corresponding to the semantic similarity threshold-number of retained modeling elements. The first relationship subgraph depicts the impact of different semantic similarity thresholds on the silhouette coefficient, and the second relationship subgraph shows the relationship between the semantic similarity threshold and the number of retained modeling elements.

[0099] Depend on Figure 6 It can be seen that when the semantic similarity threshold of Cluster0 is greater than or equal to 0.80, although the silhouette coefficient remains at the maximum value, the number of modeling elements does not change. In contrast, when the semantic similarity threshold is 0.75, the silhouette coefficient is only slightly lower than the maximum value, and the number of modeling elements is reduced by 2. This shows that when the threshold is 0.75, it not only maintains a high clustering quality, but also effectively reduces redundant data and improves the overall quality of the data. Therefore, 0.75 is selected as the semantic similarity threshold of Cluster0.

[0100] Depend on Figure 7It can be seen that when the semantic similarity threshold is greater than or equal to 0.90, the silhouette coefficient of Cluster1 decreases significantly, while the number of modeling elements is not improved. When the semantic similarity threshold is within the range, the silhouette coefficient is large and the number of modeling elements is optimized. This shows that choosing an appropriate semantic similarity threshold can make the clustering more compact, which in turn reflects the effectiveness of the optimization operation. Although the silhouette coefficient and the number of retained modeling elements are the same at the three thresholds of 0.75, 0.80, and 0.85, 0.85 is a turning point that may reflect a key change in data characteristics. Here, considering the clustering effect and the need for data de-redundancy, 0.85 is selected as the semantic similarity threshold of Cluster1.

[0101] Depend on Figure 8 It can be seen that when the semantic similarity threshold is greater than or equal to 0.80, the silhouette coefficient decreases significantly, while the number of modeling elements is not optimized. This shows that at these higher thresholds, the clustering effect becomes worse and the distribution of the data no longer has a good structure. In contrast, when the threshold is 0.75, not only is the silhouette coefficient larger and the clustering quality improved, but the number of modeling elements is effectively streamlined, reducing data redundancy, thereby improving the overall efficiency of the model. Therefore, 0.75 is selected as the semantic similarity threshold of Cluster2.

[0102] Depend on Fig. 9 It can be seen that as the threshold of semantic similarity gradually increases, the silhouette coefficient shows a fluctuation trend of first increasing, then decreasing, and then increasing again. Specifically, when the threshold increases from 0.80 to 0.85, the silhouette coefficient increases significantly. However, as the threshold continues to increase to 0.90, the silhouette coefficient decreases slightly. But when the threshold continues to increase to 0.90 and above, the silhouette coefficient increases again and reaches the highest value between the thresholds of 0.95 and 1.05. Fig. 9 It can be seen that as the threshold increases, the number of retained modeling elements also gradually increases. When the threshold is greater than or equal to 0.95, the modeling elements remain unchanged at 8. Fig. 9 From the above, we can see that for Cluster3, choosing a higher threshold (such as 0.95) can achieve better clustering effect, but it also leads to the modeling elements not being well optimized, so they are not included in the appropriate semantic similarity threshold consideration range. Although the corresponding silhouette coefficient of the lower threshold (0.75 or 0.80) is slightly lower than the silhouette coefficient of the higher threshold, the difference is not significant, and the clustering effect will not be greatly affected. At the same time, these lower thresholds can effectively reduce the number of modeling elements, remove redundant elements, and improve the compactness of clustering. Although the silhouette coefficient and the number of retained modeling elements are the same when the thresholds are 0.75 and 0.80, 0.80, as a turning point, can better reflect the subtle changes in the optimization process. Therefore, 0.80 is selected as the semantic similarity threshold of Cluster3.

[0103] Depend on Fig.10 It can be seen that as the semantic similarity threshold gradually increases, the silhouette coefficient generally shows an increasing trend. Fig.10 It can be seen that as the threshold increases, the number of retained modeling elements also gradually increases. When the threshold is greater than or equal to 0.95, the number of modeling elements remains unchanged at 12. Overall, although a higher threshold range can make the silhouette coefficient reach a larger value, it also significantly increases the number of retained modeling elements, resulting in more redundancy. In contrast, when the threshold is 0.75 and 0.80, the silhouette coefficient is slightly lower than the highest value, but still remains at a high level, and the number of retained modeling elements is small, and the redundancy removal effect is better. Fig.10 It can be seen that the threshold of 0.80 is a turning point. Further increasing the threshold will lead to increased redundancy, and further lowering the threshold does not significantly improve the clustering effect. Therefore, choosing 0.80 as the semantic similarity threshold can effectively reduce redundant elements while ensuring the clustering effect and achieve the best balance.

[0104] Based on the above analysis, considering the clustering effect and the need for redundancy removal, 0.75, 0.85, 0.75, 0.80 and 0.80 were selected as the target semantic similarity thresholds for Cluster0 to Cluster4 respectively.

[0105] After determining the target semantic similarity threshold, the modeling elements in each modeling element cluster are optimized. That is, according to the target semantic similarity threshold, it is determined whether the modeling elements can be aggregated to reduce the number of modeling elements, improve model accuracy, and thus optimize model performance.

[0106] Specifically, the similarity between different modeling elements in the same modeling element cluster is calculated, and a semantic similarity heat map is drawn to intuitively present the relationship between the elements. When the semantic similarity of two modeling elements exceeds the target semantic similarity threshold corresponding to the modeling element cluster, one of the modeling elements is retained and the other modeling element is deleted until the similarity of the modeling elements in the modeling element cluster is lower than the corresponding target semantic similarity threshold.

[0107] like Figures 11 to 15 As shown in the figure, the automatic extraction of modeling elements of CNC machine tools is used as an example to illustrate. Fig.11 It can be seen that the semantic similarity between cutting speed and spindle speed is 0.78, so we choose to keep cutting speed and delete spindle speed. Similarly, the semantic similarity between feed rate and feed adjustment rate is 0.76, so we choose to keep feed rate and delete feed adjustment rate. By optimizing the modeling elements in Cluster0, the elements that are finally retained include cutting speed, feed rate, and feed speed. Fig.12It can be seen that the semantic similarity between the running state and the working state is 0.88. The running state is retained and the working state is deleted. After optimizing the modeling elements in Cluster1, the elements that are finally retained include alarm state, running state, and running time. Fig.13 It can be seen that the semantic similarity between factory batch and serial number is 0.84. The factory batch is retained and the serial number is deleted. By optimizing the modeling elements in Cluster2, the elements that are finally retained include manufacturer, factory batch, equipment type, and production date. Fig.14 It can be seen that the semantic similarity between the servo system and the position follower system is 0.89. The servo system is retained and the position follower system is deleted. The semantic similarities between the feed servo system and the feed drive system and the feed transmission system are 0.93 and 0.84 respectively. The feed servo system is retained and the feed drive system and the feed transmission system are deleted. The semantic similarities between the spindle servo system and the main transmission system and the spindle drive system are 0.81 and 0.91 respectively. The spindle servo system is retained and the main transmission system and the spindle drive system are deleted. Here, the elements that have been deleted in the previous round will not be discussed. After optimizing the modeling elements in Cluster3, the elements that are finally retained include the servo system, the feed servo system, and the spindle servo system. Fig.15 It can be seen that the semantic similarity between the clamping system and the fixture system is 0.83, so the clamping system is retained and the fixture system is deleted. The semantic similarity between the hydraulic system and the oil pressure system is 0.92, so the hydraulic system is retained and the oil pressure system is deleted. After the modeling elements of Cluster 4 are optimized, the elements that are finally retained are CNC system, chip removal system, clamping system, lubrication system, hydraulic system, auxiliary system, cooling system, protection system, programmable controller and sensor.

[0108] In some embodiments of the present invention, a method is used that combines global optimization based on cluster analysis and local optimization combined with a semantic similarity threshold. Global optimization ensures the rational classification of the overall data and finds the most appropriate clustering structure, thereby optimizing the grouping effect of modeling elements and improving the accuracy and efficiency of subsequent analysis and modeling. Local optimization optimizes information density by removing redundant or repeated information, thereby improving classification accuracy. This not only makes the information model more accurate and efficient, but also reveals a more detailed hierarchical relationship between modeling elements, which helps to build a clearer and more reasonable hierarchical structure. By combining global optimization and local optimization, not only the accuracy and efficiency of the information model are ensured, but also a systematic and hierarchical organizational structure is provided for the modeling elements.

[0109] In addition, when the number of modeling elements in a modeling element cluster is small, there is no need to perform local optimization on the modeling element cluster. Figure 5Cluster5 in , because the number of modeling elements in Cluster5 is small and covers the core functions, it is directly retained without further local optimization.

[0110] Specifically, before determining the target semantic similarity threshold corresponding to each modeling element cluster, it also includes: determining the number of modeling elements in each modeling element cluster; for modeling element clusters with a number of modeling elements greater than 2, executing the step of determining the target semantic similarity threshold corresponding to each modeling element cluster.

[0111] In addition, after the modeling elements are extracted, in order to facilitate clustering processing and semantic similarity analysis in the subsequent modeling optimization process, each extracted modeling element needs to be converted into a corresponding word vector.

[0112] Specifically, the preset data source and model prompt information are input into the preset natural language processing model, and after the modeling elements are extracted, the method further includes: converting the modeling elements into corresponding word vector forms.

[0113] In some embodiments of the present invention, each modeling element is converted into a corresponding word vector by taking advantage of the good capture capability of Bidirectional Encoder Representations from Transformers (BERT) for complex semantic relations.

[0114] Step S103, constructing an information model based on the modeling element clusters; wherein the information model is used to indicate the relationship between the modeling elements, including the internal relationship corresponding to the modeling elements within each modeling element cluster, and the cross-cluster relationship corresponding to the modeling elements between different modeling element clusters.

[0115] In some embodiments of the present invention, taking into account the complex associations between model elements, a model representation framework of "Underground Root Structure" (URS) is proposed, including a tree structure, or a combination of a tree structure and a mesh structure. A reasonable model structure helps to clearly express the relationships between elements within and across models, improve the comprehensibility and maintainability of the model, and facilitate subsequent updates and expansions, thereby reducing maintenance costs and time. Different from the hierarchical associations in traditional tree structures, the last layer of the URS model representation framework develops a large number of roots to better express the complex, unstructured, and even intertwined extended attribute relationships in the information model. This not only retains the hierarchical organization advantages of the tree structure, but also combines the association of the mesh structure to flexibly adapt to complex extended attributes, providing a more comprehensive model representation method.

[0116] Specifically, the information model includes a tree structure layer, or a combination of a tree structure layer and a network structure layer. The tree structure layer includes an object class layer corresponding to the information model, a cluster category layer corresponding to each modeling element cluster, and a representative element layer corresponding to each modeling element cluster; the object class layer is used to indicate the modeling object corresponding to the information model; the cluster category layer is used to indicate the category corresponding to each modeling element cluster; and the representative element layer is used to indicate the representative element corresponding to each modeling element cluster.

[0117] For example, reference Fig.16 Taking the four-layer URS model representation framework as an example, the first three layers adopt a tree structure, taking advantage of its clear hierarchy and structure to make information classification and management more intuitive. The structure of the fourth layer is transformed into a horizontal branch network with a mesh structure with association relationships to enhance the model's relationship expression ability, so that the URS model representation framework can support cross-level and horizontal connections and the establishment of association relationships, allowing the model structure and connection method to be flexibly adjusted according to different needs.

[0118] At the same time, in order to more accurately represent the complexity and dynamics of the information model, the information model should be able to effectively reflect the relationship between elements within the model and across models. This embodiment refers to the Unified Modeling Language (UML) and defines relationships as inheritance, realization, dependency, association, aggregation, and composition.

[0119] Among them, the composition relationship refers to a type of whole-part relationship, which is stronger than the aggregation relationship; the aggregation relationship refers to a type of whole-part relationship, which is weaker than the composition relationship; the association relationship refers to a class knowing the properties of another class; the inheritance relationship refers to a class inheriting the functions of another class and can add additional functions; the implementation relationship refers to a class implementing the functions of one or more interfaces; the dependency relationship refers to a class depending on another class, usually through parameters, variables, etc. at runtime.

[0120] After introducing the relationship, the URS model representation framework is as follows Fig.17 As shown in the figure, the URS model representation framework defines a basic information model framework.

[0121] exist Fig.17 In the model, the first layer is the object class layer, that is, the modeling object; the second layer is the cluster category layer, which is used to represent the category corresponding to each modeling element cluster.

[0122] For example, taking industrial systems as an example, according to the different types of information and functional requirements commonly found in industrial systems, they can be subdivided into six categories: specification, process, control, management, business, and custom. Among them, the specification category includes modeling elements related to standards, technical parameters, and feature descriptions. The process category covers modeling elements related to operations, processing flows, and process steps. The control category involves modeling elements related to system control, operating instructions, and control logic. The management category includes modeling elements related to management activities, resource allocation, and organizational structure. The business category involves modeling elements related to business activities, transaction processes, and business rules. Custom categories refer to special modeling elements that cannot be classified into the aforementioned categories.

[0123] In some embodiments of the present invention, the modeling element classification involved in the cluster category layer is completed by calculating the average word vector of the modeling element in each cluster and comparing it with the semantic similarity of the preset modeling element cluster category. This can more accurately reflect the overall characteristics of the cluster, avoid the deviation that may be caused by the classification of a single element, and improve the accuracy of classification.

[0124] The third layer is the representative element layer, which is constructed by selecting appropriate representative elements in the fourth layer through the structure of the fourth layer and combining the actual structural characteristics of the modeling object, the degree of association of the modeling elements and other factors. In the case where it is impossible to select a suitable representative element in the modeling element cluster as the common superior of the remaining elements in the cluster, all modeling elements of the modeling element cluster will be elevated to the representative element layer. In actual implementation, there is a possibility that all modeling element clusters in the fourth layer are elevated to the representative element layer. At this time, the information model constructed by the URS model representation framework can only include the tree structure layer.

[0125] The fourth layer is the network structure layer. In some embodiments of the present invention, the relationship between modeling elements is pre-constructed in an artificial form based on domain knowledge and professional judgment. By manually constructing the relationship between modeling elements, it is ensured that the information model accurately reflects the actual business logic and operation process, thereby enhancing the accuracy and practicality of the model.

[0126] Specifically, an information model is constructed based on modeling element clusters, including: constructing a mesh structure layer based on modeling element clusters and association relationships; determining the average word vector of the modeling element word vectors in each modeling element cluster, and comparing the semantic similarity with the preset modeling element cluster categories to obtain a comparison result; the comparison result is used to indicate the similarity between each modeling element cluster and each preset modeling element cluster category; based on the comparison result, determining the category corresponding to each modeling element cluster, and constructing a cluster category layer; based on the internal relationship, determining the representative element corresponding to each modeling element cluster, and constructing a representative element layer.

[0127] The automatic extraction of modeling elements of CNC machine tools is taken as an example. In order to more clearly show the hierarchical relationship of modeling elements in the modeling element clusters after global optimization and local optimization, the hierarchical clustering dendrograms corresponding to the modeling element clusters are drawn respectively. By combining the dendrogram structure and the actual structural characteristics of the modeling object and other factors for analysis and judgment, the modeling element mapping of the mesh structure layer is completed.

[0128] by Figure 5 Take the modeling element clusters Cluster3 and Cluster5 in the example, refer to Fig.18 ,The feed servo system and the spindle servo system are close to each other and are first clustered into one category. Then, this category is merged with the servo system to form a higher-level cluster. Finally, the feed servo system and the spindle servo system are parallel and belong to the hierarchical structure of the servo system. The two modeling elements of Cluster5 will be directly mapped to the URS model representation framework.

[0129] After the mesh structure layer modeling element mapping is completed, the initial form of the information model is obtained as follows Fig.19 As shown in the figure, "*" indicates the part that needs to be mapped but has not been mapped yet. After that, a cluster category layer is constructed, and the modeling elements of each cluster are classified into the preset modeling element cluster category. The average word vector of the modeling element word vectors retained in the cluster is calculated and compared with the preset modeling element cluster category. The classification results are displayed through a radar chart, such as Fig. 20 As shown in Figure 1, we can intuitively see the similarity of each modeling element cluster in the six categories. Fig. 20 It can be seen that the modeling elements of Cluster0~Cluster5 belong to the control class, management class, specification class, control class, control class and business class respectively.

[0130] Based on the above classification, the modeling elements of CNC machine tools are mapped to the URS model representation framework for representation, and the corresponding relationships are constructed in combination with domain knowledge. The model architecture is composed of Fig.19 The initial form shown develops into Fig.21 The finished form is shown. According to the optimized architecture, the number of CNC machine tool modeling elements is reduced by 13 compared with the initial form, accounting for 36.11% of the total. The automated modeling optimization method proposed in this paper can not only optimize the number of modeling elements and simplify the model structure, but also ensure the integrity of information and retain necessary information and relationships, thereby replacing the traditional manual modeling method.

[0131] In actual implementation, due to the complexity and diversity of industrial systems, multiple functional modules, subject areas and hierarchical structures are usually involved. Therefore, in the modeling process, it is often necessary to split the entire system into multiple information sub-models. These sub-models can describe the characteristics of different parts or different levels of the system, such as the device layer, control layer and application layer, and each sub-model has an independent function or task. By combining these sub-models reasonably, a complete information model can be formed, which can not only simplify the modeling process, but also improve the flexibility and scalability of the system, and ultimately meet the accurate modeling of different needs.

[0132] Specifically, constructing the information model based on the modeling element cluster also includes: constructing at least two information sub-models based on the modeling element cluster; and combining the at least two information sub-models to obtain the information model.

[0133] In addition, in order to provide a more accurate and comprehensive description of the system, multiple information models can be combined to perform more complex information modeling. After the information model modeling optimization is completed, the information model is stored to ensure that when the model of the same object needs to be built in the future, the stored model and relationship can be directly called without re-modeling, which improves work efficiency and model reusability. At the same time, the consistency of the model is guaranteed.

[0134] In summary, the industrial Internet information model modeling optimization method based on natural language processing provided in this embodiment extracts modeling elements by inputting a preset data source and model prompt information into a preset natural language processing model; the model prompt information includes first-stage prompt information and second-stage prompt information; the first-stage prompt information is used to prompt the preset natural language processing model to extract entities from the preset data source according to a preset entity type list; the second-stage prompt information is used to prompt the preset natural language processing model to sort out the extracted modeling elements based on the extracted entities; the modeling elements are divided into a corresponding number of modeling element clusters based on the target number of clusters; an information model is constructed based on the modeling element clusters; the information model includes a tree structure layer and a mesh structure layer; the tree structure layer includes an object class layer corresponding to the information model, a cluster category layer corresponding to each modeling element cluster, and each modeling element cluster. The representative element layer corresponding to the model element cluster; the object class layer is used to indicate the modeling objects corresponding to the information model; the cluster category layer is used to indicate the categories corresponding to each modeling element cluster; the representative element layer is used to indicate the representative elements corresponding to each modeling element cluster; it can solve the problem of poor effect and reliability of the information model; through the preset natural language processing model, according to the preset entity type list indicated by the first-stage model prompt information, the entity type is automatically obtained from the preset data source, and then the entity information is automatically extracted from the preset data source through the second-stage model prompt information and entity type, without relying on manual collection and processing, avoiding the problem of poor data quality and integrity, thereby improving the effect and reliability of the information model; at the same time, using a unified preset entity type list to obtain entity types can overcome the compatibility barriers between different standards and improve the compatibility of information models.

[0135] In addition, after determining the target semantic similarity threshold, the modeling elements in each modeling element cluster are optimized, that is, according to the target semantic similarity threshold, it is determined whether the modeling elements can be aggregated to reduce the number of modeling elements, improve model accuracy, and thus optimize model performance.

[0136] Corresponding to the above method, the present invention also provides an industrial Internet information model modeling optimization method device based on natural language processing, including a processor, a memory and a computer program / instructions stored in the memory, the processor is used to execute the computer program / instructions, and when the computer program / instructions are executed, the device implements the steps of the aforementioned industrial Internet information model modeling optimization method based on natural language processing.

[0137] The present invention also provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the aforementioned industrial Internet information model modeling optimization method based on natural language processing.

[0138] The present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the aforementioned industrial Internet information model modeling optimization method based on natural language processing.

[0139] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0140] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0141] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or replace features of other embodiments.

[0142] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for optimizing industrial Internet information modeling based on natural language processing, characterized in that: The method comprises the following steps: Inputting model prompt information into a preset natural language processing model to extract modeling elements; the model prompt information includes first-stage prompt information and second-stage prompt information; the first-stage prompt information is used to prompt the preset natural language processing model to extract entity types from a preset data source according to a preset entity type list; the second-stage prompt information is used to prompt the preset natural language processing model to extract modeling elements based on the extracted entity types; Dividing the modeling elements into a corresponding number of modeling element clusters based on the target cluster number; An information model is constructed based on the modeling element cluster; the information model is used to indicate the relationship between the various modeling elements, including the internal relationship corresponding to the modeling elements within each modeling element cluster, and the cross-cluster relationship corresponding to the modeling elements between different modeling element clusters; the information model includes a tree structure layer, or a combination of the tree structure layer and the mesh structure layer.

2. The industrial Internet information modeling optimization method based on natural language processing according to claim 1 is characterized in that: Before dividing the modeling elements into a corresponding number of modeling element clusters based on the target cluster number, the method further includes: Performing cluster analysis on the modeling elements based on a preset cluster number set, and calculating the silhouette coefficients corresponding to clustering of the modeling elements under different cluster numbers; The target number of clusters is determined in the preset number of clusters based on the silhouette coefficient.

3. The industrial Internet information modeling optimization method based on natural language processing according to claim 1 is characterized in that: Before dividing the modeling elements into a corresponding number of modeling element clusters based on the target cluster number, the method further includes: Determine the target semantic similarity threshold corresponding to each modeling element cluster; Comparing the modeling elements in each modeling element cluster in pairs to determine the semantic similarity corresponding to each pair of modeling elements; In each of the modeling element clusters, modeling element pairs whose semantic similarity is greater than or equal to the corresponding target semantic similarity threshold are aggregated.

4. The industrial Internet information modeling optimization method based on natural language processing according to claim 3 is characterized in that: The step of determining the target semantic similarity threshold corresponding to each modeling element cluster includes: Within a preset semantic similarity threshold range, analyzing and obtaining a first relationship between the silhouette coefficient corresponding to each modeling element cluster and the semantic similarity, and a second relationship between the number of modeling elements corresponding to each modeling element cluster and the semantic similarity; Based on the first relationship and the second relationship, a target semantic similarity threshold corresponding to each modeling element cluster is determined within the preset semantic similarity threshold range.

5. The method for optimizing the industrial Internet information model based on natural language processing according to claim 3 is characterized in that: Before determining the target semantic similarity threshold corresponding to each modeling element cluster, the method further includes: Determining the number of modeling elements in each of the modeling element clusters; For the modeling element clusters whose number of modeling elements is greater than 2, the step of determining the target semantic similarity threshold corresponding to each modeling element cluster is performed.

6. The industrial Internet information modeling optimization method based on natural language processing according to claim 1 is characterized in that: The method of inputting the preset data source and model prompt information into the preset natural language processing model and extracting the modeling elements also includes: converting the modeling elements into corresponding word vector forms.

7. The method for optimizing industrial Internet information modeling based on natural language processing according to claim 1 is characterized in that: The tree structure layer includes an object class layer corresponding to the information model, a cluster category layer corresponding to each modeling element cluster, and a representative element layer corresponding to each modeling element cluster; the object class layer is used to indicate the modeling object corresponding to the information model; The cluster category layer is used to indicate the category corresponding to each modeling element cluster; the representative element layer is used to indicate the representative element corresponding to each modeling element cluster.

8. The method for optimizing the industrial Internet information model based on natural language processing according to claim 7 is characterized in that: The constructing of the information model based on the modeling element cluster includes: Based on the modeling element clusters and the association relationships, constructing the network structure layer; Determine the average word vector of the modeling element word vectors in each modeling element cluster respectively, and compare the semantic similarity with the preset modeling element cluster category to obtain a comparison result; the comparison result is used to indicate the similarity between each modeling element cluster and each preset modeling element cluster category; Based on the comparison result, determining the category corresponding to each modeling element cluster, and constructing the cluster category layer; Based on the internal relationship, the representative element corresponding to each modeling element cluster is determined, and the representative element layer is constructed.

9. The method for optimizing industrial Internet information modeling based on natural language processing according to claim 1 is characterized in that: The constructing of the information model based on the modeling element cluster further includes: At least two information sub-models are constructed based on the modeling element cluster; The at least two information sub-models are combined to obtain the information model.

10. The method for optimizing industrial Internet information modeling based on natural language processing according to claim 1 is characterized in that: The preset data source includes a text data source and / or a knowledge base corresponding to the preset natural language processing model; In the case where the preset data source includes the text data source, before inputting the model prompt information into the preset natural language processing model to extract the modeling elements, it also includes: inputting the text data source as an attachment into the preset natural language processing model.

Images