A method for constructing a digital twin model

Abstract

The present invention proposes a digital twin model construction method, which belongs to the field of artificial intelligence technology, including collecting historical data, establishing a historical database, extracting samples from the historical database through a dynamic extraction algorithm, and using the samples to build a digital twin model; performing data correction on the generated digital twin model in the model prediction stage; using the data-corrected digital twin model to predict the real-time data stream and obtain a target data set; recording parameter and data changes in the construction of the initial digital twin simulation model; and being able to reasonably allocate model optimization resources during the model construction process, and correspondingly optimize the constructed digital twin model, thereby improving the accuracy of the model. At the same time, in the model prediction stage, the prediction process of the model is tested and corrected, which can improve the accuracy of the model prediction.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and specifically relates to a method for constructing a digital twin model. Background Art

[0002] Digital twins make full use of data such as physical models, sensor updates, and operation history, integrate multi-disciplinary, multi-physical, multi-scale, and multi-probability simulation processes, and complete mapping in virtual space to reflect the entire life cycle of the corresponding physical equipment. Based on digital twins, simulation models can be quickly built to simulate different scenarios. Digital twins are a concept that transcends reality and can be regarded as a digital mapping system for one or more important, interdependent equipment systems.

[0003] At present, industrial equipment is gradually becoming more sophisticated. By building digital twin models, the operation status of industrial equipment can be accurately analyzed. During the model construction process, the model needs to be calibrated and verified to ensure the accuracy and reliability of the model. However, the existing technology cannot reasonably allocate model optimization resources during the model construction process, and cannot optimize the constructed digital twin model accordingly, thereby failing to improve the accuracy of the model. At the same time, in the model prediction stage, the model prediction process is not tested and corrected, so the accuracy of the model prediction cannot be improved. Summary of the invention

[0004] The present invention aims at solving the problems in the prior art. The technical solution adopted in the present application is: a method for constructing a digital twin model, comprising: Collect historical data, establish a historical database, extract samples from the historical database through a dynamic extraction algorithm, and use the samples to build a digital twin model; Perform data correction on the generated digital twin model during the model prediction phase; Use the data-corrected digital twin model to predict the real-time data stream and obtain the target data set; Record the parameter and data changes in the construction of the initial digital twin simulation model; evaluate the feasibility, sensitivity, practicality and stability of the initial digital twin simulation model, and use the digital twin model that meets the evaluation criteria as the final digital twin model.

[0005] Furthermore, the historical database adopts the memcached database, which is stored by continuously appending records. After being encapsulated by the memcached client, multiple instantiated clusters are obtained. Each client connects to a server host in the cluster. The server host performs assignment operations by calling the main process. When the master-slave switch of the server host occurs, the historical data in the Hmemecached memory is read and the switching operation is completed.

[0006] Furthermore, samples are extracted from the historical database through the dynamic extraction model to establish a digital twin model; the dynamic extraction model uses the DBSCAN density clustering algorithm to divide the sample point data into dense areas and sparse areas, dt=1+na / Y, na represents the number of sample points in the circular domain, and Y represents the total number of data samples. The SPARQL query associated data is constructed through the density factor. The SPARQL query associated data contains domain knowledge relationships, and then a knowledge graph overview triple is constructed for each knowledge relationship to form a knowledge graph overview G1, and then the knowledge graph overview triple of each knowledge classification is extracted to form a knowledge graph overview G2, and finally the two are merged into a complete associated data knowledge graph overview, and sample points are extracted from the associated data knowledge graph overview according to the degree of association, and the extracted sample points are associated with the knowledge graph; the extracted sample points are input into the trained large language model for feature extraction, and the feature extraction stage extracts the optimal sample point set based on the BIRCH algorithm for field unit construction, and the fields are matched into complete data through regular expressions, and the complete data is extracted as the final extracted sample features.

[0007] Furthermore, a deep learning algorithm is used to process the extracted samples to build a data model, and the corresponding entity model is loaded into the data model to form a digital twin model.

[0008] Furthermore, in the model prediction stage, data correction is performed on the generated digital twin model, the digital twin model is trained by screening valuable samples from the samples that did not participate in the modeling, a mapping analysis is performed between the three-dimensional model of the target object and the target object entity, the offset value of each offset point is obtained, the three-dimensional model is corrected, and the offset correction coefficient is calculated.

[0009] Furthermore, the digital twin model using data correction is used to predict the real-time data stream to obtain the target data set.

[0010] Furthermore, the comparison values ​​of the characteristics of the normal probability density and the characteristics of the probability density of abnormal data can be used to evaluate the acquired data of the digital twin simulation model. The higher the comparison value, the higher the model evaluation.

[0011] Compared with the prior art, the advantages and positive effects of the present invention are: The present invention adopts a digital twin model construction method, collects historical data, establishes a historical database, extracts samples from the historical database through a dynamic extraction algorithm, and uses the samples to build a digital twin model; performs data correction on the generated digital twin model in the model prediction stage; uses the data-corrected digital twin model to predict the real-time data stream and obtain a target data set; records the parameters and data changes in the construction of the initial digital twin simulation model; evaluates the feasibility, sensitivity, practicality and stability of the initial digital twin simulation model, and uses the digital twin model that meets the evaluation criteria as the final digital twin model. During the model construction process, model optimization resources can be reasonably allocated, and the constructed digital twin model is optimized accordingly, thereby improving the accuracy of the model. At the same time, in the model prediction stage, the prediction process of the model is tested and corrected, which can improve the accuracy of the model prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0013] Figure 1 A flowchart of a digital twin model construction method of the present invention; Figure 2 An architectural diagram of a digital twin model construction method of the present invention; DETAILED DESCRIPTION In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0014] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0015] Embodiment 1, as Figure 1-2As shown, the present application provides a method for building a digital twin model. At present, the accuracy of the digital twin model in the model prediction stage is low. How to improve the prediction accuracy is a technical problem that needs to be solved. The inventor found that there is a lack of optimization solutions in the model construction stage, and provided an improved technical solution for the digital twin model construction process. For the digital twin model construction process, a dynamic extraction model is first used to extract samples from a historical database, including regional division of sample point data, construction of a knowledge graph and extraction of sampled samples. In particular, a new improvement solution is proposed in the process of extracting sampled samples. The extracted sample points are input into a trained large language model for feature extraction. Feature extraction In this stage, the optimal sample point set is extracted based on the BIRCH algorithm for field unit construction. The BIRCH algorithm is used to improve the setting through similarity calculation to extract the optimal sample point feature set, which improves the sampling accuracy of sample point features. At the same time, in the model prediction stage, valuable samples are screened from the samples that do not participate in modeling, and verification samples are generated through manual verification. The digital twin model is trained with the verification samples, and the regional cross-correlation algorithm is used in the data model to repair and correct the offset points, thereby realizing data correction of the generated digital twin model. Finally, the prediction accuracy of the corrected digital twin model is greatly improved, which is suitable for current industrial model construction needs.

[0016] Specifically include: Collect historical data and establish a historical database; the historical database uses a memcached database and stores data by continuously appending records. After being encapsulated using a memcached client, multiple instantiated clusters are obtained. Each client connects to a server host in the cluster. The server host performs assignment operations by calling the main process. When the server host switches between the master and standby modes, the historical data in the Hmemecached memory is read and the switching operation is completed.

[0017] The dynamic extraction model extracts samples from the historical database and establishes a digital twin model; (1. Divide the sample point data into regions) The dynamic extraction model uses the DBSCAN density clustering algorithm to divide the sample point data into dense and sparse areas, dt = 1 + na / Y, na represents the number of sample points in the circular domain, and Y represents the total number of data samples. (2. Build a knowledge graph and extract sampling samples) SPARQL query related data is constructed through density factors. The domain knowledge relationship contained in the SPARQL query related data is then constructed for each knowledge relationship. The knowledge graph overview triple is formed and the knowledge is formed. The graph overview G1, then extract the knowledge graph overview triples of each knowledge classification and form the knowledge graph overview G2, and finally merge the two into a complete linked data knowledge graph overview, extract sample points from the linked data knowledge graph overview according to the degree of association, and the extracted sample points are associated with the knowledge graph; (3. Extract sample features) The extracted sample points are input into the trained large language model for feature extraction. In the feature extraction stage, the optimal sample point set is extracted based on the BIRCH algorithm to construct field units, and the fields are matched into complete data through regular expressions, and the complete data is extracted as the final extracted sample features. The extracted sample features are processed by deep learning algorithms to construct a data model, and the corresponding entity model is loaded into the data model to form a digital twin model.

[0018] Among them, the BIRCH algorithm is improved by similarity calculation and is set to extract the optimal sample point set. After clustering the data to be clustered, the BIRCH algorithm updates the feature information of the clustering features in the nodes according to the feature information of the data to be clustered by using the bidirectional pointers between the nodes in the preset clustering feature tree and the clustering features pointing to the nodes. For example, if the sample point A includes 3 clustering features, namely clustering feature A1, clustering feature A2 and clustering feature A3, the similarity between clustering feature A1 and the data to be clustered, the similarity between clustering feature A2 and the data to be clustered, and the similarity between clustering feature A3 and the data to be clustered are calculated based on the improved BIRCH algorithm, and the set number of sample points with the highest similarity matching are extracted to obtain the optimal sample point set.

[0019] In the BIRCH algorithm, the calculation of similarity is not part of the algorithm itself, but after clustering is completed, if the clustering results need to be analyzed or interpreted, similarity calculation may be involved. Advantages of integrating similarity calculation directly into the BIRCH algorithm: Earlier similarity evaluation: Considering similarity at the data compression stage can identify similar data points earlier, which may help to build clustering features (CFs) more efficiently. Reduced subsequent processing: If the similarity calculation is already considered in the process of building the CF tree, then additional similarity calculation may not be required in the clustering stage, thereby reducing the complexity and time of subsequent processing. Optimizing clustering quality: By considering similarity in the clustering process, you can better control the results of clustering, which may improve the quality and interpretability of clustering. Dynamic adjustment: If the algorithm can dynamically adjust the similarity threshold during the process of building the CF tree, it may be able to better adapt to changes in data, especially when processing data streams. Enhanced scalability: When processing large-scale data sets, being able to identify and merge similar data points in the early stages of the algorithm may reduce storage and processing requirements, thereby improving the scalability of the algorithm. Better noise handling: By taking similarity into account during clustering, the algorithm may be able to better distinguish between noise and real data points, thereby increasing the robustness of the clustering.

[0020] The similarity calculation formula in the improved BIRCH algorithm is:

[0021] in: is the i-th similarity result, k is the similarity number, represents the convolution weight between the data to be clustered at the kth layer, Represents the convolution weight of the clustering feature A1 of the kth layer.

[0022] In the model prediction stage, the generated digital twin model is corrected. In the model prediction stage, valuable samples are selected from samples that are not involved in modeling, and verification samples are generated through manual verification. The verification samples are used to train the digital twin model, and the data model of the target object and the target object entity are mapped and analyzed to obtain the offset value of each offset point, correct the data model, and calculate the offset correction coefficient. P(x, y) represents the pixel value of the point with coordinates (x, y), Rn-1 and Rn+1 represent the correlation coefficients corresponding to the left and right columns of the x+nth column with the largest correlation coefficient. i and j are weighted coefficients, and their values ​​should satisfy i+j=1. After experiments, i=0.8 and j=0.2 are taken, and the regional cross-correlation algorithm is used in the data model to repair and correct the offset points.

[0023] The real-time data stream is predicted by using the data-corrected digital twin model to obtain the target data set. For example, for a manipulator, a historical database can be established after collecting the operating parameters of the manipulator, a digital twin model can be generated and corrected, and the operating data of the electronic equipment for a certain period can be predicted by using the corrected digital twin model. The discrete data in the obtained manipulator operating data can be used to draw a scatter plot. Based on the distribution characteristics of each discrete point in the scatter plot, the fitting curve connecting multiple discrete points and the fitting degree of the fitting curve are determined. If the fitting degree of the fitting curve is less than the preset threshold, a probability distribution model is constructed based on the distance from each discrete point to the fitting curve. The determined discrete data is screened out from the data set to be screened using the probability distribution model, and the data set after the discrete data is screened out is determined as the target data set, and the parameter and data changes in the construction of the initial digital twin simulation model are recorded. The feasibility, sensitivity, practicality and stability of the initial digital twin simulation model are evaluated. The comparison value of the characteristics of the normal probability density and the characteristics of the probability density of the abnormal data can be used to evaluate the acquired data of the digital twin simulation model. The higher the comparison value, the higher the model evaluation. The model that meets the evaluation criteria is processed as the final digital twin model.

[0024] Record the parameter and data changes in the construction of the initial digital twin simulation model; evaluate the feasibility, sensitivity, practicality and stability of the initial digital twin simulation model; use the comparison values ​​of the characteristics of the normal probability density and the characteristics of the probability density of abnormal data to evaluate the acquired data of the digital twin simulation model. The higher the comparison value, the higher the model evaluation. The one that meets the evaluation criteria will be processed as the final digital twin model.

[0025] During the model construction process of this embodiment, model optimization resources can be reasonably allocated, and the constructed digital twin model can be optimized accordingly, thereby improving the accuracy of the model. At the same time, in the model prediction stage, the prediction process of the model is tested and corrected, which can improve the accuracy of the model prediction.

[0026] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for constructing a digital twin model, characterized in that: include: Collect historical data, establish a historical database, extract samples from the historical database through a dynamic extraction algorithm, and use the samples to build a digital twin model; Perform data correction on the generated digital twin model during the model prediction phase; Use the data-corrected digital twin model to predict the real-time data stream and obtain the target data set; Record the parameter and data changes in the construction of the initial digital twin simulation model; evaluate the feasibility, sensitivity, practicality and stability of the initial digital twin simulation model, and use the digital twin model that meets the evaluation criteria as the final digital twin model.

2. A digital twin model construction method according to claim 1, characterized in that: The historical database uses a memcached database and is stored by continuously appending records. After being encapsulated using a memcached client, multiple instantiated clusters are obtained. Each client connects to a server host in the cluster. The server host performs assignment operations by calling the main process. When the server host switches between the master and standby, it reads the historical data in the Hmemecached memory and completes the switching operation.

3. A digital twin model construction method according to claim 1, characterized in that: Samples are extracted from the historical database through a dynamic extraction model to establish a digital twin model; the dynamic extraction model uses the DBSCAN density clustering algorithm to divide the sample point data into dense areas and sparse areas, dt=1+na / Y, na represents the number of sample points in the circular domain, and Y represents the total number of data samples. The SPARQL query associated data is constructed through the density factor. The SPARQL query associated data contains domain knowledge relationships, and then a knowledge graph overview triple is constructed for each knowledge relationship to form a knowledge graph overview G1. The knowledge graph overview triple of each knowledge classification is extracted to form a knowledge graph overview G2, and finally the two are merged into a complete associated data knowledge graph overview. Sample points are extracted from the associated data knowledge graph overview according to the degree of association, and the extracted sample points are associated with the knowledge graph; the extracted sample points are input into the trained large language model for feature extraction. In the feature extraction stage, the optimal sample point set is extracted based on the BIRCH algorithm for field unit construction, and the fields are matched into complete data through regular expressions, and the complete data is extracted as the final extracted sample features.

4. A digital twin model construction method according to claim 1, characterized in that: Use deep learning algorithms to process the extracted samples to build a data model, and load the corresponding entity model into the data model to form a digital twin model.

5. A digital twin model construction method according to claim 1, characterized in that: The model prediction stage performs data correction on the generated digital twin model, trains the digital twin model by screening valuable samples from samples that do not participate in modeling, performs mapping analysis on the three-dimensional model of the target object and the target object entity, obtains the offset value of each offset point, corrects the three-dimensional model, and calculates the offset correction coefficient.

6. A digital twin model construction method according to claim 1, characterized in that: The digital twin model using data correction is used to predict the real-time data stream and obtain the target data set.

7. A digital twin model construction method according to claim 1, characterized in that: The acquired data of the digital twin simulation model can be evaluated by using the comparison values ​​of the characteristics of the normal probability density and the characteristics of the probability density of abnormal data. The higher the comparison value, the higher the model evaluation.

Images