Training method and device of digital twin model, equipment, storage medium and program product
By extracting and fusion of heterogeneous data of the target entity, building optimization functions and training digital twin models, the problem of difficult data correlation and low efficiency of optimization algorithms in traditional digital engineering is solved, and the production efficiency improvement and cost and energy consumption are optimized, and the adaptability and efficiency of the system are improved.
Patent Information
- Application Number
- CN202510295147.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-13
- Publication Date
- 2025-07-04
AI Technical Summary
Traditional digital engineering technology is difficult to fully mine data correlation when processing multi-source heterogeneous data, resulting in low decision-making accuracy, and optimization algorithms rely on experience, making it difficult to find the optimal solution under the constraints of multiple goals, and has low computing efficiency and cannot meet the needs of large-scale complex systems.
By obtaining heterogeneous data of the target entity in real time, performing feature extraction and fusion, building target optimization functions, and using convolutional neural networks, long-term memory networks and attention mechanism networks to train digital twin models to achieve multi-objective collaborative optimization.
It improves the overall performance of digital engineering systems, improves production efficiency, reduces costs and energy consumption, and ensures the system's adaptability and benefits in different application scenarios.
Smart Images
Figure CN120257068A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of digital engineering technologies, and in particular, to a method, device, equipment, storage medium, and program product for training a digital twin model. Background Art
[0002] In traditional digital engineering applications, design and optimization are usually carried out through offline modeling and analysis of a single data source. For the processing of multi-source heterogeneous data, these technologies mostly adopt the methods of decentralized data collection and separate analysis, resulting in the difficulty of fully mining and utilizing the relevance between data, thereby leading to low accuracy of subsequent decisions. In addition, the optimization algorithms in traditional digital engineering technologies mostly rely on experience and manual adjustment, making it difficult to find the optimal solution under multi-objective constraints, and having low computational efficiency and being unable to meet the requirements of large-scale and complex systems. It can be seen that traditional digital engineering technologies have many limitations when facing highly complex and variable digital engineering systems.
[0003] Therefore, how to improve the overall performance of digital engineering systems has become an urgent problem to be solved. Summary of the Invention
[0004] Embodiments of the present application provide a method, device, equipment, storage medium, and program product for training a digital twin model, which can improve the overall performance of digital engineering systems.
[0005] In a first aspect, embodiments of the present application provide a method for training a digital twin model, the method including:
[0006] Obtaining multiple data corresponding to a target entity in real time, where the multiple data are heterogeneous data;
[0007] Performing feature extraction and fusion on the multiple data to obtain a target fusion feature vector;
[0008] Analyzing the target fusion feature vector to obtain a quantization index for quantifying a pre-constructed digital twin model corresponding to the target entity; the quantization index includes the production efficiency, production cost, and energy consumption corresponding to the target entity;
[0009] Constructing a target optimization function based on the quantization index;
[0010] Taking determining the optimal solution of the target optimization function as the goal, training the digital twin model to obtain a trained digital twin model.
[0011] In one embodiment, the method further includes: selecting a pre-trained convolutional neural network and a pre-trained long short-term memory network from multiple pre-trained neural networks based on the data types of multiple data of the target entity; constructing an initial fusion model based on the pre-trained convolutional neural network, the pre-trained long short-term memory network, and the attention mechanism network; training the initial fusion model using multiple historical data corresponding to the target entity to obtain a target fusion model; the target fusion model includes a convolutional neural network, a long short-term memory network, and an attention mechanism network; extracting and fusing features of the multiple data to obtain a target fusion feature vector, including: calling the target fusion model to extract and fuse features of the multiple data to obtain a target fusion feature vector.
[0012] In one embodiment, calling the target fusion model to extract and fuse features of the multiple data to obtain a target fusion feature vector includes: using the convolutional neural network to extract features of the multiple data to obtain multiple local features corresponding to the multiple data; inputting the multiple local features into the long short-term memory network to obtain multiple temporal features corresponding to the multiple data; splicing the multiple temporal features and the multiple data to obtain an initial fusion feature vector including spatio-temporal features; inputting the initial fusion feature vector into the attention mechanism network to obtain weights corresponding to each data feature, and obtaining the target fusion feature vector based on the weights corresponding to each data feature and each data feature.
[0013] In one embodiment, analyzing the target fusion feature vector to obtain a quantization index for quantifying the pre-constructed digital twin model corresponding to the target entity includes: analyzing the target fusion feature vector to obtain operation indexes of the target entity corresponding to the features of each dimension in the target fusion feature vector; determining the correlation coefficients between the features of each dimension and the respective operation indexes; selecting target correlation coefficients with correlation coefficients greater than a preset correlation coefficient threshold, and using the operation indexes corresponding to the target correlation coefficients as the quantization indexes for quantifying the digital twin model.
[0014] In one embodiment, the method further includes: obtaining geometric shape data, material property data, and behavior pattern data corresponding to the target entity; constructing an initial digital twin model based on the geometric shape data, the material property data, and the behavior pattern data; establishing a real-time data transmission channel in the initial digital twin model, and using the digital twin model including the real-time data transmission channel as the pre-constructed digital twin model corresponding to the target entity; wherein, the real-time data transmission channel is used to obtain data corresponding to the target entity in real time, and the data is used to update the digital twin model.
[0015] In one embodiment, the method further includes: based on a plurality of data and the mapping relationship between the real-time data corresponding to the predetermined target entity and the model parameters of the digital twin model, performing real-time update on the digital twin model.
[0016] In a second aspect, the present application provides a training device for a digital twin model, the device includes:
[0017] An acquisition module, configured to acquire in real time a plurality of data corresponding to a target entity, the plurality of data being heterogeneous data;
[0018] A feature extraction and fusion module, configured to perform feature extraction and fusion on the plurality of data to obtain a target fusion feature vector;
[0019] A processing module, configured to analyze the target fusion feature vector to obtain a quantization index for quantizing the pre-constructed digital twin model corresponding to the target entity; the quantization index includes production efficiency, production cost, and energy consumption corresponding to the target entity;
[0020] A construction module, configured to construct a target optimization function based on the quantization index;
[0021] A training module, configured to train the digital twin model with the aim of determining the optimal solution of the target optimization function to obtain a trained digital twin model.
[0022] In a third aspect, the present application provides a computer device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:
[0023] Acquire in real time a plurality of data corresponding to a target entity, the plurality of data being heterogeneous data;
[0024] Perform feature extraction and fusion on the plurality of data to obtain a target fusion feature vector;
[0025] Analyze the target fusion feature vector to obtain a quantization index for quantizing the pre-constructed digital twin model corresponding to the target entity; the quantization index includes production efficiency, production cost, and energy consumption corresponding to the target entity;
[0026] Construct a target optimization function based on the quantization index;
[0027] With the aim of determining the optimal solution of the target optimization function, train the digital twin model to obtain a trained digital twin model.
[0028] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:
[0029] Obtain multiple data corresponding to the target entity in real time, where the multiple data are heterogeneous data;
[0030] Extract and fuse features from the multiple data to obtain a target fusion feature vector;
[0031] Analyze the target fusion feature vector to obtain quantization metrics for quantifying the pre-constructed digital twin model corresponding to the target entity; the quantization metrics include production efficiency, production cost, and energy consumption corresponding to the target entity;
[0032] Construct a target optimization function based on the quantization metrics;
[0033] Take determining the optimal solution of the target optimization function as the goal, and train the digital twin model to obtain the trained digital twin model.
[0034] In a fifth aspect, the present application also provides a computer program product, including a computer program, which when executed by a processor implements the following steps:
[0035] Obtain multiple data corresponding to the target entity in real time, where the multiple data are heterogeneous data;
[0036] Extract and fuse features from the multiple data to obtain a target fusion feature vector;
[0037] Analyze the target fusion feature vector to obtain quantization metrics for quantifying the pre-constructed digital twin model corresponding to the target entity; the quantization metrics include production efficiency, production cost, and energy consumption corresponding to the target entity;
[0038] Construct a target optimization function based on the quantization metrics;
[0039] Take determining the optimal solution of the target optimization function as the goal, and train the digital twin model to obtain the trained digital twin model.
[0040] The training method, device, equipment, storage medium and program product of the above digital twin model. The computer equipment can obtain multiple data corresponding to the target entity in real time, and the multiple data are heterogeneous data; extract and fuse the features of the multiple data to obtain a target fusion feature vector; analyze the target fusion feature vector to obtain a quantization index for quantizing the pre-constructed digital twin model corresponding to the target entity; the quantization index includes the production efficiency, production cost and energy consumption corresponding to the target entity; construct a target optimization function based on the quantization index; and train the digital twin model with the goal of determining the optimal solution of the target optimization function to obtain the trained digital twin model. By adopting this method, on the one hand, the computer equipment can fully explore the correlation between multiple heterogeneous data by fusing the multiple heterogeneous data corresponding to the target entity to obtain a target fusion feature vector, which is beneficial to improving the accuracy of subsequent decision-making and analysis; on the other hand, the computer equipment can obtain a quantization index (production efficiency, production cost and energy consumption) for quantizing the pre-constructed digital twin model corresponding to the target entity by analyzing the target fusion feature vector. Then, based on the quantization index, a target optimization function including a first optimization function corresponding to production efficiency, a second optimization function corresponding to the production cost of the target entity, and a third optimization function corresponding to the energy consumption of the target entity can be constructed, and the digital twin model corresponding to the target entity can be collaboratively optimized based on multiple goals such as production efficiency, cost and energy consumption, so as to achieve the comprehensive balance of multiple goals, ensure that in the actual application process using the trained digital twin model, not only the production efficiency can be improved, but also the cost and energy consumption can be minimized to the greatest extent. In this way, the adaptability and benefit maximization of the digital engineering system in different application scenarios can be ensured. Therefore, adopting this method can improve the overall performance of the digital engineering system. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required to be used in the description of the embodiments of the present application or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.
[0042] Figure 1 It is a schematic diagram of an application scenario of a training method for a digital twin model provided by an embodiment of the present application;
[0043] Figure 2 It is a schematic flowchart of a training method for a digital twin model provided by an embodiment of the present application;
[0044] Figure 3It is a schematic flowchart of another method for training a digital twin model provided by an embodiment of the present application;
[0045] Figure 4 It is a schematic structural diagram of a device for training a digital twin model provided by an embodiment of the present application;
[0046] Figure 5 It is a schematic structural diagram of a computer device provided by an embodiment of the present application. Detailed implementation manners
[0047] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0048] Next, the application scenarios of the method for training a digital twin model provided by the embodiments of the present application will be introduced.
[0049] Please refer to Figure 1 , Figure 1 It is a schematic diagram of an application scenario of a method for training a digital twin model provided by an embodiment of the present application. As Figure 1 shown, it includes a computer device 101 and a target entity 102. Among them, data is transmitted between the computer device 101 and the target entity 102 through a network.
[0050] The computer device 101 can obtain multiple data corresponding to the target entity 102 in real time, and the multiple data are heterogeneous data; extract and fuse features from the multiple data to obtain a target fusion feature vector; analyze the target fusion feature vector to obtain a quantization index for quantizing the pre-constructed digital twin model corresponding to the target entity 102; the quantization index includes the production efficiency, production cost, and energy consumption corresponding to the target entity; construct a target optimization function based on the quantization index; and train the digital twin model with the goal of determining the optimal solution of the target optimization function to obtain a trained digital twin model. By adopting this method, on the one hand, the computer device can fully explore the correlation between multiple heterogeneous data by fusing the multiple heterogeneous data corresponding to the target entity, and obtain a target fusion feature vector. In this way, it is beneficial to improve the accuracy of subsequent decision-making analysis. On the other hand, the computer device can obtain a quantization index (production efficiency, production cost, and energy consumption) for quantizing the pre-constructed digital twin model corresponding to the target entity by analyzing the target fusion feature vector. Then, based on the quantization index, a target optimization function including a first optimization function corresponding to the production efficiency, a second optimization function corresponding to the production cost of the target entity, and a third optimization function corresponding to the energy consumption of the target entity can be constructed, and the digital twin model corresponding to the target entity can be collaboratively optimized based on multiple goals such as production efficiency, cost, and energy consumption, so as to achieve the comprehensive balance of multiple goals, ensure that in the actual application process using the trained digital twin model, not only the production efficiency can be improved, but also the cost and energy consumption can be minimized to the greatest extent. In this way, the adaptability and benefit maximization of the digital engineering system in different application scenarios can be ensured. Therefore, adopting this method can improve the overall performance of the digital engineering system.
[0051] Optionally, the computer device 101 can be a terminal device or a server. Among them, the terminal device mentioned here can include, but is not limited to: smart phones, tablet computers, laptop computers, desktop computers, smart watches, smart TVs, smart vehicle terminals, etc. The server mentioned here can be an independent physical server or a server cluster or distributed system composed of multiple physical servers, etc.
[0052] Please refer to Figure 2 , Figure 2 is a schematic flowchart of a method for training a digital twin model provided by an embodiment of the present application. This method can be executed by a computer device (such as the computer device 101 in Figure 1 ). As Figure 2 shown, the method for training the digital twin model can include, but is not limited to, the following steps:
[0053] S201. Obtain multiple data corresponding to the target entity in real time, and the multiple data are heterogeneous data.
[0054] Optionally, the multiple data may include, but are not limited to, the operation data of the target entity, the environmental detection data, the business process data, etc. For example, the multiple data may include the temperature, vibration, pressure, current, etc. of the target entity.
[0055] Optionally, the operation data of the target entity can be obtained by a computer device through communication protocols such as industrial Ethernet and fieldbus, via devices such as a Programmable Logic Controller (PLC) and a sensor gateway. In this way, since communication protocols such as industrial Ethernet and fieldbus have efficient data transmission capabilities, they can ensure the real-time and low-latency data transmission. Thus, by adopting the above communication protocols, the computer device can enable the operation data of the target entity to be reflected in the digital twin system in a timely manner, providing support for the optimization and fault prediction of the digital engineering system.
[0056] In an alternative embodiment, the computer device's real-time acquisition of multiple data corresponding to the target entity can be achieved by collecting multiple data from different data sources, in different formats, and with different communication protocols through multiple sensors or data interfaces, etc.
[0057] S202. Extract and fuse features from the multiple data to obtain a target fusion feature vector.
[0058] In an alternative embodiment, when the computer device extracts and fuses features from the multiple data to obtain a target fusion feature vector, it may include: performing data cleaning, data conversion, and normalization processing on the multiple data to obtain the normalized multiple data; extracting and fusing features from the normalized multiple data to obtain the target fusion feature vector. In this way, the accuracy of the obtained target fusion feature vector can be ensured.
[0059] S203. Analyze the target fusion feature vector to obtain a quantization index for quantifying the pre-constructed digital twin model corresponding to the target entity; the quantization index includes the production efficiency, production cost, and energy consumption of the target entity.
[0060] S204. Construct a target optimization function based on the quantization index.
[0061] In an alternative embodiment, when the computer device constructs a target optimization function based on the quantization index, it may include: constructing a first optimization function corresponding to the production efficiency, constructing a second optimization function corresponding to the production cost, and constructing a third optimization function corresponding to the energy consumption; determining the target optimization function based on the first optimization function, the second optimization function, and the third optimization function.
[0062] In some embodiments, the first optimization function (denoted as f eff) It can be as shown in the following formulas (1) and (2).
[0063] (1)
[0064] (2)
[0065] In formulas (1) and (2), w i represents the weight of the i-th production link, representing the contribution degree of the i-th production link to the overall production efficiency; n represents the total number of production links; represents the production efficiency of the i-th production link; p i represents the output of the i-th production link; t i represents the time consumption of the i-th production link.
[0066] By using formula (1), the computer device can ensure that the first optimization function can more accurately reflect the impact of each production link on the overall production efficiency through weighted calculation of the production efficiencies corresponding to different production links.
[0067] In some embodiments, the second optimization function (denoted as f cost ) can be as shown in the following formula (3).
[0068] (3)
[0069] In formula (3), C procure represents the procurement cost corresponding to the target entity; C maintain represents the operation and maintenance cost corresponding to the target entity; C energy represents the energy consumption cost, which can be determined by the following formula (4).
[0070] (4)
[0071] In formula (4), E represents the energy consumption of the target entity; p energy represents the electricity price.
[0072] In some embodiments, the third optimization function (denoted as f energy ) can be as shown in the following formula (5).
[0073] (5)
[0074] In formula (5), E i represents the energy consumption of the i-th production link; n represents the number of production links.
[0075] In some embodiments, the computer device determines a target optimization function based on a first optimization function, a second optimization function, and a third optimization function, and the following formula (6) can be used.
[0076] (6)
[0077] In formula (6), F represents the target optimization function; w1 represents the weight of the first optimization function f eff ; w2 represents the weight of the second optimization function f cost ; w3 represents the weight of the third optimization function f energy .
[0078] S205. With the aim of determining the optimal solution of the target optimization function, train the digital twin model to obtain the trained digital twin model.
[0079] In the embodiments of the present application, the computer device can obtain multiple data corresponding to the target entity in real time, and the multiple data are heterogeneous data; extract and fuse the features of the multiple data to obtain a target fusion feature vector; analyze the target fusion feature vector to obtain a quantization index for quantifying the pre-constructed digital twin model corresponding to the target entity; construct a target optimization function based on the quantization index; the target optimization function includes a first optimization function corresponding to the production efficiency of the target entity, a second optimization function corresponding to the production cost of the target entity, and a third optimization function corresponding to the energy consumption of the target entity; with the aim of determining the optimal solution of the target optimization function, train the digital twin model to obtain the trained digital twin model. By adopting this method, on the one hand, the computer device can fully explore the correlation between multiple heterogeneous data by fusing the multiple heterogeneous data corresponding to the target entity to obtain a target fusion feature vector, which is beneficial to improving the accuracy of subsequent decision-making analysis; on the other hand, the computer device can obtain a quantization index (production efficiency, production cost, and energy consumption) for quantifying the pre-constructed digital twin model corresponding to the target entity by analyzing the target fusion feature vector. Then, based on the quantization index, a target optimization function including a first optimization function corresponding to production efficiency, a second optimization function corresponding to the production cost of the target entity, and a third optimization function corresponding to the energy consumption of the target entity can be constructed, and the digital twin model corresponding to the target entity can be collaboratively optimized based on multiple objectives such as production efficiency, cost, and energy consumption, so as to achieve the comprehensive balance of multiple objectives and ensure that in the actual application process using the trained digital twin model, not only the production efficiency can be improved, but also the cost and energy consumption can be minimized to the greatest extent. In this way, the adaptability and benefit maximization of the digital engineering system in different application scenarios can be ensured. Therefore, adopting this method can improve the overall performance of the digital engineering system.
[0080] In an alternative embodiment, Figure 2 In the training method of the digital twin model shown, the computer device may further select a pre-trained convolutional neural network and a pre-trained long short-term memory network from multiple pre-trained neural networks based on the data types of multiple data of the target entity; construct an initial fusion model based on the pre-trained convolutional neural network, the pre-trained long short-term memory network, and the attention mechanism network; use multiple historical data corresponding to the target entity to train the initial fusion model to obtain a target fusion model; the target fusion model includes a convolutional neural network, a long short-term memory network, and an attention mechanism network; the computer device extracts and fuses features from multiple data to obtain a target fusion feature vector, which may include: calling the target fusion model to extract and fuse features from multiple data to obtain a target fusion feature vector.
[0081] In this embodiment, the computer device calls the target fusion model to extract and fuse features from multiple data to obtain a target fusion feature vector, including: using the convolutional neural network to extract features from multiple data to obtain multiple local features corresponding to the multiple data; inputting the multiple local features into the long short-term memory network to obtain multiple temporal features corresponding to the multiple data; splicing the multiple temporal features and the multiple data to obtain an initial fusion feature vector containing spatio-temporal features; inputting the initial fusion feature vector into the attention mechanism network to obtain the weight corresponding to each data feature, and obtaining the target fusion feature vector based on the weight corresponding to each data feature and each data feature.
[0082] Optionally, the computer device uses the convolutional neural network to extract features from multiple data to obtain multiple local features corresponding to the multiple data, which may include: performing data cleaning, data transformation, and normalization processing on the multiple data to obtain the normalized multiple data; inputting the normalized multiple data into the convolutional neural network to obtain multiple local features corresponding to the multiple data.
[0083] Among them, the convolutional neural network is mainly used to extract local features and is particularly suitable for processing two-dimensional data such as image data and running waveform data corresponding to the target entity. In image data, the convolution kernel scans the input data through a sliding window to identify local features such as the edges and textures of the image, and then constructs a high-level feature representation of the image; in running waveform data, the convolution operation can identify short-term local change patterns of the data, such as mutations or periodic features in current and voltage waveforms. These local features are crucial for understanding the current state of the target entity and are helpful for further fault diagnosis and performance evaluation.
[0084] Among them, the long short-term memory network can effectively process the long-term dependencies in data through its gating mechanism, and is particularly suitable for the time series analysis of the operation data of the target entity. For example, during the operation of the target entity, data such as temperature and pressure collected by sensors usually have time delays. At this time, the computer device can use the long short-term memory network to learn the long-term trends of these data in the time dimension. In this way, the future state of the target entity can be predicted using historical data. During this process, the state update process of the long short-term memory network can be represented by the following formula (7).
[0085] (7)
[0086] In formula (7), h t represents the hidden state at the current moment; h t-1 represents the hidden state at the previous moment; x t represents the input at the current moment; W h and W x both represent weight matrices; b h represents the bias term; f represents the activation function. Through this update process, the long-term dependencies of the data captured by the long short-term memory network can be transformed into time series features that can reflect the system state.
[0087] Among them, the computer device can adopt an attention mechanism to determine the weight corresponding to each data feature through the following formula (8).
[0088] (8)
[0089] In formula (8), a i represents the weight corresponding to the i-th data feature; s i represents the score of the i-th data feature; j represents the index of all data features in the initial fusion feature vector. Through this mechanism, the attention to different data features can be dynamically adjusted, which is beneficial to improving the accuracy of the target fusion feature vector.
[0090] Adopting this implementation method, the computer device can apply the pre-trained convolutional neural network (i.e., the pre-trained convolutional neural network) and the pre-trained long short-term memory network (i.e., the pre-trained long short-term memory network) that have been trained in related fields to the current task of determining the target fusion model by using the method of transfer learning. Thus, the number of training times can be reduced, the training process can be accelerated, and the target fusion model can converge faster. In addition, by borrowing the information obtained from other tasks, the learning effect of the target fusion model can be improved. Therefore, during the application process, using the target fusion model to extract and fuse features from multiple data, a more accurate target fusion feature vector can be obtained.
[0091] In an alternative embodiment, Figure 2 In the training method of the digital twin model shown, the computer device analyzes the target fusion feature vector to obtain a quantization index for quantizing the pre-constructed digital twin model corresponding to the target entity, including: analyzing the target fusion feature vector to obtain the operation indexes of the target entity corresponding to the features of each dimension in the target fusion feature vector; determining the correlation coefficients between the features of each dimension and each operation index; selecting the target correlation coefficients with the correlation coefficients greater than the preset correlation coefficient threshold, and using the operation indexes corresponding to the target correlation coefficients as the quantization indexes for quantizing the digital twin model.
[0092] Exemplarily, assume that in the target fusion feature vector, the features of the 1st to 3rd dimensions correspond to the production efficiency of the target entity, the features of the 4th to 6th dimensions correspond to the energy consumption of the target entity, and the features of the 7th to 9th dimensions correspond to the production cost; among them, the correlation coefficient between the features of the 1st to 3rd dimensions and the production efficiency is 0.89, the correlation coefficient between the features of the 4th to 6th dimensions and the energy consumption is 0.91, and the correlation coefficient between the features of the 7th to 9th dimensions and the production cost is 0.82; and assume that the preset correlation coefficient threshold is 0.8. In this case, the computer device can use the production efficiency, production cost, and energy consumption as the quantization indexes for quantizing the digital twin model.
[0093] Adopting this embodiment, quantization indexes with a relatively large correlation with the digital twin model corresponding to the target entity can be determined, thereby facilitating the subsequent optimization of the digital twin model.
[0094] In an alternative embodiment, Figure 2 In the training method of the digital twin model shown, the computer device obtains the geometric shape data, material property data, and behavior pattern data corresponding to the target entity; constructs an initial digital twin model based on the geometric shape data, material property data, and behavior pattern data; establishes a real-time data transmission channel in the initial digital twin model, and uses the digital twin model including the real-time data transmission channel as the pre-constructed digital twin model corresponding to the target entity; wherein, the real-time data transmission channel is used to obtain the data corresponding to the target entity in real time, and the data is used to update the digital twin model.
[0095] Optionally, the geometric shape data corresponding to the target entity can be obtained by the computer device using a laser scanning method, where the geometric shape data can also be called point cloud data.
[0096] Optionally, the material property data corresponding to the target entity can be obtained by the computer device from the material analysis device. Among them, the material property data may include, but is not limited to, physical property data such as the density, hardness, and elastic modulus of the target entity.
[0097] Optionally, the behavior pattern data can be obtained by the computer device after analyzing the historical record data corresponding to the target entity. Among them, the behavior pattern data can provide a reference for subsequent behavior modeling, especially in the health state detection and fault prediction of the target entity.
[0098] In some embodiments, the computer device can construct a geometric model corresponding to the target entity based on the geometric shape data; construct a physical model corresponding to the target entity based on the material property data; construct a behavior model corresponding to the target entity based on the behavior pattern data; and perform an integration process on the geometric model, physical model, and behavior model to obtain an initial digital twin model corresponding to the target entity.
[0099] Optionally, when the computer device constructs a geometric model corresponding to the target entity based on the geometric shape data, it can use reverse engineering technology to convert the geometric shape data into mesh data, and convert the discrete point set into a continuous surface through a surface reconstruction algorithm to obtain the geometric model.
[0100] In parametric modeling, the computer device can establish a parametric formula based on the relationship between key parameters. For example, the relationship between the model size and material properties can be expressed by the following formula (9).
[0101] (9)
[0102] In formula (9), represents the key dimension of the model; represents the dimension parameter of the physical entity; represents the material property parameter. This formula indicates that the size of the target entity can change with the change of material properties.
[0103] Adopting this implementation method, the computer device can associate the geometric changes, material properties, and behavior patterns of the digital twin model, so that the digital twin model can be automatically adjusted under different operating conditions. In addition, by establishing a real-time data transmission channel, the digital twin network can obtain the real-time data corresponding to the target entity in a timely manner, and then update the digital twin network based on the real-time data, so that the digital twin model can dynamically reflect the operating state of the physical entity.
[0104] In an alternative implementation, Figure 2In the training method of the digital twin model shown, the computer device can also perform real-time updates on the digital twin model based on multiple data and the mapping relationship between the real-time data corresponding to the predetermined target entity and the model parameters of the digital twin model.
[0105] Optionally, the mapping relationship can be shown as the following formula (10).
[0106] (10)
[0107] In formula (10), represents the change amount of the model parameters of the digital twin model; represents the change amount of multiple data; the g() function defines the relationship between the change amount of multiple data and the change of the model parameter quantity.
[0108] Adopting this implementation method, the computer device performs real-time updates on the digital twin model based on multiple data and the mapping relationship between the real-time data corresponding to the predetermined target entity and the model parameters of the digital twin model, which can ensure the continuous consistency between the digital twin model and the target entity throughout the entire life cycle. Thus, it can accurately reflect the changes and operating states of the target entity. Furthermore, it is beneficial for the digital twin model to provide more accurate decision support and prediction capabilities subsequently.
[0109] Please refer to Figure 3 , Figure 3 which is a schematic flowchart of another training method of the digital twin model provided by an embodiment of the present application. Different from the training method of the digital twin model shown in Figure 2 , in the training method of the digital twin model shown in Figure 3 , it specifically elaborates on how the computer device extracts and fuses features from multiple data to obtain the target fusion feature vector. As shown in Figure 3 , the training method of this digital twin model may include but is not limited to the following steps:
[0110] S301. Real-time obtain multiple data corresponding to the target entity, and the multiple data are heterogeneous data.
[0111] In an optional implementation method, the relevant description of step S301 can refer to the description in the foregoing step S201, and will not be elaborated here.
[0112] S302. Perform data cleaning, data conversion, and normalization processing on the multiple data to obtain the normalized multiple data.
[0113] In an alternative embodiment, the computer device extracts and fuses features of multiple data to obtain a target fusion feature vector, which may include: performing data cleaning processing on the multiple data to obtain the multiple cleaned data; performing data conversion processing on the multiple cleaned data to obtain the multiple converted data; performing normalization processing on the multiple converted data to obtain the multiple normalized data.
[0114] In some embodiments, the computer device performs data cleaning processing on the multiple data to obtain the multiple cleaned data, which may include: determining the abnormal data in the multiple data, and removing the abnormal data from the multiple data to obtain the multiple data after removing the abnormal data; determining the duplicate data in the multiple data after removing the abnormal data, and removing the duplicate data from the multiple data after removing the abnormal data to obtain the multiple cleaned data.
[0115] Optionally, the computer device determines the abnormal data in the multiple data by using an outlier detection algorithm. Among them, the outlier detection algorithm may be a Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm.
[0116] The DBSCAN algorithm can effectively process data with noise and outliers. Specifically, assuming there is a set of data points {x1, x2,..., x n}, the computer device can use the DBSCAN algorithm to calculate the density around each data point, and mark and remove the data points with density less than the preset density threshold, so as to improve the quality of the data. Among them, when the computer device calculates the density around each data point, the following formula (11) can be used.
[0117] (11)
[0118] In formula (11), represents the density of the data point x i ; represents the set of neighbor data points whose distance from the data point x i is less than or equal to ; represents the number of neighbor data points.
[0119] Optionally, the computer device may determine duplicate data among the multiple pieces of data after removing abnormal data. It can calculate the hash value corresponding to each piece of data among the multiple pieces of data after removing abnormal data by using the hash algorithm, compare the multiple hash values, and determine the duplicate data from the multiple pieces of data after removing abnormal data based on the comparison result. In this way, not only can the high computational cost of comparing each record one by one be avoided, enabling rapid processing of a large amount of data, but also the uniqueness of the data can be ensured, providing support for subsequent analysis.
[0120] In some embodiments, when the computer device performs data conversion processing on the multiple pieces of data after cleaning to obtain multiple pieces of data after conversion, it may include: converting the unstructured data among the multiple pieces of data after cleaning into structured data, and converting the structured data obtained by conversion and the structured data among the multiple pieces of data after cleaning into multiple pieces of data in a target format; using the multiple pieces of data in the target format as the multiple pieces of data after conversion.
[0121] Exemplarily, assuming that the unstructured data among the multiple pieces of data after cleaning are text data and image data, for the text data, the computer device may use natural language processing technology to process the text data to obtain text information corresponding to the text data, such as keywords, sentiment tendencies, etc., and convert the text information into structured data; for the image data, the computer device may use image processing technology to extract image information in the image data, such as object type, location, etc., and convert the image information into structured data.
[0122] Optionally, when the computer device converts the structured data obtained by conversion and the structured data among the multiple pieces of data after cleaning into multiple pieces of data in a target format, it may use data mapping technology to convert the structured data obtained by conversion and the structured data among the multiple pieces of data after cleaning into multiple pieces of data in a target format.
[0123] Exemplarily, assuming that the structured data are data in a relational database, the computer device may use data mapping technology to uniformly convert the data in different database table structures into a unified format. For example, by formulating mapping rules for data fields, unifying field names and types, ensuring seamless docking of structured data from different sources, and unifying them into a data format suitable for subsequent analysis.
[0124] In some embodiments, when the computer device performs normalization processing on the multiple pieces of data after conversion to obtain multiple pieces of data after normalization, it may use the minimum-maximum or Z-score normalization method to perform normalization processing on the multiple pieces of data after conversion to obtain multiple pieces of data after normalization. In this way, by mapping data of different scales into the unified [0,1] interval, the dimensionality differences between different data sources and dimensions can be eliminated, ensuring that feature extraction and fusion are performed on the data at the same scale, thereby improving the accuracy of the data features after fusion.
[0125] Among them, the min-max normalization formula can be shown as the following formula (12).
[0126] (12)
[0127] In formula (12), x i represents the i-th data among the multiple converted data; x max represents the maximum value among the multiple converted data; x min represents the minimum value among the multiple converted data; represents the i-th normalized data.
[0128] S303. Invoke the target fusion model to perform feature extraction and fusion on the normalized multiple data to obtain a target fusion feature vector; the target fusion model includes a convolutional neural network, a long short-term memory network, and an attention mechanism network, and the convolutional neural network and the long short-term memory network are determined by using the method of transfer learning.
[0129] In an optional implementation manner, before step S303, the computer device can also select a pre-trained convolutional neural network and a pre-trained long short-term memory network from multiple pre-trained neural networks based on the data types of the multiple data of the target entity; construct an initial fusion model based on the pre-trained convolutional neural network, the pre-trained long short-term memory network, and the attention mechanism network; and use the multiple historical data corresponding to the target entity to train the initial fusion model to obtain the target fusion model.
[0130] In some embodiments, when the computer device invokes the target fusion model to perform feature extraction and fusion on multiple data to obtain a target fusion feature vector, it may include: using the convolutional neural network to perform feature extraction on the multiple data to obtain multiple local features corresponding to the multiple data; inputting the multiple local features into the long short-term memory network to obtain multiple temporal features corresponding to the multiple data; splicing the multiple temporal features and the multiple data to obtain an initial fusion feature vector containing spatio-temporal features; inputting the initial fusion feature vector into the attention mechanism network to obtain the weight corresponding to each data feature, and obtaining the target fusion feature vector based on the weight corresponding to each data feature and each data feature.
[0131] S304. Analyze the target fusion feature vector to obtain a quantization index for quantifying the pre-constructed digital twin model corresponding to the target entity; the quantization index includes the production efficiency, production cost, and energy consumption of the target entity.
[0132] S305. Construct a target optimization function based on the quantization index.
[0133] In an alternative embodiment, the relevant description in step S305 can be referred to the description in the aforementioned step S204, and will not be elaborated here.
[0134] S306. Taking the determination of the optimal solution of the target optimization function as the goal, train the digital twin model to obtain the trained digital twin model.
[0135] In an alternative embodiment, the computer device takes the determination of the optimal solution of the target optimization function as the goal, trains the digital twin model to obtain the trained digital twin model. It can be to use the genetic algorithm to train the digital twin model with the determination of the optimal solution of the target optimization function as the goal to obtain the trained digital twin model; or it can be to use the particle swarm optimization algorithm to train the digital twin model with the determination of the optimal solution of the target optimization function as the goal to obtain the trained digital twin model.
[0136] Among them, using the genetic algorithm, in each iteration process, encode the parameters of the digital twin model, encode the operating parameters of the target entity, the control parameters of the system, etc. into gene sequences, and each gene sequence represents a candidate solution; through genetic operations, such as selection, crossover, mutation, etc., generate new parameter combinations, and gradually explore a more optimized solution space. Among them, the computer device can determine the individuals (solutions) with higher fitness as the selected solutions entering the next generation through the selection operation, and eliminate the individuals with lower fitness. The computer device can exchange the genes of two individuals through the crossover operation to generate new offspring. The computer device can randomly change some genes of an individual through the mutation operation to increase the diversity of the solution space. Through the above genetic operations, the optimal solution of the optimization can be continuously approximated.
[0137] Among them, in the particle swarm optimization algorithm, each particle represents a possible solution, its position represents the parameter combination, and the velocity controls the search direction and pace of the particle. In the iteration process, each particle will adjust its flight speed and direction according to its own historical optimal position and the group historical optimal position. In each iteration, the position update formula of the particle is as follows in formulas (13) and (14).
[0138] (13)
[0139] (14)
[0140] In formulas (13) and (14), represents the velocity of the i-th particle at the t-th moment; represents the position of the i-th particle at the t-th moment; represents the historical optimal position of the i-th particle; It represents the group historical optimal position; w represents the inertia weight; c1 and c2 represent the learning factors; r1 and r2 are random numbers. By adjusting the positions and velocities of the particles, the optimal solution can be determined.
[0141] In the embodiments of the present application, the computer device can obtain multiple data corresponding to the target entity in real time, and the multiple data are heterogeneous data; perform data cleaning, data transformation, and normalization processing on the multiple data to obtain the normalized multiple data; call the target fusion model to perform feature extraction and fusion on the normalized multiple data to obtain the target fusion feature vector; the target fusion model includes a convolutional neural network, a long short-term memory network, and an attention mechanism network, and the convolutional neural network and the long short-term memory network are determined by using the method of transfer learning; analyze the target fusion feature vector to obtain a quantization index for quantifying the pre-constructed digital twin model corresponding to the target entity; the quantization index includes the production efficiency, production cost, and energy consumption of the target entity; construct a target optimization function based on the quantization index; aim at determining the optimal solution of the target optimization function, train the digital twin model to obtain the trained digital twin model. By using this method, on the one hand, the computer device can fully explore the correlation between multiple heterogeneous data by performing data cleaning, data transformation, and normalization processing on the multiple data corresponding to the target entity, and calling the target fusion model to perform feature extraction and fusion on the normalized multiple data, and obtain a more accurate target fusion feature vector. In this way, it is beneficial to improve the accuracy of subsequent decision-making analysis; on the other hand, the computer device can obtain a quantization index (production efficiency, production cost, and energy consumption) for quantifying the pre-constructed digital twin model corresponding to the target entity by analyzing the target fusion feature vector. Then, based on the quantization index, a target optimization function including a first optimization function corresponding to production efficiency, a second optimization function corresponding to the production cost of the target entity, and a third optimization function corresponding to the energy consumption of the target entity can be constructed, and the digital twin model corresponding to the target entity can be collaboratively optimized based on multiple objectives such as production efficiency, cost, and energy consumption, so as to achieve the comprehensive balance of multiple objectives, ensure that in the actual application process of the trained digital twin model, not only the production efficiency can be improved, but also the cost and energy consumption can be minimized to the greatest extent. In this way, the adaptability and benefit maximization of the digital engineering system in different application scenarios can be ensured. Therefore, using this method can improve the overall performance of the digital engineering system.
[0142] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown according to the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.
[0143] Based on the same inventive concept, an embodiment of the present application also provides a training device for a digital twin model for implementing the training method of the digital twin model involved above. The solution provided by this device for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the training device for the digital twin model provided below can refer to the limitations on the training method of the digital twin model in the above text, and will not be repeated here.
[0144] Please refer to Figure 4 , Figure 4 which is a schematic structural diagram of a training device for a digital twin model provided by an embodiment of the present application. As Figure 4 shown, the training device for the digital twin model may include but is not limited to:
[0145] An acquisition module 401, configured to acquire multiple pieces of data corresponding to a target entity in real time, and the multiple pieces of data are heterogeneous data;
[0146] A feature extraction and fusion module 402, configured to perform feature extraction and fusion on the multiple pieces of data to obtain a target fusion feature vector;
[0147] A processing module 403, configured to analyze the target fusion feature vector to obtain a quantization index for quantifying a digital twin model corresponding to a pre-constructed target entity; the quantization index includes the production efficiency, production cost, and energy consumption corresponding to the target entity;
[0148] A construction module 404, configured to construct a target optimization function based on the quantization index;
[0149] A training module 405, configured to train the digital twin model with the goal of determining the optimal solution of the target optimization function to obtain a trained digital twin model.
[0150] In one embodiment, the device may further include a selection module. The selection module is configured to select a pre-trained convolutional neural network and a pre-trained long short-term memory network from multiple pre-trained neural networks based on the data types of multiple data of the target entity; the construction module 404 is further configured to construct an initial fusion model based on the pre-trained convolutional neural network, the pre-trained long short-term memory network, and the attention mechanism network; the training module 405 is further configured to use multiple historical data corresponding to the target entity to train the initial fusion model to obtain a target fusion model; the target fusion model includes a convolutional neural network, a long short-term memory network, and an attention mechanism network; when the feature extraction and fusion module 402 is configured to perform feature extraction and fusion on multiple data to obtain a target fusion feature vector, it is specifically configured to: call the target fusion model to perform feature extraction and fusion on multiple data to obtain a target fusion feature vector.
[0151] In one embodiment, when the feature extraction and fusion module 402 is configured to call the target fusion model to perform feature extraction and fusion on multiple data to obtain a target fusion feature vector, it is specifically configured to: use the convolutional neural network to perform feature extraction on multiple data to obtain multiple local features corresponding to the multiple data; input the multiple local features into the long short-term memory network to obtain multiple temporal features corresponding to the multiple data; splice the multiple temporal features and the multiple data to obtain an initial fusion feature vector including spatio-temporal features; input the initial fusion feature vector into the attention mechanism network to obtain the weights corresponding to each data feature, and based on the weights corresponding to each data feature and each data feature, obtain a target fusion feature vector.
[0152] In one embodiment, when the processing module 403 is configured to analyze the target fusion feature vector to obtain a quantization index for quantifying the pre-constructed digital twin model corresponding to the target entity, it is specifically configured to: analyze the target fusion feature vector to obtain the operation indexes of the target entity corresponding to the features of each dimension in the target fusion feature vector; determine the correlation coefficients between the features of each dimension and the respective operation indexes; select the target correlation coefficients with the correlation coefficients greater than the preset correlation coefficient threshold, and use the operation indexes corresponding to the target correlation coefficients as the quantization indexes for quantifying the digital twin model.
[0153] In one embodiment, the obtaining module 401 is further configured to obtain geometric shape data, material property data, and behavior pattern data corresponding to the target entity; the constructing module 404 is further configured to construct an initial digital twin model based on the geometric shape data, the material property data, and the behavior pattern data; establish a real-time data transmission channel in the initial digital twin model, and use the digital twin model including the real-time data transmission channel as the digital twin model corresponding to the pre-constructed target entity; wherein, the real-time data transmission channel is used to obtain the data corresponding to the target entity in real time, and the data is used to update the digital twin model.
[0154] In one embodiment, the processing module 403 is further configured to: perform real-time update on the digital twin model based on multiple data and the mapping relationship between the real-time data corresponding to the pre-determined target entity and the model parameters of the digital twin model.
[0155] Each module in the above digital twin model training device can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in or independent of the processor in the terminal device in the form of hardware, or stored in the memory in the terminal device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0156] In an exemplary embodiment, the embodiment of the present application provides a computer device, which can be a terminal device, and its internal structure diagram can be as Figure 5As shown in the figure. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, near field communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a method for training a digital twin model. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad set on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.
[0157] Those skilled in the art can understand that Figure 5 the structure shown in the figure is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.
[0158] In an exemplary embodiment, the present application provides a computer device, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, it implements the steps in the above method for training a digital twin model.
[0159] In an exemplary embodiment, the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, it implements the steps in the above method for training a digital twin model.
[0160] In an exemplary embodiment, the present application provides a computer program product, including a computer program. When the computer program is executed by the processor, it implements the steps in the above method for training a digital twin model.
[0161] It should be noted that the data involved in this application (including but not limited to multiple data corresponding to the target entity, the target fusion feature vector, quantization indicators, etc.) are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.
[0162] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., and are not limited thereto.
[0163] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this application.
[0164] The above-described embodiments merely represent several implementation manners of this application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of this application. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of this application shall be subject to the appended claims.
Claims
1. A training method for a digital twin model, characterized in that The method includes: Obtaining multiple data corresponding to the target entity in real time, where the multiple data are heterogeneous data; Performing feature extraction and fusion on the multiple data to obtain a target fusion feature vector; Analyzing the target fusion feature vector to obtain a quantization index for quantifying the pre-constructed digital twin model corresponding to the target entity; the quantization index includes the production efficiency, production cost, and energy consumption corresponding to the target entity; Constructing a target optimization function based on the quantization index; Training the digital twin model with the goal of determining the optimal solution of the target optimization function to obtain a trained digital twin model.
2. The method according to claim 1, characterized in that, The method further includes: Based on the data types of the multiple data of the target entity, selecting a pre-trained convolutional neural network and a pre-trained long short-term memory network from multiple pre-trained neural networks; Constructing an initial fusion model based on the pre-trained convolutional neural network, the pre-trained long short-term memory network, and the attention mechanism network; Training the initial fusion model with the multiple historical data corresponding to the target entity to obtain a target fusion model; the target fusion model includes a convolutional neural network, a long short-term memory network, and an attention mechanism network; The performing feature extraction and fusion on the multiple data to obtain a target fusion feature vector includes: Invoking the target fusion model to perform feature extraction and fusion on the multiple data to obtain a target fusion feature vector.
3. The method according to claim 2, wherein The invoking the target fusion model to perform feature extraction and fusion on the multiple data to obtain a target fusion feature vector includes: Using the convolutional neural network to perform feature extraction on the multiple data to obtain multiple local features corresponding to the multiple data; Inputting the multiple local features into the long short-term memory network to obtain multiple temporal features corresponding to the multiple data; Concatenating the multiple temporal features and the multiple data to obtain an initial fusion feature vector containing spatio-temporal features; Inputting the initial fusion feature vector into the attention mechanism network to obtain the weight corresponding to each data feature, and obtaining a target fusion feature vector based on the weight corresponding to each data feature and each data feature.
4. The method according to claim 1, characterized in that, The analyzing the target fusion feature vector to obtain a quantization index for quantifying the pre-constructed digital twin model corresponding to the target entity includes: Analyzing the target fusion feature vector to obtain the operation indexes of the target entity corresponding to the features of each dimension in the target fusion feature vector; Determining the correlation coefficients between the features of each dimension and the respective operation indexes; Selecting target correlation coefficients with correlation coefficients greater than a preset correlation coefficient threshold, and using the operation indexes corresponding to the target correlation coefficients as the quantization indexes for quantifying the digital twin model.
5. The method according to claim 1, characterized in that, The method further includes: Obtaining the geometric shape data, material property data, and behavior pattern data corresponding to the target entity; Construct an initial digital twin model based on the geometric shape data, the material property data, and the behavior pattern data; Establish a real-time data transmission channel in the initial digital twin model, and use the digital twin model including the real-time data transmission channel as the digital twin model corresponding to the pre-constructed target entity; Among them, the real-time data transmission channel is used to obtain the data corresponding to the target entity in real time, and the data is used to update the digital twin model.
6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Based on the multiple pieces of data and the mapping relationship between the real-time data corresponding to the target entity and the model parameters of the digital twin model determined in advance, the digital twin model is updated in real time.
7. A training device for a digital twin model, characterized in that, The device includes: An acquisition module, configured to acquire multiple pieces of data corresponding to a target entity in real time, and the multiple pieces of data are heterogeneous data; A feature extraction and fusion module, configured to perform feature extraction and fusion on the multiple pieces of data to obtain a target fusion feature vector; A processing module, configured to analyze the target fusion feature vector to obtain a quantization index for quantifying the digital twin model corresponding to the pre-constructed target entity; the quantization index includes the production efficiency, production cost, and energy consumption corresponding to the target entity; A construction module, configured to construct a target optimization function based on the quantization index; A training module, configured to train the digital twin model with the goal of determining the optimal solution of the target optimization function to obtain a trained digital twin model.
8. A computer device, characterized in that, Including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.