High-precision digital twinning method for LFO-FCAL data model deviation compensation based on geometric behavior model

Abstract

The invention discloses a high-precision digital twinning method and device for LFO-FCAL data model deviation compensation based on a geometric behavior model. The method comprises the following steps: acquiring real-time data of a physical entity at the current moment and historical data of each moment in a first preset time period; the real-time data at the current moment are input into a geometric behavior model, initial control data are obtained, and the geometric behavior model is a digital twinborn model of a physical entity; inputting the historical data of each moment in the first preset time period into the target LFO-FCAL model to obtain deviation data corresponding to the real-time data; and compensating the initial control data by using the deviation data to obtain target control data for digital twinning. According to the method, the accuracy of the target control data is improved, the requirement of a user for the output accuracy is met, and then prediction, diagnosis and full-life-cycle management of a simulation result through digital twinning are met.

Description

Technical Field

[0001] The present invention relates to the technical field of digital twins, and particularly to a high-precision digital twin method and device for compensating the deviation of the LFO-FCAL data model based on a geometric behavior model. Background Art

[0002] Geometric behavior modeling of physical entities is the basis for constructing digital twin technology. Through the digital twin system, users can monitor and make decisions on physical entities more intuitively and comprehensively. However, the geometric behavior models in the prior art are modeled based on physical entity models, and the input quantities considered by the physical entity models are limited by the domain knowledge of experts and cannot traverse all external factors, resulting in a certain deviation between the output of the geometric behavior model and the output of the real physical entity, making the results of the digital twin system inaccurate and unable to meet the user's requirements for output accuracy. Based on this, how to improve the accuracy of the digital twin system is an urgent problem to be solved. Summary of the Invention

[0003] The present invention aims to solve at least one of the technical problems in the related art to some extent.

[0004] To this end, an object of the present invention is to propose a high-precision digital twin method for compensating the deviation of the LFO-FCAL data model based on a geometric behavior model. The method inputs the historical data at each moment within a first preset time period into the target LFO-FCAL model to obtain the deviation data corresponding to the real-time data, and uses the deviation data to compensate for the deviation between the initial control data and the real-time data to obtain the target control data, thereby improving the accuracy of the target control data, meeting the user's requirements for output accuracy, and further meeting the prediction, diagnosis, and full-life cycle management through the simulation results of the digital twin.

[0005] Another object of the present invention is to propose a high-precision digital twin device for compensating the deviation of the LFO-FCAL data model based on a geometric behavior model.

[0006] To achieve the above object, an embodiment of one aspect of the present invention proposes a high-precision digital twin method for compensating the deviation of the LFO-FCAL data model based on a geometric behavior model, including:

[0007] Obtain the real-time data of the physical entity at the current moment and the historical data at each moment within a first preset time period;

[0008] Input the real-time data at the current moment into the geometric behavior model to obtain the initial control data, where the geometric behavior model is the digital twin model of the physical entity;

[0009] Input the historical data at each moment within the first preset time period into the target LFO-FCAL model to obtain the deviation data corresponding to the real-time data;

[0010] Use the deviation data to compensate the initial control data to obtain the target control data for digital twin.

[0011] The high-precision digital twin method for LFO-FCAL data model deviation compensation based on geometric behavior model in the embodiments of the present invention may further have the following additional technical features:

[0012] Further, the target LFO-FCAL model includes an input layer, a fully connected layer, an attention layer, an LSTM layer, and an output layer; the step of inputting the historical data at each moment within the first preset time period into the target LFO-FCAL model to obtain the deviation data corresponding to the real-time data includes:

[0013] The historical data at each moment within the first preset time period is transmitted to the fully connected layer through the input layer;

[0014] The fully connected layer expands the dimension of the historical data at each moment to obtain a corresponding first feature matrix;

[0015] The attention layer processes the first feature matrix to obtain a second feature matrix;

[0016] The LSTM layer makes a prediction based on the second feature matrix to obtain a third feature matrix;

[0017] The output layer obtains the deviation data corresponding to the real-time data based on the third feature matrix.

[0018] Further, the step that the attention layer processes the first feature matrix to obtain a second feature matrix includes:

[0019] The attention layer performs a matrix transformation on the first feature matrix to obtain a query matrix, a key matrix, and a value matrix;

[0020] The query matrix and the key matrix perform matrix multiplication to obtain an attention score matrix;

[0021] The attention score matrix is divided by a preset value and then normalized to obtain an attention weight matrix;

[0022] The attention weight matrix and the value matrix are summed to obtain a second feature matrix.

[0023] Further, the method further includes:

[0024] Obtain the output data of the physical entity;

[0025] Compare the target control data and the output data at the same moment based on a preset time interval to obtain a comparison result;

[0026] If the comparison results at a preset number of consecutive moments exceed a preset threshold, perform an online update on the target LFO-FCAL model to obtain an updated LFO-FCAL model;

[0027] Use the updated LFO-FCAL model to obtain deviation data corresponding to the real-time data.

[0028] Further, the performing an online update on the target LFO-FCAL model to obtain an updated LFO-FCAL model includes:

[0029] Obtain a first historical data set within a second preset time period of the geometric behavior model;

[0030] Obtain a second historical data set within a second preset time period of the physical entity;

[0031] Based on the first historical data set and the second historical data set, perform an online update on the target LFO-FCAL model by freezing network layers to obtain an updated LFO-FCAL model.

[0032] To achieve the above object, another embodiment of the present invention proposes a high-precision digital twin device for LFO-FCAL data model deviation compensation based on a geometric behavior model, and the device includes:

[0033] An acquisition module, configured to acquire real-time data of a physical entity at the current moment and historical data at each moment within a first preset time period;

[0034] A first processing module, configured to input the real-time data at the current moment into a geometric behavior model to obtain initial control data, where the geometric behavior model is a digital twin model of the physical entity;

[0035] A second processing module, configured to input the historical data at each moment within the first preset time period into a target LFO-FCAL model to obtain deviation data corresponding to the real-time data;

[0036] A compensation module, configured to compensate the initial control data by using the deviation data to obtain target control data for digital twin.

[0037] The high-precision digital twin method and device for LFO-FCAL data model deviation compensation based on a geometric behavior model proposed by the present invention input historical data at each moment within a first preset time period into a target LFO-FCAL model to obtain deviation data corresponding to real-time data, and use the deviation data to compensate for the deviation between initial control data and real-time data, thereby obtaining target control data, improving the accuracy of the target control data, meeting the user's requirements for output accuracy, and further meeting prediction, diagnosis, and full-life cycle management through the simulation results of digital twins.

[0038] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the following description of embodiments in conjunction with the accompanying drawings, in which:

[0040] Figure 1 FIG. is a flowchart of a high-precision digital twin method for LFO-FCAL data model deviation compensation based on a geometric behavior model according to an embodiment of the present invention;

[0041] Figure 2 FIG. is a schematic structural diagram of a target LFO-FCAL model according to an embodiment of the present invention;

[0042] Figure 3 FIG. is a schematic diagram of the data flow of a digital twin system according to an embodiment of the present invention;

[0043] Figure 4 FIG. is a graph comparing the accuracy of the output of a digital twin model with integrated deviation compensation according to an embodiment of the present invention and that of a traditional geometric behavior model;

[0044] Figure 5 FIG. is a schematic structural diagram of a high-precision digital twin device for LFO-FCAL data model deviation compensation based on a geometric behavior model according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0046] The high-precision digital twin method and device for LFO-FCAL data model deviation compensation based on a geometric behavior model according to an embodiment of the present invention will be described below with reference to the accompanying drawings.

[0047] First, the high-precision digital twin method for LFO-FCAL data model deviation compensation based on a geometric behavior model according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0048] Figure 1 It is a flowchart of the high-precision digital twin method for LFO-FCAL data model deviation compensation based on a geometric behavior model according to an embodiment of the present invention.

[0049] As Figure 1 shown, the high-precision digital twin method for LFO-FCAL data model deviation compensation based on a geometric behavior model includes the following steps:

[0050] Step S1, obtaining the real-time data of the physical entity at the current moment and the historical data at each moment within a first preset time period;

[0051] In an embodiment of the present invention, the above-mentioned first preset time period may include n moments before the current moment. Among them, in an embodiment of the present invention, the above-mentioned n can be set as needed. For example, n is 5.

[0052] Among them, in an embodiment of the present invention, the real-time data of the physical entity at the current moment and the historical data at each moment within a first preset time period can be obtained through sensors on the physical entity.

[0053] Step S2, inputting the real-time data at the current moment into the geometric behavior model to obtain initial control data;

[0054] Among them, in an embodiment of the present invention, the geometric behavior model is a digital twin model of the physical entity.

[0055] Step S3, inputting the historical data at each moment within a first preset time period into the target LFO-FCAL model to obtain deviation data corresponding to the real-time data;

[0056] In an embodiment of the present invention, the above-mentioned target LFO-FCAL model is trained.

[0057] Also, in an embodiment of the present invention, before inputting the historical data at each moment within the first preset time period into the target LFO-FCAL model to obtain the deviation data corresponding to the real-time data, the above method further includes: training the initial LFO-FCAL model to obtain the target LFO-FCAL model. Among them, in an embodiment of the present invention, the initial LFO-FCAL model can be trained with the historical data set of physical entities and the historical data set of geometric behavior models to obtain the target LFO-FCAL model. In an embodiment of the present invention, the method for training the initial LFO-FCAL model is the same as the prior art, and the embodiments of the present disclosure will not elaborate here.

[0058] Among them, in an embodiment of the present invention, Figure 2 is a schematic structural diagram of a target LFO-FCAL model proposed by an embodiment of the present invention. As Figure 2 shown, the above target LFO-FCAL model includes an input layer, a fully connected layer, an attention layer, an LSTM layer, and an output layer.

[0059] Also, in an embodiment of the present invention, the method of inputting the historical data at each moment within the first preset time period into the target LFO-FCAL model to obtain the deviation data corresponding to the real-time data may include the following steps:

[0060] Step S31, the historical data at each moment within the first preset time period is transmitted to the fully connected layer through the input layer;

[0061] Step S32, the fully connected layer expands the dimension of the historical data at each moment to obtain a corresponding first feature matrix;

[0062] Step S33, the attention layer processes the first feature matrix to obtain a second feature matrix;

[0063] Step S34, the LSTM layer makes a prediction based on the second feature matrix to obtain a third feature matrix;

[0064] Step S35, the output layer obtains the deviation data corresponding to the real-time data based on the third feature matrix.

[0065] Among them, in an embodiment of the present invention, the method of the attention layer processing the first feature matrix to obtain the second feature matrix may include the following steps:

[0066] Step S331, the attention layer performs a matrix transformation on the first feature matrix to obtain a query matrix, a key matrix, and a value matrix;

[0067] Step S332, the query matrix and the key matrix perform matrix multiplication to obtain an attention score matrix;

[0068] In step S333, the attention score matrix is divided by a preset value and then normalized to obtain an attention weight matrix.

[0069] In step S334, the attention weight matrix is summed with the value matrix to obtain a second feature matrix.

[0070] In an embodiment of the present invention, after the attention layer performs matrix transformation on the first feature matrix with the same dimension but different weights, a query matrix Q, a key matrix K, and a value matrix V are respectively obtained.

[0071] Among them, in an embodiment of the present invention, the above-mentioned query matrix Q and key matrix K are multiplied to obtain an attention score matrix, thereby reflecting the correlation between data at each moment. And, in an embodiment of the present invention, after the attention score matrix is divided by a preset value, it is normalized through the softmax function to obtain an attention weight matrix. After that, the attention weight matrix is added to the value matrix V and summed to obtain a second feature matrix, so that the relationship between each time-step vector and other time steps can be reflected through the second feature matrix. In an embodiment of the present invention, the above-mentioned preset value can be

[0073] And, in an embodiment of the present invention, in the LSTM layer, the current output is calculated through the data of the previous moment and the current information. The input gate and forget gate are used to control the influence degree of the current data and historical data on the output respectively, and the output gate is used to control the completeness of extracting data from the memory unit. The introduction of the gating mechanism can effectively avoid the problems of gradient explosion and gradient disappearance, enabling the network to remember data from a long time ago and having a good effect on predicting time-series data. Among them, in an embodiment of the present invention, based on the explicit feature data extracted by Attention, LSTM can more accurately obtain the relationship between the current moment output and the inputs at the current moment and historical moments.

[0074] Further, in an embodiment of the present invention, the dimension of the third feature matrix output by the LSTM layer is the same as that of the input data. Based on this, the output layer can determine the last output in the third feature matrix as the deviation data corresponding to the real-time data. As Figure 2 shown, Y in the third feature matrix t-1 is determined as the deviation data corresponding to the real-time data X t

[0075] In an embodiment of the present invention, Table 1 is a performance comparison table of the proposed target LFO-FCAL model and other typical models in this application.

[0076] Table 1

[0077] ​

[0078] As described in Table 1, the above-mentioned target LFO-FCAL model has better prediction accuracy on this dataset than the classical LSTM and LSTM-Attention structures. Among them, Attention is more suitable for feature extraction. In the LSTM-Attention structure, LSTM can only extract the correlation between the current moment and historical moments, resulting in the loss of some relevant information. Since the subsequent Attention module cannot recognize the features of the missing part, the performance of this model is limited. In the above-mentioned target LFO-FCAL model, the fully connected layer elevates the dimension of low-dimensional features and refines the input features. The Attention layer is located at the front side, which can not only extract the correlation with historical moments, but also extract the correlation with future moments within the time window, making it more suitable for feature extraction. The LSTM layer located at the back takes the explicit temporal feature data as input and recognizes the relationship between historical features and the output. Compared with the LSTM-Attention structure, the target LFO-FCAL model structure can give full play to the advantages of each network layer. Based on this, not only can higher accuracy be obtained on the test set, but also shorter training time is required.

[0079] Step S4: Compensate the initial control data with the deviation data to obtain the target control data for digital twin.

[0080] Among them, in an embodiment of the present invention, after obtaining the deviation data and the initial control data through the above steps, the deviation data can be used to compensate the initial control data to obtain the target control data for digital twin.

[0081] Moreover, in an embodiment of the present invention, after obtaining the target control data through the above steps, the target control data can be used for the simulation or emulation of digital twin, so as to assist the user in simulation, prediction, decision-making, and full life cycle management.

[0082] Further, in an embodiment of the present invention, the above method may further include the following steps:

[0083] Step S5: Obtain the output data of the physical entity;

[0084] Step S6: Compare the target control data and the output data at the same moment based on a preset time interval to obtain a comparison result;

[0085] Step S7: If the comparison results at continuously preset numerical moments exceed a preset threshold, perform online update on the target LFO-FCAL model to obtain an updated LFO-FCAL model;

[0086] Step S8: Using the updated LFO-FCAL model, obtain the deviation data corresponding to the real-time data.

[0087] Among them, in an embodiment of the present invention, the above preset time interval can be set as needed. And, in an embodiment of the present invention, the target control data at the same moment can be subtracted from the output data of the physical entity and the absolute value can be taken to obtain a comparison result.

[0088] Furthermore, in an embodiment of the present invention, if the comparison results at continuously preset numerical moments exceed a preset threshold, it indicates that the target LFO-FCAL model needs to be updated online to ensure that the deviation data of the target LFO-FCAL model can meet the accuracy requirements for compensating the geometric behavior model. Among them, in an embodiment of the present invention, the above preset numerical value and preset threshold can be set as needed. For example, the preset numerical value can be 4.

[0089] Among them, in an embodiment of the present invention, the method for updating the target LFO-FCAL model online to obtain the updated LFO-FCAL model may include the following steps:

[0090] Step S71: Obtain the first historical data set within the second preset time period of the geometric behavior model;

[0091] Step S72: Obtain the second historical data set within the second preset time period of the physical entity;

[0092] Step S73: Based on the first historical data set and the second historical data set, realize the online update of the target LFO-FCAL model by freezing the network layer to obtain the updated LFO-FCAL model.

[0093] In an embodiment of the present invention, the above second preset time period may include m moments before the current moment. Among them, in an embodiment of the present invention, the above m can be set as needed. For example, m is 6.

[0094] Moreover, in an embodiment of the present invention, based on the first historical data set and the second historical data set, the target LFO-FCAL model is updated online by fine-tuning the network weights of the last layer of Attention and LSTM, and the remaining network layers in the target LFO-FCAL model are frozen, so as to obtain an updated LFO-FCAL model. In an embodiment of the present invention, the target LFO-FCAL model is updated online with small sample data within a second preset time period, so that the target LFO-FCAL model can consider the influence of environmental changes, and ensure that the deviation data of the target LFO-FCAL model can meet the accuracy requirements for compensating the geometric behavior model.

[0095] Based on the above description, Figure 3 is a schematic diagram of the data flow of a digital twin system proposed by the present invention. As Figure 3 shown, the data flow indicated by the black line is the historical data set for training the initial LFO-FCAL model, including the historical data sets of the physical entity and the geometric behavior model. On the one hand, the data of the physical entity is used as the input feature for model training. On the other hand, it forms a deviation data set with the output of the geometric behavior model and is used as the label for model training. By training and fine-tuning the initial LFO-FCAL model on the historical data set, the optimal hyperparameter combination is obtained; the data flow indicated by the green color is the real-time data flow. On the one hand, the real-time data of the sensor is sent to the geometric behavior model for real-time behavior mapping of the model. On the other hand, the input values of the first n moments collected by the physical entity are input to the target LFO-FCAL model to obtain the deviation data at the current moment. This deviation data is compensated in real time to the output of the geometric behavior model to obtain the output target control data; the data flow indicated by the yellow color is that the output target control data is compared with the output data of the physical entity at a certain time interval. When the statistically multiple differences exceed the threshold, the online update of the target LFO-FCAL model is triggered. Based on the small sample data with a time length of m, the trained target LFO-FCAL model is fine-tuned by the network weights of the last layer of Attention and LSTM to realize the online update of the target LFO-FCAL model, so that the model always considers the influence of environmental changes on the model output and ensures that the deviation of the model prediction always meets the accuracy requirements for compensating the geometric behavior model.

[0096] Moreover, in an embodiment of the present invention, Figure 4 is a graph comparing the accuracy of the output of a digital twin model with fusion deviation compensation proposed in an embodiment of the present invention and the output of a traditional geometric behavior model. As Figure 4As shown in the figure, by comparing the output results of the geometric behavior model of the yellow line with the true values represented by the blue line, it can be seen that the geometric behavior model can better reflect the output of physical entities, but there is still a deviation of about 3%. However, due to the limitations of the modeling accuracy and mechanism integrity of the geometric behavior model, there are systematic and random errors between the model and the true values. The target LFO-FCAL model can well predict the above deviations through training. As Figure 4 shown, the output results shown by the green line of the digital twin model with fusion deviation compensation are basically consistent with the true values. Based on this, the accuracy of the digital twin model with fusion deviation compensation is significantly improved and the credibility is higher.

[0097] According to the high-precision digital twin method for LFO-FCAL data model deviation compensation based on the geometric behavior model proposed in the embodiments of the present invention, the historical data at each moment within the first preset time period is input into the target LFO-FCAL model to obtain the deviation data corresponding to the real-time data, and the deviation between the initial control data and the real-time data is compensated by using the deviation data to obtain the target control data, thereby improving the accuracy of the target control data, meeting the user's requirements for output accuracy, and further meeting the prediction, diagnosis, and full-life cycle management through the simulation results of digital twin.

[0098] Next, a high-precision digital twin device for LFO-FCAL data model deviation compensation based on the geometric behavior model proposed in the embodiments of the present invention is described with reference to the accompanying drawings.

[0099] Figure 5 FIG. is a schematic structural diagram of a high-precision digital twin device for LFO-FCAL data model deviation compensation based on the geometric behavior model according to an embodiment of the present invention.

[0100] As Figure 5 shown, the high-precision digital twin device 10 for LFO-FCAL data model deviation compensation based on the geometric behavior model includes: an acquisition module 501, a first processing module 502, a second processing module 503, and a compensation module 504, where

[0101] The acquisition module 501 is configured to acquire the real-time data of the physical entity at the current moment and the historical data at each moment within the first preset time period;

[0102] The first processing module 502 is configured to input the real-time data at the current moment into the geometric behavior model to obtain the initial control data, where the geometric behavior model is a digital twin model of the physical entity;

[0103] The second processing module 503 is configured to input the historical data at each moment within the first preset time period into the target LFO-FCAL model to obtain the deviation data corresponding to the real-time data;

[0104] A compensation module 504, configured to compensate the initial control data by using the deviation data to obtain target control data for the digital twin.

[0105] Further, the above-mentioned target LFO-FCAL model includes an input layer, a fully connected layer, an attention layer, an LSTM layer, and an output layer; the above-mentioned second processing module 503 is specifically configured to:

[0106] The historical data at each moment within the first preset time period is transmitted to the fully connected layer through the input layer;

[0107] The fully connected layer expands the dimension of the historical data at each moment to obtain a corresponding first feature matrix;

[0108] The attention layer processes the first feature matrix to obtain a second feature matrix;

[0109] The LSTM layer makes a prediction based on the second feature matrix to obtain a third feature matrix;

[0110] The output layer obtains the deviation data corresponding to the real-time data based on the third feature matrix.

[0111] Further, the above-mentioned second processing module 503 is further configured to:

[0112] The attention layer performs a matrix transformation on the first feature matrix to obtain a query matrix, a key matrix, and a value matrix;

[0113] The query matrix and the key matrix perform matrix multiplication to obtain an attention score matrix;

[0114] The attention score matrix is divided by a preset value and then normalized to obtain an attention weight matrix;

[0115] The attention weight matrix and the value matrix are summed to obtain a second feature matrix.

[0116] Further, the above-mentioned device is further configured to:

[0117] Obtain the output data of the physical entity;

[0118] Compare the target control data and the output data at the same moment based on a preset time interval to obtain a comparison result;

[0119] If the comparison results at a continuous preset number of moments exceed a preset threshold, the target LFO-FCAL model is updated online to obtain an updated LFO-FCAL model;

[0120] Use the updated LFO-FCAL model to obtain the deviation data corresponding to the real-time data.

[0121] Further, the above-mentioned device is further configured to:

[0122] Obtain the first historical data set within the second preset time period of the geometric behavior model;

[0123] Obtain the second historical data set within the second preset time period of the physical entity;

[0124] Based on the first historical data set and the second historical data set, online update of the target LFO-FCAL model is realized by freezing network layers, and the updated LFO-FCAL model is obtained.

[0125] According to the high-precision digital twin device for LFO-FCAL data model deviation compensation based on the geometric behavior model proposed in the embodiment of the present invention, the historical data at each moment within the first preset time period is input into the target LFO-FCAL model to obtain the deviation data corresponding to the real-time data, and the deviation between the initial control data and the real-time data is compensated by using the deviation data to obtain the target control data, thereby improving the accuracy of the target control data, meeting the user's requirement for output accuracy, and further meeting the prediction, diagnosis, and full-life cycle management through the simulation results of digital twin.

[0126] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, "a plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0127] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0128] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A high-precision digital twin method for LFO-FCAL data model deviation compensation based on a geometric behavior model, characterized in that: The method comprises: Acquire the real-time data of the physical entity at the current moment and the historical data of each moment in the first preset time period; Inputting the real-time data at the current moment into a geometric behavior model to obtain initial control data, wherein the geometric behavior model is a digital twin model of the physical entity; Inputting the historical data at each moment in the first preset time period into the target LFO-FCAL model to obtain deviation data corresponding to the real-time data; The initial control data is compensated using the deviation data to obtain target control data for digital twins.

2. The method according to claim 1, characterized in that The target LFO-FCAL model includes an input layer, a fully connected layer, an attention layer, an LSTM layer and an output layer; the historical data at each moment in the first preset time period is input into the target LFO-FCAL model to obtain the deviation data corresponding to the real-time data, including: The historical data at each moment in the first preset time period is transmitted to the fully connected layer through the input layer; The fully connected layer expands the dimension of the historical data at each moment to obtain a corresponding first feature matrix; The attention layer processes the first feature matrix to obtain a second feature matrix; The LSTM layer performs prediction based on the second feature matrix to obtain a third feature matrix; The output layer obtains deviation data corresponding to the real-time data based on the third characteristic matrix.

3. The method according to claim 2, characterized in that The attention layer processes the first feature matrix to obtain a second feature matrix, including: The attention layer performs a matrix transformation on the first feature matrix to obtain a query matrix, a key matrix, and a value matrix; Perform matrix multiplication on the query matrix and the key matrix to obtain an attention score matrix; The attention score matrix is ​​divided by a preset value and then normalized to obtain an attention weight matrix; The attention weight matrix is ​​summed with the value matrix to obtain a second feature matrix.

4. The method according to claim 1, characterized in that: The method further comprises: Obtaining output data of the physical entity; Comparing the target control data with the output data at the same time based on a preset time interval to obtain a comparison result; If the comparison result of a preset value of consecutive moments exceeds a preset threshold, the target LFO-FCAL model is updated online to obtain an updated LFO-FCAL model; The updated LFO-FCAL model is used to obtain deviation data corresponding to the real-time data.

5. The method according to claim 4, characterized in that The online updating of the target LFO-FCAL model to obtain an updated LFO-FCAL model includes: Acquire a first historical data set within a second preset time period of the geometric behavior model; Acquire a second historical data set within a second preset time period of the physical entity; Based on the first historical data set and the second historical data set, the target LFO-FCAL model is updated online by freezing the network layer to obtain an updated LFO-FCAL model.

6. A high-precision digital twin device for LFO-FCAL data model deviation compensation based on a geometric behavior model, characterized in that: The device comprises: An acquisition module, used to acquire the real-time data of the physical entity at the current moment and the historical data of each moment in a first preset time period; A first processing module is used to input the real-time data at the current moment into a geometric behavior model to obtain initial control data, wherein the geometric behavior model is a digital twin model of the physical entity; A second processing module, used for inputting the historical data at each moment in the first preset time period into the target LFO-FCAL model to obtain deviation data corresponding to the real-time data; A compensation module is used to compensate the initial control data using the deviation data to obtain target control data for digital twins.

7. The device according to claim 6, characterized in that The device is also used for: Obtaining output data of the physical entity; Comparing the target control data with the output data at the same time based on a preset time interval to obtain a comparison result; If the comparison result of a preset value of consecutive moments exceeds a preset threshold, the target LFO-FCAL model is updated online to obtain an updated LFO-FCAL model; The updated LFO-FCAL model is used to obtain deviation data corresponding to the real-time data.

8. The device according to claim 7, characterized in that The device is also used for: Acquire a first historical data set within a second preset time period of the geometric behavior model; Acquire a second historical data set within a second preset time period of the physical entity; Based on the first historical data set and the second historical data set, the target LFO-FCAL model is updated online by freezing the network layer to obtain an updated LFO-FCAL model.

9. An electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 5.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images