A pipeline multi-fidelity digital twin method, system, device, product, and medium

Abstract

The application relates to the technical field of pipeline safety, and provides a pipeline multi-fidelity digital twin method, system, device, product and medium, the method comprising the following steps: building a full-size pipeline test platform, performing a pipeline load test on the pipeline through the full-size pipeline test platform to obtain high-fidelity data; constructing a finite element simulation network to obtain original low-fidelity data through the finite element simulation network; obtaining first low-fidelity data; constructing a first regression tree set to obtain second low-fidelity data; obtaining pretreated second low-fidelity data through the second low-fidelity data, constructing an initial multi-fidelity digital twin network, training the pretreated second low-fidelity data through the initial multi-fidelity digital twin network to obtain third low-fidelity data, and iteratively training the initial multi-fidelity digital twin network to complete the training of the initial multi-fidelity digital twin network. Through training of the multi-fidelity digital twin network, the precision of simulation data of the pipeline is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pipeline safety, in particular to a pipeline multi-fidelity digital twin method, system, device, product and medium. BACKGROUND

[0002] One of the key links of offshore oil and gas resource development is the oil and gas transportation system mainly composed of thin-walled pipelines. However, the complex and harsh environment of deep sea makes the submarine pipeline prone to local failure due to external water pressure, collision, vibration and its own ovality defects, etc., which may further lead to serious production safety accidents. Therefore, it is necessary to accurately measure and simulate the mechanical state of the submarine pipeline.

[0003] The current mainstream measurement and simulation methods include physical sensor monitoring and numerical simulation analysis. Among them, the physical sensor monitoring obtains real-time monitoring of key parts by arranging sensors at local positions of the physical structure. Although the first-hand monitoring data obtained by the sensor has high fidelity, the monitoring information obtained by this method cannot reflect the full-field mechanical response of the pipeline due to the density and cost of sensor arrangement. At the same time, the sensor has the risk of failure in complex and harsh environments, which greatly reduces the stability and reliability of the monitoring information. In contrast, the numerical model simulation technology, although with lower cost, can obtain the global mechanical response of the structure under the corresponding working condition. However, due to the idealized assumptions and simplifications of the structure in modeling and analysis of the numerical simulation technology mainly based on finite element analysis, the analysis results have low fidelity. SUMMARY

[0004] The present application aims to at least solve one of the technical problems in the related art. To this end, the present application provides a pipeline multi-fidelity digital twin method, system, device, product and medium, which realizes the establishment and training of a multi-fidelity digital twin network of a deep-sea pipeline.

[0005] The present application provides a pipeline multi-fidelity digital twin method, comprising:

[0006] S1: building a full-size pipeline test platform, performing pipeline load test on the pipeline through the full-size pipeline test platform, and obtaining high-fidelity data;

[0007] S2: constructing a finite element simulation network, constructing a pipeline model, simulating the pipeline model through the finite element simulation network, and obtaining original low-fidelity data;

[0008] S3: clustering the original low-fidelity data through a hierarchical clustering method, and obtaining first low-fidelity data;

[0009] S4: constructing a first regression tree set, obtaining a regression tree residual by the first low-fidelity data, substituting the regression tree residual into the first regression tree set to obtain an initial second regression tree set, constructing a loss function, constructing an intermediate regression tree by the loss function, merging the intermediate regression tree into the initial second regression tree set and iterating the initial second regression tree set to obtain a target second regression tree set, and predicting the first low-fidelity data by the target second regression tree set to obtain second low-fidelity data;

[0010] S5: preprocessing the second low-fidelity data to obtain preprocessed second low-fidelity data, constructing an initial multi-fidelity digital twin network, training the preprocessed second low-fidelity data by the initial multi-fidelity digital twin network to obtain third low-fidelity data, constructing a minimax game training network, and iterating the initial multi-fidelity digital twin network by the minimax game training network, the high-fidelity data and the third low-fidelity data to obtain a target multi-fidelity digital twin network.

[0011] According to the pipeline multi-fidelity digital twin method provided by the application, step S4 further comprises:

[0012] S41: constructing a first regression tree set, determining an initial prediction, an actual target and a preset regression tree number by the first low-fidelity data, calculating the residual of the initial prediction and the actual target to obtain a regression tree residual, and substituting the regression tree residual into the first regression tree set to obtain an initial second regression tree set;

[0013] S42: constructing a loss function, and calculating the sample residual of the initial second regression tree set by the loss function;

[0014] S43: constructing an intermediate regression tree by the sample residual, merging the intermediate regression tree into the initial second regression tree set and iterating the initial second regression tree set until the number of regression trees in the initial second regression tree set reaches the preset regression tree number, and taking the iterated initial second regression tree set as the target second regression tree set;

[0015] S44: substituting the first low-fidelity data into the target second regression tree set, and predicting the first low-fidelity data by the target second regression tree set to obtain second low-fidelity data.

[0016] According to the pipeline multi-fidelity digital twin method provided by the application, step S5 further comprises:

[0017] S51: preprocessing the second low-fidelity data by a stacking algorithm and K-fold cross-validation to obtain preprocessed second low-fidelity data;

[0018] S52: constructing a feedforward neural network, and taking the feedforward neural network as an initial multi-fidelity digital twin network;

[0019] S53: the pre-processed second low-fidelity data is substituted into the initial multi-fidelity digital twin network for training to obtain third low-fidelity data;

[0020] S54: a minimax game training network is constructed, the difference between the third low-fidelity data and the high-fidelity data is calculated, and the initial multi-fidelity digital twin network is iterated through the minimax game training network until the difference between the third low-fidelity data and the high-fidelity data converges, and the initial multi-fidelity digital twin network after iteration is taken as the target multi-fidelity digital twin network.

[0021] According to the pipeline multi-fidelity digital twin method provided by the application, step S3 further comprises:

[0022] S31: filling the internal hollow area of the pipeline model as a solid, obtaining nodes in the pipeline model, and obtaining pipeline layers by layering the pipeline model;

[0023] S32: obtaining the original low-fidelity data of the nodes in the same pipeline layer as a layer node data set, and clustering the layer node data set layer by layer through geodesic distance to obtain the first low-fidelity data.

[0024] The application further provides a pipeline multi-fidelity digital twin system, comprising:

[0025] The high-fidelity data acquisition module is used for building a full-size pipeline test platform, performing pipeline load test on the pipeline through the full-size pipeline test platform, and obtaining high-fidelity data.

[0026] The original low-fidelity data acquisition module is used for constructing a finite element simulation network, constructing a pipeline model, simulating the pipeline model through the finite element simulation network, and obtaining original low-fidelity data.

[0027] The first low-fidelity data acquisition module is used for clustering the original low-fidelity data through a hierarchical clustering method to obtain first low-fidelity data.

[0028] The second low-fidelity data acquisition module is used for constructing a first regression tree set, obtaining a regression tree residual error through the first low-fidelity data, substituting the regression tree residual error into the first regression tree set to obtain an initial second regression tree set, constructing a loss function, constructing an intermediate regression tree through the loss function, merging the intermediate regression tree into the initial second regression tree set and iterating the initial second regression tree set to obtain a target second regression tree set, and predicting the first low-fidelity data through the target second regression tree set to obtain second low-fidelity data.

[0029] The multi-fidelity digital twin network training module is configured to preprocess the second low-fidelity data to obtain preprocessed second low-fidelity data, construct an initial multi-fidelity digital twin network, train the preprocessed second low-fidelity data through the initial multi-fidelity digital twin network to obtain third low-fidelity data, construct a minimax game training network, and iteratively train the initial multi-fidelity digital twin network through the minimax game training network, the high-fidelity data and the third low-fidelity data to obtain a target multi-fidelity digital twin network.

[0030] The application further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the pipeline multi-fidelity digital twin method according to any one of the above when executing the computer program.

[0031] The application further provides a non-transitory computer-readable storage medium having a computer program stored thereon, and the computer program implements the steps of the pipeline multi-fidelity digital twin method according to any one of the above when executed by a processor.

[0032] The application further provides a computer program product, including a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions executable by a computer to perform the steps of the pipeline multi-fidelity digital twin method according to any one of the above.

[0033] The one or more technical solutions described above in the embodiments of the application have at least one of the following technical effects:

[0034] The pipeline multi-fidelity digital twin method, system, device, product and medium provided by the application can complete the training of a multi-fidelity digital twin network through digital twin technology. The trained multi-fidelity digital twin network can process simulation data of a pipeline structure and convert low-fidelity data obtained by simulation into data with higher prediction accuracy, thereby solving the problem of low fidelity of simulation data of the pipeline and inconsistency with actual conditions.

[0035] The application uses regression tree iteration and a multi-fidelity digital twin network to fuse low-fidelity data obtained by numerical simulation and high-fidelity data obtained by full-scale testing, trains a multi-fidelity digital twin network, and improves the accuracy of structure response prediction.

[0036] Additional aspects and advantages of the application will be made apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to make the technical solutions in the present application or the prior art clearer, the accompanying drawings needed in the embodiments or the prior art description will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present application, and other accompanying drawings can be obtained by those skilled in the art without any creative effort on the basis of these accompanying drawings.

[0038] Figure 1 is a flowchart of a pipeline multi-fidelity digital twin method provided by the present application.

[0039] Figure 2 is a schematic diagram of the installation method of the pipeline in the full-size pipeline test platform of the pipeline multi-fidelity digital twin method provided by the present application.

[0040] Figure 3 is a schematic diagram of the installation method of the pipeline in the full-size pipeline test platform of the pipeline multi-fidelity digital twin method provided by the present application.

[0041] Figure 4 is a schematic diagram of a pipeline model of the pipeline multi-fidelity digital twin method provided by the present application.

[0042] Figure 5 is a schematic diagram of the clustering result of the layer-by-layer node data set of the pipeline multi-fidelity digital twin method provided by the present application.

[0043] Figure 6 is a structural schematic diagram of a pipeline multi-fidelity digital twin system provided by the present application.

[0044] Figure 7 is a structural schematic diagram of a pipeline multi-fidelity digital twin device provided by the present application.

[0045] The accompanying drawings are as follows: 1, test pipe; 11, tail end cover; 12, pressure liquid; 13, vibration clamp; 2, pipeline; 100, high-fidelity data acquisition module; 200, original low-fidelity data acquisition module; 300, first low-fidelity data acquisition module; 400, second low-fidelity data acquisition module; 500, multi-fidelity digital twin network training module; 810, processor; 820, communication interface; 830, memory; 840, communication bus. DETAILED DESCRIPTION

[0046] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be clearly and completely described below. Obviously, the described embodiments are only a part of, but not all of the embodiments of the present application. Based upon the embodiments in the present application, all of the other embodiments obtained by those ordinarily skilled in the art without creative efforts should fall into the protective scope of the present application. The following embodiments are used to illustrate the present application, but should not be used to limit the scope of the present application.

[0047] In addition, the terms "first", "second", "third" are only used for descriptive purpose, and should not be understood as indicating or implying relative importance.

[0048] In the description of the embodiments of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "connected", "connected" should be understood in a broad sense, for example, can be fixedly connected, or can be detachably connected, or integrally connected; can be mechanically connected, or can be electrically connected; can be directly connected, or can be indirectly connected through an intermediate medium. For those ordinarily skilled in the art, the specific meanings of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0049] In the description of the present application, the description of the terms "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 application. In the present application, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.

[0050] The embodiments of the present application will be described below in combination with Figure 1 The embodiments of the present application are described.

[0051] Figure 1For a schematic diagram of the process of a pipeline multi-fidelity digital twin method, the preferred embodiment of the present application, the specific implementation process is divided into 5 steps, wherein S1 is the acquisition of high-fidelity data, including full-size deep-sea pipeline test, sensor data acquisition and processing, and finally obtaining high-fidelity data. S2 to S4 are the acquisition of low-fidelity data, including building a finite element simulation network for finite element analysis, clustering the finite element result data, and obtaining the second low-fidelity data through regression tree iteration, S5 is to build a multi-fidelity digital twin network, including preprocessing the second low-fidelity data through K-fold cross-validation and stacking algorithm ensemble learning, building a multi-fidelity digital twin network, obtaining third low-fidelity data, and completing training through third low-fidelity data and high-fidelity data.

[0052] The present application provides a pipeline multi-fidelity digital twin method, comprising

[0053] S1: Build a full-size pipeline test platform, and perform pipeline load test on the pipeline through the full-size pipeline test platform to obtain high-fidelity data;

[0054] Further, the purpose of this stage is to build a full-size pipeline test platform, and perform pipeline load test on the pipeline through the full-size pipeline test platform, and collect data from the measurement points of the pipeline during the pipeline load test to obtain high-fidelity data.

[0055] Firstly, a full-size pipeline test platform is built, and the full-size pipeline test platform built in the present embodiment has a total length of 11.5m, an inner diameter of 1.25m, and a design pressure capacity of 120MPa, can accommodate full-size test pipe fittings with a length of 8m or less, and has the loading functions of pipeline external water pressure, axial force, torque, bending moment and vibration. The full-size pipeline test platform comprises a sensor system composed of strain gauges, acceleration sensors, wires and water-tight cabin interfaces, and further comprises a data acquisition system composed of data acquisition instruments, data lines, computers and data acquisition software. In addition, the full-size pipeline test platform further comprises a test support system, an automatic pressurization system and a cabin cover automatic opening and closing system, and can realize automatic operation and remote control of the test process.

[0056] Subsequently, the pipeline 2 is obtained and installed in the full-size pipeline test platform, Figure 2 for the overall schematic diagram of the installation mode of the pipeline 2 in the full-size pipeline test platform, Figure 3Figure 1 is a schematic diagram of a cross section of a pipe 2 in a full-scale pipe test platform. In this embodiment, the pipe 2 is a full-scale X65 seamless thin-walled pipe with a length of 7755 mm, a diameter of 406 mm, and a wall thickness of 10 mm. The geometric parameters of the pipe 2 are measured and the pipe 2 is polished. The positions of a plurality of measurement points on the pipe 2 and the performance parameters of the pipe 2 are determined. The strain gauges and acceleration sensors in the sensor system are glued to the measurement points on the pipe 2 and the wires are welded. The pipe 2 is clamped with a vibration clamp 13, and a tail end cover 11 is installed at the tail of the pipe 2. The pipe 2 is fixed in a test pipe 1 through the vibration clamp 13 and the tail end cover 11, wherein the test pipe 1 is fixed on the full-scale pipe test platform. Then, the pressure liquid 12 is injected into the test pipe 1, and a pre-test is performed. After the pre-test, the test parameters of the full-scale pipe test platform are determined according to the performance parameters. The pipe load test is performed on the pipe 2 according to the test parameters to obtain high-fidelity data until the pipe 2 is crushed and fails. The sensor system and the data acquisition system can continuously obtain high-fidelity data at the measurement points during the pipe load test.

[0057] S2: constructing a finite element simulation network, constructing a pipe model, simulating the pipe model through the finite element simulation network, and obtaining original low-fidelity data;

[0058] Further, in this stage, the finite element simulation network is constructed, and the pipe model is constructed, wherein Figure 4 Figure 2 is a schematic diagram of a pipe model. According to the load and geometric symmetry, a 1 / 2 symmetric model of the pipe and the fluid in the cabin is established as the pipe model. The pipe model is divided by 8-node non-conforming elements C3D8I. The fixed structure attached to the pipe model is set as a symmetric rigid constraint. For the finite element simulation network, the liquid environment in which the pipe model is located is established, and Euler calculation domain is used for simulation. The normal zero flow velocity boundary is used on the symmetric surface of the fluid, the constant pressure boundary condition is used on the outer surface of the fluid to simulate the deep sea constant pressure condition, and the general contact algorithm is used to simulate the interaction of Euler bodies. The Euler calculation domain is meshed by 8-node linear simplified integral elements EC3D8R.

[0059] The strain, stress, displacement, and other data of the pipe model are obtained by finite element simulation of the pipe model through the finite element simulation network. The strain, stress, displacement, and other data of the pipe model are extracted, exported, and arranged, thereby obtaining the original low-fidelity data.

[0060] S3: clustering the original low-fidelity data by a hierarchical clustering method to obtain first low-fidelity data;

[0061] Further, the purpose of this stage is to cluster the original low-fidelity data layer by layer to obtain first low-fidelity data, so as to reduce the data volume of the original low-fidelity data, thereby reducing the time and computing resources consumed in the subsequent process, and solving the problem that the geodesic distance of the internal nodes of the model is difficult to calculate, and avoiding the phenomenon of local optimization, that is, the clustering result is more biased towards local optimization rather than global optimization, and also avoiding the clustering center being located outside the model range.

[0062] The step S3 further comprises:

[0063] S31: filling the internal hollow region of the pipeline model as a solid, obtaining nodes in the pipeline model, and layering the pipeline model to obtain pipeline layers;

[0064] S32: obtaining the original low-fidelity data of the nodes in the same pipeline layer as a layer node data set, and clustering the layer node data set layer by layer through the geodesic distance to obtain the first low-fidelity data.

[0065] For the above steps, the specific implementation in this embodiment is as follows:

[0066] First, the internal hollow region of the pipeline model is filled as a solid, then nodes in the pipeline model are obtained, and the original low-fidelity data on the nodes is extracted; the pipeline model is layered from outside to inside according to the modeling accuracy of the pipeline model to obtain pipeline layers, and the original low-fidelity data of the nodes in the same pipeline layer is taken as a layer node data set. Finally, the layer node data set is clustered layer by layer from outside to inside through the geodesic distance, until all nodes of the pipeline model are clustered, to obtain the first low-fidelity data. The geodesic distance takes the shortest distance between two points along the surface path of the pipeline model as the distance between nodes. Figure 5 The result of clustering the layer node data set layer by layer is shown in the schematic diagram.

[0067] S4: constructing a first regression tree set, obtaining a regression tree residual error through the first low-fidelity data, substituting the regression tree residual error into the first regression tree set to obtain an initial second regression tree set, constructing a loss function, constructing an intermediate regression tree through the loss function, merging the intermediate regression tree into the initial second regression tree set and iterating the initial second regression tree set to obtain a target second regression tree set, and predicting the first low-fidelity data through the target second regression tree set to obtain second low-fidelity data;

[0068] Further, the purpose of this stage is to obtain second low-fidelity data. First, a first regression tree set is constructed, an initial second regression tree set is constructed by the first low-fidelity data and the first regression tree set, then an intermediate regression tree is constructed by the sample residual of the child node of the initial second regression tree set, the intermediate regression tree is continuously merged into the initial second regression tree set and the initial second regression tree set is iterated to obtain a target second regression tree set, and finally the first low-fidelity data is converted into second low-fidelity data by using the target second regression tree set.

[0069] The step S4 further comprises:

[0070] S41: Constructing a first regression tree set, determining an initial prediction, an actual target and a preset regression tree number by the first low-fidelity data, calculating the residual of the initial prediction and the actual target to obtain a regression tree residual, and substituting the regression tree residual into the first regression tree set to obtain an initial second regression tree set;

[0071] S42: Constructing a loss function, and calculating the sample residual of the initial second regression tree set by the loss function;

[0072] S43: Constructing an intermediate regression tree by the sample residual, merging the intermediate regression tree into the initial second regression tree set and iterating the initial second regression tree set until the number of regression trees in the initial second regression tree set reaches the preset regression tree number, and taking the iterated initial second regression tree set as the target second regression tree set;

[0073] S44: Substituting the first low-fidelity data into the target second regression tree set, and predicting the first low-fidelity data by the target second regression tree set to obtain second low-fidelity data.

[0074] For the above steps, the specific implementation in this embodiment is as follows:

[0075] First, the first regression tree set is constructed by fitting the regression target function through the second-order Taylor expansion and the regression tree regularization term, then the initial prediction of the first low-fidelity data and the actual target are calculated, the preset regression tree number is determined according to the first low-fidelity data, the difference between the initial prediction and the actual target can be obtained to obtain the regression tree residual, then the regression tree residual is substituted into each child node of the first regression tree set to obtain the initial second regression tree set, then the loss function is constructed, and the sample residual of the initial second regression tree set is calculated by the loss function.

[0076] Further, a blank intermediate regression tree is established and the sample residual error is substituted into each sub-node of the intermediate regression tree to construct an intermediate regression tree, and the intermediate regression tree is merged into the initial second regression tree set; the process of “establishing a blank intermediate regression tree and substituting a sample residual error into each sub-node of the intermediate regression tree to construct an intermediate regression tree, and merging the intermediate regression tree into the initial second regression tree set” is repeatedly executed and iterated until the number of regression trees in the initial second regression tree set reaches the preset number of regression trees, and the initial second regression tree set after iteration is taken as a target second regression tree set. Finally, the first low-fidelity data is substituted into the target second regression tree set for prediction to obtain second low-fidelity data.

[0077] Further, the first low-fidelity data can also be divided into a training set and a test set. In this embodiment, the training set accounts for 80% of the first low-fidelity data, and the test set accounts for 20%. The training set is used to replace the first low-fidelity data to obtain the second low-fidelity data, and the difference between the test set and the second low-fidelity data is calculated to verify the effectiveness of the second low-fidelity data. When the difference between the second low-fidelity data and the test set is less than a threshold value, it is considered that the second low-fidelity data is effective.

[0078] S5: preprocessing the second low-fidelity data to obtain preprocessed second low-fidelity data, constructing an initial multi-fidelity digital twin network, training the initial multi-fidelity digital twin network with the preprocessed second low-fidelity data to obtain third low-fidelity data, constructing a minimax game training network, and iterating the initial multi-fidelity digital twin network with the minimax game training network, high-fidelity data and third low-fidelity data to obtain a target multi-fidelity digital twin network.

[0079] Further, in order to obtain the target multi-fidelity digital twin network, after preprocessing the second low-fidelity data to obtain preprocessed second low-fidelity data, the third low-fidelity data is obtained by the initial multi-fidelity digital twin network and the preprocessed second low-fidelity data, and the difference between the third low-fidelity data and the high-fidelity data is calculated. At the same time, a minimax game training network is constructed, and the initial multi-fidelity digital twin network is iterated by using the minimax game training network until the difference between the third low-fidelity data and the high-fidelity data converges, the training of the multi-fidelity digital twin network is completed, and the target multi-fidelity digital twin network is obtained.

[0080] Further, the step S5 further comprises:

[0081] S51: preprocessing the second low-fidelity data by a stacking algorithm and K-fold cross-validation to obtain preprocessed second low-fidelity data;

[0082] S52: Construct a feedforward neural network, and take the feedforward neural network as an initial multi-fidelity digital twin network;

[0083] S53: Substitute the preprocessed second low-fidelity data into the initial multi-fidelity digital twin network for training to obtain third low-fidelity data;

[0084] S54: Construct a minimax game training network, calculate the difference between the third low-fidelity data and high-fidelity data, and iterate the initial multi-fidelity digital twin network through the minimax game training network until the difference between the third low-fidelity data and high-fidelity data converges, and take the iterated initial multi-fidelity digital twin network as a target multi-fidelity digital twin network.

[0085] For the above steps, the specific implementation in the embodiment is as follows:

[0086] First, the second low-fidelity data is preprocessed through a stacking algorithm and K-fold cross-validation, and the preprocessing process is divided into two steps, that is, the second low-fidelity data is divided by K-fold cross-validation, and then the second low-fidelity data after data division is integrated by a stacking algorithm for learning, and the preprocessed second low-fidelity data is taken as preprocessed second low-fidelity data. Then a feedforward neural network is constructed, and the feedforward neural network is taken as an initial multi-fidelity digital twin network, and then the preprocessed second low-fidelity data is substituted into the initial multi-fidelity digital twin network for training to obtain third low-fidelity data;

[0087] Further, a minimax game training network is constructed, the third low-fidelity data and high-fidelity data are made to game through the minimax game training network, the initial multi-fidelity digital twin network is iterated according to the game result, and the difference between the third low-fidelity data and high-fidelity data is calculated until the difference between the third low-fidelity data and high-fidelity data is less than a threshold to achieve convergence, that is, the training of the multi-fidelity digital twin network is completed, and the iterated initial multi-fidelity digital twin network is taken as a target multi-fidelity digital twin network.

[0088] The present application also verifies the effectiveness of a pipeline multi-fidelity digital twin method, and in the verification process, the global accuracy of the multi-fidelity digital twin network is evaluated by a root mean square error RRMSE, and the local accuracy of the multi-fidelity digital twin network is evaluated by a maximum relative error RE max The expressions of the root mean square error RRMSE and the maximum relative error RE max are as follows:

[0089]

[0090]

[0091] wherein, a quantity of high-fidelity data, a value of high-fidelity data at a first measurement point, a value of third low-fidelity data at a first measurement point obtained by the target multi-fidelity digital twin network, a mean value of high-fidelity data, a value of high-fidelity data at a most dangerous test point, a value of third low-fidelity data at the most dangerous test point, wherein the most dangerous test point is a measurement point corresponding to a maximum value of a Mises stress value in the high-fidelity data. Table 1 shows a comparison of root mean square errors RRMSE, maximum relative errors RE max and training time of the present application and other proxy models, wherein the multi-fidelity digital twin network is the method proposed in the present application:

[0092] Table 1 Comparison of prediction accuracy and training time of proxy models of different methods

[0093]

[0094] It can be seen that, compared with the proxy model based on single-fidelity data and other multi-fidelity models, the multi-fidelity digital twin network has higher global accuracy and local accuracy while having a reasonable training time.

[0095] A pipeline multi-fidelity digital twin system provided by the present application is described below, and the pipeline multi-fidelity digital twin system described below can be mutually corresponding with reference to the pipeline multi-fidelity digital twin method described above.

[0096] Figure 6 An example of a structural schematic diagram of a pipeline multi-fidelity digital twin system is shown in Figure 6 for executing the pipeline multi-fidelity digital twin method as described above, comprising:

[0097] The high-fidelity data acquisition module 100 is configured to build a full-size pipeline test platform, perform a pipeline load test on the pipeline through the full-size pipeline test platform, and obtain high-fidelity data.

[0098] The original low-fidelity data acquisition module 200 is configured to build a finite element simulation network, build a pipeline model, simulate the pipeline model through the finite element simulation network, and obtain original low-fidelity data.

[0099] The first low-fidelity data acquisition module 300 is configured to cluster the original low-fidelity data through a hierarchical clustering method, and obtain first low-fidelity data.

[0100] The second low-fidelity data acquisition module 400 is used to construct a first set of regression trees, obtain regression tree residuals from the first low-fidelity data, substitute the regression tree residuals into the first set of regression trees to obtain an initial second set of regression trees, construct a loss function, construct intermediate regression trees through the loss function, merge the intermediate regression trees into the initial second set of regression trees and iterate the initial second set of regression trees to obtain a target second set of regression trees, and use the target second set of regression trees to predict the first low-fidelity data to obtain the second low-fidelity data.

[0101] Multi-fidelity digital twin network training module 500: This module is used to preprocess the second low-fidelity data to obtain preprocessed second low-fidelity data, construct an initial multi-fidelity digital twin network, train the preprocessed second low-fidelity data using the initial multi-fidelity digital twin network to obtain third low-fidelity data, construct a minimax game training network, and iterate the initial multi-fidelity digital twin network using the minimax game training network, high-fidelity data, and third low-fidelity data to obtain the target multi-fidelity digital twin network.

[0102] on the other hand, Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a pipelined multi-fidelity digital twin method, which includes:

[0103] S1: Build a full-size pipeline test platform and conduct pipeline load tests on the pipeline through the full-size pipeline test platform to obtain high-fidelity data;

[0104] S2: Construct a finite element simulation network, construct a pipeline model, and simulate the pipeline model through the finite element simulation network to obtain the original low-fidelity data;

[0105] S3: Cluster the original low-fidelity data using hierarchical clustering to obtain the first low-fidelity data;

[0106] S4: Construct a first set of regression trees, obtain regression tree residuals from the first low-fidelity data, substitute the regression tree residuals into the first set of regression trees to obtain an initial second set of regression trees, construct a loss function, construct intermediate regression trees using the loss function, merge the intermediate regression trees into the initial second set of regression trees and iterate over the initial second set of regression trees to obtain a target second set of regression trees, and use the target second set of regression trees to predict the first low-fidelity data to obtain the second low-fidelity data;

[0107] S5: preprocessing the second low-fidelity data to obtain preprocessed second low-fidelity data, constructing an initial multi-fidelity digital twin network, training the preprocessed second low-fidelity data through the initial multi-fidelity digital twin network to obtain third low-fidelity data, constructing a minimax game training network, and iteratively training the initial multi-fidelity digital twin network through the minimax game training network, the high-fidelity data and the third low-fidelity data to obtain a target multi-fidelity digital twin network.

[0108] In addition, the logic instructions in the memory 830 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in the embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various storage program codes.

[0109] On the other hand, the present application also provides a computer program product, which comprises a computer program stored on a non-transitory computer readable storage medium, and the computer program comprises program instructions, when the program instructions are executed by a computer, the computer can execute a pipeline multi-fidelity digital twin method provided by the above-mentioned method, the method comprises:

[0110] S1: building a full-size pipeline test platform, performing pipeline load test on the pipeline through the full-size pipeline test platform, and obtaining high-fidelity data;

[0111] S2: constructing a finite element simulation network, constructing a pipeline model, simulating the pipeline model through the finite element simulation network, and obtaining original low-fidelity data;

[0112] S3: clustering the original low-fidelity data through a hierarchical clustering method to obtain first low-fidelity data;

[0113] S4: Construct a first regression tree set, obtain a regression tree residual through the first low-fidelity data, substitute the regression tree residual into the first regression tree set to obtain an initial second regression tree set, construct a loss function, construct an intermediate regression tree through the loss function, merge the intermediate regression tree into the initial second regression tree set, and iterate the initial second regression tree set to obtain a target second regression tree set, and predict the first low-fidelity data through the target second regression tree set to obtain second low-fidelity data;

[0114] S5: Preprocess the second low-fidelity data to obtain preprocessed second low-fidelity data, construct an initial multi-fidelity digital twin network, train the preprocessed second low-fidelity data through the initial multi-fidelity digital twin network to obtain third low-fidelity data, construct a minimax game training network, and iterate the initial multi-fidelity digital twin network through the minimax game training network, the high-fidelity data and the third low-fidelity data to obtain a target multi-fidelity digital twin network.

[0115] In another aspect, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement a pipeline multi-fidelity digital twin method provided by each of the above methods, the method comprising:

[0116] S1: Build a full-size pipeline test platform, perform a pipeline load test on the pipeline through the full-size pipeline test platform, and obtain high-fidelity data;

[0117] S2: Construct a finite element simulation network, construct a pipeline model, and simulate the pipeline model through the finite element simulation network to obtain original low-fidelity data;

[0118] S3: Cluster the original low-fidelity data through a hierarchical clustering method to obtain first low-fidelity data;

[0119] S4: Construct a first regression tree set, obtain a regression tree residual through the first low-fidelity data, substitute the regression tree residual into the first regression tree set to obtain an initial second regression tree set, construct a loss function, construct an intermediate regression tree through the loss function, merge the intermediate regression tree into the initial second regression tree set, and iterate the initial second regression tree set to obtain a target second regression tree set, and predict the first low-fidelity data through the target second regression tree set to obtain second low-fidelity data;

[0120] S5: preprocessing the second low-fidelity data to obtain preprocessed second low-fidelity data, constructing an initial multi-fidelity digital twin network, training the preprocessed second low-fidelity data through the initial multi-fidelity digital twin network to obtain third low-fidelity data, constructing a minimax game training network, and iteratively training the initial multi-fidelity digital twin network through the minimax game training network, the high-fidelity data, and the third low-fidelity data to obtain a target multi-fidelity digital twin network.

[0121] The system, device, product and medium embodiments described above are merely illustrative, and the units described as separate units may or may not be physically separated, and the units shown as units may or may not be physical units, i.e., they may be located in one place or distributed on multiple network units. Some or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0122] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course, can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0123] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for creating a multi-fidelity digital twin of pipelines, characterized in that, include: S1: Build a full-size pipeline test platform and conduct pipeline load tests on the pipeline through the full-size pipeline test platform to obtain high-fidelity data; S2: Construct a finite element simulation network, construct a pipeline model, and simulate the pipeline model through the finite element simulation network to obtain the original low-fidelity data; S3: Cluster the original low-fidelity data using hierarchical clustering to obtain the first low-fidelity data; S4: Construct a first set of regression trees, obtain regression tree residuals from the first low-fidelity data, substitute the regression tree residuals into the first set of regression trees to obtain an initial second set of regression trees, construct a loss function, construct intermediate regression trees using the loss function, merge the intermediate regression trees into the initial second set of regression trees and iterate over the initial second set of regression trees to obtain a target second set of regression trees, and use the target second set of regression trees to predict the first low-fidelity data to obtain the second low-fidelity data; S5: Preprocess the second low-fidelity data to obtain preprocessed second low-fidelity data, construct an initial multi-fidelity digital twin network, train the preprocessed second low-fidelity data using the initial multi-fidelity digital twin network to obtain third low-fidelity data, construct a minimax game training network, iterate the initial multi-fidelity digital twin network using the minimax game training network, high-fidelity data, and third low-fidelity data to obtain the target multi-fidelity digital twin network.

2. The pipeline multi-fidelity digital twin method according to claim 1, characterized in that, Step S4 further includes: S41: Construct the first set of regression trees. Determine the initial prediction, actual target and preset number of regression trees through the first low-fidelity data. Calculate the residuals of the initial prediction and actual target to obtain the regression tree residuals. Substitute the regression tree residuals into the first set of regression trees to obtain the initial second set of regression trees. S42: Construct a loss function and calculate the sample residuals of the initial second regression tree set using the loss function; S43: Construct intermediate regression trees through sample residuals, merge intermediate regression trees into the initial second regression tree set, and iterate the initial second regression tree set until the number of regression trees in the initial second regression tree set reaches the preset number of regression trees. Use the iterated initial second regression tree set as the target second regression tree set. S44: Substitute the first low-fidelity data into the target second regression tree set, and use the target second regression tree set to predict the first low-fidelity data to obtain the second low-fidelity data.

3. The pipeline multi-fidelity digital twin method according to claim 1, characterized in that, Step S5 further includes: S51: The second low-fidelity data is preprocessed using a stacking algorithm and K-fold cross-validation to obtain preprocessed second low-fidelity data; S52: Construct a feedforward neural network and use it as the initial multifidelity digital twin network; S53: Substitute the preprocessed second low-fidelity data into the initial multi-fidelity digital twin network for training to obtain the third low-fidelity data; S54: Construct a minimax game training network, calculate the difference between the third low-fidelity data and the high-fidelity data, and iterate the initial multifidelity digital twin network through the minimax game training network until the difference between the third low-fidelity data and the high-fidelity data converges. Use the iterated initial multifidelity digital twin network as the target multifidelity digital twin network.

4. The pipeline multi-fidelity digital twin method according to claim 1, characterized in that, Step S3 further includes: S31: Fill the internal void area of ​​the pipe model with solids, obtain nodes in the pipe model, and divide the pipe model into layers to obtain pipe layers. S32: Obtain the original low-fidelity data of the nodes in the same pipeline layer as the layer node data set, and cluster the layer node data set layer by layer by geodesic distance to obtain the first low-fidelity data.

5. A pipeline multifidelity digital twin system, used to execute a pipeline multifidelity digital twin method as described in any one of claims 1 to 4, characterized in that, include: High-fidelity data acquisition module: used to build a full-size pipeline test platform, and to conduct pipeline load tests on the pipeline through the full-size pipeline test platform to obtain high-fidelity data; Raw low-fidelity data acquisition module: used to construct a finite element simulation network, construct a pipeline model, and simulate the pipeline model through the finite element simulation network to obtain raw low-fidelity data; First low-fidelity data acquisition module: used to cluster the original low-fidelity data using hierarchical clustering method to obtain first low-fidelity data; The second low-fidelity data acquisition module is used to construct a first set of regression trees, obtain regression tree residuals from the first low-fidelity data, substitute the regression tree residuals into the first set of regression trees to obtain an initial second set of regression trees, construct a loss function, construct intermediate regression trees through the loss function, merge the intermediate regression trees into the initial second set of regression trees and iterate the initial second set of regression trees to obtain a target second set of regression trees, and use the target second set of regression trees to predict the first low-fidelity data to obtain the second low-fidelity data. The multi-fidelity digital twin network training module is used to preprocess the second low-fidelity data to obtain preprocessed second low-fidelity data, construct an initial multi-fidelity digital twin network, train the preprocessed second low-fidelity data through the initial multi-fidelity digital twin network to obtain third low-fidelity data, construct a minimax game training network, and iterate the initial multi-fidelity digital twin network through the minimax game training network, high-fidelity data, and third low-fidelity data to obtain the target multi-fidelity digital twin network.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of a pipeline multifidelity digital twin method as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a pipeline multifidelity digital twin method as described in any one of claims 1 to 4.

8. A computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, characterized in that, When the program instructions are executed by a computer, the computer is able to perform the steps of a pipeline multifidelity digital twin method as described in any one of claims 1 to 4.

Images