CN114564878A - Structural dynamics model generation method based on graph neural network - Google Patents

Structural dynamics model generation method based on graph neural network Download PDF

Info

Publication number
CN114564878A
CN114564878A CN202111674213.9A CN202111674213A CN114564878A CN 114564878 A CN114564878 A CN 114564878A CN 202111674213 A CN202111674213 A CN 202111674213A CN 114564878 A CN114564878 A CN 114564878A
Authority
CN
China
Prior art keywords
matrix
structural
neural network
response data
generating
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202111674213.9A
Other languages
Chinese (zh)
Other versions
CN114564878B (en
Inventor
王迎
张俊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Graduate School Harbin Institute of Technology
Original Assignee
Shenzhen Graduate School Harbin Institute of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Graduate School Harbin Institute of Technology filed Critical Shenzhen Graduate School Harbin Institute of Technology
Priority to CN202111674213.9A priority Critical patent/CN114564878B/en
Publication of CN114564878A publication Critical patent/CN114564878A/en
Application granted granted Critical
Publication of CN114564878B publication Critical patent/CN114564878B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a method for generating a structural dynamics model based on a graph neural network, which comprises the following steps: determining a target structure, and acquiring structure response data and topological information of the target structure; training a preset artificial neural network according to the structural response data and the topological information; and taking the trained artificial neural network as a structural dynamics model corresponding to the target structure. The structure dynamics model generated by the invention can be used for calculating (predicting) the dynamics response of the structure. According to the invention, the topological information of the target structure is introduced into the frame of the deep learning network, so that the generalization capability of the model can be improved to a certain extent while the extraction capability of the deep learning network on the physical characteristics of the target structure is not lost, and the conversion of the artificial neural network model in different topological structures is realized. The problem of the deep learning network that is used for simulating structure dynamics among the prior art generalizes the ability poor is solved.

Description

Structural dynamics model generation method based on graph neural network
Technical Field
The invention relates to the field of deep learning, in particular to a structure dynamics model generation method based on a graph neural network.
Background
The traditional structure dynamics simulation method (such as a finite element method) aims to be established based on a mathematical model, however, the mathematical model of a complex structure is difficult to obtain, and therefore, the traditional method cannot reflect the real physical characteristics of the structure. To overcome the limitations of conventional simulation methods, machine learning methods, such as SVM [1], and deep learning methods, such as ANN [2,3], CNN [4], RNN [5], LSTM [6,7], etc., have been introduced into the simulation of structural dynamics. However, the artificial intelligence model requires a large amount of measured data for training, and the generalization capability of the trained model is often poor. Therefore, the artificial intelligence algorithm based structural dynamics simulation method also has its own limitations.
Thus, there is still a need for improvement and development of the prior art.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a structure dynamics model generation method based on a graph neural network, aiming at solving the problem that the deep learning network used for simulating structure dynamics in the prior art has poor generalization capability.
The technical scheme adopted by the invention for solving the problems is as follows:
in a first aspect, an embodiment of the present invention provides a structural dynamics model generation method based on a graph neural network, where the method includes:
determining a target structure, and acquiring structure response data and topological information of the target structure;
training a preset artificial neural network according to the structural response data and the topological information;
and taking the trained artificial neural network as a structural dynamics model corresponding to the target structure.
In one embodiment, the obtaining structural response data of the target structure includes:
acquiring displacement response data, speed response data and acceleration response data respectively corresponding to a plurality of degrees of freedom on a target structure according to a preset sampling time point;
taking the displacement response data, the velocity response data, and the acceleration response data as the structural response data.
In one embodiment, the training a preset artificial neural network according to the structural response data and the topological information includes:
generating a sending matrix and a receiving matrix according to the topology information, wherein the combination of the sending matrix and the receiving matrix can reflect the interaction topology information among a plurality of degrees of freedom of the target structure;
generating a structural displacement vector according to the displacement response data corresponding to the degrees of freedom respectively, generating a structural speed vector according to the speed response data corresponding to the degrees of freedom respectively, and generating training label data according to the acceleration response data corresponding to the degrees of freedom respectively;
and training the artificial neural network according to the sending matrix, the receiving matrix, the structure displacement vector, the structure speed vector and the training label data.
In one embodiment, the training of the artificial neural network based on the transmit matrix, the receive matrix, the structure displacement vector, and the structure velocity vector and the training label data comprises:
generating a restoring force calculation matrix according to the sending matrix, the receiving matrix and the structural displacement vector, and inputting the restoring force calculation matrix into the first multilayer perceptron to obtain a restoring force prediction matrix;
generating a damping force calculation matrix according to the sending matrix, the receiving matrix and the structural velocity vector, and inputting the damping force calculation matrix into the second multilayer perceptron to obtain a damping force prediction matrix;
obtaining a resultant force calculation matrix according to the restoring force prediction matrix and the damping force prediction matrix, and inputting the resultant force calculation matrix into the third multilayer perceptron to obtain a resultant force prediction matrix, wherein the resultant force prediction matrix is used for reflecting the vector sum of the restoring force and the damping force;
acquiring external excitation load data, generating an acceleration response calculation matrix according to the receiving matrix, the resultant force prediction matrix and the external excitation load data, and inputting the acceleration response calculation matrix into the fourth multilayer perceptron to obtain acceleration response prediction data;
and training the artificial neural network according to the acceleration response prediction data and the training label data.
In one embodiment, the generating a resilience calculation matrix from the transmit matrix, the receive matrix, and the structural displacement vector includes:
multiplying the sending matrix by the structure displacement vector to obtain a first matrix;
multiplying the receiving matrix by the structure displacement vector to obtain a second matrix;
and combining the first matrix and the second matrix according to columns to obtain the restoring force calculation matrix.
In one embodiment, the generating a damping force calculation matrix from the transmit matrix, the receive matrix, and the structural velocity vector comprises:
multiplying the sending matrix by the structural velocity vector to obtain a third matrix;
multiplying the receiving matrix and the structural velocity vector to obtain a fourth matrix;
and combining the third matrix and the fourth matrix in columns to obtain the damping force calculation matrix.
In one embodiment, the obtaining a calculation matrix of a resultant force according to the restoring force prediction matrix and the damping force prediction matrix includes:
and combining the restoring force prediction matrix and the damping force prediction matrix according to columns to obtain a calculation matrix of the resultant force.
In one embodiment, the generating an acceleration response calculation matrix from the receive matrix, the total force prediction matrix, and the external excitation load data includes:
multiplying the left side of the resultant prediction matrix by the transposition of the receiving matrix to obtain a fifth matrix;
and combining the fifth matrix and the external excitation load data in columns to obtain the acceleration response calculation matrix.
In a second aspect, an embodiment of the present invention further provides a structural dynamics model generation system based on a graph neural network, where the system includes:
the data acquisition module is used for determining a target structure and acquiring structure response data and topological information of the target structure;
and the network training module is used for training a preset artificial neural network according to the structural response data and the topological information, and taking the trained artificial neural network as a structural dynamic model corresponding to the target structure.
In a third aspect, an embodiment of the present invention further provides a structural dynamics model based on a graph neural network, where the structural dynamics model based on the graph neural network is generated by using any of the above-mentioned structural dynamics model generation methods based on the graph neural network.
The invention has the beneficial effects that: the embodiment of the invention obtains the structural response data and the topological information of the target structure by determining the target structure; training a preset artificial neural network according to the structural response data and the topological information; and taking the trained artificial neural network as a structural dynamics model corresponding to the target structure. The structure dynamics model generated by the invention can be used for calculating (predicting) the dynamics response of the structure. According to the invention, the topological information of the target structure is introduced into the frame of the deep learning network, so that the generalization capability of the model can be improved to a certain extent while the extraction capability of the deep learning network on the physical characteristics of the target structure is not lost, and the conversion of the artificial neural network model in different topological structures is realized. The problem of the deep learning network that is used for simulating structure dynamics among the prior art generalizes the ability poor is solved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present invention, and it is also possible for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
Fig. 1 is a schematic flow chart of a structural dynamics model generation method based on a graph neural network according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of a spring-mass system provided by an embodiment of the present invention.
Fig. 3 is a schematic diagram of a training process of the artificial neural network according to the embodiment of the present invention.
FIG. 4 is a schematic diagram of the relationship between seismic acceleration and time provided by the embodiment of the invention.
Fig. 5 is a schematic diagram of a prediction result of a first degree of freedom in the three-degree-of-freedom system according to the embodiment of the present invention.
Fig. 6 is a schematic diagram illustrating a prediction result of a second degree of freedom in the three-degree-of-freedom system according to the embodiment of the present invention.
Fig. 7 is a schematic diagram of a prediction result of a third degree of freedom in the three-degree-of-freedom system according to the embodiment of the present invention.
Fig. 8 is a diagram illustrating a predicted result of the first degree of freedom in the four-degree-of-freedom system according to the embodiment of the present invention.
FIG. 9 is a diagram illustrating predicted results of a second degree of freedom in a four degree of freedom system according to an embodiment of the present invention.
Fig. 10 is a diagram illustrating a predicted result of the third degree of freedom in the four-degree-of-freedom system according to the embodiment of the present invention.
Fig. 11 is a diagram illustrating a predicted result of the fourth degree of freedom in the four-degree-of-freedom system according to the embodiment of the present invention.
Fig. 12 is a schematic diagram of internal modules of a structural dynamics model generation system based on a graph neural network according to an embodiment of the present invention.
Fig. 13 is a functional block diagram of a terminal according to an embodiment of the present invention.
Detailed Description
The invention discloses a structure dynamics model generation method based on a graph neural network, and in order to make the purpose, technical scheme and effect of the invention clearer and clearer, the invention is further described in detail below by referring to the attached drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. As used herein, the term "and/or" includes all or any element and all combinations of one or more of the associated listed items.
It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
The traditional structure dynamics simulation method (such as a finite element method) aims to be established based on a mathematical model, however, the mathematical model of a complex structure is difficult to obtain, and therefore, the traditional method cannot reflect the real physical characteristics of the structure. To overcome the limitations of conventional simulation methods, machine learning methods, such as SVM [1], and deep learning methods, such as ANN [2,3], CNN [4], RNN [5], LSTM [6,7], etc., have been introduced into the simulation of structural dynamics. However, the artificial intelligence model requires a large amount of measured data for training, and the generalization capability of the trained model is often poor. Therefore, the artificial intelligence algorithm based structural dynamics simulation method also has its own limitations.
In view of the above-mentioned drawbacks of the prior art, the present invention provides a structural dynamics model generation method based on a graph neural network, the method including: determining a target structure, and acquiring structure response data and topological information of the target structure; training a preset artificial neural network according to the structural response data and the topological information; and taking the trained artificial neural network as a structural dynamics model corresponding to the target structure. The invention introduces the topological information of the target structure into the training process of the deep learning network. The structure dynamics model generated by the invention can be used for calculating (predicting) the dynamics response of the structure. According to the invention, the topological information of the target structure is introduced into the frame of the deep learning network, so that the generalization capability of the model can be improved to a certain extent while the extraction capability of the deep learning network on the physical characteristics of the target structure is not lost, and the conversion of the artificial neural network model in different topological structures is realized. The problem of the deep learning network that is used for simulating structure dynamics among the prior art generalizes the ability poor is solved.
As shown in fig. 1, the method comprises the steps of:
step S100, determining a target structure, and acquiring structure response data and topology information of the target structure.
Specifically, the target structure in this embodiment may be any complex structure that requires a structure dynamics simulation process. In order to generate a structural dynamics model corresponding to a target structure, in this embodiment, structural response data and topology information of the target structure need to be acquired, where the structural response data may reflect responses of nodes of the target structure under external excitation, and the topology information may reflect physical characteristics of the target structure.
In one implementation, the step S100 specifically includes:
s101, acquiring displacement response data, speed response data and acceleration response data respectively corresponding to a plurality of degrees of freedom on a target structure according to a preset sampling time point;
and S102, taking the displacement response data, the speed response data and the acceleration response data as the structural response data.
Specifically, in order to obtain the structural response data of the target structure, the present embodiment needs to acquire the displacement, velocity, and acceleration responses of the target structure in each degree of freedom at a preset sampling time point, so as to obtain the displacement response data, the velocity response data, and the acceleration response data. These three types of data are then taken as structural response data of the target structure.
For example, taking the spring-mass system shown in fig. 2 as an example, the model belongs to a one-dimensional dynamic system, that is, each node has only one translational degree of freedom, and is often applied to a building structure in which interlayer shear is considered in calculation, and is an important theoretical basis in the calculation of earthquake-resistant design of the building structure. The structural response data for the structure shown in fig. 2 can be calculated by the following formula (1), wherein,
Figure BDA0003450393440000081
y is acceleration response, speed response and displacement response of respective degree of freedom respectively; k. c, m and f are respectively rigidity, damping, mass and external excitation; i is an element of [1, ndof]Indicating a degree of freedom number.
Figure BDA0003450393440000082
Wherein the content of the first and second substances,
Figure BDA0003450393440000091
Figure BDA0003450393440000092
Figure BDA0003450393440000093
equation (1) can also be written in the form of equation (2):
Figure BDA0003450393440000094
as shown in fig. 1, the method further comprises:
and S200, training a preset artificial neural network according to the structural response data and the topological information.
Specifically, in order to improve the generalization ability of the artificial neural network, the present embodiment needs to input the topology information of the structure into the framework of the artificial neural network. Because the topological information can reflect the physical characteristics of the target structure, the artificial neural network obtained after the training of the structural response data and the topological information can improve the generalization capability of the model and realize the conversion of the artificial neural network model in different topological structures without losing the characteristic extraction capability of the deep learning network.
In one implementation, the step S200 specifically includes:
step S201, generating a sending matrix and a receiving matrix according to the topology information, wherein the combination of the sending matrix and the receiving matrix can reflect the interaction topology information among a plurality of degrees of freedom of the target structure;
step S202, generating a structural displacement vector according to the displacement response data respectively corresponding to the degrees of freedom, generating a structural velocity vector according to the velocity response data respectively corresponding to the degrees of freedom, and generating training label data according to the acceleration response data respectively corresponding to the degrees of freedom;
step S203, training the artificial neural network according to the sending matrix, the receiving matrix, the structure displacement vector, the structure speed vector and the training label data.
Specifically, in the present embodiment, a pair of relationship matrices, i.e., a transmission matrix and a reception matrix, is generated based on the structural topology information, and the interaction topology information of all complex structures can be represented by modifying internal elements of the pair of relationship matrices, and the physical meaning represented by combining the internal elements is similar to the physical meaning of the element positions in the finite element stiffness matrix. And then generating a structure displacement vector according to the displacement response data of each degree of freedom, generating a structure speed vector according to the speed response data, and generating training label data (used for training and testing) according to the acceleration response data. And finally, generating training data of the artificial neural network according to the sending matrix, the receiving matrix, the structural displacement vector and the structural speed vector, evaluating the difference between the result output by the artificial neural network and the real result by adopting training label data, and then carrying out iterative updating on parameters in the artificial neural network by taking the difference between the result output by the artificial neural network and the real result as the guide until the difference between the result output by the artificial neural network and the real result reaches a preset training target.
For example, the dynamic response of the three-degree-of-freedom system shown in fig. 2 can be obtained by the following formula (3), wherein the expression represents that vectors are multiplied by each other in bit, and if there is no sign between the two vectors, the expression represents that the matrix multiplication is normal. Wherein, Msr and MrrA pair of relation matrixes is formed and can be used for representing interaction topology information of all complex structures, Msr is called a sending matrix, MrrReferred to as a receive matrix, the two matrices combined together represent a physical meaning similar to the physical meaning of the element positions in the finite element stiffness matrix.
Figure BDA0003450393440000111
Figure BDA0003450393440000112
Wherein, equal sign right side is composed of three parts, respectively represent the kinetic system: restoring force, damping force and external excitation.
In one implementation manner, the artificial neural network includes a first multilayer perceptron, a second multilayer perceptron, a third multilayer perceptron, and a fourth multilayer perceptron, and the step S203 specifically includes:
step S2031, generating a restoring force calculation matrix according to the sending matrix, the receiving matrix and the structural displacement vector, and inputting the restoring force calculation matrix into the first multilayer perceptron to obtain a restoring force prediction matrix;
step S2032, generating a damping force calculation matrix according to the sending matrix, the receiving matrix and the structural velocity vector, and inputting the damping force calculation matrix into the second multilayer perceptron to obtain a damping force prediction matrix;
step S2033, a resultant force calculation matrix is obtained according to the restoring force prediction matrix and the damping force prediction matrix, and the resultant force calculation matrix is input into the third multilayer perceptron to obtain a resultant force prediction matrix, wherein the resultant force prediction matrix is used for reflecting the vector sum of the restoring force and the damping force;
step S2034, obtaining external excitation load data, generating an acceleration response calculation matrix according to the receiving matrix, the resultant force prediction matrix and the external excitation load data, and inputting the acceleration response calculation matrix into the fourth multilayer perceptron to obtain acceleration response prediction data;
and S2035, training the artificial neural network according to the acceleration response prediction data and the training label data.
It is to be understood that, as shown in the above equation (3), even if the external excitation fiAre known, but the restoring and damping forces are unknown because the stiffness and damping of the structure are not measurable. In addition to this, traditionThe spring-mass system of (a) assumes that the mass of each node is proportionally distributed to each node by the total mass of the structure, which assumption lacks basis. Therefore, in the present embodiment, the above three unknown parameters are indirectly obtained by four multi-layer perceptrons (MLPs). The first multilayer perceptron is used for predicting restoring force, the second multilayer perceptron is used for predicting damping force, the third multilayer perceptron is used for predicting the resultant force of the restoring force and the damping force, and the fourth multilayer perceptron is used for predicting acceleration response. Specifically, since the transmission matrix and the reception matrix are generated based on the topology information of the target structure and can reflect the physical characteristics of the target structure, the input data of the first multi-layer sensing machine, i.e. the restoring force calculation matrix, is generated according to the transmission matrix, the reception matrix and the structure displacement vector, and the first multi-layer sensing machine outputs a restoring force prediction matrix based on the input data. Similarly, the input data of the second multi-layer sensor, namely the damping force calculation matrix, is generated according to the transmitting matrix, the receiving matrix and the structural velocity vector, and the second multi-layer sensor outputs a damping force prediction matrix based on the input data. Then, input data of the third multi-layer sensor, namely a resultant force calculation matrix, is generated based on the predicted restoring force prediction matrix and the damping force prediction matrix, and the third multi-layer sensor outputs a resultant force prediction matrix based on the input data. And then, generating input data of a fourth multi-layer perceptron based on the known external excitation load data, the receiving matrix and the resultant force prediction matrix, namely an acceleration response calculation matrix, wherein the fourth multi-layer perceptron outputs acceleration response prediction data based on the input data and takes the acceleration response prediction data as final output data of the artificial neural network.
In one implementation, the step S2031 specifically includes:
step S20311, multiplying the sending matrix by the structure displacement vector to obtain a first matrix;
step S20312, multiplying the receiving matrix by the structure displacement vector to obtain a second matrix;
step S20313, combining the first matrix and the second matrix in rows to obtain the restoring force calculation matrix.
Specifically, the transmitting matrix and the receiving matrix are multiplied by the structural displacement vector respectively to obtain a first matrix and a second matrix, the two matrices are combined into a 2-column matrix according to columns, and then the restoring force calculation matrix is obtained and is used as input data of the first multi-layer sensor (as shown in fig. 3).
In one implementation, the step S2032 specifically includes:
step S20321, multiplying the sending matrix by the structure velocity vector to obtain a third matrix;
step S20322, multiplying the receiving matrix by the structure velocity vector to obtain a fourth matrix;
step S20323, combining the third matrix and the fourth matrix in columns to obtain the damping force calculation matrix.
Specifically, the transmitting matrix and the receiving matrix are multiplied by the structural velocity vector to obtain a third matrix and a fourth matrix, the two matrices are combined into a 2-column matrix according to columns to obtain a damping force calculation matrix, and then the damping force calculation matrix is used as input data of the second multi-layer sensor (as shown in fig. 3).
In one implementation, the step S2033 specifically includes:
and S20331, combining the restoring force prediction matrix and the damping force prediction matrix in columns to obtain a calculation matrix of the resultant force.
Specifically, the acquired restoring force prediction matrix and damping force prediction matrix are combined into a 2-column matrix in columns, that is, a resultant force calculation matrix is obtained, and then the resultant force calculation matrix is used as input data of the third multi-layer sensor (as shown in fig. 3).
In an implementation manner, the step S2034 specifically includes:
step S20341, the left side of the resultant prediction matrix is multiplied by the transposition of the receiving matrix to obtain a fifth matrix;
and S20342, combining the fifth matrix and the external excitation load data in columns to obtain the acceleration response calculation matrix.
Specifically, the resultant force prediction matrix is multiplied by the receiving matrix MrrAnd transposing to obtain a fifth matrix, combining the fifth matrix and external excitation load data into a 2-column matrix according to columns to obtain an acceleration response calculation matrix, and then using the acceleration response calculation matrix as input data of a fourth multilayer sensor (as shown in fig. 3).
In order to prove the technical effect of the invention, the inventor performs the following experiments:
the three-degree-of-freedom system shown in fig. 2 is used as a training model, data are generated according to a formula (1), a neural network is trained, and then the trained neural network model is used for predicting the acceleration response of the four-degree-of-freedom system (the initial speed and the displacement of each degree of freedom of the system are respectively set to be 0m/s2And 0m, the time domain response acceleration of the structure can be predicted only by inputting the external excitation values corresponding to the sampling time points). The parameters of the three-degree-of-freedom system and the four-degree-of-freedom system are shown in tables 1 and 2. A set of seismic accelerations (BAJA _020787_251) whose wave forms are shown in fig. 4 are selected as external stimuli. In order to calculate the time series, the acceleration response needs to be numerically integrated to solve the velocity and displacement of the node degree of freedom, and the numerical integration method is shown as formula (4). The sampling frequency used in this example was 1000Hz, and the training results and the test results are shown in fig. 5, 6, and 7 and fig. 8, 9, 10, and 11, respectively. As can be seen from the result graph, the method can accurately predict the acceleration response of the structure under the condition that the structure parameters are unknown. And because of the input of the topological information, the generalization capability of the neural network model is greatly improved.
TABLE 1 three-DOF spring-Mass System parameters
Figure BDA0003450393440000151
TABLE 2 four degree of freedom spring-proof mass system parameters
Figure BDA0003450393440000152
Figure BDA0003450393440000153
Figure BDA0003450393440000154
Based on the above embodiment, the present invention further provides a structural dynamics model generation system based on a graph neural network, as shown in fig. 12, the system includes:
the data acquisition module 01 is used for determining a target structure and acquiring structure response data and topological information of the target structure;
and the network training module 02 is configured to train a preset artificial neural network according to the structural response data and the topology information, and use the trained artificial neural network as a structural dynamics model corresponding to the target structure.
Based on the above embodiments, the present invention further provides a terminal, and a schematic block diagram thereof may be as shown in fig. 13. The terminal comprises a processor, a memory, a network interface and a display screen which are connected through a system bus. Wherein the processor of the terminal is configured to provide computing and control capabilities. The memory of the terminal comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the terminal is used for connecting and communicating with an external terminal through a network. The computer program is executed by a processor to implement a method for generating a structural dynamics model based on a graph neural network. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.
It will be understood by those skilled in the art that the block diagram of fig. 13 is only a block diagram of a portion of the structure associated with the solution of the present invention, and does not constitute a limitation of the terminal to which the solution of the present invention is applied, and a specific terminal may include more or less components than those shown in the drawings, or may combine some components, or have a different arrangement of components.
In one implementation, one or more programs are stored in a memory of the terminal and configured to be executed by one or more processors include instructions for performing a method of structural dynamics model generation based on a graph neural network.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, databases, or other media used in embodiments provided herein may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
In summary, the present invention discloses a structural dynamics model generation method based on a graph neural network, the method includes: determining a target structure, and acquiring structure response data and topological information of the target structure; training a preset artificial neural network according to the structural response data and the topological information; and taking the trained artificial neural network as a structural dynamics model corresponding to the target structure. The invention introduces the topological information of the target structure into the framework of the deep learning network. The structure dynamics model generated by the invention can be used for calculating (predicting) the dynamics response of the structure. According to the invention, the topological information of the target structure is introduced into the frame of the deep learning network, so that the generalization capability of the model can be improved to a certain extent while the extraction capability of the deep learning network on the physical characteristics of the target structure is not lost, and the conversion of the artificial neural network model in different topological structures is realized. The problem of the deep learning network that is used for simulating structure dynamics among the prior art generalizes the ability poor is solved.
It is to be understood that the invention is not limited to the examples described above, but that modifications and variations may be effected thereto by those of ordinary skill in the art in light of the foregoing description, and that all such modifications and variations are intended to be within the scope of the invention as defined by the appended claims.

Claims (10)

1. A structural dynamics model generation method based on a graph neural network is characterized by comprising the following steps:
determining a target structure, and acquiring structure response data and topological information of the target structure;
training a preset artificial neural network according to the structural response data and the topological information;
and taking the trained artificial neural network as a structural dynamics model corresponding to the target structure.
2. The method of claim 1, wherein the obtaining structural response data of the target structure comprises:
acquiring displacement response data, speed response data and acceleration response data respectively corresponding to a plurality of degrees of freedom on a target structure according to a preset sampling time point;
taking the displacement response data, the velocity response data, and the acceleration response data as the structural response data.
3. The method for generating a structural dynamics model based on a graph neural network according to claim 2, wherein the training of the preset artificial neural network according to the structural response data and the topology information comprises:
generating a sending matrix and a receiving matrix according to the topology information, wherein the combination of the sending matrix and the receiving matrix can reflect the interaction topology information among a plurality of degrees of freedom of the target structure;
generating a structural displacement vector according to the displacement response data corresponding to the degrees of freedom respectively, generating a structural speed vector according to the speed response data corresponding to the degrees of freedom respectively, and generating training label data according to the acceleration response data corresponding to the degrees of freedom respectively;
and training the artificial neural network according to the sending matrix, the receiving matrix, the structure displacement vector, the structure speed vector and the training label data.
4. The method of generating a structural dynamical model based on a graph neural network of claim 3, wherein the artificial neural network comprises a first multi-layered perceptron, a second multi-layered perceptron, a third multi-layered perceptron, and a fourth multi-layered perceptron, and wherein training the artificial neural network according to the transmit matrix, the receive matrix, the structural displacement vector, and the structural velocity vector and the training label data comprises:
generating a restoring force calculation matrix according to the sending matrix, the receiving matrix and the structural displacement vector, and inputting the restoring force calculation matrix into the first multilayer perceptron to obtain a restoring force prediction matrix;
generating a damping force calculation matrix according to the sending matrix, the receiving matrix and the structural velocity vector, and inputting the damping force calculation matrix into the second multilayer perceptron to obtain a damping force prediction matrix;
obtaining a resultant force calculation matrix according to the restoring force prediction matrix and the damping force prediction matrix, and inputting the resultant force calculation matrix into the third multilayer perceptron to obtain a resultant force prediction matrix, wherein the resultant force prediction matrix is used for reflecting the vector sum of the restoring force and the damping force;
acquiring external excitation load data, generating an acceleration response calculation matrix according to the receiving matrix, the resultant force prediction matrix and the external excitation load data, and inputting the acceleration response calculation matrix into the fourth multilayer perceptron to obtain acceleration response prediction data;
and training the artificial neural network according to the acceleration response prediction data and the training label data.
5. The method of claim 4, wherein the generating a restorative force computation matrix from the transmit matrix, the receive matrix, and the structural displacement vector comprises:
multiplying the sending matrix by the structure displacement vector to obtain a first matrix;
multiplying the receiving matrix by the structure displacement vector to obtain a second matrix;
and combining the first matrix and the second matrix according to columns to obtain the restoring force calculation matrix.
6. The method of generating a structural dynamics model based on a graphical neural network of claim 4, wherein the generating a damping force calculation matrix from the transmit matrix, the receive matrix, and the structural velocity vector comprises:
multiplying the sending matrix by the structural velocity vector to obtain a third matrix;
multiplying the receiving matrix and the structural velocity vector to obtain a fourth matrix;
and combining the third matrix and the fourth matrix in columns to obtain the damping force calculation matrix.
7. The method of claim 4, wherein the obtaining a computation matrix of resultant forces from the restorative force prediction matrix and the damping force prediction matrix comprises:
and combining the restoring force prediction matrix and the damping force prediction matrix according to columns to obtain a calculation matrix of the resultant force.
8. The method of generating a structural dynamics model based on a graph neural network of claim 4, wherein the generating an acceleration response calculation matrix from the receive matrix, the resultant force prediction matrix, and the external excitation load data comprises:
multiplying the left side of the resultant prediction matrix by the transposition of the receiving matrix to obtain a fifth matrix;
and combining the fifth matrix and the external excitation load data in columns to obtain the acceleration response calculation matrix.
9. A structural dynamics model generation system based on a graph neural network, the system comprising:
the data acquisition module is used for determining a target structure and acquiring structure response data and topological information of the target structure;
and the network training module is used for training a preset artificial neural network according to the structural response data and the topological information, and taking the trained artificial neural network as a structural dynamic model corresponding to the target structure.
10. A structural dynamics model based on a graph neural network, characterized in that the structural dynamics model is generated by using the method for generating the structural dynamics model based on the graph neural network according to any of the claims 1 to 8.
CN202111674213.9A 2021-12-31 2021-12-31 Structural dynamics model generation method based on graph neural network Active CN114564878B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111674213.9A CN114564878B (en) 2021-12-31 2021-12-31 Structural dynamics model generation method based on graph neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111674213.9A CN114564878B (en) 2021-12-31 2021-12-31 Structural dynamics model generation method based on graph neural network

Publications (2)

Publication Number Publication Date
CN114564878A true CN114564878A (en) 2022-05-31
CN114564878B CN114564878B (en) 2022-10-28

Family

ID=81711195

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111674213.9A Active CN114564878B (en) 2021-12-31 2021-12-31 Structural dynamics model generation method based on graph neural network

Country Status (1)

Country Link
CN (1) CN114564878B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116305461A (en) * 2023-03-13 2023-06-23 清华大学 Structure response calculation method, device, electronic equipment and storage medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150220311A1 (en) * 2014-02-03 2015-08-06 Richard Salter Computer implemented modeling system and method
CN104898414A (en) * 2015-03-31 2015-09-09 西南交通大学 Key parameter identification method for high-speed train dynamics performance design
US20210049314A1 (en) * 2019-08-15 2021-02-18 Global Energy Interconnection Research Institute Co. Ltd Systems and Methods of Composite Load Modeling for Electric Power Systems
CN113158543A (en) * 2021-02-02 2021-07-23 浙江工商大学 Intelligent prediction method for software defined network performance
CN113177626A (en) * 2021-06-29 2021-07-27 中国科学院自动化研究所 Data-driven automatic learning method, system and equipment for complex system mechanism

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150220311A1 (en) * 2014-02-03 2015-08-06 Richard Salter Computer implemented modeling system and method
CN104898414A (en) * 2015-03-31 2015-09-09 西南交通大学 Key parameter identification method for high-speed train dynamics performance design
US20210049314A1 (en) * 2019-08-15 2021-02-18 Global Energy Interconnection Research Institute Co. Ltd Systems and Methods of Composite Load Modeling for Electric Power Systems
CN113158543A (en) * 2021-02-02 2021-07-23 浙江工商大学 Intelligent prediction method for software defined network performance
CN113177626A (en) * 2021-06-29 2021-07-27 中国科学院自动化研究所 Data-driven automatic learning method, system and equipment for complex system mechanism

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
TOLGA BAKLACIOGLU等: ""Dynamic modeling of exergy efficiency of turboprop engine components using hybrid genetic algorithm-artificial neural networks"", 《ENERGY》 *
俞阿龙: "机器人系统中操作环境动力学模型研究", 《计量学报》 *
张介明: ""利用神经网络建立轮胎动力学模型"", 《上海汽车》 *
袁军社等: "基于神经网络的火箭发动机结构动力学优化", 《火箭推进》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116305461A (en) * 2023-03-13 2023-06-23 清华大学 Structure response calculation method, device, electronic equipment and storage medium
CN116305461B (en) * 2023-03-13 2023-10-13 清华大学 Structure response calculation method, device, electronic equipment and storage medium

Also Published As

Publication number Publication date
CN114564878B (en) 2022-10-28

Similar Documents

Publication Publication Date Title
Zhang et al. Physics-informed multi-LSTM networks for metamodeling of nonlinear structures
Miller et al. Total Lagrangian explicit dynamics finite element algorithm for computing soft tissue deformation
Rumelhart et al. A general framework for parallel distributed processing
Naveros et al. A spiking neural simulator integrating event-driven and time-driven computation schemes using parallel CPU-GPU co-processing: a case study
Vlachas et al. A local basis approximation approach for nonlinear parametric model order reduction
Nouri et al. A digital neuromorphic realization of the 2-D Wilson neuron model
Wang et al. Global asymptotic stability of bidirectional associative memory neural networks with S-type distributed delays
Gupta et al. A hybrid partitioned deep learning methodology for moving interface and fluid–structure interaction
CN114564878B (en) Structural dynamics model generation method based on graph neural network
Song et al. Data-driven and active learning of variance-based sensitivity indices with Bayesian probabilistic integration
Chang et al. Selection of training samples for model updating using neural networks
Yousef Stability and further analytical bifurcation behaviors of Moran–Ricker model with delayed density dependent birth rate regulation
Ringwood et al. Identifying models using recorded data
Ballesteros et al. Matrix representation and simulation algorithm of numerical spiking neural P systems
Pentland et al. GParareal: a time-parallel ODE solver using Gaussian process emulation
Kapoor et al. Transfer learning for improved generalizability in causal physics-informed neural networks for beam simulations
de la Fuente An efficient procedure to obtain exact solutions in random vibration analysis of linear structures
EP4109351A1 (en) Inference program, inference method, and information processing apparatus
CN114358260A (en) Neural network model building method, device, equipment and storage medium
de la Fuente Von Mises stresses in random vibration of linear structures
Liang Removal of hidden neurons in multilayer perceptrons by orthogonal projection and weight crosswise propagation
Dong et al. Parameter reconstruction of vibration systems from partial eigeninformation
Hamrani et al. Machine learning surrogate modeling for meshless methods: Leveraging universal approximation
CN112989683A (en) Vectorization parallel computing method and device of SPH
Joghataie et al. Neuro-Skins: Dynamics, Plasticity and Effect of Neuron Type and Cell Size on Their Response

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant