CN115794447A - Grid data transmission method for multi-physical field coupling - Google Patents
Grid data transmission method for multi-physical field coupling Download PDFInfo
- Publication number
- CN115794447A CN115794447A CN202310075304.3A CN202310075304A CN115794447A CN 115794447 A CN115794447 A CN 115794447A CN 202310075304 A CN202310075304 A CN 202310075304A CN 115794447 A CN115794447 A CN 115794447A
- Authority
- CN
- China
- Prior art keywords
- grid
- nodes
- program
- interface
- node
- 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
Links
Images
Abstract
The invention provides a grid data transmission method for multi-physical field coupling, which comprises the following steps: storing grids, acquiring two solvers for multi-physical field coupling, and extracting data on interface grids and interface grid nodes of the two solvers; establishing a mapping relation, determining the number M of search nodes, and dividing the conditions into two conditions of small grid node number ratio and large grid node number ratio according to the grid number of an interface, wherein the former adopts a nearest point interpolation method to establish the corresponding relation of grid nodes between programs a and b, and the latter adopts a neural network method to fit a data distribution relation to establish a BP neural network model; and obtaining the node interpolation physical quantity and then sending the node interpolation physical quantity to a solver. The invention defines a judging method for optimal multi-physical-field coupling data transmission, which adopts the grid node number to compare with an interpolation method used in multi-physical-field weak coupling grid data transmission to judge and can select the interpolation method by self-adapting to the difference of grid scales.
Description
Technical Field
The invention relates to the field of physical field coupling, in particular to a grid data transmission method for multi-physical field coupling.
Background
There are two solving methods for the multi-physical field coupling problem, which are a strong coupling method and a weak coupling method. The strong coupling method requires establishing a simultaneous equation containing the variables of each physical field of the problem to be solved, and then solving the variables in the equation simultaneously in the same time step. The weak coupling method decomposes the multi-field coupling problem into independent solutions to each physical field. In the solution process, the coupling is achieved by information transfer at each interaction time step over the interfaces of the different physical fields. Due to the great difference of the requirements of different physical fields on the grid type and the density, for example, in fluid-solid coupling, a finite volume/finite difference solver is mostly used for CFD, the grid type is structured and unstructured, and finite elements are mostly used for CSD as a solving tool. And the multi-physical field weak coupling calculation usually needs to transmit parameters at an interface, for example, the fluid transmits parameters such as pressure and temperature to the structure, and the structure transmits parameters such as displacement and heat flow density to the fluid. This creates a problem of inter-grid data transfer between different physical fields, which typically requires the use of interpolation.
The nearest point interpolation method is that the node of one grid is searched for the node which is closest to the node on the other grid as the corresponding node, and the data on the corresponding node is used as the interpolation data on the node. The radial basis function interpolation method uses coordinates and data fitting equations of all nodes on one grid, the equation set is solved to obtain equation set coefficients, and then the nodes on the other grid can obtain interpolation data by inputting the coordinates into the fitted equation set. At present, the data transfer method adopted in the multi-physical field coupling is often a definite method, for example, a nearest neighbor interpolation method is directly adopted, or a radial basis function interpolation method is directly adopted. If a coupling method is selected at will without considering the influence caused by the difference of the number of grid nodes, sometimes the coupling precision or efficiency is not high because the selected interpolation algorithm is not suitable. The nearest point interpolation method also has the problem of low precision, and the radial basis function integral interpolation method has the problems of difficult solution and low efficiency.
Disclosure of Invention
The invention aims to provide a grid data transmission method for multi-physical-field coupling, which not only can adaptively select a proper interpolation algorithm to carry out data transmission and exchange among grids of a boundary surface, but also respectively improves the efficiency and the precision when the number of grid nodes is larger or smaller.
In order to achieve the purpose, the invention is realized by the following technical scheme:
a method for multi-physics coupled mesh data transfer, comprising the steps of:
(1) The method comprises the steps of storing grids, acquiring two solvers for multi-physical field coupling, extracting data on interface grids and interface grid nodes of the two solvers, defining the solver A with the larger interface grid number and the solver B with the larger interface grid number, and respectively storing the extracted data in adaptation programs a and B of the solvers A and B;
(2) Establishing a mapping relation, determining the number M of search nodes, and dividing the conditions into two conditions of small grid node number ratio and large grid node number ratio according to the interface grid number, wherein the former adopts a nearest neighbor interpolation method to establish a corresponding relation of grid nodes between the programs a and b; the data distribution relation is fitted by adopting a neural network method, and a BP neural network model is constructed;
(3) Data transmission, under the condition that the number ratio of the grid nodes is small, according to the constructed corresponding relation of the grid nodes, when the program a transmits physical quantity to the program b, the physical quantity of the interface grid node on the program b is found, the corresponding interface grid node of the program a is found, the physical quantities of the nodes are added to obtain interpolated physical quantity, and if one node is corresponding to a plurality of nodes, the physical quantity on the node is uniformly distributed to the corresponding nodes; when the program b transmits the physical quantity to the program a, the interface grid nodes on the program a find all the interface grid nodes on the program b corresponding to the interface grid nodes, and the nodes on the program b divide the physical quantity equally according to the number of the corresponding program a nodes and add the divided physical quantity to obtain the interpolation physical quantity of the program a grid nodes; under the condition that the number ratio of grid nodes is large, a fitting equation is used, and the grid node coordinates of an interface are input to obtain node interpolation physical quantity;
(4) And the programs a and b update the interface grid physical quantity and then send the interface grid physical quantity to a solver.
The method for judging the small grid node number ratio and the large grid node number ratio comprises the following steps:
(1) Grid division, wherein solvers A and B start a plurality of processes, interface grids are divided and stored in different processes, one of the processes is defined as a master process, and the other processes are defined as slave processes;
(2) Grid communication, namely, a main process of the program a sends a complete interface grid of the program a to all processes of the program b in an MPI parallel communication mode, and a main process of the program b sends the complete interface grid of the program a to all processes of the program a;
(3) Calculating the number ratio of grid nodes of interface, defining the number of grid nodes of complete interface on the main process of the program a and b as AN and BN, respectively calculating the number ratio of grid nodes of interface N = AN/BN on the main process of the program a and b and broadcasting to other slave processes, defining the threshold value N of the number of grid nodes 0 When N does not exceed N 0 When N exceeds N, the number ratio of grid nodes is considered to be small 0 When the number of the grid nodes is larger than the total number of the grid nodes.
Preferably, the value of the number M of search nodes is a value obtained by rounding up the number of the interface grid nodes compared with the number N.
Preferably, the threshold value N 0 =10。
Under the condition that the number ratio of grid nodes is small, a Kdtree search method is adopted, each process of the program b traverses nodes on a partition interface grid, M nearest nodes on the complete interface grid of the program a are searched for each node, the number of times of searching the nodes is recorded, the partitioned grid nodes of each process of the program b and the corresponding grid nodes of the program a are uniformly stored in the main process of the program b, the stored corresponding relation is broadcasted to the program a, and therefore the corresponding relation of the grid nodes between the program a and the program b is established.
Under the condition that the number ratio of grid nodes is large, fitting a data distribution relation by adopting a neural network method, wherein the data distribution relation comprises that the programs a and b respectively fit the data distribution on the grid of the other party on each process by using the node coordinates and the physical quantity of the complete grid of the other party; in the program b with a large number of mesh nodes of the fitting surface, a part of the nodes obtained by sampling are used for constructing a neural network model, and in the program a with a large number of mesh nodes of the interpolation surface, all the nodes are directly used for constructing the neural network model; and the program a takes the coordinates of all nodes of the complete interface grid on the program b as input, and takes physical quantity as output to construct a BP neural network model.
The construction and training steps of the BP neural network model are as follows: according to the number ratio of the grid nodes, adopting Latin hypercube sampling, extracting the number of sample nodes to be 1/M of the number of all interface grid nodes of the program a, taking the coordinates of the obtained nodes as input, taking physical quantity as output, and constructing a BP neural network; setting parameters of a neural network, and obtaining a trained BP neural network model through a certain number of iterations; the setting of the parameters comprises: the activation function adopts a Sigmoid function, the loss function adopts a mean square error, the learning rate is set to be 0.01 to 0.1, and the initial weight and the bias are both set to be random numbers of [0,1 ];
after a model is trained and built, the model is interpolated to obtain the predicted physical quantity of the residual nodes which are not used as samples, the predicted physical quantity is subtracted from the original real physical quantity to obtain an error, a maximum error node is obtained by searching, whether the maximum error meets the precision requirement or not is judged, if yes, the model is built, if not, the node is added into the sample, a BP neural network is built again, a new model is obtained, and the process is repeated until the maximum error meets the precision requirement.
The invention has the advantages that: a judgment method for optimal multi-physical-field coupling data transmission is defined, the number of grid nodes is compared with an interpolation method used in multi-physical-field weak coupling grid data transmission for judgment, the interpolation method can be selected by self-adapting to the difference of grid scales, and when the difference of the number of the grid nodes is not large, the accuracy of grid data transmission can be improved by adopting a multi-point nearest point interpolation method; when the difference of the number of grid nodes is too large, the adaptive neural network fitting data is adopted to carry out interpolation of grid data, and the efficiency of data transmission can be improved. In the method, parallel processing is adopted to start a plurality of processes, the interface grids are respectively stored in different processes, the respective processing of the programs a and b can be simultaneously carried out, and the coupling speed is improved.
Drawings
FIG. 1 is a flow chart of mesh data transfer according to the present invention;
FIG. 2 is a flowchart of a multipoint nearest neighbor interpolation method according to embodiment 1 of the present invention;
fig. 3 is a flowchart of an adaptive neural network model interpolation method according to embodiment 2 of the present invention.
Detailed Description
The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
Example 1
The embodiment discloses a grid data transmission method for multi-physical field coupling, and the specific flow refers to fig. 1, which includes the following steps.
S1, storage grid
The two solvers which are subjected to multi-physical field weak coupling are respectively called a solver A and a solver B, an interface grid is extracted from the two solvers which are subjected to coupling, data on nodes of the interface grid, namely physical quantities, are also extracted, and the data are respectively stored in adaptation programs a and B of the solvers A and B.
S2, grid division
And a parallel coupling mode is adopted, so that the coupling efficiency is improved. The solvers A and B can start a plurality of processes, interface grids extracted from the programs a and B need to be divided and stored in different processes, one process is defined as a main process, the other processes are defined as slave processes, and the main process stores complete interface grids besides the blocked interface grids.
S3, grid communication
Taking the example that the interface grid number of the solver A is greater than that of the solver B, the physical quantities of the interface grids of the solver A and the solver B are exchanged. And the main process of the program a sends the own complete interface grid to all processes of the program b, and the main process of the program b sends the own complete interface grid to all processes of the program a in an MPI parallel communication mode.
S4, calculating the number ratio of the grid nodes of the interface
All the processes of the program a and the program b have complete interface grids of each other, and each process has a blocked interface grid. And calculating the node number ratio N of the interface grids, setting the node number of the complete interface grid on the program a main process as AN, setting the node number of the complete interface grid on the program b as BN, calculating N = AN/BN, and broadcasting the value to other slave processes after the value is respectively calculated and obtained on the program a main process and the program b main process.
S5, establishing a mapping relation
Performing rounding calculation on the N to obtain the number M of the search nodes, namely when N =1, M =1; when 1 is constructed with n < =2, M =2; when 2 were less than n < =3, M =3, the processes were performed in this order.
(1) The number of grid nodes is small
When the number of grid nodes is greater than 1< = N < =10, a Kdtree search method is adopted, please refer to fig. 2, each process of the program b traverses nodes on the divided interface grid, M nearest nodes on the complete interface grid of the program a are searched for each node, and the searched nodes record the searched times. And uniformly storing the partitioned grid nodes of each process of the program b and the grid nodes of the program a corresponding to the partitioned grid nodes in the main process of the program b, and broadcasting the stored corresponding relation to the program a, thereby establishing the corresponding relation of the grid nodes between the program a and the program b.
(2) The number of grid nodes is larger
When the number ratio N of the grid nodes is greater than 10, a data distribution relationship is fitted by using a neural network method, please refer to fig. 3, where data distribution on the grid of the other side needs to be fitted in each process of the programs a and b, and the node coordinates and physical quantities of the complete grid of the other side need to be used for fitting. And if the number of the mesh nodes of the fitting surface is large, namely the program b, constructing a neural network model by using part of the nodes obtained by sampling. And if the number of the grid nodes of the interpolation surface is large, namely the program a, directly using all the nodes to construct a neural network model. And (c) taking the coordinates of all nodes of the complete interface grid on the program b as input and taking the physical quantity as output by the program a, and constructing a BP neural network model.
And (b) sampling by adopting Latin hypercube according to the number ratio of the grid nodes, extracting the number of the sample nodes to be 1/M of the number of all interface grid nodes of the program (a), and constructing the BP neural network by taking the coordinates of the obtained nodes as input and the physical quantity as output. And (3) setting the number of hidden layers of the neural network, the number of neurons, an activation function, a loss function, a learning rate and the like, and obtaining the trained BP neural network model through a certain number of iterations. The activation function adopts a Sigmoid function, the loss function adopts a Mean Square Error (MSE), and the learning rate can be set to be 0.01 to 0.1. The initial weight and bias are both set to random numbers of [0,1 ].
After the model is trained and built, the model is interpolated to obtain the residual predicted physical quantity of the nodes which are not taken as samples, and the difference between the predicted physical quantity and the original real physical quantity is calculated to obtain the error. And searching to obtain a maximum error node, judging whether the maximum error meets the precision requirement, if so, completing model construction, if not, adding the node into a sample, reconstructing a BP neural network to obtain a new model, and repeating the processes until the maximum error meets the precision requirement.
S6, data transmission
(1) The number of grid nodes is small
When the program a transmits the physical quantity to the program b, the interface grid node of the program a corresponding to the physical quantity of the interface grid node on the program b is found according to the constructed corresponding relation of the grid nodes, the physical quantities of the nodes are added to obtain an interpolated physical quantity, and if one node is corresponding to a plurality of nodes, the physical quantity on the node is uniformly distributed to the corresponding nodes.
When the program b transmits the physical quantity to the program a, the interface grid nodes on the program a find all the interface grid nodes on the program b corresponding to the interface grid nodes according to the constructed corresponding relation of the grid nodes, and the nodes on the program b divide the physical quantity evenly according to the number of the corresponding program a nodes and add the divided physical quantity to obtain the interpolated physical quantity of the program a grid nodes.
(2) The number of grid nodes is larger
And (5) inputting the coordinates of the grid nodes of the interface by the program a and the program b according to the constructed BP neural network model to obtain the node interpolation physical quantity.
And S7, after the interface grid physical quantity is updated, the program a and the program b send the interface grid physical quantity to a solver.
Finally, it should be noted that: the above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that are within the spirit and principle of the present invention are intended to be included in the scope of the present invention.
Claims (7)
1. A method for multi-physics coupled mesh data transfer, comprising the steps of:
(1) The method comprises the steps of storing grids, acquiring two solvers for multi-physical field coupling, extracting data on interface grids and interface grid nodes of the two solvers, defining the solver A with the larger interface grid number and the solver B with the larger interface grid number, and respectively storing the extracted data in adaptation programs a and B of the solvers A and B;
(2) Establishing a mapping relation, determining the number M of search nodes, and dividing the conditions into two conditions of small grid node number ratio and large grid node number ratio according to the grid number of an interface, wherein the former adopts a nearest point interpolation method to establish a corresponding relation of grid nodes between programs a and b; the data distribution relation is fitted by adopting a neural network method, and a BP neural network model is constructed;
(3) Data transmission, under the condition that the number ratio of the grid nodes is small, according to the constructed corresponding relation of the grid nodes, when the program a transmits physical quantity to the program b, the physical quantity of the interface grid node on the program b is found, the corresponding interface grid node of the program a is found, the physical quantities of the nodes are added to obtain interpolated physical quantity, and if one node is corresponding to a plurality of nodes, the physical quantity on the node is uniformly distributed to the corresponding nodes; when the program b transmits the physical quantity to the program a, the interface grid nodes on the program a find all the interface grid nodes on the program b corresponding to the interface grid nodes, and the nodes on the program b divide the physical quantity equally according to the number of the corresponding program a nodes and add the divided physical quantity to obtain the interpolation physical quantity of the program a grid nodes; under the condition that the number ratio of the grid nodes is large, a fitting equation is used, and the grid node coordinates of the interface are input to obtain node interpolation physical quantity;
(4) And the programs a and b update the interface grid physical quantity and then send the interface grid physical quantity to a solver.
2. The method of claim 1, wherein the determination method that the grid node number ratio is small and the grid node number ratio is large is:
(1) Grid division, wherein solvers A and B start a plurality of processes, interface grids are divided and stored in different processes, one of the processes is defined as a master process, and the other processes are defined as slave processes;
(2) Grid communication, through MPI parallel communication mode, program a main process sends its complete interface grid to all program b processes, program b main process sends its complete interface grid to all program a processes
(3) Calculating the number ratio of grid nodes of interface, defining the number of grid nodes of complete interface on the main process of the program a and b as AN and BN, respectively calculating the number ratio of grid nodes of interface N = AN/BN on the main process of the program a and b and broadcasting to other slave processes, defining the threshold value N of the number of grid nodes 0 When N does not exceed N 0 When N exceeds N, the number ratio of grid nodes is considered to be small 0 When the number of the grid nodes is larger, the grid nodes are considered to be larger.
3. The method as claimed in claim 2, wherein the number of search nodes M is a value obtained by rounding up the number of interface grid nodes to N.
4. The method of claim 2, wherein the threshold N is set to 0 =10。
5. The method according to any one of claims 1 to 4, wherein, in the case that the number ratio of mesh nodes is small, using Kdtree search method, each process of the program b traverses the nodes on the divided interface mesh, searches each node for M nearest nodes on the complete interface mesh of the program a and records the number of times the node is searched, stores the partitioned mesh nodes of each process of the program b and the corresponding mesh nodes of the program a in the main process of the program b, and broadcasts the stored correspondence to the program a, thereby establishing the mesh node correspondence between the program a and the program b.
6. The method for transmitting grid data for coupling of multiple physical fields according to any one of claims 1 to 4, wherein in case of a large number ratio of grid nodes, fitting the data distribution relationship by using a neural network method, including fitting the data distribution on the grid of the other party on each process by using node coordinates and physical quantities of the complete grid of the other party in the programs a and b; in a program b with a large number of mesh nodes of a fitting surface, constructing a neural network model by using part of nodes obtained by sampling, and in a program a with a large number of mesh nodes of an interpolation surface, directly constructing the neural network model by using all the nodes; and the program a takes the coordinates of all nodes of the complete interface grid on the program b as input, and takes physical quantity as output to construct a BP neural network model.
7. The method of claim 6, wherein the BP neural network model is constructed and trained by the steps of: according to the number ratio of the grid nodes, adopting Latin hypercube sampling, extracting the number of sample nodes to be 1/M of the number of all interface grid nodes of the program a, taking the coordinates of the obtained nodes as input, taking physical quantity as output, and constructing a BP neural network; setting parameters of a neural network, and obtaining a trained BP neural network model through a certain number of iterations; the setting of the parameters comprises: the activation function adopts a Sigmoid function, the loss function adopts a mean square error, the learning rate is set to be 0.01 to 0.1, and the initial weight and the bias are set to be random numbers of [0,1 ];
after a model is trained and constructed, the model is interpolated to obtain the predicted physical quantity of the residual nodes which are not taken as samples, the predicted physical quantity is subtracted from the original real physical quantity to obtain an error, the maximum error node is obtained by searching, whether the maximum error meets the precision requirement or not is judged, if yes, the model construction is finished, if not, the node is added into the sample, a BP neural network is reconstructed, a new model is obtained, and the processes are repeated until the maximum error meets the precision requirement.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310075304.3A CN115794447B (en) | 2023-02-07 | 2023-02-07 | Grid data transmission method for multi-physical field coupling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310075304.3A CN115794447B (en) | 2023-02-07 | 2023-02-07 | Grid data transmission method for multi-physical field coupling |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115794447A true CN115794447A (en) | 2023-03-14 |
| CN115794447B CN115794447B (en) | 2023-04-28 |
Family
ID=85430280
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310075304.3A Active CN115794447B (en) | 2023-02-07 | 2023-02-07 | Grid data transmission method for multi-physical field coupling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115794447B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117094268A (en) * | 2023-10-17 | 2023-11-21 | 北京智芯微电子科技有限公司 | Inter-grid data transmission method and device, storage medium and electronic equipment |
| CN117131707A (en) * | 2023-10-25 | 2023-11-28 | 青岛哈尔滨工程大学创新发展中心 | Multi-physical field abnormal grid mapping method based on feature extraction |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB9014427D0 (en) * | 1989-07-18 | 1990-08-22 | Amoco Corp | Estimation of error of data values with increasing distance from control data points |
| CN106547964A (en) * | 2016-10-21 | 2017-03-29 | 北京宇航系统工程研究所 | It is a kind of to be used for isomery lattice interface data mapping method |
| US20170231601A1 (en) * | 2014-08-15 | 2017-08-17 | Hitachi, Ltd. | Ultrasonic image processing device |
| CN112364442A (en) * | 2020-12-22 | 2021-02-12 | 北京机电工程研究所 | Bidirectional intelligent selection fluid-solid coupling analysis method |
| CN113034694A (en) * | 2021-03-26 | 2021-06-25 | 浙江大学 | Method and device for automatically generating measurement orthogonal grid adaptive to flow field characteristics |
| CN114757124A (en) * | 2022-04-21 | 2022-07-15 | 哈尔滨工程大学 | CFD workflow modeling method and device based on XML, computer and storage medium |
| CN114841066A (en) * | 2022-04-29 | 2022-08-02 | 浙江工业大学 | Flying dust particle drift trajectory model and modeling method thereof |
-
2023
- 2023-02-07 CN CN202310075304.3A patent/CN115794447B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB9014427D0 (en) * | 1989-07-18 | 1990-08-22 | Amoco Corp | Estimation of error of data values with increasing distance from control data points |
| US20170231601A1 (en) * | 2014-08-15 | 2017-08-17 | Hitachi, Ltd. | Ultrasonic image processing device |
| CN106547964A (en) * | 2016-10-21 | 2017-03-29 | 北京宇航系统工程研究所 | It is a kind of to be used for isomery lattice interface data mapping method |
| CN112364442A (en) * | 2020-12-22 | 2021-02-12 | 北京机电工程研究所 | Bidirectional intelligent selection fluid-solid coupling analysis method |
| CN113034694A (en) * | 2021-03-26 | 2021-06-25 | 浙江大学 | Method and device for automatically generating measurement orthogonal grid adaptive to flow field characteristics |
| CN114757124A (en) * | 2022-04-21 | 2022-07-15 | 哈尔滨工程大学 | CFD workflow modeling method and device based on XML, computer and storage medium |
| CN114841066A (en) * | 2022-04-29 | 2022-08-02 | 浙江工业大学 | Flying dust particle drift trajectory model and modeling method thereof |
Non-Patent Citations (5)
| Title |
|---|
| HONGWEI LI,KAI GUO,ZANYUN ZHANG,DANDAN YU,JIANXIN ZHANG,PINGFAN NING: "High-Power LED Photoelectrothermal Analysis Based on Backpropagation Artificial Neural Networks" * |
| YING LIU,YAN ZHANG,BIN TANG,MINGYUAN GAO,JUN DAI: "Introducing the thermal field into multi-physics coupling for the modeling of MR fluid-based micro-brake" * |
| 刘刚;池骋;孙立鹏;马永强;李琳;: "基于线性插值函数的二维异构网格数据插值方法研究" * |
| 刘振宇;傅云;谭建荣;: "基于异构网格耦合的产品多物理场有限元数据集成与可视化仿真" * |
| 孟小红,侯建全,梁宏英,胡朝顺: "离散光滑插值方法在地球物理位场中的快速实现" * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117094268A (en) * | 2023-10-17 | 2023-11-21 | 北京智芯微电子科技有限公司 | Inter-grid data transmission method and device, storage medium and electronic equipment |
| CN117094268B (en) * | 2023-10-17 | 2024-01-19 | 北京智芯微电子科技有限公司 | Inter-grid data transmission method and device, storage medium and electronic equipment |
| CN117131707A (en) * | 2023-10-25 | 2023-11-28 | 青岛哈尔滨工程大学创新发展中心 | Multi-physical field abnormal grid mapping method based on feature extraction |
| CN117131707B (en) * | 2023-10-25 | 2024-01-16 | 青岛哈尔滨工程大学创新发展中心 | Multi-physical field abnormal grid mapping method based on feature extraction |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115794447B (en) | 2023-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115794447B (en) | Grid data transmission method for multi-physical field coupling | |
| CN106156810B (en) | General-purpose machinery learning algorithm model training method, system and calculate node | |
| CN112180734B (en) | Multi-agent consistency method based on distributed adaptive event triggering | |
| CN110276442B (en) | Searching method and device of neural network architecture | |
| CN109241093A (en) | A kind of method of data query, relevant apparatus and Database Systems | |
| CN102945333B (en) | Key protein predicating method based on prior knowledge and network topology characteristics | |
| CN109977972B (en) | Intelligent feature identification method based on STEP | |
| CN110262954A (en) | Method based on the automatic learning system reliability model of Condition Monitoring Data | |
| CN115862319A (en) | Traffic flow prediction method for space-time diagram self-encoder | |
| CN114967460A (en) | Distributed cooperative control method of time-lag nonlinear multi-agent system | |
| CN111638648A (en) | Distributed pulse quasi-synchronization method with proportional delay complex dynamic network | |
| CN103177291A (en) | Variable-search-space ribonucleic acid (RNA) genetic algorithm modeling method for continuous stirred tank reactor | |
| CN109754008B (en) | High-dimensional symmetric sparse network missing information estimation method based on matrix decomposition | |
| CN111260056B (en) | Network model distillation method and device | |
| CN111985162A (en) | Replacement flow shop control method and system based on deep learning | |
| CN109033603B (en) | Intelligent substation secondary system simulation method based on source flow path chain | |
| Radicchi et al. | Renormalization flows in complex networks | |
| CN115455302A (en) | Knowledge graph recommendation method based on optimized graph attention network | |
| CN113824802B (en) | Decentralized distributed training topological structure, training system and method | |
| CN116760837B (en) | Distributed consensus method and system based on federal learning and equity election | |
| CN117332543B (en) | Distribution processing method for heterogeneous data sources of power grid | |
| Zhou et al. | DRL-Based Workload Allocation for Distributed Coded Machine Learning | |
| CN117557870B (en) | Classification model training method and system based on federal learning client selection | |
| CN116319364B (en) | MPI virtual graph topology communication method and system suitable for wave numerical mode | |
| CN111967590B (en) | Heterogeneous multi-XPU machine learning system oriented to recommendation system matrix decomposition method |
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 |