CN109766588A - A kind of parallelization time domain hybrid electromagnetic algorithm based on alternative manner - Google Patents
A kind of parallelization time domain hybrid electromagnetic algorithm based on alternative manner Download PDFInfo
- Publication number
- CN109766588A CN109766588A CN201811539924.3A CN201811539924A CN109766588A CN 109766588 A CN109766588 A CN 109766588A CN 201811539924 A CN201811539924 A CN 201811539924A CN 109766588 A CN109766588 A CN 109766588A
- Authority
- CN
- China
- Prior art keywords
- node
- grid
- iteration
- mesh
- fvtd
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of parallelization time domain hybrid electromagnetic algorithm based on alternative manner, it include: the calculating of size of mesh opening, zoning is divided by two parts using square dough sheet, use tetrahedron mesh generation enveloping surface interior zone to realize Accurate Model, using hexahedral mesh subdivision enveloping surface to the part on zoning boundary to save computing resource；Using equally loaded as principle, the node for participating in calculating is divided into FDTD iteration and FVTD iteration two parts, calculates the grid amount on each node；Grid is assigned to realization Optimal Parallel scheme on specific node with data traffic at least for design object by the data traffic between the adjacent mesh physical quantity estimation node for needing to use according to iteration normal on grid；Electromagnetic problem is solved using parallel FDTD-FVTD mixed method, obtains calculated result.The invention avoids the big problems of ladder subdivision and grid amount, efficiently use computer resource and shorten the calculating time.
Description
Technical field
The present invention relates to Numerical Calculation of Electromagnetic Fields technical field more particularly to a kind of parallelization time domains based on alternative manner
Hybrid electromagnetic algorithm.
Background technique
Radar system needs to provide by the radar cross section (RCS) of target to identification target and remotely predicting alarm
Data supporting and accurate judgement, big-size complicated shape target RCS computational problem can be emulated by time domain electromagnetic and once be counted
It calculates and obtains broadband simulation result, parallelization is then conducive to the expansion of calculation scale and calculates the shortening of time.It is simple to use simultaneously
When rowization Finite-Difference Time-Domain Method (FDTD) calculates electromagnetic problem, entire zoning is a cuboid, the uniform subdivision of grid
For hexahedral mesh, net lattice control is divided into each process by virtual topology mode, to balance each process load, so that grid is handed over
It is at least Optimal Parallel scheme that the grid amount of interface, which reaches,.And parallelization finite volume method (FVTD) method is used merely
When, the mesh generation of zoning is tetrahedral grid, by neighboring region mean allocation grid to each process, and optimizes boundary
Surface grids number makes it at least can guarantee optimal parallel speedup ratio.
It in the prior art, merely can be by the boundary ladder of model using hexahedral mesh subdivision for complicated electromagnetic structure
Change leads to modeling error, can realize Accurate Model using tetrahedron mesh generation merely but often grid amount is huge.Using mixing
Mesh modeling calculating is then an effective means, uses tetrahedron mesh generation, model for bending, tiny or incline structure
Rest part use hexahedral mesh subdivision, not only can guarantee modeling accuracy grid amount but also be unlikely to excessive.Hybrid grid calculates
Method has the advantage in modeling, but due to being related to two kinds of grids, two kinds of calculation methods cause its parallelization process to become multiple
It is miscellaneous.The meeting of use of concurrent technique as the number of processes for participating in calculating is continuously increased the calculating time so that reduce rapidly.And it calculates
When scale increases, the Border-zone infarcts total amount of field value communication increases, so that replacing communication mechanism benefit to become influences parallel efficiency
Principal element.Moreover, it is very big that parallel scheme does not calculate time difference simultaneously in the case where for same process number.Therefore,
How best parallel scheme is obtained under limited computing resource as the major issue for needing to solve in mixed method.
Summary of the invention
For model ladder subdivision, grid amount is huge, parallel procedure is complicated and there are the two methods of two kinds of grids
Computational problem, the purpose of the present invention is to provide a kind of parallelization time domain mixing electricity based on alternative manner solved these problems
Magnetic algorithm.
In order to achieve the above objectives, the invention adopts the following technical scheme:
Step 1, frequency is solved according to the highest of target and calculates size of mesh opening；
Step 2, target surface is surrounded using closed square dough sheet enveloping surface；
Step 3, zoning is divided into two parts: use tetrahedron mesh generation enveloping surface to the region of target surface with
Accurate Model is realized, using hexahedral mesh subdivision enveloping surface to the part on zoning boundary to save computing resource；
It step 4,, will using equally loaded as principle according to alternative manner used on grid and consumed calculation amount
The node for participating in calculating is divided into different iteration classifications, calculates the grid amount on each node；
Step 5, the data communication between node is estimated according to the adjacent mesh physical quantity that iteration normal on grid needs to use
Grid is assigned to realization Optimal Parallel scheme on specific node with data traffic at least for design object by amount；
Step 6, it is iteratively solved using parallel FDTD-FVTD mixed method, obtains calculated result.
It is further, described that frequency calculating size of mesh opening, method are solved according to the highest of target are as follows:
For target in [fmin, fmax] radar cross section RCS in frequency range, it is calculated using the following equation FDTD method
Size of mesh opening Δ:
Δ=(c0/fmax)/10
Wherein, C0=3.0 × 108M/s is the light velocity.
It is further, described to surround target surface using closed square dough sheet enveloping surface, comprising:
According to the size of size of mesh opening, triangular mesh subdivision is carried out to target surface, is recycled in triangular mesh dough sheet
Heart point coordinate 2-3 size of mesh opening of extrapolation establishes the encirclement being made of the square dough sheet that side length is the size of mesh opening size
Face.
Further, the node for participating in calculating is divided into different iteration classifications, comprising:
All tetrahedral grids only carry out FVTD iteration, and the hexahedral mesh adjacent with tetrahedral grid also only carries out FVTD
Iteration, the hexahedral mesh for only carrying out FVTD iteration with these have the hexahedral mesh in public face to be referred to as transition region, this part
Not only FDTD iteration had been carried out on grid but also has carried out FVTD iteration, and in addition to this, the hexahedral mesh outside transition region then only carries out FDTD
Iteration.
Further, the grid amount on each node of calculating, comprising:
If the hexahedral mesh number for only carrying out FDTD method iteration is N1, only carry out the tetrahedron net of FVTD method iteration
Lattice number is N2, the hexahedral mesh number for only carrying out FVTD method iteration is N3, not only carried out FDTD iteration but also carried out FVTD iteration
Hexahedral mesh number be N4, then:
Calculation amount on all grids in FDTD iteration region is C1=7 × 6 × N1；
Calculation amount C on rest part grid2=97 × 6 × N2+145×6×N3+(145+7)×6×N4。
Further, step 4 further include: by all grids according to the calculation amount distribution node of estimation:
D node is distributed for FDTD iteration region, its calculation formula is:
D=[C1×M/(C1+C2)]
V node is distributed for FVTD iteration region, its calculation formula is:
V=[C2×M/(C1+C2)]
Meet relationship between D and V: D+V=M, M are node total number；
By N1A hexahedral mesh for participating in FDTD iteration is evenly distributed on 0-D-1 node；By (N2+N3+N4) a
The net lattice control for participating in FVTD iteration is distributed on D-M-1 node.
Further, between the adjacent mesh physical quantity estimation node for needing to use according to iteration normal on grid
Data traffic, comprising:
The grid being distributed on node i (0≤i≤D-1) and its adjacent node j (0≤j≤D-1, and i ≠ j) carries out FDTD
Method iteration, if having adjacent mesh n on two node public boundaries1It is a, the data traffic of the Partial Mesh are as follows: 2 × n1×4；
The grid being distributed in node i (0≤i≤D-1) carries out FDTD method iteration, if its adjacent node j satisfaction (D≤j≤
M-1, and i ≠ j), i.e. grid on node j carries out FVTD method iteration, if having adjacent mesh to have on two node public boundaries
n2It is a, the data traffic of the Partial Mesh are as follows: 2 × n2×4；
The grid being distributed in node i (D≤i≤M-1) carries out FVTD method iteration, adjacent node j (D≤j≤M-1, and i
≠ j) on grid also carry out FVTD method iteration, if thering is adjacent mesh to have n on two node public boundaries3It is a, the part net
The data traffic of lattice are as follows: 2 × n3×26。
Further, it is described with data traffic at least for design object, grid is assigned on specific node and is realized
Optimal Parallel scheme, comprising:
0-D-1 node is subjected to three-dimensional topology model construction, that is, sets a node of x directional spreding, y directional spreding b is a
Node, c node of z directional spreding, a, b, c is integer, and is determined by following two condition:
a:b:c≈(Lxmax-Lxmin):(Lymax-Lymin):(Lzmax-Lzmin)
A × b × c=D
The three-dimensional topology coordinate of node are as follows: (0,0,0), (1,0,0) ... (a-1, b-1, c-1)；
Center point coordinate (x, y, z) is met into condition:
Lxmin+(o-1)Δ≤x≤Lxmin+oΔ
Lymin+(p-1)Δ≤y≤Lymin+pΔ
Lzmin+(q-1)Δ≤z≤LzminThe grid of+q Δ is distributed in advance on the node that topological coordinate is (o, p, q)；
Start from (0,0,0) topological node, traverse each node and execute following steps:
If the number of grids being distributed in advance on node is greater than N1Grid on the node boundary face is then fully allocated to phase by/D
On neighbors；
If the number of grids being distributed in advance on node is less than N1Grid in adjacent node boundary face is then fully allocated to by/D
On the node；
Node is distributed in such a way that syntople traverses to the periphery by numbering small grid according to grid number order
On D-M-1；
It estimates the data traffic between different topology structure lower node, selects the smallest allocation plan of data traffic for most
Whole parallel scheme.
The present invention has following technical effect that compared with prior art
1. the present invention proposes a kind of Electromagnetic Calculation method of parallelization, avoids ladder subdivision and grid amount is big
Problem, Parallelization Scheme have maximally utilized computer resource and have shortened the calculating time.
2. the present invention is according to the alternative manner difference carried out in the grid characteristic distributions and grid of concrete model to calculating
Region is divided, and using process load balance as design criteria, improves parallel speedup ratio to reduce the data traffic between process
For design object, a kind of Parallelization Scheme being simple and efficient is provided for the solution of large-scale complex geometrical model electromagnetic problem.
Detailed description of the invention
Fig. 1 is the square dough sheet enveloping surface schematic diagram of spherical object；
Fig. 2 is using tetrahedron mesh generation enveloping surface to the zone profile figure between spherical object surface；
Fig. 3 is using hexahedral mesh subdivision enveloping surface to the zone profile figure between the boundary of zoning；
Fig. 4 is used alternative manner schematic diagram on the mesh generation and grid of entire zoning；
Fig. 5 is the alternative manner schematic diagram of grid near enveloping surface；
Fig. 6 is by the schematic diagram of alternative manner used on grid division zoning；
Fig. 7 is the schematic diagram that zoning is divided by grid syntople；
Fig. 8 is the dual station RCS calculated result of spherical object；
Fig. 9 is the flow diagram of the method for the present invention.
Specific embodiment
The present invention uses parallelization to there is bending, fine or incline structure extensive electromagnetic field numerical simulation problem
The mixed method of Finite-Difference Time-Domain Method and finite volume method (FDTD-FVTD) is related to two kinds of grids, two kinds of iterative manners when solving
The Parallel Scheme Design of numerical method.Zoning subdivision is hexahedral mesh by widely used Parallel FDTD Method, according to
The simple direct effect of the parallel scheme that virtual topology mode designs is good, but for there are the calculating of the two methods of two kinds of grids to ask
The design studies of topic, parallel scheme are insufficient.The alternative manner difference carried out the present invention is based on grid draws zoning
Point, it using process load balance as design object, reduces the data traffic between process to the greatest extent, improves parallel speedup ratio, Jin Erda
To before calculating use this method can get under Limited computational resources be suitable for FDTD-FVTD mixed method calculate it is optimal simultaneously
Row scheme.
As shown in Figures 1 to 8, the invention discloses a kind of parallelization time domain hybrid electromagnetic algorithm based on alternative manner,
To calculate a target in [fmin, fmax] in frequency range for RCS result, due to the limitation of computer capacity, numerical value is calculated
It can only be carried out in finite region.In order to simulate open domain electromagnetic problem, ABSORPTION EDGE must be used at the cutoff boundary of zoning
Boundary.Problem space determined by absorbing boundary condition is limited, be rationally arranged (present invention in mixed method continue to use
Classical perfect domination set PML in FDTD method) after the confined space can be equivalent with infinite space；This confined space is to count
Region is calculated, it is established in tri- directions along reference axis x, y, z, and the cuboid sky that including target, boundary is made of PML is included
Between, if its vertex bound in three directions is respectively [Lxmin,Lxmax], [Lymin,Lymax], [Lzmin,Lzmax].This method
Detailed step is as follows:
Step 1, frequency is solved according to the highest of target and calculates size of mesh opening；
For target in [fmin, fmax] radar cross section RCS in frequency range, it is calculated using the following equation FDTD method
Size of mesh opening Δ:
Δ=(c0/fmax)/10
Wherein, C0=3.0 × 108M/s is the light velocity.
Step 2, target surface is surrounded using closed square dough sheet enveloping surface；
The realization of mixed method needs entire zoning subdivision to be two kinds of grids, which is partially used in zoning
Tetrahedron mesh generation, which partially needs to determine before subdivision using hexahedral mesh subdivision.Therefore, the mesh of the step for
Be that zoning is divided into two parts, with the subdivision of two kinds of grids later.
According to size of mesh opening size (side length) Δ that step 1 obtains, three are carried out to target surface using mesh generation software
Angular subdivision, circulation triangular mesh dough sheet center point coordinate (the approximate geometric coordinate for obtaining target surface by this method) extrapolation
- 3 Δ distance of 2 Δ establishes the enveloping surface being made of the square that side length is Δ.The enveloping surface will be cutd open as grid in next step
The interface of timesharing tetrahedral grid and hexahedral mesh.Fig. 1 show the enveloping surface of spherical object model.
Step 3, zoning is divided into two parts: use tetrahedron mesh generation enveloping surface to the region of target surface with
Accurate Model is realized, using hexahedral mesh subdivision enveloping surface to the part on zoning boundary to save computing resource；
Entire zoning is divided into two parts: enveloping surface to target surface using the enveloping surface obtained in previous step
With enveloping surface to zoning boundary.
Step 3.1, the region between enveloping surface and target surface tetrahedral grid is carried out using mesh generation software to cut open
Point, tetrahedral grid can accurately describe the geometry of target, and only 2-3 Δ limits four sides to the distance between enveloping surface and target
The a wide range of use of volume mesh, so that grid amount is unlikely to excessively, this is an advantage of mixed method.As shown in Figure 2
It is the sectional view of tetrahedral grid by subdivision for enveloping surface to the region between spherical object surface.
Step 3.2, the part of enveloping surface periphery to zoning boundary is used into hexahedron net using mesh generation software
Lattice subdivision.It is cuing open for hexahedral mesh that spherical object enveloping surface periphery to zoning borderline region, which is illustrated in figure 3, by subdivision
Face figure.Another advantage of mixed method is that the grid used on a large scale in zoning is hexahedral mesh using simple direct
FDTD method iteration, taken into account computational efficiency.
So far, the mesh generation of entire zoning is completed, and carries out the same of one-dimensional number (1,2 ...) to hexahedral mesh
When calculate its three-dimensional number, if the coordinate of certain grid element center is (x, y, z), then the three-dimensional number of the grid is (i, j, k), specifically
Calculation formula are as follows:
WhereinRepresentative rounds up to its built-in variable.It is illustrated in figure 4 cuing open for the mesh generation of spherical object
Face schematic diagram.
It step 4,, will using equally loaded as principle according to alternative manner used on grid and consumed calculation amount
The node for participating in calculating is divided into two class of FDTD method iteration and FVTD method iteration, calculates the grid amount on each node；
Step 4.1, two methods of FDTD and FVTD disclose the local characteristics of field, and the mixing of the two is only needed at two kinds
Value exchange in field is carried out at method interface between nets can be realized.As shown in figure 5, all tetrahedral grids only carry out FVTD iteration, with
The adjacent hexahedral mesh of tetrahedral grid also only carries out FVTD iteration, and the hexahedral mesh for only carrying out FVTD iteration with these has
The hexahedral mesh in public face is referred to as transition region, and FDTD iteration had not only been carried out on this Partial Mesh but also has carried out FVTD iteration, has removed this
Except, the hexahedral mesh (including PML layers) outside transition region then only carries out FDTD iteration.It is whole in spherical object RCS computational problem
Used alternative manner is as shown in Figure 4 on each grid in a zoning.
Step 4.2, on node grid amount calculating
If the hexahedral mesh number for only carrying out FDTD method iteration is N1, only carry out the tetrahedron net of FVTD method iteration
Lattice number is N2, the hexahedral mesh number for only carrying out FVTD method iteration is N3, not only carried out FDTD iteration but also carried out FVTD iteration
Hexahedral mesh number be N4。
In FDTD method, needed on a hexahedral mesh unit 6 field values (three components of electric field: Ex, Ey, Ez with
And three components in magnetic field: Hx, Hy, Hz), realizing that an iteration needs the calculation times carried out is about 7 × 6.FVTD method
In, when being reconstructed using Second-order Up-wind format, consider that the syntople between unit, a tetrahedron element realize an iteration
The average computation number needed are as follows: 97 × 6, a hexahedral element realizes that the average computation number that an iteration needs is 145
×6。
Therefore, if the calculation amount on all grids in FDTD iteration region is C1, its calculation formula is:
C1=7 × 6 × N1
Rest part grid computing amount C2Calculation formula are as follows:
C2=97 × 6 × N2+145×6×N3+(145+7)×6×N4
Step 4.3, if the node number for participating in parallel computation is M (number is respectively as follows: 0,1,2 ... M-1), consider between grid
Syntople and grid on used alternative manner, by all grids according to the calculation amount distribution node of estimation:
It is set as FDTD iteration region and distributes D node, its calculation formula is:
D=[C1×M/(C1+C2)]
It is set as FVTD iteration region and distributes V node, its calculation formula is:
V=[C2×M/(C1+C2)]
Meet relationship between D and V: D+V=M, wherein [], which represents, carries out round to its built-in variable.
By N1A hexahedral mesh for participating in FDTD iteration is evenly distributed on 0-D-1 node, is divided on each node
Cloth about N1/ D grid；By (N2+N3+N4) it is a participate in FVTD iteration net lattice control be distributed on D-M-1 node, each
About (N is distributed on node2+N3+N4The grid of)/V.
The step has determined the number of grids being distributed on each node, and calculation amount can be made to be evenly distributed on each node,
Achieve the effect that load balancing.
Step 5, the data communication between node is estimated according to the adjacent mesh physical quantity that iteration normal on grid needs to use
Grid is assigned to realization Optimal Parallel scheme on specific node with data traffic at least for design object by amount.
There are the volume mesh in public face to be referred to as adjacent mesh.It needs to use in adjacent mesh when the normal iteration of one grid
Field amount and other physical quantitys, when grid iteration on node interface needs its adjacent mesh (to be distributed in the adjacent of the node
On node) field value and physical quantity need these values and physical quantity passing through communication mode and send out to be normally carried out iteration
It send.Data traffic is more, and the ratio that call duration time accounts for the calculating time is bigger, seriously affects parallel speedup ratio, therefore, is counting
Estimated data's traffic will be conducive to the design of Optimal Parallel scheme before calculating.
It is located at each grid on node interface in FDTD method and needs to send 2 field Value Datas, while receives 2 fields
Value Data.It is located at each grid at node interface in FVTD method and needs to send 13 values, while receives 13 values.Using
The interface between nets face value traffic of the interface between nets face value traffic of FVTD method iteration much larger than FDTD method iteration.Cause
This, the child node division numbers for reducing FVTD method iteration region can substantially reduce the consumption of call duration time.
Step 5.1, three kinds of situations of calculating point of data traffic or less:
1. the grid being distributed on node i (0≤i≤D-1) and its adjacent node j (0≤j≤D-1, and i ≠ j) carries out
FDTD method iteration, if having adjacent mesh n on two node public boundaries1It is a, the data traffic of the Partial Mesh are as follows: 2 × n1
×4。
2. the grid being distributed in node i (0≤i≤D-1) carries out FDTD method iteration, if its adjacent node j meets (D≤j
≤ M-1, and i ≠ j), i.e. grid on node j carries out FVTD method iteration, if having adjacent mesh on two node public boundaries
There is n2It is a, the data traffic of the Partial Mesh are as follows: 2 × n2×4。
3. the grid progress FVTD method iteration being distributed in node i (D≤i≤M-1), adjacent node j (D≤j≤M-1, and
I ≠ j) on grid also carry out FVTD method iteration, if thering is adjacent mesh to have n on two node public boundaries3It is a, the part net
The data traffic of lattice are as follows: 2 × n3×26。
The design of parallel scheme will be minimum with the data traffic summation of all grids on node interface in next step
As design object.
Step 5.2, the design of parallel scheme
The number of grids on each node being calculated according to step 4.2 is at least that target will be every with field value traffic summation
A grid be assigned on specific node be Parallel Scheme Design final step.Specific allocation plan is as follows:
The allocation plan of step 5.2.1, FDTD iteration area grid:
0-D-1 node is subjected to three-dimensional topology model construction, that is, sets a node of x directional spreding, y directional spreding b is a
Node, c node of z directional spreding, a, b, c is integer, and is determined by following two condition:
a:b:c≈(Lxmax-Lxmin):(Lymax-Lymin):(Lzmax-Lzmin)
A × b × c=D
The three-dimensional topology coordinate of node are as follows: (0,0,0), (1,0,0) ... (a-1, b-1, c-1).
Center point coordinate (x, y, z) is met into condition:
Lxmin+(o-1)Δ≤x≤Lxmin+oΔ
Lymin+(p-1)Δ≤y≤Lymin+pΔ
Lzmin+(q-1)Δ≤z≤LzminThe grid of+q Δ is distributed in advance on the node that topological coordinate is (o, p, q)；
As shown in fig. 6, according to 3 calculating sections of x directional spreding when calculating spherical object dual station RCS using 6 calculate nodes
Point, 2 calculate nodes of y directional spreding, the grid distribution of 1 calculate node of z directional spreding and the topological coordinate of each node.
Consider that the number of grids being distributed on node wants balanced, by step 4.3 it is found that N will be distributed on each node1/ D net
Lattice.Due to being not fully the hexahedral mesh of proper alignment in zoning, some tetrahedral grid, therefore by opening up
Flutter on each node of structure distribution the number of grids and uneven, it is also necessary to it adjusts, so that the calculation amount on each node is balanced:
Step 5.2.2, starts from (0,0,0) topological node, traverses each node and executes following steps:
If the number of grids being distributed in advance on node is greater than N1Grid on the node boundary face is then fully allocated to phase by/D
On neighbors；
If the number of grids being distributed in advance on node is less than N1Grid in adjacent node boundary face is then fully allocated to by/D
On the node.
Step 5.2.3, by step 4.3 it is found that (N will be distributed in node i (D≤i≤M-1)2+N3+N4The FVTD iterative region of)/V
The grid in domain is distributed to node in such a way that syntople traverses to the periphery by numbering small grid according to grid number order
On D-M-1；
Finally, according to the data traffic calculation estimation different topology structure (i.e. different a, b, c combination) of step 5.1
Data communication total amount between lower node selects the smallest allocation plan of data traffic for final parallel scheme.
For the RCS computational problem of spherical object, when such as being calculated using the parallel FDTD-FVTD mixed method of 6 nodes,
It is illustrated in figure 7 and divides sectional view using the zoning that parallel scheme proposed by the present invention obtains.According to used by grid
Tetrahedral grid region and hexahedral mesh region are respectively divided alternative manner, and this method considers tetrahedral grid and six faces
Influence of the difference and data traffic that volume mesh consumes computing resource to parallel efficiency.
As shown in fig. 6, the interface between nets face of different nodes has 3 inside FVTD iteration region.Fig. 6 is according to grid point
All net lattice controls are distributed on all nodes by cloth, alternative manner used by not considering grid according to syntople merely
Grid dividing situation, the interface between nets face of different nodes has 6 inside FVTD iteration region.It can by the 3rd kind of situation of step 5.2
Know, compared to first two situation, the increase of FVTD iteration Area Node interface can greatly increase data traffic, influence parallel
Efficiency.
Step 6, it is iteratively solved using parallel FDTD-FVTD mixed method, obtains calculated result.
Grid is distributed in M calculate node by the parallel scheme obtained using step 5, then uses parallel FDTD-
The iterative solution of FVTD mixed method.The source of the parallel FDTD-FVTD mixed method are as follows: He Zhili, based on the parallel of GiD
FDTD and its mixed method research, doctoral thesis, in June, 2012.
If Fig. 8 is that the spherical object that radius is 1m is counted in the dual station RCS calculated result that frequency is 1GHz with moment method (MOM)
Result is calculated to coincide well.
Claims (8)
1. a kind of parallelization time domain hybrid electromagnetic algorithm based on alternative manner, which comprises the following steps:
Step 1, frequency is solved according to the highest of target and calculates size of mesh opening；
Step 2, target surface is surrounded using closed square dough sheet enveloping surface；
Step 3, zoning is divided into two parts: uses tetrahedron mesh generation enveloping surface to the region of target surface to realize
Accurate Model, using hexahedral mesh subdivision enveloping surface to the part on zoning boundary to save computing resource；
Step 4, it according to alternative manner used on grid and consumed calculation amount, using equally loaded as principle, will participate in
The node of calculating is divided into different iteration classifications, calculates the grid amount on each node；
Step 5, the data traffic between node is estimated according to the adjacent mesh physical quantity that iteration normal on grid needs to use, with
Data traffic is at least design object, and grid is assigned to realization Optimal Parallel scheme on specific node；
Step 6, it is iteratively solved using parallel FDTD-FVTD mixed method, obtains calculated result.
2. the parallelization time domain hybrid electromagnetic algorithm based on alternative manner as described in claim 1, which is characterized in that described
Frequency, which is solved, according to the highest of target calculates size of mesh opening, method are as follows:
For target in [fmin, fmax] radar cross section RCS in frequency range, it is calculated using the following equation the net of FDTD method
Lattice size Δ:
Δ=(c0/fmax)/10
Wherein, C0=3.0 × 108M/s is the light velocity.
3. the parallelization time domain hybrid electromagnetic algorithm based on alternative manner as described in claim 1, which is characterized in that described
Target surface is surrounded using closed square dough sheet enveloping surface, comprising:
According to the size of size of mesh opening, triangular mesh subdivision is carried out to target surface, recycles triangular mesh dough sheet central point
Coordinate 2-3 size of mesh opening of extrapolation establishes the enveloping surface being made of the square dough sheet that side length is the size of mesh opening size.
4. the parallelization time domain hybrid electromagnetic algorithm based on alternative manner as described in claim 1, which is characterized in that will participate in
The node of calculating is divided into different iteration classifications, comprising:
All tetrahedral grids only carry out FVTD iteration, and the hexahedral mesh adjacent with tetrahedral grid also only carries out FVTD and change
Generation, the hexahedral mesh for only carrying out FVTD iteration with these have the hexahedral mesh in public face to be referred to as transition region, this part net
Not only FDTD iteration had been carried out on lattice but also has carried out FVTD iteration, and in addition to this, the hexahedral mesh outside transition region then only carries out FDTD and changes
Generation.
5. the parallelization time domain hybrid electromagnetic algorithm based on alternative manner as described in claim 1, which is characterized in that described
Calculate the grid amount on each node, comprising:
If the hexahedral mesh number for only carrying out FDTD method iteration is N1, only carry out the tetrahedral grid of FVTD method iteration
Number is N2, the hexahedral mesh number for only carrying out FVTD method iteration is N3, not only carried out FDTD iteration but also carried out the six of FVTD iteration
Face volume mesh number is N4, then:
Calculation amount on all grids in FDTD iteration region is C1=7 × 6 × N1；
Calculation amount C on rest part grid2=97 × 6 × N2+145×6×N3+(145+7)×6×N4。
6. the parallelization time domain hybrid electromagnetic algorithm based on alternative manner as described in claim 1, which is characterized in that step 4
Further include: by all grids according to the calculation amount distribution node of estimation:
D node is distributed for FDTD iteration region, its calculation formula is:
D=[C1×M/(C1+C2)]
V node is distributed for FVTD iteration region, its calculation formula is:
V=[C2×M/(C1+C2)]
Meet relationship between D and V: D+V=M, M are node total number；
By N1A hexahedral mesh for participating in FDTD iteration is evenly distributed on 0-D-1 node；By (N2+N3+N4) a participation
The net lattice control of FVTD iteration is distributed on D-M-1 node.
7. the parallelization time domain hybrid electromagnetic algorithm based on alternative manner as described in claim 1, which is characterized in that described
Data traffic between the adjacent mesh physical quantity estimation node for needing to use according to iteration normal on grid, comprising:
The grid being distributed on node i (0≤i≤D-1) and its adjacent node j (0≤j≤D-1, and i ≠ j) carries out FDTD method
Iteration, if having adjacent mesh n on two node public boundaries1It is a, the data traffic of the Partial Mesh are as follows: 2 × n1×4；
The grid being distributed in node i (0≤i≤D-1) carries out FDTD method iteration, if its adjacent node j satisfaction (D≤j≤M-1,
And i ≠ j), i.e. grid on node j carries out FVTD method iteration, if having adjacent mesh to have n on two node public boundaries2It is a,
The data traffic of the Partial Mesh are as follows: 2 × n2×4；
The grid being distributed in node i (D≤i≤M-1) carries out FVTD method iteration, adjacent node j (D≤j≤M-1, and i ≠ j)
On grid also carry out FVTD method iteration, if thering is adjacent mesh to have n on two node public boundaries3It is a, the Partial Mesh
Data traffic are as follows: 2 × n3×26。
8. the parallelization time domain hybrid electromagnetic algorithm based on alternative manner as described in claim 1, which is characterized in that described
With data traffic at least for design object, grid is assigned to realization Optimal Parallel scheme on specific node, comprising:
0-D-1 node is subjected to three-dimensional topology model construction, that is, sets a node of x directional spreding, b node of y directional spreding,
C node of z directional spreding, a, b, c are integer, and are determined by following two condition:
a:b:c≈(Lxmax-Lxmin):(Lymax-Lymin):(Lzmax-Lzmin)
A × b × c=D
The three-dimensional topology coordinate of node are as follows: (0,0,0), (1,0,0) ... (a-1, b-1, c-1)；
Center point coordinate (x, y, z) is met into condition:
Lxmin+(o-1)Δ≤x≤Lxmin+oΔ
Lymin+(p-1)Δ≤y≤Lymin+pΔ
Lzmin+(q-1)Δ≤z≤Lzmin+qΔ
Grid be distributed in advance topological coordinate be (o, p, q) node on；
Start from (0,0,0) topological node, traverse each node and execute following steps:
If the number of grids being distributed in advance on node is greater than N1Grid on the node boundary face is then fully allocated to adjacent segments by/D
Point on；
If the number of grids being distributed in advance on node is less than N1Grid in adjacent node boundary face is then fully allocated to the section by/D
Point on；
Node D-M- is distributed in such a way that syntople traverses to the periphery by numbering small grid according to grid number order
On 1；
Estimate different topology structure lower node between data traffic, select the smallest allocation plan of data traffic for finally
Parallel scheme.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811539924.3A CN109766588B (en) | 2018-12-17 | 2018-12-17 | Parallelization time domain hybrid electromagnetic method based on iteration method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811539924.3A CN109766588B (en) | 2018-12-17 | 2018-12-17 | Parallelization time domain hybrid electromagnetic method based on iteration method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109766588A true CN109766588A (en) | 2019-05-17 |
| CN109766588B CN109766588B (en) | 2022-10-14 |
Family
ID=66450818
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811539924.3A Active CN109766588B (en) | 2018-12-17 | 2018-12-17 | Parallelization time domain hybrid electromagnetic method based on iteration method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109766588B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079357A (en) * | 2019-10-22 | 2020-04-28 | 重庆邮电大学 | Electromagnetic interference time domain analysis method of microstrip line on PCB under action of electromagnetic wave |
| CN112581625A (en) * | 2020-12-28 | 2021-03-30 | 中国科学院电工研究所 | Method for embedding overlapped grid boundary |
| CN113720244A (en) * | 2021-09-01 | 2021-11-30 | 航宇智造（北京）工程技术有限公司 | Strain testing method based on augmented reality technology |
| CN114528742A (en) * | 2022-04-20 | 2022-05-24 | 广州中望龙腾软件股份有限公司 | Wave port modeling and S parameter calculation method based on CFDTD algorithm, terminal and medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2287757A1 (en) * | 2009-08-20 | 2011-02-23 | Fujitsu Limited | Multilevel-Multigrid simulation techniques |
| CN106446470A (en) * | 2016-11-17 | 2017-02-22 | 上海无线电设备研究所 | Efficient concurrent inhomogeneous medium frequency domain finite difference method |
| CN107845141A (en) * | 2017-11-27 | 2018-03-27 | 山东大学 | A kind of transient electromagnetic three-dimensional FDTD forward modeling multi-resolution meshes subdivision methods |
-
2018
- 2018-12-17 CN CN201811539924.3A patent/CN109766588B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2287757A1 (en) * | 2009-08-20 | 2011-02-23 | Fujitsu Limited | Multilevel-Multigrid simulation techniques |
| CN106446470A (en) * | 2016-11-17 | 2017-02-22 | 上海无线电设备研究所 | Efficient concurrent inhomogeneous medium frequency domain finite difference method |
| CN107845141A (en) * | 2017-11-27 | 2018-03-27 | 山东大学 | A kind of transient electromagnetic three-dimensional FDTD forward modeling multi-resolution meshes subdivision methods |
Non-Patent Citations (1)
| Title |
|---|
| 葛德彪等: "各向异性介质FDTD分析及其并行计算", 《系统工程与电子技术》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079357A (en) * | 2019-10-22 | 2020-04-28 | 重庆邮电大学 | Electromagnetic interference time domain analysis method of microstrip line on PCB under action of electromagnetic wave |
| CN112581625A (en) * | 2020-12-28 | 2021-03-30 | 中国科学院电工研究所 | Method for embedding overlapped grid boundary |
| CN112581625B (en) * | 2020-12-28 | 2023-11-10 | 中国科学院电工研究所 | Method for embedding overlapped grid boundary |
| CN113720244A (en) * | 2021-09-01 | 2021-11-30 | 航宇智造（北京）工程技术有限公司 | Strain testing method based on augmented reality technology |
| CN113720244B (en) * | 2021-09-01 | 2024-03-22 | 航宇智造（北京）工程技术有限公司 | Strain testing method based on augmented reality technology |
| CN114528742A (en) * | 2022-04-20 | 2022-05-24 | 广州中望龙腾软件股份有限公司 | Wave port modeling and S parameter calculation method based on CFDTD algorithm, terminal and medium |
| CN114528742B (en) * | 2022-04-20 | 2022-08-02 | 广州中望龙腾软件股份有限公司 | Wave port modeling and S parameter calculation method based on CFDTD algorithm, terminal and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109766588B (en) | 2022-10-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109766588A (en) | A kind of parallelization time domain hybrid electromagnetic algorithm based on alternative manner | |
| AU2019442319B2 (en) | Structural topology optimization method based on material-field reduction series expansion | |
| CN112989680B (en) | FVFD far-field integration boundary condition calculation method for reducing grid usage | |
| CN110069800B (en) | Three-dimensional structure topology optimization design method and equipment with smooth boundary expression | |
| CN103428282B (en) | Online energy-saving control method and the device of a kind of cloud computing data center | |
| CN107067091B (en) | Urban ecological land space planning model based on ant colony optimization algorithm | |
| Wang et al. | SSIEA: a hybrid evolutionary algorithm for supporting conceptual architectural design | |
| CN110348068A (en) | Method and device for calculating shielding efficiency of electromagnetic shielding net and storage medium | |
| CN109614638B (en) | Indirect modeling urban wind environment CFD simulation method | |
| CN108268696A (en) | A kind of FSS antenna house modeling methods suitable for high order MoM | |
| CN106981097A (en) | A kind of T spline surface approximating methods based on subregion Local Fairing weight factor | |
| CN106227599A (en) | The method and system of scheduling of resource in a kind of cloud computing system | |
| CN108647777A (en) | A kind of data mapped system and method for realizing that parallel-convolution calculates | |
| CN109726433A (en) | The three-dimensional method for numerical simulation streamed without viscous low speed based on surface boundary condition | |
| CN106777517A (en) | Aero-engine high-pressure turbine disk Optimum Design System and method based on population | |
| CN108196951A (en) | GPU runoff simulations distributed scheduling system and method | |
| CN113221370B (en) | FSS radome modeling method based on conical surface projection | |
| CN112766609A (en) | Power consumption prediction method based on cloud computing | |
| CN105205299B (en) | The quick Dimension Reduction Analysis method of TV university electromagnetic characteristic of scattering | |
| CN112765884B (en) | Phononic crystal material structure design method based on extended isogeometric analysis method | |
| CN103391557B (en) | A kind ofly automatically get the method for determining domestic aerial demand and layout based on grid | |
| CN109117545A (en) | Antenna Fast design method neural network based | |
| CN111046542B (en) | Method for evaluating LE grid computing characteristics of icosahedron and dispersion method | |
| CN103440540B (en) | A kind of parallel method of land utilization space layout artificial immunity Optimized model | |
| CN110349265B (en) | Tetrahedral topological mesh generation method and electronic equipment |
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 |