CN109447245A - Equivalent model generation method and modeling method neural network based - Google Patents
Equivalent model generation method and modeling method neural network based Download PDFInfo
- Publication number
- CN109447245A CN109447245A CN201811268685.2A CN201811268685A CN109447245A CN 109447245 A CN109447245 A CN 109447245A CN 201811268685 A CN201811268685 A CN 201811268685A CN 109447245 A CN109447245 A CN 109447245A
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- equivalent model
- target object
- training
- 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.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 108
- 238000000034 method Methods 0.000 title claims abstract description 75
- 238000012549 training Methods 0.000 claims abstract description 54
- 238000013507 mapping Methods 0.000 claims abstract description 48
- 230000000153 supplemental effect Effects 0.000 claims abstract description 37
- 238000005259 measurement Methods 0.000 claims abstract description 17
- 230000008569 process Effects 0.000 claims description 22
- 210000005036 nerve Anatomy 0.000 claims description 18
- 230000008859 change Effects 0.000 claims description 7
- 239000000463 material Substances 0.000 claims description 7
- 210000004218 nerve net Anatomy 0.000 claims description 2
- 238000003062 neural network model Methods 0.000 claims 1
- 238000004088 simulation Methods 0.000 abstract description 10
- 238000013461 design Methods 0.000 description 28
- 238000010586 diagram Methods 0.000 description 18
- 241000208340 Araliaceae Species 0.000 description 6
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 6
- 235000003140 Panax quinquefolius Nutrition 0.000 description 6
- 238000004590 computer program Methods 0.000 description 6
- 235000008434 ginseng Nutrition 0.000 description 6
- 238000012545 processing Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000013506 data mapping Methods 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 239000010409 thin film Substances 0.000 description 2
- 230000002463 transducing effect Effects 0.000 description 2
- 241000406668 Loxodonta cyclotis Species 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000010408 film Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
Classifications
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the present application provides a kind of equivalent model generation method neural network based and modeling method, wherein, equivalent model generation method neural network based comprises determining that the corresponding sample data of target object, sample data include the multiple groups supplemental characteristic and actual measurement working characteristic data corresponding with every group of supplemental characteristic of target object;According to sample data training neural network, so that the mapping relations between neural network learning supplemental characteristic and working characteristic data;The equivalent model of target object is determined according to the neural network that training is completed.Equivalent model by schemes generation provided by the embodiments of the present application is more accurate, and obtained equivalent model can be adapted for a variety of simulation softwares, i.e., the equivalent model that the application obtains is limited lower by simulation software.
Description
Technical field
The invention relates to field of artificial intelligence more particularly to a kind of equivalent model neural network based are raw
At method and modeling method.
Background technique
The component converted for electromagnetic wave and mechanical wave is commonly known as energy converter, and general energy converter is divided into several classes
Not, one kind is the energy converter (such as loudspeaker) for converting electromagnetic wave to mechanical wave, and another kind of is to convert electromagnetic wave for mechanical wave
Energy converter (such as microphone);It in addition to this, further include that can carry out electromagnetic wave-mechanical wave-electromagnetic wave conversion transducing
Device (such as SAW and FBAR) etc..
Under normal conditions, when designing energy converter, the every of energy converter sample first is tested out using related equipment and is joined
Then number establishes equivalent model to simulate the parameters of energy converter, then by design of Simulation work according to the parameters tested out
Tool is calculated, continuous to adjust equivalent model if calculated result is unsatisfactory for design objective, until calculated result satisfaction is set
Count index.It is typical design cycle above, this process is scattered, and is manually implemented.
However, existing equivalent model is not sufficient to meet the needs of design.Such as design thin film bulk acoustic wave resonator
When (film bulk acoustic resonator, FBAR), using equivalent model can be equivalent-circuit model, equivalent mathematical
Model etc., wherein the equivalent-circuit model of FBAR can specifically include Mason model, MBVD model.But Mason model master
Thin film bulk acoustic wave resonator to be described using material parameter and physical structure, but Mason model and existing circuit simulation
Design software is simultaneously incompatible, and MBVD model describes FBAR using 36 lumped-parameter elements, but it can only simulate resonance frequency
Impedance operator near rate can not provide the characteristic impedance of entire frequency domain.
In view of this, the technical issues of urgent need to resolve, is to provide a kind of new equivalent model.
Summary of the invention
In view of this, the main purpose of the present invention is to provide a kind of equivalent model generation method neural network based with
And modeling method, solve the problems, such as that equivalent model in the prior art is insufficient for design requirement.
In a first aspect, the embodiment of the present application provides a kind of equivalent model generation method neural network based comprising:
Determine the corresponding sample data of target object, the sample data include target object multiple groups supplemental characteristic and with every group of ginseng
The corresponding actual measurement working characteristic data of number data;According to sample data training neural network, so that the Neural Network Science
Practise the mapping relations between the supplemental characteristic and the working characteristic data;It is determined according to the neural network that training is completed
The equivalent model of the target object.
Second aspect, the embodiment of the present application provide a kind of equivalent model generating means neural network based comprising:
Sample determining module, for determining that the corresponding sample data of target object, the sample data include the multiple groups ginseng of target object
Number data and actual measurement working characteristic data corresponding with every group of supplemental characteristic;Training module, for according to the sample data
Training neural network, so that the mapping between supplemental characteristic described in the neural network learning and the working characteristic data is closed
System;Model determining module, the neural network for being completed according to training determine the equivalent model of the target object.
The third aspect, the embodiment of the present application provide a kind of computer-readable medium, are stored thereon with equivalent model, described
Equivalent model is to determine that the neural network is used for the basic parameter of learning objective object according to the neural network that training is completed
Mapping relations between the working characteristic data of the target object.
Fourth aspect, the embodiment of the present application provide a kind of modeling method comprising:
Raw process parameter data is input to equivalent model, so that the initial parameter number of the equivalent model according to input
According to output prediction work performance data, wherein the equivalent model is determined according to the neural network that training is completed, the mind
Through network for the mapping relations between the basic parameter of learning objective object and the working characteristic data of the target object;If
Difference between the prediction work performance data and target operation performance data then adjusts the original ginseng outside preset range
Number data, so that the equivalent model updates the prediction work performance data according to the raw process parameter data adjusted;
Alternatively, if the difference between the prediction work performance data and the target operation performance data is in preset range,
According to the current corresponding basic parameter of the prediction work performance data, the product model of the target object is established.
5th aspect, the embodiment of the present application provide a kind of modeling method comprising: working characteristic data is input to
Model is imitated, so that the equivalent model exports basic parameter data according to the working characteristic data of input, wherein described etc.
Effect model be according to training complete neural network determine, the neural network for learning objective object basic parameter and
Mapping relations between the working characteristic data of the target object;According to the basic parameter data of output, described in foundation
The product model of target object.
From the foregoing, it will be observed that passing through supplemental characteristic described in neural network learning and the work in scheme provided herein
Then mapping relations between performance data determine the equivalent mould of the target object according to the neural network that training is completed
Type, so that equivalent model is that the neural network completed based on training is determined, and while training neural network is to pass through supplemental characteristic
It is trained, so that obtained equivalent model is more accurate;And the output of neural network is adjustable, so that is obtained is equivalent
Model can be adapted for a variety of simulation softwares, i.e., the equivalent model that the application obtains is limited lower by simulation software.
Detailed description of the invention
The some specific of the embodiment of the present application is described in detail by way of example and not limitation with reference to the accompanying drawings hereinafter
Embodiment.Identical appended drawing reference denotes same or similar part or part in attached drawing.Those skilled in the art should manage
Solution, the drawings are not necessarily drawn to scale.In attached drawing:
Fig. 1 is a kind of equivalent model generation method process signal neural network based that the embodiment of the present application one provides
Figure;
Fig. 2 is a kind of equivalent model generation method flow diagram for basic unit that the embodiment of the present application two provides;
Fig. 3 is a kind of equivalent model generation method flow diagram for target product that the embodiment of the present application three provides;
Fig. 4 a is a kind of electrical block diagram for energy converter that the embodiment of the present application three provides;
Fig. 4 b is a kind of characteristic working curve for energy converter that the embodiment of the present application three provides;
Fig. 5 is a kind of structural representation for equivalent model generating means neural network based that the embodiment of the present application four provides
Figure;
Fig. 6 is a kind of flow diagram for modeling method that the embodiment of the present application six provides;
Fig. 7 is a kind of flow diagram for modeling method that the embodiment of the present application seven provides.
Specific embodiment
Any technical solution for implementing the embodiment of the present invention must be not necessarily required to reach simultaneously above all advantages.
In order to make those skilled in the art more fully understand the technical solution in the embodiment of the present invention, below in conjunction with the present invention
Attached drawing in embodiment, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described reality
Applying example only is a part of the embodiment of the embodiment of the present invention, instead of all the embodiments.Based on the implementation in the embodiment of the present invention
The range of protection of the embodiment of the present invention all should belong in example, those of ordinary skill in the art's every other embodiment obtained.
Below with reference to attached drawing of the embodiment of the present invention the embodiment of the present invention will be further explained specific implementation.
Fig. 1 is a kind of equivalent model generation method process signal neural network based that the embodiment of the present application one provides
Figure, as shown in Figure 1, comprising:
S11, the corresponding sample data of target object is determined.Wherein, the sample data includes the multiple groups ginseng of target object
Number data and actual measurement working characteristic data corresponding with every group of supplemental characteristic.
In the present embodiment, target object is the object of equivalent model to be generated, concrete kind of the present embodiment to target object
Type without limit, as long as it needs to generate equivalent model, be specifically as follows electronic component equivalent model or its
His mechanical, material equivalent model.
In the present embodiment, the sample data of target object is the data of existing target object, and sample data can be direct
It is determined, can also be determined by measuring the sample of target object according to the historical data of target object.For example, when design FBAR,
The data of the FBAR of completion will can directly have been designed as sample data;Can with the multiple FBAR entity samples of pre-production,
Then FBAR entity sample is measured, using measurement result as sample data, the entity sample standard deviation of each FBAR can be true
One characteristic working curve of fixed one group of basic parameter and its corresponding FBAR.
Specifically, the supplemental characteristic for including in sample data can be the parameter pair for influencing the working characteristics of target object
The data answered.For example, supplemental characteristic is specifically as follows the ginseng such as each thickness degree of FBAR, material property when target object is FBAR
The corresponding value of number;When target object is electronic product, supplemental characteristic can be the corresponding value of topological structure etc. of electronic product.
S12, neural network is trained according to the sample data, so that supplemental characteristic described in the neural network learning and institute
State the mapping relations between working characteristic data.
Specifically, the mapping relations between the supplemental characteristic that neural network is learnt and working characteristics, can specifically include
By the mapping relations of maps parameter data to working characteristic data, or working characteristic data is mapped to supplemental characteristic
Mapping relations, the present embodiment is to this without limiting.
Below by taking neural network learning is by maps parameter data to working characteristic data as an example, to the training of neural network into
Row explanation, the training method of neural network can have following three kinds.
A) using the corresponding target characteristic data of every group of supplemental characteristic as its label, the training mind by way of supervised learning
Through network.
B) it determines the delta data between multiple groups supplemental characteristic, and determines that the corresponding work of the delta data of supplemental characteristic is special
Property data delta data, then as sample, the training neural network by way of enhancing study.
C) noise data determined according to sample data and basic model data is input to the generator of neural network
Network, generator network generate reconstruction model;The basic model of target object and reconstruction model are input to neural network
Arbiter network, so that arbiter network updates generator network according to basic model and reconstruction model, to complete nerve
The training of generator network in network.
S13, the equivalent model that the target object is determined according to the neural network that training is completed.
In the present embodiment, after the completion of neural metwork training, the neural network that can complete training is as target object
Equivalent model;Or handled on the basis of the neural network that training is completed, determine the equivalent model of target object.
In the present embodiment, by determining the corresponding sample data of target object, the sample data includes target object
Multiple groups supplemental characteristic and actual measurement working characteristic data corresponding with every group of supplemental characteristic;According to sample data training nerve
Network, so that the mapping relations between supplemental characteristic described in the neural network learning and the working characteristic data;According to instruction
Practice the equivalent model that the neural network completed determines the target object, so that equivalent model is the mind completed based on training
It when being determined through network, and training neural network is trained by supplemental characteristic, so that obtained equivalent model is more
Accurately;And the output of neural network is adjustable, and so that obtained equivalent model can be adapted for a variety of simulation softwares, i.e. this Shen
The equivalent model that please be obtained is limited lower by simulation software.
In addition, existing target object can be divided into two classes, one kind is primary element, what one kind was made of primary element
Target product, in the following, the method flow for the equivalent model that the application generates primary element by two pairs of embodiment is illustrated, leads to
The method flow for crossing the equivalent model of three pairs of embodiment generation target products is illustrated.
Embodiment two
The present embodiment carries out specifically the method flow for generating equivalent model so that target object is primary element as an example
It is bright.As shown in Fig. 2, the equivalent model generation method of basic unit provided in this embodiment includes:
S21, the corresponding first sample data of the primary element are determined.
Wherein, in the first sample data include multiple groups described in primary element physical parameter and every group described in physics
The corresponding actual measurement working characteristic data of parameter.In the present embodiment, the physical parameter of primary element may include the material of primary element
Expect characteristic, style characteristic etc..
Specifically, the primary element may include basic inverting element, the physical parameter of the inverting element include with
It is at least one lower: the structural parameters of the inverting element, material parameter, the working characteristic data packet of the basic inverting element
It includes: the change data of the input and output wave of the inverting element.
S22, first nerves network is trained according to the first sample data.
Wherein, the first nerves network is for learning between the physical parameter and the corresponding working characteristic data
Mapping relations.
Specifically, same as the previously described embodiments, the mapping relations of the first nerves e-learning in the present embodiment can
To include at least one: the mapping relations of map physical parameters to working characteristic data, working characteristic data map to object
Manage the mapping relations of parameter.
Specifically, in the present embodiment, when primary element is basic inverting element, the working characteristics of the basic inverting element
Data include: the change data of the input and output wave of the inverting element, corresponding, and first nerves network is for learning the object
Mapping relations between reason parameter and the corresponding working characteristic data include: that first nerves network is described basic for learning
Mapping relations between the physical parameter of inverting element and the change data of input and output wave.
S23, the equivalent model that the primary element is determined according to the first nerves network that training is completed.
In the present embodiment, if first nerves e-learning to be map physical parameters close to the mapping of working characteristic data
System then can obtain more accurate Adjusted Option according to the equivalent model determined based on first nerves network.For example, if base
This element is FBAR, and the physical parameter of FBAR may include: the electrode of the thickness of each layer of FBAR, the material property of FBAR, FBAR
Shape, the electrode size of FBAR, external cavity volume of FBAR etc., the working characteristic data of FBAR may include the resonance of FBAR
Frequency etc..The influence of the volume change of FBAR to the resonance frequency of FBAR can be determined by the learning process of first nerves network
Maximum then can be determining according to the mapping relations of first nerves e-learning, can be quickly by adjusting the volume of FBAR
The resonance frequency for adjusting FBAR, to obtain more accurate Adjusted Option.
If first nerves e-learning is mapping relations that working characteristic data maps to physical parameter, can be direct
Corresponding target operation characteristic is determined according to design objective, and the equivalent model then determined by neural network directly determines institute
The physical parameter needed, and then the design of primary element can be completed, greatly improve design efficiency.
Embodiment three
The present embodiment carries out specifically the method flow for generating equivalent model so that target object is target product as an example
It is bright, wherein to include multiple primary elements in the target product.As shown in figure 3, target product provided in this embodiment is equivalent
Model generating method includes:
S31, corresponding second sample data of the target product is determined.
Wherein, second sample data includes the multiple data splittings and each data splitting pair of the target product
The actual measurement working characteristic data answered, the data splitting are the data splitting of multiple primary elements in the target product.
In the present embodiment, data splitting refers to the corresponding data splitting of multiple primary element composition target products, number of combinations
According to may include at least one: the position where the equivalent model of each primary element, each primary element in target product
It sets, the connection relationship between each primary element etc..
In the present embodiment, the equivalent of the method generation provided according to embodiment two is can be used in the equivalent model of primary element
Other mathematical models etc. also can be used in model, and the present embodiment is to this without limiting.Where primary element in target product
Position be specifically as follows the data such as its number of plies in target product, coordinate.
Specifically, in the present embodiment, target product can be electronic product, then the primary element for including in electronic product is
Electronic component;Corresponding, the data splitting includes at least one: each electronic component that electronic product includes
Equivalent model data, the position in the electronic product of electronic product each electronic component for including, electronics
The connection relationship between each electronic component that product includes.
For example, electronic component is energy converter, it may include multiple FBAR resonators in energy converter, in conjunction with Fig. 4 a it is found that changing
The base components that can include in device may include four FBAR, respectively X1, X2, X3, X4, wherein (1) indicate input port,
(2) output port is indicated.Then the circuit diagram shown in Fig. 4 a can be used as one group of data splitting of energy converter, this group of data splitting pair
The actual measurement working characteristic data answered can be the corresponding actual measurement characteristic working curve of circuit diagram shown in Fig. 4 b.Similar, may be used also
Using determine energy converter other circuit diagrams and corresponding actual measurement characteristic working curve as data splitting and actual measurement work
Performance data.
S32, nervus opticus network is trained according to second sample data.
The nervus opticus network is for learning reflecting between the data splitting and working characteristic data of the target product
Penetrate relationship.
Specifically, same as the previously described embodiments, the mapping relations of the nervus opticus e-learning in the present embodiment can
To include at least one: data splitting maps to the mapping relations of working characteristic data, working characteristic data maps to group
Close the mapping relations of data.
Specifically training method is referred to embodiment one, and details are not described herein for the present embodiment.
S33, the equivalent model that the target product is determined according to the nervus opticus network that training is completed.
In the present embodiment, if nervus opticus e-learning be data splitting map to working characteristic data mapping close
System then can obtain more accurate Adjusted Option according to the equivalent model determined based on nervus opticus network.For example, if mesh
Mark product is electronic product, then the variation of electronic product connection relationship can be determined to electronics by the study of nervus opticus network
The influence of product resonance frequency then can directly adjust electronic product resonance frequency by increasing or decreasing corresponding connection relationship
Rate, to obtain more accurate Adjusted Option.
If nervus opticus e-learning is mapping relations that working characteristic data maps to data splitting, can be direct
Corresponding target operation characteristic is determined according to design objective, and the equivalent model then determined by neural network directly determines institute
The data splitting needed, and then the design of target product can be completed, greatly improve design efficiency.
Example IV
Fig. 5 is a kind of structural representation for equivalent model generating means neural network based that the embodiment of the present application four provides
Figure, as shown in figure 5, comprising:
Sample determining module 501, for determining that the corresponding sample data of target object, the sample data include target pair
The multiple groups supplemental characteristic of elephant and actual measurement working characteristic data corresponding with every group of supplemental characteristic;
Training module 502, for training neural network according to the sample data, so that described in the neural network learning
Mapping relations between supplemental characteristic and the working characteristic data;
Model determining module 503, the neural network for being completed according to training determine the equivalent of the target object
Model.
Optionally, in the present embodiment, the target object includes primary element, corresponding, and the neural network includes the
One sub-neural network, first sub-neural network be used to learn the primary element physical parameter and the primary element
Mapping relations between working characteristic data.
Optionally, in the present embodiment, the target object includes target product, includes multiple basic in the target product
Element, corresponding, the neural network includes the second sub-neural network, wherein second sub-neural network is for learning institute
The mapping relations between the data splitting of target product and the working characteristic data of the target product are stated, the data splitting is
The data splitting of multiple primary elements in the target product.
Embodiment five
The embodiment of the present application five provides a kind of computer-readable medium, is stored thereon with equivalent model, the equivalent model
It is to determine that the neural network is used for the basic parameter and the mesh of learning objective object according to the neural network that training is completed
Mark the mapping relations between the working characteristic data of object.
Embodiment six
Fig. 6 is a kind of flow diagram for modeling method that the embodiment of the present application six provides, as shown in fig. 6, comprising:
S61, raw process parameter data is input to equivalent model, so that the original ginseng of the equivalent model according to input
Number data export prediction work performance data.
Wherein, the equivalent model is to determine that the neural network is for learning according to the neural network that training is completed
Mapping relations between the basic parameter of target object and the working characteristic data of the target object.
In scheme provided in this embodiment, the mapping relations of neural network learning are specially that basic parameter maps to work spy
The mapping relations of property data.
In the present embodiment, when designing a product, it is required to be designed according to design objective, finally determines the product
Modeling result.Then in the present embodiment, initial parameter can be what designer determined according to experience combination design objective.If
Modeling object is primary element, then supplemental characteristic may include the specific data of each physical parameter of primary element, if modeling
Object is target product, then supplemental characteristic may include the data splitting of target product.
Optionally, in the present embodiment, modeling object can be target product, then for determining the neural network of equivalent model
Including the second sub-neural network, second sub-neural network be used for learn the target product data splitting and the target
Mapping relations between the working characteristic data of product, the data splitting are the group of multiple primary elements in the target product
Close data.It should be noted that the second sub-neural network in the present embodiment can be the nervus opticus in above-described embodiment three
Network.
It further, include the equivalent model of each primary element in the data splitting of the target product, it is corresponding, it uses
It can also include the first sub-neural network in the neural network for determining equivalent model, first sub-neural network is for learning institute
The mapping relations between the physical parameter of primary element and the working characteristic data of the primary element are stated, it is each basic with determination
The equivalent model of element.It should be noted that the first sub-neural network in the present embodiment can be in above-described embodiment two
First nerves network.
S62, it determines difference between prediction work performance data and target operation performance data, and whether determines the difference
Within a preset range.If thening follow the steps S63 outside preset range;If within a preset range, thening follow the steps S64.
In the present embodiment, target operation performance data is determined according to design objective.For example, modeling object is transducing
Device, then design objective may include the Wave data of input, the target waveform data for needing energy converter output, then is referred to according to design
Mark, can determine target operation performance data, such as its bandwidth, resonance frequency, the coefficient of coup of energy converter etc..
In the present embodiment, the corresponding preset range of difference is also to be determined according to design objective.For example, modeling object is to change
Energy device, then design objective may include the Wave data of input, the target waveform data for needing energy converter to export and be subjected to
Error range, then the corresponding preset range of difference can be determined according to error range.
S63, the adjustment raw process parameter data, so that the equivalent model is according to the raw process parameter data adjusted
Update the prediction work performance data.Then step S62 is executed again.
In the present embodiment, the process and above-mentioned steps of prediction work performance data are updated according to initial parameter adjusted
The process for exporting prediction work performance data according to initial parameter in S61 is similar, and details are not described herein for the present embodiment.
The current corresponding basic parameter of the prediction work performance data of S64, basis, establishes the production of the target object
Product model.Terminate process.
In the present embodiment, when the difference between prediction work performance data and target operation performance data within a preset range
When, it can primarily determine that the model established according to the current corresponding basic parameter of the prediction work performance data meets design
Index, at this point it is possible to establish the product model of the target object.Then product model is put into simulation software, if emulation
As a result meet the requirement of design objective, then physical product can be made according to the product model of foundation, if it is determined that physical product
The requirement of design objective is also corresponded to, then product design is completed.
In the present embodiment, since the equivalent model used is that the neural network completed based on training is determined, and trains mind
It is to be trained by supplemental characteristic when network, so that obtained equivalent model is more accurate, so that passing through this Shen
The product model that the modeling method that please be provided determines is more nearly with actual product;And the output of neural network is adjustable, makes
The equivalent model that must be obtained can be adapted for a variety of simulation softwares, i.e., modeling method provided by the present application is limited by simulation software
It is lower.
Embodiment seven
Fig. 7 is a kind of flow diagram for modeling method that the embodiment of the present application seven provides, as shown in fig. 7, comprising:
S71, working characteristic data is input to equivalent model, so that the equivalent model is special according to the work of input
Property data export basic parameter data.
Wherein, the equivalent model is to determine that the neural network is for learning according to the neural network that training is completed
Mapping relations between the basic parameter of target object and the working characteristic data of the target object.
In scheme provided in this embodiment, the mapping relations of neural network learning are specially that working characteristic data maps to base
The mapping relations of this parameter.
In the present embodiment, working characteristic data can be determined according to design objective, specific to determine method and above-mentioned determining mesh
The method for marking working characteristic data is similar.
S72, the basic parameter data according to output, establish the product model of the target object.Terminate process.
In the present embodiment, after establishing product model according to basic parameter data, product model can be put into simulation software
It is interior, if simulation result meets the requirement of design objective, physical product can be made according to the product model of foundation, if it is determined that
Physical product also corresponds to the requirement of design objective, then product design is completed.
In the present embodiment, since the equivalent model used is that the neural network completed based on training is determined, and trains mind
It is to be trained by supplemental characteristic when network, so that obtained equivalent model is more accurate, so that passing through this Shen
The product model that the modeling method that please be provided determines is more nearly with actual product;And the output of neural network is adjustable, makes
The equivalent model that must be obtained can be adapted for a variety of simulation softwares, i.e., modeling method provided by the present application is limited by simulation software
It is lower.
Embodiment eight
A kind of structural schematic diagram for model building device that the embodiment of the present application eight provides comprising:
First input module, for raw process parameter data to be input to equivalent model, so that the equivalent model is according to defeated
The raw process parameter data output prediction work performance data entered, wherein the equivalent model is the mind completed according to training
It is determined through network, the neural network is for the basic parameter of learning objective object and the working characteristics number of the target object
Mapping relations between;
The first adjustment module, if the difference between the prediction work performance data and target operation performance data exists
Outside preset range, then the raw process parameter data is adjusted, so that the equivalent model is according to the initial parameter number adjusted
According to the update prediction work performance data;
First modeling module, if for the difference between the prediction work performance data and the target operation performance data
It is different to be in preset range, then according to the current corresponding basic parameter of the prediction work performance data, establish the target
The product model of object.
In the present embodiment, since the equivalent model used is that the neural network completed based on training is determined, and trains mind
It is to be trained by supplemental characteristic when network, so that obtained equivalent model is more accurate, so that passing through this Shen
The product model that the modeling method that please be provided determines is more nearly with actual product;And the output of neural network is adjustable, makes
The equivalent model that must be obtained can be adapted for a variety of simulation softwares, i.e., modeling method provided by the present application is limited by simulation software
It is lower.
Embodiment nine
A kind of structural schematic diagram for model building device that the embodiment of the present application nine provides comprising:
Second input module, for working characteristic data to be input to equivalent model, so that the equivalent model is according to defeated
The working characteristic data output basic parameter data entered, wherein the equivalent model is the nerve net completed according to training
Network determine, the neural network for learning objective object basic parameter and the target object working characteristic data it
Between mapping relations;
Second modeling module establishes the product mould of the target object for the basic parameter data according to output
Type.
In the present embodiment, since the equivalent model used is that the neural network completed based on training is determined, and trains mind
It is to be trained by supplemental characteristic when network, so that obtained equivalent model is more accurate, so that passing through this Shen
The product model that the modeling method that please be provided determines is more nearly with actual product;And the output of neural network is adjustable, makes
The equivalent model that must be obtained can be adapted for a variety of simulation softwares, i.e., modeling method provided by the present application is limited by simulation software
It is lower.
The apparatus embodiments described above are merely exemplary, wherein described, module can as illustrated by the separation member
It is physically separated with being or may not be, the component shown as module may or may not be physics mould
Block, it can it is in one place, or may be distributed on multiple network modules.It can be selected according to the actual needs
In some or all of the modules achieve the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying creativeness
Labour in the case where, it can understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It realizes by means of software and necessary general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be embodied in the form of software products in other words, should
Computer software product may be stored in a computer readable storage medium, the computer readable recording medium include for
Any mechanism of the readable form storage of computer (such as computer) or transmission information.For example, machine readable media includes only
Read memory (ROM), random access memory (RQM), magnetic disk storage medium, optical storage media, flash medium, electricity, light,
Sound or the transmitting signal (for example, carrier wave, infrared signal, digital signal etc.) of other forms etc., which includes
Some instructions are used so that a computer equipment (can be personal computer, server or the network equipment etc.) executes respectively
Method described in certain parts of a embodiment or embodiment.
Finally, it should be noted that above embodiments are only to illustrate the technical solution of the embodiment of the present application, rather than it is limited
System;Although the application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: its
It is still possible to modify the technical solutions described in the foregoing embodiments, or part of technical characteristic is equal
Replacement;And these are modified or replaceed, each embodiment technical solution of the application that it does not separate the essence of the corresponding technical solution
Spirit and scope.
It will be understood by those skilled in the art that the embodiment of the embodiment of the present invention can provide as method, apparatus (equipment) or
Computer program product.Therefore, the embodiment of the present invention can be used complete hardware embodiment, complete software embodiment or combine soft
The form of the embodiment of part and hardware aspect.Moreover, it wherein includes to calculate that the embodiment of the present invention, which can be used in one or more,
Computer-usable storage medium (including but not limited to magnetic disk storage, CD-ROM, the optical memory of machine usable program code
Deng) on the form of computer program product implemented.
The embodiment of the present invention referring to according to the method for the embodiment of the present invention, device (equipment) and computer program product
Flowchart and/or the block diagram describes.It should be understood that can be realized by computer program instructions every in flowchart and/or the block diagram
The combination of process and/or box in one process and/or box and flowchart and/or the block diagram.It can provide these computers
Processor of the program instruction to general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices
To generate a machine, so that generating use by the instruction that computer or the processor of other programmable data processing devices execute
In the dress for realizing the function of specifying in one or more flows of the flowchart and/or one or more blocks of the block diagram
It sets.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Claims (12)
1. a kind of equivalent model generation method neural network based characterized by comprising
Determine the corresponding sample data of target object, the sample data include target object multiple groups supplemental characteristic and with it is every
The corresponding actual measurement working characteristic data of group supplemental characteristic;
According to sample data training neural network, so that supplemental characteristic described in the neural network learning and the work are special
Mapping relations between property data;
The equivalent model of the target object is determined according to the neural network that training is completed.
2. the method according to claim 1, wherein if the target object be primary element, it is corresponding,
Determine that the corresponding sample data of target object comprises determining that the corresponding first sample data of the primary element, wherein institute
State in first sample data include multiple groups described in primary element physical parameter and every group described in the corresponding actual measurement of physical parameter
Working characteristic data;
It include: to train first nerves network according to the first sample data according to sample data training neural network,
In, the mapping that the first nerves network is used to learn between the physical parameter and the corresponding working characteristic data is closed
System;
The equivalent model that the target object is determined according to the neural network that training is completed includes: the completed according to training
One neural network determines the equivalent model of the primary element.
3. according to the method described in claim 2, it is characterized in that, the primary element includes basic inverting element, the base
The working characteristic data of this inverting element includes: the change data of the input and output wave of the inverting element;
Corresponding, first nerves network is used to learn the mapping between the physical parameter and the corresponding working characteristic data
Relationship includes: first nerves network for learning the physical parameter of the basic inverting element and the change data of input and output wave
Between mapping relations.
4. according to the method described in claim 3, it is characterized in that, the physical parameter of the inverting element include it is following at least it
One: the structural parameters of the inverting element, material parameter.
5. the method according to claim 1, wherein the target produces if the target object is target product
It include multiple primary elements in product, then it is corresponding,
Determine that the corresponding sample data of target object comprises determining that corresponding second sample data of the target product, wherein institute
State the multiple data splittings and the corresponding actual measurement work spy of each data splitting that the second sample data includes the target product
Property data, the data splitting be the target product in multiple primary elements data splitting;
It include: according to second sample data training nervus opticus network, institute according to sample data training neural network
Nervus opticus network is stated for learning the mapping relations between the data splitting and working characteristic data of the target product;
The equivalent model that the target object is determined according to the neural network that training is completed includes: the completed according to training
Two neural networks determine the equivalent model of the target product.
6. according to the method described in claim 5, the target produces it is characterized in that, the target product includes electronic product
The primary element for including in product is electronic component, corresponding, and the data splitting includes at least one: case for electronic organizer
Each electronic component that the equivalent model data of each electronic component included, electronic product include described
The connection relationship between each electronic component that position, electronic product in electronic product include.
7. a kind of equivalent model generating means neural network based characterized by comprising
Sample determining module, for determining that the corresponding sample data of target object, the sample data include the more of target object
Group supplemental characteristic and actual measurement working characteristic data corresponding with every group of supplemental characteristic;
Training module, for training neural network according to the sample data, so that parameter number described in the neural network learning
According to the mapping relations between the working characteristic data;
Model determining module, the neural network for being completed according to training determine the equivalent model of the target object.
8. device according to claim 7, which is characterized in that the target object includes primary element, corresponding, described
Neural network include the first sub-neural network, first sub-neural network be used for learn the primary element physical parameter and
Mapping relations between the working characteristic data of the primary element.
9. device according to claim 7, which is characterized in that the target object includes target product, and the target produces
It include multiple primary elements in product, corresponding, the neural network model includes the second sub-neural network, wherein described second
Sub-neural network is for learning reflecting between the data splitting of the target product and the working characteristic data of the target product
Relationship is penetrated, the data splitting is the data splitting of multiple primary elements in the target product.
10. a kind of computer-readable medium, which is characterized in that be stored thereon with equivalent model, the equivalent model is according to instruction
Practice what the neural network completed determined, basic parameter and the target object of the neural network for learning objective object
Mapping relations between working characteristic data.
11. a kind of modeling method characterized by comprising
Raw process parameter data is input to equivalent model, so that the equivalent model is defeated according to the raw process parameter data of input
Prediction work performance data out, wherein the equivalent model is determined according to the neural network that training is completed, the nerve net
Network is for the mapping relations between the basic parameter of learning objective object and the working characteristic data of the target object;
If the difference between the prediction work performance data and target operation performance data is outside preset range, described in adjustment
Raw process parameter data, so that the equivalent model updates the prediction work characteristic according to the raw process parameter data adjusted
Data;
Alternatively, if the difference between the prediction work performance data and the target operation performance data is in preset range
It is interior, then according to the current corresponding basic parameter of the prediction work performance data, establish the product model of the target object.
12. a kind of modeling method characterized by comprising
Working characteristic data is input to equivalent model, so that the equivalent model is defeated according to the working characteristic data of input
Basic parameter data out, wherein the equivalent model is to determine that the neural network is used according to the neural network that training is completed
Mapping relations between the basic parameter of learning objective object and the working characteristic data of the target object;
According to the basic parameter data of output, the product model of the target object is established.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811268685.2A CN109447245A (en) | 2018-10-29 | 2018-10-29 | Equivalent model generation method and modeling method neural network based |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811268685.2A CN109447245A (en) | 2018-10-29 | 2018-10-29 | Equivalent model generation method and modeling method neural network based |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109447245A true CN109447245A (en) | 2019-03-08 |
Family
ID=65548979
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811268685.2A Pending CN109447245A (en) | 2018-10-29 | 2018-10-29 | Equivalent model generation method and modeling method neural network based |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109447245A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110197016A (en) * | 2019-05-15 | 2019-09-03 | 南京星火技术有限公司 | Equivalent model method for building up, device, computer-readable medium and application method |
CN112927701A (en) * | 2021-02-05 | 2021-06-08 | 商汤集团有限公司 | Sample generation method, neural network generation method, audio signal generation method and device |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102983819A (en) * | 2012-11-08 | 2013-03-20 | 南京航空航天大学 | Imitating method of power amplifier and imitating device of power amplifier |
CN106446405A (en) * | 2016-09-23 | 2017-02-22 | 北京大学深圳研究生院 | Integrated circuit device neural network modeling sample selecting method and device |
US20170091616A1 (en) * | 2015-09-29 | 2017-03-30 | International Business Machines Corporation | Scalable architecture for analog matrix operations with resistive devices |
CN107748809A (en) * | 2017-09-20 | 2018-03-02 | 苏州芯智瑞微电子有限公司 | A kind of semiconductor devices modeling method based on nerual network technique |
CN108416167A (en) * | 2018-03-27 | 2018-08-17 | 成都海威华芯科技有限公司 | A kind of GaN HEMT devices multiple physical field coupling large-signal model method for building up |
CN108595815A (en) * | 2018-04-18 | 2018-09-28 | 石家庄创天电子科技有限公司 | Artificial intelligence body training system and passive circuit Optimum Design System and method |
-
2018
- 2018-10-29 CN CN201811268685.2A patent/CN109447245A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102983819A (en) * | 2012-11-08 | 2013-03-20 | 南京航空航天大学 | Imitating method of power amplifier and imitating device of power amplifier |
US20170091616A1 (en) * | 2015-09-29 | 2017-03-30 | International Business Machines Corporation | Scalable architecture for analog matrix operations with resistive devices |
CN106446405A (en) * | 2016-09-23 | 2017-02-22 | 北京大学深圳研究生院 | Integrated circuit device neural network modeling sample selecting method and device |
CN107748809A (en) * | 2017-09-20 | 2018-03-02 | 苏州芯智瑞微电子有限公司 | A kind of semiconductor devices modeling method based on nerual network technique |
CN108416167A (en) * | 2018-03-27 | 2018-08-17 | 成都海威华芯科技有限公司 | A kind of GaN HEMT devices multiple physical field coupling large-signal model method for building up |
CN108595815A (en) * | 2018-04-18 | 2018-09-28 | 石家庄创天电子科技有限公司 | Artificial intelligence body training system and passive circuit Optimum Design System and method |
Non-Patent Citations (4)
Title |
---|
ZHAOCHENG LIU ET AL.: "Generative Model for the Inverse Design of Metasurfaces", 《NANO LETTERS》 * |
张宇等: "集成电容设计的逆向神经网络建模", 《微波学报》 * |
张涛: "《压电薄膜材料与器件制备技术》", 30 September 2012, 西北工业大学出版社 * |
王凤美: "基于神经网络的射频和微波无源器件建模与仿真实现", 《万方数据库》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110197016A (en) * | 2019-05-15 | 2019-09-03 | 南京星火技术有限公司 | Equivalent model method for building up, device, computer-readable medium and application method |
CN112927701A (en) * | 2021-02-05 | 2021-06-08 | 商汤集团有限公司 | Sample generation method, neural network generation method, audio signal generation method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106503802A (en) | A kind of method of utilization genetic algorithm optimization BP neural network system | |
CN103561377B (en) | Optimized setting method for electroacoustic products | |
CN101411213A (en) | Calibration method and device in an audio system | |
CN109447245A (en) | Equivalent model generation method and modeling method neural network based | |
Bissell et al. | Telling tales: models, stories and meanings | |
CN110032751A (en) | The constraint Non-linear parameter estimation of steady non-linear speaker modeling | |
CN107729621B (en) | A kind of verification tool of statical model | |
CN109344969B (en) | Neural network system, training method thereof, and computer-readable medium | |
CN110427650A (en) | The Numerical Simulation Analysis method of moving iron type loud-speaker fundamental characteristics | |
CN109670257A (en) | A kind of converter station sound field emulation mode and system | |
CN109284541A (en) | A kind of more Method of Physical Modeling of neural network for microwave passive component | |
CN109992885A (en) | A kind of casting limited fatigue life member prediction technique and system | |
CN107484069A (en) | The determination method and device of loudspeaker present position, loudspeaker | |
CN101516055B (en) | Method and device capable of simulating three-dimensional echo sound effect in different acoustic environments | |
CN108062446A (en) | Electrical practice training experimental bench analog simulation method and system | |
CN108759648A (en) | Ground Penetrating Radar detection method based on machine learning | |
CN110197016A (en) | Equivalent model method for building up, device, computer-readable medium and application method | |
CN115408949B (en) | Load model parameter identification method System, apparatus and medium | |
CN110188039A (en) | The method and system of software test, software optimization | |
CN108597612B (en) | Virtual cutting method for simulating bleeding model | |
CN105893667A (en) | Heterogeneous simulation data unified integrated visual method | |
CN102789594B (en) | Voice generation method based on DIVA neural network model | |
Juvela et al. | End-to-end amp modeling: from data to controllable guitar amplifier models | |
CN110661511B (en) | Matrix type adaptive filter hysteresis control system and method | |
CN104954179B (en) | A kind of analog device networking simulation node information exchange optimized relation system |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190308 |