CN109101712A - Product model designing system and method based on figure network - Google Patents

Product model designing system and method based on figure network Download PDF

Info

Publication number
CN109101712A
CN109101712A CN201810846434.1A CN201810846434A CN109101712A CN 109101712 A CN109101712 A CN 109101712A CN 201810846434 A CN201810846434 A CN 201810846434A CN 109101712 A CN109101712 A CN 109101712A
Authority
CN
China
Prior art keywords
design
network
parameter
product model
model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201810846434.1A
Other languages
Chinese (zh)
Other versions
CN109101712B (en
Inventor
不公告发明人
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shijiazhuang Chuang Tian Electronic Technology Co Ltd
Original Assignee
Shijiazhuang Chuang Tian Electronic Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shijiazhuang Chuang Tian Electronic Technology Co Ltd filed Critical Shijiazhuang Chuang Tian Electronic Technology Co Ltd
Priority to CN201810846434.1A priority Critical patent/CN109101712B/en
Publication of CN109101712A publication Critical patent/CN109101712A/en
Application granted granted Critical
Publication of CN109101712B publication Critical patent/CN109101712B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

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

Abstract

The embodiment of the present application provides a kind of product model designing system and method based on figure network, by the way that product model is converted to corresponding figure network, to enable deep neural network carry out learning training according to the individual features that figure network is mapped out, so that deep neural network has the function of product model the Automation Design, so that deep neural network can go out corresponding optimization design movement according to the design objective parameter prediction of product model yet to be built, and automatically generate the product model for meeting the design objective parameter.

Description

Product model designing system and method based on figure network
Technical field
The invention relates to product model design method and its systems, espespecially a kind of to be realized based on figure network technology Product model designing system and method.
Background technique
Currently, the design of product model is realized by using Computer simulation design tool, mainly includes leading to It crosses design of Simulation tool and establishes simulation model in a computer, related running environment condition is added, then again by design of Simulation work Tool is calculated, if the calculated result of simulation model is unable to satisfy the design requirement of model index parameter, is needed continuous Simulation model is adjusted, until calculated result meets the design requirement of model index parameter.
It in addition, the design cycle of current production model is also scattered, and is realized by artificial means.This Outside, highly relevant data can be generated in the design process of product model, however, can currently without a system and method The rule of data caused by during automatic learning scene, leading to current product model design work, there are working efficiencies The technical problems such as low, development cycle length.
In view of this, existing variety of problems, as this case are to be solved during how overcoming existing product modelling Technical task.
Summary of the invention
In order to solve defect of the existing technology, the main purpose of the present invention is to provide a kind of productions based on figure network Product modelling system and method realizes the optimization design of product model by using deep neural network.
Another object of the present invention is to provide a kind of product model designing systems and method based on figure network, are based on The frame of artificial intelligence can be abstracted into product model a kind of relational structure, and the reasoning association therein from this structure Relationship.
Another object of the present invention is to provide a kind of product model designing system and method based on figure network, by making With relationship induction bias (relational inductive biases), promotes the entity abstracted to product model and its close System, and the composition rule between above-mentioned each is learnt.
For the above-mentioned purpose and other related purposes, embodiments herein provide a kind of product mould based on figure network Type designing system characterized by comprising data acquisition module is used for collecting sample model its corresponding sample pattern and refers to Parameter and sample optimization design parameter are marked, to establish data bank, wherein the sample optimization design parameter expression is directed to The movement of optimization design performed by the sample pattern；Setting module is used to set the design objective of product model yet to be built Parameter；Figure network generation module is used to generate corresponding figure network for the sample pattern in the data bank； Figure e-learning module, be used for by the corresponding figure network of the sample pattern, the sample pattern index parameter and The sample optimization design parameter is inputted in the first deep neural network and is trained, to construct first deep neural network Deep learning model, to enable first deep neural network based on the constructed deep learning model, and be directed to institute Design objective parameter prediction is stated to generate the first optimal design parameter and exported；And product model generation module, it is used for First optimal design parameter is received, executes corresponding optimization design movement accordingly, and generates and meets the design objective ginseng Several product models.
Optionally, in embodiments herein, the figure network generation module is further used for by by the sample The second deep neural network of mode input, so that second deep neural network is reflected accordingly according to sample pattern output Characteristic pattern is penetrated, and using the mappings characteristics figure as the figure network for corresponding to the sample pattern.
Optionally, in embodiments herein, the figure network generation module is further used for from the sample pattern Middle extraction nodal information, side information and global information, to be corresponded to using the generation of the nodal information, side information and global information The figure network of the sample pattern；The figure e-learning module be further used for analysis when update the nodal information, When one of them in the side information and the global information, the nodal information, the side information and the overall situation Variation caused by other the two in information, and analyze the nodal information, the side information and the global information it Between incidence relation；And the product model generation module is further used for by adjusting the node in the figure network One of them in information, the side information and the global information, to enable the nodal information, the side information and institute The other the two stated in global information generate corresponding change, to execute the optimization design movement.
Optionally, in embodiments herein, the system further comprises analysis module, is used to analyze the production Product model generation module executes the optimization design movement product model generated according to first optimal design parameter It is no to meet the design objective parameter, and when analyzing result is to meet, the product model is exported, and when analysis result is not When meeting, output analysis signal to the figure e-learning module, to enable first depth by the figure e-learning module Neural network is spent based on the constructed deep learning model, and predicts to generate newly again for the design objective parameter First optimal design parameter is simultaneously exported.
Optionally, in embodiments herein, the system further comprises figure network intensified learning module, is used for When first optimal design parameter generation that can not be exported by the figure e-learning module meets the design objective parameter The product model when, from the product model generation module according to first optimal design parameter multiple institutes generated It states in product model, chooses the product model closest to the design objective parameter, and by by the product model Corresponding figure network inputs are to third deep neural network, to enable the third deep neural network be joined according to the design objective Number generates the second optimal design parameter and is exported；And wherein, the product model generation module is further used for receiving institute The second optimal design parameter is stated, executes corresponding optimization design movement accordingly, and generates the product model；The analysis module It is further used for analyzing the product model generation module according to second optimal design parameter execution optimization design movement institute Whether the product model generated meets the design objective parameter, and when analyzing result is to meet, exports the product Model, and when analyzing result is not meet, output analysis signal to the figure network intensified learning module, to pass through the figure Network intensified learning module enables the third deep neural network again according to design objective parameter generation one newly Second optimal design parameter is simultaneously exported.
Optionally, in embodiments herein, the analysis module is further used for raw when analyzing the product model The optimization design movement product model generated, which is executed, according to second optimal design parameter at module meets described set When counting index parameter, the product model, the design objective parameter and second optimal design parameter are input to described Using as the sample pattern and its corresponding sample pattern index parameter and the sample optimization in data bank Design parameter is stored.
Another embodiment of the application provides a kind of product model design method based on figure network, which is characterized in that The described method includes: its corresponding sample pattern index parameter of collecting sample model and sample optimization design parameter, to establish Data bank, wherein the sample optimization design parameter is indicated for the movement of optimization design performed by the sample pattern； Corresponding figure network is generated for the sample pattern stored in the data bank；By the corresponding institute of the sample pattern State figure network, the sample pattern index parameter and the sample optimization design parameter input in the first deep neural network into Row training, to construct the deep learning model of first deep neural network；The design objective parameter of input product model, with It enables first deep neural network be based on the deep learning model, and generates the first optimization for the design objective parameter Design parameter is simultaneously exported；And corresponding optimization design is executed according to first optimal design parameter by emulation tool Movement generates the product model for meeting the design objective parameter.
Optionally, it in embodiments herein, is generated for the sample pattern stored in the data bank Corresponding figure network includes: to input the second deep neural network through by the sample pattern, to enable the second depth nerve Network exports corresponding mappings characteristics figure according to the sample pattern, and as the figure for corresponding to the sample pattern Network.
Optionally, in embodiments herein, the method further includes: from the sample pattern extract node Information, side information and global information correspond to the sample mould to generate using the nodal information, side information and global information The figure network of type；Analysis is when the one of them in the update nodal information, the side information and the global information When, variation caused by other the two in the nodal information, the side information and the global information, and described in learning Incidence relation between nodal information, the side information and the global information；And the optimization design movement is executed, with By adjusting the one of them in the nodal information, the side information and the global information in the figure network, and Other the two in the nodal information, the side information and the global information are enabled to generate corresponding change.
Optionally, described to generate the product model for meeting the design objective parameter in embodiments herein The step of further comprise: analyze the emulation tool according to first optimal design parameter execute optimization design movement give birth to At the product model whether meet the design objective parameter, and when analyzing result is to meet, export the product mould Type, and when analyzing result is not meet, enable first deep neural network be based on the deep learning model and be directed to institute Design objective parameter is stated to regenerate the first new optimal design parameter and exported.
Optionally, described to generate the product model for meeting the design objective parameter in embodiments herein The step of further comprise: when analysis can not according to first optimal design parameter generate meet the design objective parameter When the product model, by third deep neural network to execute intensified learning step, with from multiple productions generated In product model, the product model closest to the design objective parameter is chosen, and by the figure network of the product model It is input to the third deep neural network, to enable the third deep neural network generate the according to the design objective parameter Two optimal design parameters are simultaneously exported；It is corresponding excellent according to second optimal design parameter execution by the emulation tool Change design action to generate the new product model；And the analysis emulation tool is according to second optimal design parameter It executes optimization design and acts whether the product model generated meets the design objective parameter, and when analysis result is symbol When conjunction, the product model is exported, and when analyzing result is not meet, then enable third deep neural network re-execute described Intensified learning step.
Optionally, in embodiments herein, the method further includes when the analysis emulation tool is according to institute When stating the second optimal design parameter and executing optimization design and act the product model generated and meet the design objective parameter, Described in the product model generated, the design objective parameter and the third deep neural network are exported Second optimal design parameter is input in the data bank, using as the sample pattern and its corresponding sample mould Type index parameter and the sample optimization design parameter are stored.
Another embodiment of the application provides a kind of product model designing system based on figure network, which is characterized in that Include: setting module, is used to set the design objective parameter of product model yet to be built；Product model generation module, is used for Basic model is generated according to the design objective parameter, and receives the second optimal design parameter and is directed to the basic model accordingly It executes corresponding optimization design movement and generates the product model；Figure network generation module is used for for the basic mould Type generates corresponding figure network；Figure network intensified learning module, is used to construct the intensified learning mould of third deep neural network Type, and by third deep neural network described in the figure network generation module figure network inputs generated, described in enabling Third deep neural network is based on the intensified learning model, and it is excellent to generate described second for the design objective parameter prediction Change design parameter and is exported；And analysis module, it is used to analyze the product model generation module according to described second Whether the optimal design parameter product model generated meets the design objective parameter, and when analysis result is to meet When, the product model is exported, and when analyzing result is not meet, output analysis signal to the figure network intensified learning mould Block, to enable the figure network intensified learning module by the third deep neural network again according to the design objective parameter It generates new second optimal design parameter and is exported, until generating the production for meeting the design objective parameter Product model, and wherein, the figure network intensified learning module is further used for the analysis according to the analysis module as a result, being directed to The third deep neural network is trained, and updates the intensified learning model with adjustment.
Optionally, in embodiments herein, the figure network generation module is further used for by by the basis The second deep neural network of mode input, so that second deep neural network exports corresponding mappings characteristics figure accordingly, and Using the mappings characteristics figure as the figure network for corresponding to the basic model.
Optionally, in embodiments herein, the figure network generation module is further used for from the basic model Middle extraction nodal information, side information and global information, to be corresponded to using the generation of the nodal information, side information and global information The figure network of the sample pattern；The product model generation module is further used for by adjusting in the figure network One of them in the nodal information, the side information and the global information, to enable the nodal information, side letter Other the two in breath and the global information generate corresponding change, to execute the optimization design movement.
Another embodiment of the application provides a kind of product model design method based on figure network, which is characterized in that It include: the design objective parameter of setting product model yet to be built；It is raw according to the design objective parameter by the emulation tool At basic model；Corresponding figure network is generated for the basic model；By the corresponding figure network inputs of the basic model Three deep neural networks generate second optimization to enable the third deep neural network according to the design objective parameter Design parameter is simultaneously exported；The basic model is executed according to second optimal design parameter by the emulation tool Corresponding optimization design movement is to generate the product model；Analyze whether the product model generated meets the design Index parameter, and when analyzing result is to meet, the product model is exported, and when analyzing result is not meet, described in order Third deep neural network generates new second optimal design parameter according to the design objective parameter again and gives Output, until the product model generated meets the design objective parameter.
Optionally, in embodiments herein, generating corresponding figure network for the basic model further comprises: The basic model is inputted into the second deep neural network, so that second deep neural network is defeated according to the basic model Corresponding mappings characteristics figure out, using as correspond to the sample pattern the figure network.
Optionally, in embodiments herein, the method further includes: from the sample pattern extract node Information, side information and global information correspond to the sample mould to generate using the nodal information, side information and global information The figure network of type；And execute optimization design movement, with by adjust the nodal information in the figure network, One of them in the side information and the global information, to enable the nodal information, the side information and described complete Other the two in office's information generate corresponding change.
From the foregoing, it will be observed that the product model designing system and method provided herein based on figure network, by by sample Model is converted to corresponding figure network in a manner of relational structure, to be learnt using deep neural network for figure network Model optimization design rule is generated, with the design objective parameter using deep neural network according to the product model inputted, is held Row optimization design movement, to automatically generate the product model for meeting design objective parameter.It is given birth in addition, the application can be more directed to At model optimization design rule carry out intensified learning, with according to product model design objective parameter and generate the second optimization and set Parameter is counted, to automatically generate the product model for meeting design objective parameter.Whereby, the application can realize that product model is automatic Advantage high with more design efficiency while changing design, that reusable degree is high and design cost is cheap.
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 the structural schematic diagram for showing the product model designing system of first embodiment of the application；
Fig. 2 is the embodiment schematic diagram for showing the product model designing system of Fig. 1；
Fig. 3 is the flow diagram for showing the product model design method of second embodiment of the application；
Fig. 4 is the embodiment schematic diagram for showing the product model design method of Fig. 3；
Fig. 5 is the structural schematic diagram for showing the product model designing system of 3rd embodiment of the application；
Fig. 6 is the flow diagram for showing the product model design method of fourth embodiment of the application；
Fig. 7 is composed by nodal information (P), side information (E) and global information (u) in display the embodiment of the present application The schematic diagram of figure network；And
Fig. 8 A to Fig. 8 B and Fig. 9 A to Fig. 9 B is the different application for showing the product model designing system and method for the application Embodiment schematic diagram.
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.
Figure network (Graph Network) is a kind of generalized regression neural network based on graph structure, can be directly to true Real problem is modeled, and is also learnt using automatic differential technology, has powerful relationship induction bias, and support relationship pushes away It manages and combines extensive, lay the foundation for more complicated, interpretable and flexible reasoning pattern.In view of this, the invention proposes one The product model design method that kind is realized based on figure network technology.Those skilled in the art are it should be apparent that the present invention mentions Product model designing technique out is suitable for all product models that can be designed by analog simulation tool, related production Product model may include mechanical model and circuit model, but be not limited only to this.
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 the structural schematic diagram for showing the product model designing system of first embodiment of the application.As shown, this The product model designing system 100 of embodiment mainly includes setting module 105, data acquisition module 110, figure network generation module 120, figure e-learning module 130 and product model generation module 140.
Setting module 105 is used to set the design objective parameter of product model yet to be built.
Data acquisition module 110 is for establishing data bank 101, for its corresponding sample pattern of stored samples model Index parameter and sample optimization design parameter.Specifically, sample pattern stored in data bank 101 refer to The product model for meeting sample pattern index parameter that manual type is established, for example, using veteran designer, based on production The requirement such as product shape (product size etc.) and product performance index (electrical characteristics, power etc.), by one or more computers Product simulation model is established in design of Simulation tool, debug and optimizes product simulation model, and is finally obtained and met model index The product simulation model of parameter request is as sample pattern.Wherein, each step during debugging (optimization) is generated The sample data of product simulation model can be used as sample pattern acquired in data acquisition module 110.Furthermore the sample Model index parameter may include but be not limited to, such as the input and output-index requirement of design of electronic products, believe dielectric material The information such as breath, component information, conductor information, input interface, output interface.
Above-mentioned " sample optimization design parameter " refers to be acted for optimization design performed by the sample pattern.Specifically For, it is assumed that stored different sample patterns include M0, M1, M2 in data bank 101,, M (n-1), Mn, wherein When executing optimization design movement (such as being indicated with A1) for sample pattern M0, sample mould can be generated by sample pattern M0 Type M1；When executing optimization design movement (such as being indicated with A2) for sample pattern M1, can be generated by sample pattern M1 Sample pattern M2；And so on, when generating sample pattern Mn by sample pattern M (n-1), performed optimization design movement It is then An, above-mentioned optimization design acts A1, A2,, An is " sample optimization design parameter " defined herein.
Figure network generation module 120 is used to generate for stored each sample pattern in data bank 101 and correspond to Figure network.Specifically, figure network generation module 120 is through each sample mode input that will be stored in data bank 101 Second deep neural network, so that the second deep neural network exports corresponding mappings characteristics according to the sample pattern inputted Figure, and using exported mappings characteristics figure as the figure network for corresponding to sample pattern, to realize to come in a manner of relational structure Indicate sample pattern.In an embodiment, the second deep neural network is, for example, deep neural network (the convolution mind of CNN framework Through network), it is so not limited thereto, is also achieved using other types of deep neural network.
Please refer to Fig. 7, in an embodiment, figure network generation module 120 further includes that section is extracted from sample pattern Point information (P), side information (E) and global information (u), to be formed using nodal information (P), side information (E) and global information (u) Figure network corresponding to sample pattern.
Figure e-learning module 130 is used for the corresponding figure network of the sample pattern, the sample pattern index Parameter and the sample optimization design parameter, which input in the first deep neural network, to be trained, to construct the first depth nerve The deep learning model of network, that is, the training for the first deep neural network progress product model optimization design function. And after the training of the first deep neural network is completed, figure e-learning module 130 is also used to enable the first deep neural network base The first optimal design parameter is generated for inputted design objective parameter prediction and is given in constructed deep learning model Output.
In an embodiment, the figure e-learning module 130 based on the first deep neural network algorithm is worked as by analysis When updating the one of them in nodal information, side information and global information, in nodal information, side information and global information Variation caused by other the two, and the incidence relation between above three information is analyzed, for example, figure e-learning module 130 Through analysis when more new change occurs for side information, nodal information and the corresponding generated variation of global information, on analyzing State the incidence relation between three.
As shown in Figure 7, wherein by including node in any one the figure network generated of network generation module 120 Three information, side information and global information big information units, and wherein, each information unit may be characterized as one Vector, therefore, the processing of analytic learning performed by the figure e-learning module 130 based on the first deep neural network algorithm can be with It betides in each link, the update of update, node state including opposite side state and the update to global information.For example, Figure e-learning module 130 based on the first deep neural network algorithm can carry out backpropagation after finishing feed forward operation Study, therefore side information, nodal information and global information then can be adjusted in feedback procedure, this may make figure network Overall architecture becomes to learn.
Further, the figure e-learning module 130 based on the first deep neural network algorithm can be based on deep learning frame Framework and the neural net layer for building out multilayer, to be fitted the feature that complicated calculating process carrys out enhanced products model, wherein figure Input of the network as the first deep neural network layer, output are the optimization of product model required execution during design optimization Design action parameter, and the study of figure network is update for the side information state in figure network, nodal information state Update and the update of global information.
Further, based on the figure e-learning module 130 of the first deep neural network algorithm for learning model optimization The optimization of first deep neural network of design rule is trained for offline Training, used deep learning algorithm packet It includes but is not limited only to: deepness belief network, Recognition with Recurrent Neural Network, convolution deepness belief network, large memories and memory nerve Network, depth Boltzmann machine stack autocoder, deep accumulation network, tensor depth stacking network scheduling algorithm.
Further, such as figure e-learning module 130 is neural for the first depth of learning model optimization design rule Network is built based on Recognition with Recurrent Neural Network, and figure network state is put into the mind of the first depth based on Recognition with Recurrent Neural Network framework Through to complete the prediction to network state, and further using the output of network as its input, and carrying out next round in network Prediction and generation.
Further, such as in figure e-learning module 130 it is used to the first depth mind of learning model optimization design rule Through introducing the attention mechanism of network in network, in this way, can carried out more for each node when updating nodal information every time Which part should be more paid close attention in new process, also that is, which partially for the node even more important offer with reference to letter Breath.
Product model generation module 140 executes corresponding optimization accordingly and sets for receiving first optimal design parameter Meter movement, and generate the product model for meeting the design objective parameter.In practical application, product model generation module 140 the first optimal design parameters that can be exported by being repeated as many times reception figure e-learning module 130, by executing repeatedly Optimization design movement, and automatically generate the product model for meeting the design objective parameter.
In an embodiment, product model generation module 140 is by adjusting the nodal information in figure network, the side One of them in information and the global information, to enable the nodal information, the side information and the global information In it is both other generate corresponding changes, to execute the optimization design movement.In other words, product model generation module 140 can Based on product index, product simulation model is established in Computer simulation design tool, and debugs and optimize product model, In, the debugging and optimization operation of product model include being worked as current simulation model via figure network generation module 120 The figure network of preceding product model, by the figure network inputs to via the figure e-learning mould based on the first deep neural network algorithm In trained first deep neural network of block 130, and the required optimization executed is set during obtaining product model design optimization Meter movement, until generating the product model for meeting design objective requirement.
Please refer to Fig. 2, in an embodiment, product model designing system 100 can also further comprise analysis module 150, it is generated according to the execution optimization design movement of the first optimal design parameter to be used to analyze product model generation module 140 Whether product model meets the design objective parameter, for example, analysis module 150 can generate mould by model running product model The product model generated of block 140 and obtain corresponding dry run as a result, and analyzing the dry run knot of the product model Whether fruit matches with the design objective parameter of the product model inputted.Wherein, when analyzing result is to meet, then production is exported The product model generated of product model generation module 140.And when analyzing result is not meet, then exportable analysis signal is with anti- Figure e-learning of feeding module 130, to enable the first deep neural network be based on the depth by figure e-learning module 130 Learning model, and predict to generate new first optimal design parameter again and give defeated for the design objective parameter Out, to enable product model generation module 150 re-execute optimization design movement accordingly, and new product model is generated.
Please continue to refer to Fig. 2, in another embodiment, the application can be when product model generation module 140 is according to first When the optimal design parameter execution optimization design movement product model generated can not meet the design objective parameter, When exactly can not meet the product model of design objective parameter by the generation of figure e-learning module 130, it can start and execute the last one Change mode of learning.
Specifically, product model designing system 100 also further has figure network intensified learning module 160, it is used for When judging that product model generation module 140 executes optimization design according to the first optimal design parameter and act product model generated It can not generate when meeting design objective parameter (for example, can pass through setting figure e-learning module 130 generates the first optimization design ginseng Several number or setting figure e-learning module 130 generate the execution time of the first optimal design parameter, in judging figure network Study module 130 reach the set upper limit and to get arriving when not designing the product model for meeting design objective parameter yet Figure e-learning module 130 can not generate the judging result for meeting the product model of design objective parameter), in this case, will Starting executes figure network intensified learning module 160, is generated from according to model generation module 140 according to the first optimal design parameter Multiple product models in, the product model closest to design objective parameter is chosen, and by figure network generation module 120 Corresponding figure network is generated for the product model, so that figure network intensified learning module 160 is by by the product model Corresponding figure network inputs are to third deep neural network, to enable third deep neural network be joined according to the design objective inputted Number generates one second optimal design parameter, to enable product model generation module 140 according to the second optimal design parameter, executes corresponding Optimization design movement, and generate new product model.
In addition, analysis module 150 can also be further after figure network intensified learning module 160 generates new product model Whether analysis meets according to the second optimal design parameter product model generated that figure network intensified learning module 160 is exported Design objective parameter, and when analyzing result is to meet, then product model generated is exported, and when analysis result is not meet When, output analysis signal simultaneously feeds back to figure network intensified learning module 160, to enable figure network intensified learning module 160 by the Three deep neural networks generate new second optimal design parameter according to design objective parameter again and are exported, with It enables product model generation module 140 re-execute an optimization design movement accordingly, and generates new product model, for analysis Module 150 is analyzed again.
Specifically, the state that figure network intensified learning module 160 can learn figure network as deeply, passes through The individual features of the mapped out figure network of two deep neural networks, transfer to tactful network to be analyzed, provide to current production mould The prediction optimization design action (i.e. the corresponding optimization design movement of the second optimal design parameter) of type；Then value network is transferred to again The quality for the prediction optimization design action that assessment strategy network is made automatically generates symbol to realize to find out optimal policy Close the technical effect of the product model of design objective parameter.
In addition, excellent according to the execution of the second optimal design parameter working as the analysis product model generation module 140 of analysis module 150 Change design action product model generated when meeting the design objective parameter, then it can be by product model generated, described The second optimal design parameter that design objective parameter and figure network intensified learning module 160 are exported is input to data information In library 101, using as the sample pattern and its corresponding sample pattern index parameter and the sample optimization Design parameter is stored, to constantly optimize instruction for deep learning model constructed by the first deep neural network Practice.
Please refer to Fig. 3, for the process signal of the product model design method of the second embodiment of display the application Figure.As shown, step S31 is first carried out, its corresponding sample pattern index parameter of collecting sample model and sample optimization Design parameter, to establish data bank, wherein sample optimization design parameter refers to be set for optimization performed by sample pattern Meter movement, then carries out step S32.
In step S32, corresponding figure network is generated for the sample pattern stored in data bank.In the present embodiment In, it can pass through and sample pattern is input to the second deep neural network, to enable the second deep neural network defeated according to sample pattern Corresponding mappings characteristics figure out, and as the figure network for corresponding to inputted sample pattern.Specifically, can by from Nodal information, side information and global information are extracted in sample pattern, with raw using the nodal information, side information and global information At the figure network for corresponding to sample pattern, step S33 is then carried out.
In step S33, by the corresponding figure network of sample pattern, the sample pattern index parameter and the sample This optimal design parameter is inputted in the first deep neural network and is trained, to construct the deep learning of the first deep neural network Model.It, can be by analysis when in the update nodal information, the side information and the global information in an embodiment When one of them, variation caused by other the two in the nodal information, the side information and the global information, and The incidence relation between the nodal information, the side information and the global information is analyzed, thus to the figure of sample pattern Network is learnt, and step S34 is then carried out.
In step S34, the design objective parameter of input product model, to enable trained first deep neural network base In the deep learning model, and the first optimal design parameter is generated for the design objective parameter and is exported, then Carry out step S35.
In step S35, corresponding optimization design is executed according to the first optimal design parameter by emulation tool and is acted, from And generate the product model for meeting design objective parameter.In this present embodiment, executing the optimization design movement can be such as By adjusting the one of them in the nodal information, the side information and the global information in the figure network, with Other the two in the nodal information, the side information and the global information are enabled to generate corresponding change and realize.
It please cooperate Fig. 4, in an embodiment, the step S33 and step S34 in Fig. 3 can also be further refined as following several A operating procedure, specifically includes:
Step S41 enables the first deep neural network be based on the deep learning model, and is directed to inputted design objective Parameter prediction generates the first optimal design parameter and is exported.In an embodiment, it can first be referred to according to design by emulation tool It marks parameter and generates a basic model, to enable the first deep neural network based on basic model generated and the design inputted Index parameter, prediction generate the first optimal design parameter and are exported, and then carry out step S42.
In step S42, emulation tool is enabled to execute optimization design movement according to the first optimal design parameter, it is corresponding to generate Product model, then carry out step S43.
In step S43, whether analysis meets according to the first optimal design parameter product model generated is inputted Design objective parameter, wherein when analyzing result is to meet, then product model generated is exported, when analysis result is not to be inconsistent When conjunction, then step S44 is carried out.
In step S44, analysis meets the design objective according to whether first optimal design parameter can generate The product model of parameter can export the first optimization design ginseng in the present embodiment by the first deep neural network of analysis Whether the execution time that several numbers or the first deep neural network generate the first optimal design parameter reaches preset upper limit value To realize, wherein when the number or the first depth mind for judging the first optimal design parameter of current first deep neural network output When the operating time for generating the first optimal design parameter through network is not up to upper limit value, then S41 is returned to step, to enable first Deep neural network is predicted to generate the first new optimal design parameter again, and works as and judge current first deep neural network output The operating time that the number of first optimal design parameter or the first deep neural network generate the first optimal design parameter has reached When upper limit value, S45 is thened follow the steps.
In step S45, by third deep neural network to execute intensified learning step, from current generated multiple (i.e. according to the first optimal design parameter multiple product models generated) in product model, chooses closest to design objective and join A several product models, and the product model is converted into corresponding figure network to be input in third deep neural network, with It enables third deep neural network according to design objective parameter, generates the second optimal design parameter and exported, then to execute Step S42 enables emulation tool execute optimization design movement according to the second optimal design parameter, to generate new product model, then Then step S43 is executed, whether design objective ginseng is met according to the second optimal design parameter product model generated with analysis Number exports product model generated when analyzing result is to meet, and when analyzing result is not meet, then via step Step S45 is reentered after S44, to enable third deep neural network repeat intensified learning step.
In another embodiment, when being analyzed in step S43 according to the second optimal design parameter (i.e. intensified learning step) And the product model generated is when meeting design objective parameter, it can also be by product model generated, design objective parameter, Yi Ji Two optimal design parameters are input in data bank, using as the sample pattern and its corresponding sample pattern index Parameter and the sample optimization design parameter are stored, thus the correlated samples data constantly in expanding data data bank, These correlated samples data expanded in data bank, which can be further input into the first deep neural network, to be trained, To continue to optimize the deep learning model of the first deep neural network of adjustment.
Please continue to refer to Fig. 5, for the structural representation of the product model designing system of the 3rd embodiment of display the application Figure.As shown, the product model designing system 200 of the present embodiment mainly includes setting module 205, product model generation module 210, figure network generation module 220, the figure based on deeply learning algorithm (DeepReinforcementLearning, DRL) Network intensified learning module 230 and analysis module 240.
Setting module 205 is used to set the design objective parameter of product model yet to be built.
Product model generation module 210 is used to generate basic mould according to design objective parameter set by setting module 205 Type, or be directed to the corresponding optimization design movement of the basic model execution accordingly for the second optimal design parameter of reception and give birth to At the product model.In this present embodiment, described in product model generation module 210 can be established by design of Simulation tool software Basic model or the product model.
Figure network generation module 220 is used to the basic model generated of product model generation module 210 being converted to correspondence Figure network.In an embodiment, figure network generation module 220, which is penetrated, is input to the second deep neural network for basic model (such as deep neural network of CNN framework), so that the second deep neural network is special according to the corresponding mapping of basic model output Sign figure, and using the mappings characteristics figure exported as the figure network for corresponding to inputted basic model.In another embodiment, Figure network generation module 220 from basic model by extracting nodal information, side information and global information, to utilize the node Information, side information and global information generate the figure network for corresponding to sample pattern.
Figure network intensified learning module 230 is used to construct the intensified learning model of third deep neural network, and will be described The figure network inputs third deep neural network generated of figure network generation module 220, to enable third deep neural network Based on constructed intensified learning model, and the second optimization design ginseng is generated for inputted design objective parameter prediction It counts and is exported.Specifically, figure network intensified learning module 230 passes through the figure generated of figure network generation module 220 The third deep neural network that network inputs are built based on deeply learning algorithm, to enable third deep neural network foundation set Design objective parameter set by cover half block 205 generates the second optimal design parameter and is exported, that is, exports basic mould The optimization design action parameter of type required execution during design optimization.
In addition, the second optimization design that product model generation module 210 will be exported according to figure network intensified learning module 230 Parameter executes corresponding optimization design to basic model and acts, and generates new product model.Specifically, being based on deeply The figure e-learning module 230 of learning algorithm can be by analysis when updating the nodal information, the side information and described complete When one of them in office's information, both other in the nodal information, the side information and the global information are produced Raw variation, and the incidence relation between the nodal information, the side information and the global information is analyzed, to generate Two optimal design parameters.
Further, the figure e-learning module 230 based on deeply learning algorithm is online deeply study rank Section, deeply learning algorithm used include but are not limited to: Deep Q Learning, Double Q-Network, Deep Deterministic PolicyGradien, Actor-Critic etc..
Specifically, the figure e-learning module 230 based on deeply learning algorithm can be strong as depth using figure network The state that chemistry is practised transfers to tactful network to be divided by the individual features of the mapped out figure network of the second deep neural network Analysis, provides the prediction optimization design action (i.e. the second optimal design parameter) to current production model (i.e. basic model)；Then The quality for the prediction optimization design action for transferring to value network assessment strategy network to be made again, to find out optimal policy, thus Realize the technical effect for automatically generating the product model for meeting design objective parameter.
Analysis module 240 is used to analyze the product model of product model generation module 210, that is, strong based on depth Whether the product model generated of figure e-learning module 230 for changing learning algorithm meets inputted design objective parameter, when Analysis result is when meeting, to export the product model；When analyzing result is not meet, then exports analysis signal and feed back to Figure e-learning module 230 based on deeply learning algorithm, to enable the figure e-learning based on deeply learning algorithm Module 230 regenerates the second new optimal design parameter, and re-executes prediction optimization design action accordingly.
In addition, figure network intensified learning module 230 is further used for the analysis according to analysis module 240 as a result, being directed to Third deep neural network is trained, to adjust the intensified learning model for updating third deep neural network.
As shown in fig. 6, it is the flow diagram for showing the product model design method of fourth embodiment of the application.
As shown, step S61 is first carried out, the design objective parameter of product model yet to be built is set, is then walked Rapid S62.
In step S62, a product is established according to set design objective parameter by design of Simulation tool software Model then carries out step S63 as basic model.
In step S63, basic model generated is converted into corresponding figure network.In an embodiment, this step It can pass through and basic model is inputted into the second deep neural network, so that the second deep neural network is answered according to sample pattern output phase Mappings characteristics figure, using as correspond to basic model figure network.Specifically, can be by extracting node from basic model Information, side information and global information correspond to sample pattern to generate using the nodal information, side information and global information Figure network.Then step S64 is carried out.
In step S64, the state input that figure network generated learns as deeply is based on deeply The third deep neural network of algorithm is practised, to enable third deep neural network according to design objective parameter, the second optimization is generated and sets Meter parameter is simultaneously exported.Specifically, by the individual features of the mapped out figure network of the second deep neural network, the is transferred to The tactful network of three deep neural networks is analyzed, and the prediction optimization design to current production model (i.e. basic model) is provided Movement (i.e. the corresponding optimization design movement of the second optimal design parameter)；Then the value network of third deep neural network is transferred to again The quality for the prediction optimization design action that network assessment strategy network is made, to find out optimal policy, to generate deisgn product The optimal movement of model.Then step S65 is carried out.
In step S65, using the second optimal design parameter of design of Simulation tool software foundation step S64 output to step The basic model that rapid S62 is generated is debugged, to generate new model.Then step S66 is carried out.
In step S66, analyze whether product model generated meets design objective parameter, and when analysis result is symbol When conjunction, output products model, and when analyzing result is not meet then enables third deep neural network again according to being inputted Design objective parameter prediction generates a second new optimal design parameter and is exported, until product model generated meets The design objective parameter.
Further illustrate that the product model of the application is set below with reference to the design embodiment of two specific product models Meter systems and method.
Please refer to Fig. 8 A and Fig. 8 B, wherein Fig. 8 A shows that the product model of filter circuit, Fig. 8 B show figure The corresponding figure network (i.e. relation structure diagram) of filter circuit product model shown in 8A.Present embodiment discloses utilize the application Product model designing system and method and the product model designing technique of filter circuit realized, specifically include: firstly, defeated Enter the index parameter of filter circuit；Then, the basic model of a filter circuit is established by design of Simulation tool software, It recycles the second deep neural network that basic model generated is abstracted into one and includes input interface, output interface, resonance The relational structure (i.e. figure network shown in Fig. 8 B) of 2 four device 1, resonator nodes, wherein each node is produced with other nodes Raw electromagnetic relationship.First deep neural network or third deep neural network (the possible body of artificial intelligence) can be based on interfaces and emulation Interaction between tool, and the relationship between node and node (i.e. adjusting side information) is adjusted, thus analysis and learning filters The principle of the basic model of circuit.Specifically, typical filter circuit is by input interface, output interface, resonator institute Composition, wherein only input interface, keep specific electromagnetic coupling relationship between output interface resonator, and resonator In the case that resonance frequency meets special value, entire circuit is just able to achieve the function of filter.First deep neural network can It is configured as figure network corresponding to learning filters circuit, extracts resonance frequency and electricity from simulation model and emulation data Magnetic coupling relation, and these numerical value are adjusted, and then realize the Automatic Optimal Design of filter circuit, it is disclosed in this embodiment Design framework be intended to reach using the combination of deep learning, intensified learning or deep learning and intensified learning about filter The optimizing decision of wave device circuit model design and implementation.
Please refer to Fig. 9 A and Fig. 9 B, wherein Fig. 9 A shows that the product model of tripod table, Fig. 9 B show Fig. 9 A institute The corresponding figure network (i.e. relation structure diagram) of the tripod table product model shown.Present embodiment discloses the product moulds using the application Type designing system and method and the designing technique of tripod table mechanical model realized, using figure network technology by tripod table model It is abstracted into a kind of relational structure, and therefrom study and reasoning, realizes the best decision of the design and implementation about tripod table model, It specifically includes: firstly, the design objective parameter based on tripod table establishes three feet by controlling design of Simulation tool software The fundamental mechanics model of table；Then, the basic model is abstracted into one using the second deep neural network and includes desktop, table leg 1, the relational structure (i.e. figure network shown in Fig. 9 B) of 3 four table leg 2, table leg nodes, wherein each node with other nodes Generate stress relationship.Then, the first deep neural network can pass through the interaction between emulation tool, adjust node and node it Between relationship, to understand the principle of the tripod table model.For example, the given Impact direction and stress size the case where Under, there is an optimizing decision (i.e. optimization design movements), and the thickness of desktop cannot be too thick first, and table leg cannot be too thick, no The then rising of the excessive waste that will lead to material of surplus, the increase of weight and cost.Therefore, artificial intelligence intelligence (the first depth mind Through network) it is configured as learning from interaction, it is continuous to adjust the parameters such as desktop thickness, table leg size, position legs, realize three The design of foot table.The design framework disclosed in this embodiment is intended to utilize deep learning, intensified learning or deep learning and strong The combination that chemistry is practised reaches the optimizing decision about tripod table modelling and realization, utilizes least material, most light weight Amount, minimum cost realize maximum stress.
In conclusion product model designing system provided herein and method, pass through collecting sample model and its sample This model index parameter and sample optimization design parameter to establish data bank, and by by sample pattern with relational structure Change mode is converted to corresponding figure network, and deep neural network is recycled to be learnt for figure network and generate model optimization and set Meter rule can find out the model optimization design rule to match according to the design objective parameter of the product model of input for subsequent Optimization design movement is then executed, to realize the technical effect for automatically generating the product model for meeting design objective parameter.
In addition, the application more can carry out intensified learning for model optimization design rule generated, the production according to input The design objective parameter of product model and generate the second optimal design parameter, and execute prediction optimization design action accordingly, and it is automatic Generate the product model for meeting design objective parameter.
Whereby, the rule that the present invention is designed in deep learning frame using figure network algorithm learning model, and then make to produce The design technology of product model is intelligent, can be stepped up by the study of continuous relation inference to product model design rule The accuracy rate of cognition and design.Therefore, the application can not only realize product model the Automation Design, and product model also can be improved Design efficiency, and have the advantages that reusable degree is high and design cost is cheap.
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 (RAM), 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 (18)

1. a kind of product model designing system based on figure network characterized by comprising
Data acquisition module is used for its corresponding sample pattern index parameter of collecting sample model and sample optimization design ginseng Number, to establish data bank, wherein the sample optimization design parameter is indicated for optimization performed by the sample pattern Design action；
Setting module is used to set the design objective parameter of product model yet to be built；
Figure network generation module is used to generate corresponding figure network for the sample pattern in the data bank；
Figure e-learning module is used for the corresponding figure network of the sample pattern, the sample pattern index parameter And the sample optimization design parameter is inputted in the first deep neural network and is trained, to construct the first depth nerve The deep learning model of network, to enable first deep neural network based on the constructed deep learning model, and needle First optimal design parameter is generated to the design objective parameter prediction and is exported；And
Product model generation module is used to receive first optimal design parameter, and it is dynamic to execute corresponding optimization design accordingly Make, and generates the product model for meeting the design objective parameter.
2. product model designing system according to claim 1, which is characterized in that the figure network generation module is further For by by the sample pattern input the second deep neural network, so that second deep neural network is according to the sample This model exports corresponding mappings characteristics figure, and using the mappings characteristics figure as the figure net for corresponding to the sample pattern Network.
3. product model designing system according to claim 2, which is characterized in that
The figure network generation module is further used for extracting nodal information, side information and global letter from the sample pattern Breath, to generate the figure network for corresponding to the sample pattern using the nodal information, side information and global information；
The figure e-learning module is further used for analysis when the update nodal information, the side information and the overall situation When one of them in information, produced by other the two in the nodal information, the side information and the global information Variation, and analyze the incidence relation between the nodal information, the side information and the global information；And
The product model generation module is further used for by adjusting the nodal information in the figure network, side letter One of them in breath and the global information, to enable in the nodal information, the side information and the global information It is both other generate corresponding changes, to execute the optimization design movement.
4. product model designing system according to claim 1 or 3, which is characterized in that the system further comprises:
Analysis module is used to analyze the product model generation module and sets according to first optimal design parameter execution optimization Meter acts whether the product model generated meets the design objective parameter, and when analyzing result is to meet, output The product model, and when analyzing result is not meet, output analysis signal is to the figure e-learning module, to pass through Stating figure e-learning module enables first deep neural network based on the constructed deep learning model, and for described Design objective parameter is predicted to generate new first optimal design parameter again and be exported.
5. product model designing system according to claim 4, which is characterized in that the system further comprises:
Figure network intensified learning module is used for when first optimization design that can not be exported by the figure e-learning module Parameter generation is when meeting the product model of the design objective parameter, from the product model generation module according to described the In one optimal design parameter multiple product models generated, the production closest to the design objective parameter is chosen Product model, and by by the corresponding figure network inputs of the product model to third deep neural network, to enable the third deep Degree neural network generates the second optimal design parameter according to the design objective parameter and is exported；And wherein,
The product model generation module is further used for receiving second optimal design parameter, executes corresponding optimization accordingly Design action, and generate the product model；
The analysis module is further used for analyzing the product model generation module to be held according to second optimal design parameter Row optimization design acts whether the product model generated meets the design objective parameter, and when analysis result is to meet When, the product model is exported, and when analyzing result is not meet, output analysis signal to the figure network intensified learning mould Block, to enable the third deep neural network again according to the design objective parameter by the figure network intensified learning module It generates new second optimal design parameter and is exported.
6. product model designing system according to claim 5, which is characterized in that the analysis module is further used for working as It is generated described according to second optimal design parameter execution optimization design movement to analyze the product model generation module When product model meets the design objective parameter, by the product model, the design objective parameter and second optimization Design parameter is input in the data bank to join as the sample pattern and its corresponding sample pattern index The several and described sample optimization design parameter is stored.
7. a kind of product model design method based on figure network characterized by comprising
Its corresponding sample pattern index parameter of collecting sample model and sample optimization design parameter, to establish data information Library, wherein the sample optimization design parameter is indicated for the movement of optimization design performed by the sample pattern；
Corresponding figure network is generated for the sample pattern stored in the data bank；
The corresponding figure network of the sample pattern, the sample pattern index parameter and the sample optimization are designed and joined It is trained in number the first deep neural network of input, to construct the deep learning model of first deep neural network；
The design objective parameter of input product model, to enable first deep neural network be based on the deep learning model, And the first optimal design parameter is generated for the design objective parameter and is exported；And by emulation tool according to described in First optimal design parameter executes corresponding optimization design movement, generates the product mould for meeting the design objective parameter Type.
8. product model design method according to claim 7, which is characterized in that for being stored in the data bank The sample pattern generate corresponding figure network and include:
The second deep neural network is inputted through by the sample pattern, to enable second deep neural network according to the sample This model exports corresponding mappings characteristics figure, and as the figure network for corresponding to the sample pattern.
9. product model design method according to claim 8, which is characterized in that the method further includes:
Nodal information, side information and global information are extracted from the sample pattern, with using the nodal information, side information and Global information generates the figure network for corresponding to the sample pattern；
It analyzes when updating the one of them in the nodal information, the side information and the global information, the node Variation caused by other the two in information, the side information and the global information, and learn the nodal information, institute State the incidence relation between side information and the global information；And
Execute optimization design movement, with by adjust the nodal information in the figure network, the side information and One of them in the global information, and enable other in the nodal information, the side information and the global information The two generates corresponding change.
10. the product model design method according to claim 7 or 9, which is characterized in that the generation meets the design The step of product model of index parameter, further comprises:
It analyzes the emulation tool and executes the optimization design movement product generated according to first optimal design parameter Whether model meets the design objective parameter, and when analyzing result is to meet, exports the product model, and when analysis knot Fruit is when not meeting, first deep neural network to be enabled to be based on the deep learning model and be directed to the design objective parameter It regenerates the first new optimal design parameter and is exported.
11. product model design method according to claim 10, which is characterized in that the generation meets the design and refers to The step of marking the product model of parameter further comprises:
When analysis can not meet the product model of the design objective parameter according to first optimal design parameter generation When, by third deep neural network to execute intensified learning step, to choose from multiple product models generated Closest to a product model of the design objective parameter, and by the figure network inputs of the product model to the third Deep neural network, to enable the third deep neural network generate the second optimal design parameter according to the design objective parameter And it is exported；
It is new to generate that according to second optimal design parameter movement of corresponding optimization design is executed by the emulation tool The product model；And
It analyzes the emulation tool and executes the optimization design movement product generated according to second optimal design parameter Whether model meets the design objective parameter, and when analyzing result is to meet, exports the product model, and when analysis knot Fruit is when not meeting, then third deep neural network to be enabled to re-execute the intensified learning step.
12. product model design method according to claim 11, which is characterized in that the method further includes working as to divide It analyses the emulation tool and executes the optimization design movement product model symbol generated according to second optimal design parameter When closing the design objective parameter, by the product model generated, the design objective parameter and the third depth Second optimal design parameter that neural network is exported is input in the data bank, using as the sample pattern And its corresponding sample pattern index parameter and the sample optimization design parameter are stored.
13. a kind of product model designing system based on figure network characterized by comprising
Setting module is used to set the design objective parameter of product model yet to be built；
Product model generation module is used to generate basic model according to the design objective parameter, and receives the second optimization Design parameter is directed to the basic model accordingly and executes corresponding optimization design movement and generate the product model；
Figure network generation module is used to generate corresponding figure network for the basic model；
Figure network intensified learning module, is used to construct the intensified learning model of third deep neural network, and by the figure net Third deep neural network described in the network generation module figure network inputs generated, to enable the third deep neural network Based on the intensified learning model, and second optimal design parameter is generated for the design objective parameter prediction and is given Output；And
Analysis module is used to analyze the product model generation module according to second optimal design parameter institute generated It states whether product model meets the design objective parameter, and when analyzing result is to meet, exports the product model, and work as Analysis result is when not meeting, and output analysis signal is to the figure network intensified learning module, to enable the figure network extensive chemical It practises module and new second optimization is generated according to the design objective parameter again by the third deep neural network Design parameter is simultaneously exported, until the product model for meeting the design objective parameter is generated, and wherein,
The figure network intensified learning module is further used for the analysis according to the analysis module as a result, deep for the third Degree neural network is trained, and updates the intensified learning model with adjustment.
14. product model designing system according to claim 13, which is characterized in that the figure network generation module is into one Step for by by the basic model input the second deep neural network, so that second deep neural network exports accordingly Corresponding mappings characteristics figure, and using the mappings characteristics figure as the figure network for corresponding to the basic model.
15. product model designing system according to claim 14, which is characterized in that
The figure network generation module is further used for extracting nodal information, side information and global letter from the basic model Breath, to generate the figure network for corresponding to the sample pattern using the nodal information, side information and global information；And
The product model generation module is further used for by adjusting the nodal information in the figure network, side letter One of them in breath and the global information, to enable in the nodal information, the side information and the global information It is both other generate corresponding changes, to execute the optimization design movement.
16. a kind of product model design method based on figure network characterized by comprising
Set the design objective parameter of product model yet to be built；
Basic model is generated according to the design objective parameter by the emulation tool；
Corresponding figure network is generated for the basic model；
By the corresponding figure network inputs third deep neural network of the basic model, with enable the third deep neural network according to According to the design objective parameter, generates second optimal design parameter and exported；
Corresponding optimization design is executed to the basic model according to second optimal design parameter by the emulation tool Movement is to generate the product model；And
Analyze whether the product model generated meets the design objective parameter, and when analyzing result is to meet, it is defeated The product model out, and when analyzing result is not meet, enable the third deep neural network again according to the design Index parameter generates new second optimal design parameter and is exported, until the product model generated meets The design objective parameter.
17. product model design method according to claim 16, which is characterized in that for basic model generation pair The figure network answered further comprises:
The basic model is inputted into the second deep neural network, so that second deep neural network is according to the basic mould Type exports corresponding mappings characteristics figure, using as the figure network for corresponding to the sample pattern.
18. product model design method according to claim 17, which is characterized in that the method further includes:
Nodal information, side information and global information are extracted from the sample pattern, with using the nodal information, side information and Global information generates the figure network for corresponding to the sample pattern；And
Execute optimization design movement, with by adjust the nodal information in the figure network, the side information and One of them in the global information, it is other in the nodal information, the side information and the global information to enable The two generates corresponding change.
CN201810846434.1A 2018-07-27 2018-07-27 Product model design system and method based on graph network Active CN109101712B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810846434.1A CN109101712B (en) 2018-07-27 2018-07-27 Product model design system and method based on graph network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810846434.1A CN109101712B (en) 2018-07-27 2018-07-27 Product model design system and method based on graph network

Publications (2)

Publication Number Publication Date
CN109101712A true CN109101712A (en) 2018-12-28
CN109101712B CN109101712B (en) 2023-06-20

Family

ID=64847929

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810846434.1A Active CN109101712B (en) 2018-07-27 2018-07-27 Product model design system and method based on graph network

Country Status (1)

Country Link
CN (1) CN109101712B (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110188039A (en) * 2019-05-20 2019-08-30 南京星火技术有限公司 The method and system of software test, software optimization
CN110210654A (en) * 2019-05-20 2019-09-06 南京星火技术有限公司 Product model designing system and method
CN110262511A (en) * 2019-07-12 2019-09-20 同济人工智能研究院(苏州)有限公司 Biped robot's adaptivity ambulation control method based on deeply study
CN110414012A (en) * 2019-07-29 2019-11-05 腾讯科技（深圳）有限公司 A kind of encoder construction method and relevant device based on artificial intelligence
CN111259603A (en) * 2020-01-17 2020-06-09 南京星火技术有限公司 Electronic device, model designing apparatus, and computer-readable medium
CN111291432A (en) * 2020-03-06 2020-06-16 南京航空航天大学 Intelligent product appearance design method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1979496A (en) * 2005-12-02 2007-06-13 中国科学院金属研究所 Copper-alloy pipe-material casting-milling technology parameter designing and optimizing method
CN101226606A (en) * 2007-11-16 2008-07-23 燕山大学 Method for optimization of socket-shaped part machinery expanding technological parameter
CN101320400A (en) * 2008-07-16 2008-12-10 桂林电子科技大学 Optimization design method of micro-electron packaging device based on artificial neural network
CN106503800A (en) * 2016-10-11 2017-03-15 天津大学 Deep learning model based on complex network and the application in measurement signal analysis
GB201713419D0 (en) * 2016-08-26 2017-10-04 Ford Global Tech Llc Physics modeling for radar and ultrasonic sensors

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1979496A (en) * 2005-12-02 2007-06-13 中国科学院金属研究所 Copper-alloy pipe-material casting-milling technology parameter designing and optimizing method
CN101226606A (en) * 2007-11-16 2008-07-23 燕山大学 Method for optimization of socket-shaped part machinery expanding technological parameter
CN101320400A (en) * 2008-07-16 2008-12-10 桂林电子科技大学 Optimization design method of micro-electron packaging device based on artificial neural network
GB201713419D0 (en) * 2016-08-26 2017-10-04 Ford Global Tech Llc Physics modeling for radar and ultrasonic sensors
CN106503800A (en) * 2016-10-11 2017-03-15 天津大学 Deep learning model based on complex network and the application in measurement signal analysis

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
YUJIA LI ET AL.: "GATED GRAPH SEQUENCE NEURAL NETWORKS", 《HTTPS://ARXIV.ORG/ABS/1511.05493》 *
赵明峰: "小型制冷装置冷凝器的仿真与实验研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110188039A (en) * 2019-05-20 2019-08-30 南京星火技术有限公司 The method and system of software test, software optimization
CN110210654A (en) * 2019-05-20 2019-09-06 南京星火技术有限公司 Product model designing system and method
CN110262511A (en) * 2019-07-12 2019-09-20 同济人工智能研究院(苏州)有限公司 Biped robot's adaptivity ambulation control method based on deeply study
CN110262511B (en) * 2019-07-12 2022-08-09 同济人工智能研究院(苏州)有限公司 Biped robot adaptive walking control method based on deep reinforcement learning
CN110414012A (en) * 2019-07-29 2019-11-05 腾讯科技（深圳）有限公司 A kind of encoder construction method and relevant device based on artificial intelligence
CN110414012B (en) * 2019-07-29 2022-12-09 腾讯科技（深圳）有限公司 Artificial intelligence-based encoder construction method and related equipment
CN111259603A (en) * 2020-01-17 2020-06-09 南京星火技术有限公司 Electronic device, model designing apparatus, and computer-readable medium
CN111259603B (en) * 2020-01-17 2024-01-30 南京星火技术有限公司 Electronic device, model design apparatus, and computer-readable medium
CN111291432A (en) * 2020-03-06 2020-06-16 南京航空航天大学 Intelligent product appearance design method

Also Published As

Publication number Publication date
CN109101712B (en) 2023-06-20

Similar Documents

Publication Publication Date Title
CN109101712A (en) Product model designing system and method based on figure network
Golilarz et al. A new automatic method for control chart patterns recognition based on ConvNet and harris hawks meta heuristic optimization algorithm
US11120365B2 (en) For hierarchical decomposition deep reinforcement learning for an artificial intelligence model
CN106096727A (en) A kind of network model based on machine learning building method and device
CN110427006A (en) A kind of multi-agent cooperative control system and method for process industry
CN108595815A (en) Artificial intelligence body training system and passive circuit Optimum Design System and method
CN110210654A (en) Product model designing system and method
CN109344969B (en) Neural network system, training method thereof, and computer-readable medium
Jafar et al. Hyperparameter optimization for deep residual learning in image classification
Cheng et al. Parallel cyber-physical-social systems based smart energy robotic dispatcher and knowledge automation: Concepts, architectures, and challenges
CN112528591A (en) Automatic PCB wiring method based on joint Monte Carlo tree search
CN115169235A (en) Super surface unit structure inverse design method based on improved generation of countermeasure network
Łapa et al. Aspects of structure and parameters selection of control systems using selected multi-population algorithms
CN109255148A (en) Mechanics product design method and its system
Ge et al. Stability switches and fold-Hopf bifurcations in an inertial four-neuron network model with coupling delay
Anochi et al. Self-configuring two types of neural networks by mpca
Michel et al. Weak interaction and strong interaction in agent based simulations
Qu et al. Adaptive planning of human–robot collaborative disassembly for end-of-life lithium-ion batteries based on digital twin
CN105893667A (en) Heterogeneous simulation data unified integrated visual method
JP2023076513A5 (en)
CN110188039A (en) The method and system of software test, software optimization
Kamada et al. Knowledge extracted from recurrent deep belief network for real time deterministic control
CN115859780A (en) Power equipment vibration simulation method and system based on physical constraint neural network
CN110210838B (en) Government affair system design method and government affair system
Shanthi et al. The Blue Brain Technology using Machine Learning

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