CN108536976A - A kind of multidisciplinary optimization software platform based on agent model - Google Patents
A kind of multidisciplinary optimization software platform based on agent model Download PDFInfo
- Publication number
- CN108536976A CN108536976A CN201810337669.8A CN201810337669A CN108536976A CN 108536976 A CN108536976 A CN 108536976A CN 201810337669 A CN201810337669 A CN 201810337669A CN 108536976 A CN108536976 A CN 108536976A
- Authority
- CN
- China
- Prior art keywords
- model
- submodule
- data
- optimization
- type
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of multidisciplinary optimization software platform based on agent model, belongs to Optimization Platform technical field.One multidisciplinary optimization platform based on agent model, this platform are write based on Python, and interface is built using PyQt.The platform is divided into three big modules:DOE modules, Surrogate models modules, Optimization modules.Data are chosen from DOE modules first, into Surrogate models modules, the data of selection is imported, builds agent model, can be optimized using Optimization modules if model meets required precision.This platform is easy to operate, occupy little space and can be independently installed advantage, be suitble to unprofessional user application proxy model to optimize and be also suitable professional user, be conducive to the popularization of agent model.
Description
Technical field
The present invention relates to a kind of multidisciplinary optimization software platform based on agent model, belongs to Optimization Platform technical field.
Background technology
Although technical development of computer is rapid, the stabilization and lasting growth of computing capability and speed, many analyses are still
It is so computationally costly.And complex engineering design problem increasingly becomes more, needs to solve in a short time.
And the important link that optimization is mechanical design field is designed to machine components and system.Parametric designing needs
Parameter is varied multiple times and finds results change trend, experimental cost is excessively high, mostly uses the modes such as computer simulation now and replaces
Experiment.But on complex engineering design problem, computer simulation experiment model is complicated, and single operation usually may require that a few hours very
Time to a couple of days completes.This makes Design and optimization and model analysis become hardly possible completion.Agent model can then substitute
Emulation experiment can effectively improve computational efficiency on the basis of ensureing computational accuracy, it is made to quickly grow.Agent model
(Surrogate Models) is that a kind of calculation amount is small but its result of calculation is approximate with similar in the result of calculation of high-precision model
Substitution analysis model.When being optimized with agent model, mainly protected by experimental design (Design of Experience)
The uniform and randomness for demonstrate,proving data, establishes agent model (Surrogate Models), to building with the data that experimental design obtains
The agent model stood carries out precision evaluation, is being optimized (Optimization) if meeting required precision.But it is answering
The ability of certain Fundamentals of Mathematics and application software is needed with agent model, is limited layman in this way and is done and optimizes,
And it is unfavorable for promoting the use of for agent model.
The software of multidisciplinary optimization is less on the market at present, and wherein Isight is applied than wide, is by the rich of MIT earliest
Scholar Siu S.Tong are proposed in or so the last century 80's and exploitation are led to complete, and are had become by development these years
Outstanding person in similar software.The Optimization Toolbox that nineteen ninety depends on Matlab is also issued so that agent model obtains further
Use.Although Isight can provide multidisciplinary design optimization and different levels optimisation technique and optimization process management energy
Power, complicated, the troublesome in poeration, elapsed time but Isight software learnings get up;Matlab Optimization Toolboxes can only installed
It uses, depends on and Matlab in the case of Matlab, occupied space.So in the case where ensureing precise manner, need design one right
Non-professional people is easy to operate, installs the Optimization Platform that occupies little space and can be done for professional people simple task, not only
The optimization of amateur people can be met, it helps the popularization of agent model technology.
Invention content
The present invention provides one and is suitable for the multidisciplinary optimization platform that amateur people's application proxy model does to optimize, to overcome
Multidisciplinary optimization software is complicated for operation, depends on the shortcomings that other software.
Technical scheme of the present invention:
A kind of multidisciplinary optimization software platform based on agent model, this platform are write based on Python, application
PyQt sets for this platform control interfaces, including DOE modules, Surrogate models modules and Optimization modules;
Module one:Experimental design module
Experimental design (DOE) module is to choose data module, and based on different strategies, DOE modules include sampling type
Module, input parameter submodule, output sample point and figure submodule;
The sampling type submodule includes four options, is selected according to user demand, including Latin hypercube
Test option (Latin Hypercube Sampling design), BBD experiment option (Box-Behnken design), in
The heart tests option (Center Composite design) and full factorial test option (Full Factorial Sampling
design);
The input parameter submodule, according to the sampling selected option of type submodule, corresponding different input ginseng
Number;Under the sampling type selected, input parameter determines dimension, number and the distribution in space of output data;
The output sample point and figure submodule shows distribution situation in its space, i.e. display pair according to selected dimension
The point diagram answered;It is up to three-dimensional;
Module two:Agent model module
Agent model (Surrogate Model) is the nucleus module for building model module and platform, is divided into five
Submodule:Selection establish agent model type submodule, data source and input parameter submodule, output model figure submodule,
Model accuracy evaluates submodule and sequence and adds some points submodule;
It includes three options that agent model type submodule is established in the selection:Kriging model options, radial base letter
Number (RBF) model options and polynomial response surface method (PRS) model options;According to user demand, determine that a model type selecting is built
Found corresponding agent model;
The data source and input parameter submodule includes two parts:Select data source part and input parameter
Part;Selection data source part includes the data options and user's legacy data option that experimental design module is chosen;Input ginseng
Number part inputs corresponding parameter according to the agent model to be established;
The output model and figure submodule is by inputting training data, generating model, can establish multiple models;Choosing
The variable to be drawn and target are selected, two-dimentional contour map and diagram of block are obtained;The training data of variable is up to bidimensional, institute
The target of choosing is one-dimensional;
The described model accuracy evaluation submodule be by incoming inspection data, comprising variable data and target value data,
Evaluate built model accuracy;Meet if user requires if model accuracy and apply the model, is otherwise added some points submodule by sequence
Block, which improves model accuracy or chooses data again, establishes agent model;
Sequence submodule of adding some points is added some points submodule by sequence when the precision of model is unsatisfactory for user and requires
Improve the accuracy variable data of model;The data that will obtain adding some points are imported into training data, re -training model;
Module three:Optimization module
Optimize (Optimization) module based on the agent model having built up, the model having had built up is carried out excellent
Change, obtains optimum results;It is divided into three submodules:Selection optimization type submodule, setting Optimal Parameters submodule and output are excellent
Change result submodule;
The selection optimizes type submodule, is selected according to user, with Different Optimization type to the agent model of foundation
It optimizes;
The setting Optimal Parameters submodule, the optimization type of corresponding choosing input corresponding parameter, including initial point,
The model of optimization, precision, maximum iteration required for selection;
The output result submodule, after optimization, display is as a result, including each at iterative steps, optimal value, optimal value
The value of a variable.
Beneficial effects of the present invention:The present invention provides an Optimization Platform, can make operation under the premise of ensureing precision
Simply, it is only necessary to which optimization task can be completed by click control step by step;It is independently installed without depending on other software,
It occupies little space.Suitable layman does optimization task, and very positive effect is played to the popularization of agent model.
Description of the drawings
Fig. 1 is the structure subordinate figure of the multidisciplinary optimization platform based on agent model.
The data transfer flow chart of multidisciplinary optimization platforms of the Fig. 2 based on agent model.
The DOE module construction flow charts of multidisciplinary optimization platforms of the Fig. 3 based on agent model.
The Surrogate model module construction flow charts of multidisciplinary optimization platforms of the Fig. 4 based on agent model.
The Optimization module construction flow charts of multidisciplinary optimization platforms of the Fig. 5 based on agent model.
Specific implementation mode
Below in conjunction with attached drawing and technical solution, the specific implementation mode that further illustrates the present invention.
A kind of multidisciplinary optimization software platform based on agent model, this platform are write based on Python, application
PyQt sets for this platform control interfaces, including DOE modules, Surrogate models modules and Optimization modules;
Module one:Experimental design module
Experimental design (DOE) module is to choose data module, and based on different strategies, DOE modules include sampling type
Module, input parameter submodule, output sample point and figure submodule;
The sampling type submodule includes four options, is selected according to user demand, including Latin hypercube
Test option (Latin Hypercube Sampling design), BBD experiment option (Box-Behnken design), in
The heart tests option (Center Composite design) and full factorial test option (Full Factorial Sampling
design);
The input parameter submodule, according to the sampling selected option of type submodule, corresponding different input ginseng
Number;Under the sampling type selected, input parameter determines dimension, number and the distribution in space of output data;
The output sample point and figure submodule shows distribution situation in its space, i.e. display pair according to selected dimension
The point diagram answered;It is up to three-dimensional;
(1.1) according to the demand of user, an option is chosen in sampling type submodule;
(1.2) option determined according to step (1.1) inputs corresponding parameter, including up-and-down boundary, sampling number, dimension
Number;When input, whether system automatic decision input by user is number, if not then re-entering;
(1.3) obtained data point is shown in Table tables;According to user demand, dimension, display corresponding two are determined
Dimension or three-dimensional scatter plot;If user is dissatisfied to the quality of selected data point, all data points are emptied, step (1.1) is returned to
Restart;User can click save button and the data point of generation is saved in Excel;
Module two:Agent model module
Agent model (Surrogate Model) is the nucleus module for building model module and platform, is divided into five
Submodule:Selection establish agent model type submodule, data source and input parameter submodule, output model figure submodule,
Model accuracy evaluates submodule and sequence and adds some points submodule;
It includes three options that agent model type submodule is established in the selection:Kriging model options, radial base letter
Number (RBF) model options and polynomial response surface method (PRS) model options;According to user demand, a types of models is selected to build
Found corresponding agent model;
The data source and input parameter submodule includes two parts:Select data source part and input parameter
Part;Selection data source part includes the data and user's legacy data that experimental design module is chosen;Input parameter part root
Corresponding parameter is inputted according to the type of the agent model to be established;
The output model figure submodule is by inputting training data, generating model, can establish multiple models simultaneously;
The variable to be drawn and target are selected, two-dimentional contour map and diagram of block are obtained;The dimension of selected variable is up to two
Dimension, selected target are one-dimensional;
The described model accuracy evaluation submodule be by incoming inspection data, comprising variable data and target value data,
Evaluate built model accuracy;Meet if user requires if model accuracy and apply the model, is otherwise added some points submodule by sequence
Block, which improves model accuracy or chooses data again, establishes agent model;
Sequence submodule of adding some points is added some points submodule by sequence when the precision of model is unsatisfactory for user and requires
Improve the accuracy variable data of model;The data that will obtain adding some points are imported into training data, re -training model;
(2.1) according to the demand of user, it is established in selection and determines a model options in agent model type submodule;
(2.2) when user does not have data, the data chosen in experimental design module of data source are selected;When user has number
According to when, be introduced directly into;After data input, input is according to the determining corresponding parameter of model of step (2.1);
(2.3) model is established:It clicks button and establishes model, after the completion of model foundation, select the dimension and target of variable, structure
The two-dimentional contour map and diagram of block of established model;Variable is to more options bidimensional;Structure agent model builds multiple simultaneously
Model determines that the number of model is established in input according to user demand in data source and input parameter submodule;In display mould
When type figure, if only building a model, then target to be shown need not be selected;If build multiple models, it is determined that aobvious
Show any of which target;
(2.4) inspection data is separately input in two tables, shows variable data and corresponding desired value number respectively
According to;Prediction Value Data is further obtained, and is output in corresponding table;It is obtained by model accuracy evaluation submodule relevant
Evaluation of estimate determines whether model meets required precision;
(2.5) if model accuracy is unsatisfactory for, according to user demand, in sequence adds some points submodule input add some points number with
And the point of model accuracy is improved after determining type of adding some points;Point is added in the training data of step (2.2), returns to step
(2.3) model is established again;
Module three:Optimization module
Optimize (Optimization) module based on the agent model having built up, the model having had built up is carried out excellent
Change, obtains optimum results;It is divided into three submodules:Selection optimization type submodule, setting Optimal Parameters submodule and output are excellent
Change result submodule;
The selection optimizes type submodule, is selected according to user, with Different Optimization type to the agency that has built up
Model optimizes;
The setting Optimal Parameters submodule, the optimization type of corresponding choosing input corresponding parameter, including initial point,
The model of optimization, precision, maximum iteration required for selection;
The output result submodule, after optimization, display is as a result, including each at iterative steps, optimal value, optimal value
The value of a variable;
(3.1) according to user demand, selection optimization type;
(3.2) based on selected optimization type, input and select corresponding parameter;
(3.3) optimize, obtain optimum results;If result is not optimum point, re-optimization is returned.
Claims (1)
1. a kind of multidisciplinary optimization software platform based on agent model, which is characterized in that this platform is write based on Python
, it sets for this platform control interfaces using PyQt;The multidisciplinary optimization software platform includes experimental design module, agent model
And optimization module;
Module one:Experimental design module
Experimental design module is to choose data module, and based on different strategies, DOE modules include sampling type submodule, input
Parameter sub-module, output sample point and figure submodule;
The sampling type submodule includes four options, is selected according to user demand, including Latin hypercube experiment
Option, BBD experiments option, multi-center trial option and full factorial test option;
The input parameter submodule, according to the sampling selected option of type submodule, corresponding different input parameter;
Under the sampling type selected, input parameter determines dimension, number and the distribution in space of output data;
The output sample point and figure submodule shows distribution situation in its space, that is, shows corresponding according to selected dimension
Point diagram;It is up to three-dimensional;
1.1 according to the demand of user, and an option is chosen in sampling type submodule;
1.2 options determined according to step 1.1 input corresponding parameter, including up-and-down boundary, sampling number, dimension;Input
When, whether system automatic decision input by user is number, if not then re-entering;
1.3 are shown to obtained data point in table;According to user demand, dimension is determined, show corresponding two dimension or three-dimensional scattered
Point diagram;If user is dissatisfied to the quality of selected data point, all data points are emptied, step 1.1 is returned to and restarts;User
Save button can be clicked the data point of generation is saved in Excel;
Module two:Agent model module
Agent model is the nucleus module for building model module and platform, is divided into five submodules:Selection establishes and acts on behalf of mould
Type type submodule, data source and input parameter submodule, output model figure submodule, model accuracy evaluation submodule and
Sequence is added some points submodule;
It includes three options that agent model type submodule is established in the selection:Kriging model options, radial basis function mould
Type option and polynomial response surface method model options;According to user demand, selects a types of models to establish and corresponding act on behalf of mould
Type;
The data source and input parameter submodule includes two parts:Select data source part and input parameter portion
Point;Selection data source part includes the data and user's legacy data that experimental design module is chosen;Input parameter part according to
The type of the agent model to be established inputs corresponding parameter;
The output model figure submodule is by inputting training data, generating model, can establish multiple models simultaneously;Selection
The variable and target to be drawn, obtains two-dimentional contour map and diagram of block;The dimension of selected variable is up to bidimensional, institute
The target of choosing is one-dimensional;
The model accuracy evaluation submodule is by incoming inspection data, comprising variable data and target value data, evaluation
Built model accuracy;Meet if user requires if model accuracy and apply the model, is otherwise carried by sequence submodule of adding some points
High model accuracy or again choose data establish agent model;
Sequence submodule of adding some points is improved by sequence submodule of adding some points when the precision of model is unsatisfactory for user and requires
The accuracy variable data of model;The data that will obtain adding some points are imported into training data, re -training model;
2.1 according to the demand of user, is established in selection and determines a type for establishing model in agent model type submodule;
2.2 when user does not have data, the data for selecting data source to be chosen in experimental design module;When user has data,
It is introduced directly into;After data input, the corresponding parameter of model determined according to step 2.1 is inputted;
2.3 establish model:Model is established in click, after the completion of model foundation, is selected the dimension and target of variable, is built the two of model
Tie up contour map and diagram of block;Variable is to more options bidimensional;Structure agent model builds multiple models simultaneously, in data
According to user demand in source and input parameter submodule, determine that the number of model is established in input;In display model figure, if
A model is only built, then need not select target to be shown;If build multiple models, it is determined that display any of which mesh
Mark;
2.4 are separately input to inspection data in corresponding table, show variable data and corresponding target value data;Further
Prediction Value Data is obtained, and is output in corresponding table;Submodule is evaluated by model accuracy and obtains relevant evaluation of estimate, i.e.,
Determine whether model meets required precision;
If 2.5 model accuracies are unsatisfactory for, according to user demand, add some points number and determination are inputted in sequence adds some points submodule
It adds some points type, you can obtain to improve the point of model accuracy;Point is added in the data of step 2.2, again back to step 2.3
It is secondary to establish model;
Module three:Optimization module
Optimization module optimizes the model having had built up, is obtained optimum results based on the agent model having built up;Point
For three submodules:Selection optimization type submodule, setting Optimal Parameters submodule and output optimum results submodule;
The selection optimizes type submodule, is selected according to user, with Different Optimization type to the agent model that has built up
It optimizes;
The setting Optimal Parameters submodule, the optimization type of corresponding choosing input corresponding parameter, including initial point, selection
The model of required optimization, precision, maximum iteration;
The output result submodule, after optimization, display is as a result, including each change at iterative steps, optimal value, optimal value
The value of amount;
3.1 according to user demand, selection optimization type;
3.2, based on selected optimization type, input and select corresponding parameter;
3.3 optimizations, obtain optimum results;If result is not optimum point, re-optimization is returned.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810337669.8A CN108536976B (en) | 2018-04-11 | 2018-04-11 | Multidisciplinary optimization software platform based on agent model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810337669.8A CN108536976B (en) | 2018-04-11 | 2018-04-11 | Multidisciplinary optimization software platform based on agent model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108536976A true CN108536976A (en) | 2018-09-14 |
CN108536976B CN108536976B (en) | 2021-07-06 |
Family
ID=63481057
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810337669.8A Active CN108536976B (en) | 2018-04-11 | 2018-04-11 | Multidisciplinary optimization software platform based on agent model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108536976B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120109606A1 (en) * | 2010-10-31 | 2012-05-03 | International Business Machines Corporation | Automated interactive multi-objective optimization-based systems design tool |
CN104834774A (en) * | 2015-04-29 | 2015-08-12 | 西北工业大学 | Comprehensive optimization design method and design platform for stratospheric composite material propeller |
CN105183987A (en) * | 2015-09-07 | 2015-12-23 | 北京航空航天大学 | Complex engineering multidisciplinary synthetical design optimization software platform system |
CN106777616A (en) * | 2016-12-05 | 2017-05-31 | 安徽农业大学 | A kind of high-speed rice pot seedling planting machine seedling box mover Multidisciplinary Optimization method |
-
2018
- 2018-04-11 CN CN201810337669.8A patent/CN108536976B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120109606A1 (en) * | 2010-10-31 | 2012-05-03 | International Business Machines Corporation | Automated interactive multi-objective optimization-based systems design tool |
CN104834774A (en) * | 2015-04-29 | 2015-08-12 | 西北工业大学 | Comprehensive optimization design method and design platform for stratospheric composite material propeller |
CN105183987A (en) * | 2015-09-07 | 2015-12-23 | 北京航空航天大学 | Complex engineering multidisciplinary synthetical design optimization software platform system |
CN106777616A (en) * | 2016-12-05 | 2017-05-31 | 安徽农业大学 | A kind of high-speed rice pot seedling planting machine seedling box mover Multidisciplinary Optimization method |
Non-Patent Citations (2)
Title |
---|
JOHN T. HWANG 等: "《A fast-prediction surrogate model for large datasets》", 《AEROSPACE SCIENCE AND TECHNOLOGY》 * |
金伟楠: "《核主泵水力模型设计与优化一体化平台构建》", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Also Published As
Publication number | Publication date |
---|---|
CN108536976B (en) | 2021-07-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107480789A (en) | The efficient conversion method and device of a kind of deep learning model | |
JP7285977B2 (en) | Neural network training methods, devices, electronics, media and program products | |
CN102024067A (en) | Method for technology transplant of analog circuit | |
Venkataramani et al. | Logic synthesis of approximate circuits | |
CN109857459A (en) | A kind of E grades of supercomputer ocean model transplants optimization method and system automatically | |
Fu et al. | Auto-NBA: Efficient and effective search over the joint space of networks, bitwidths, and accelerators | |
CN110222407A (en) | A kind of fusion method and device of BIM data | |
CN114676522B (en) | Pneumatic shape optimization design method, system and equipment integrating GAN and migration learning | |
CN111143038B (en) | RISC-V architecture microprocessor kernel information model modeling and generating method | |
CN110648768B (en) | POM ocean mode optimization method and device | |
CN103763302A (en) | Web service combination generating method | |
CN105631135A (en) | Rapid modeling analysis system based on ABAQUS foundation pit excavation and application system thereof | |
CN108536976A (en) | A kind of multidisciplinary optimization software platform based on agent model | |
CN117193988A (en) | Task scheduling method and medium for wafer-level framework AI acceleration chip | |
CN111460552A (en) | Automatic design method between sample plates in toilet | |
WO2023284130A1 (en) | Chip and control method for convolution calculation, and electronic device | |
CN107862132B (en) | Automatic node deletion method for circuit approximate calculation | |
CN116432477B (en) | BIM-based construction simulation method, system and storage medium | |
CN110580517A (en) | Feature extraction method and device based on stacked self-encoder and terminal equipment | |
CN118092918B (en) | Website front-end development method and device based on line block diagram conversion and storage medium | |
CN116187265A (en) | Chip design method and terminal | |
CN109086537A (en) | Particle swarm algorithm accelerated method based on FPGA | |
CN108920755A (en) | Wiring method based on award mechanism | |
CN112800669B (en) | Method for evaluating various approximate technical errors based on probability map model in special accelerator | |
Wang et al. | Hardgbm: A framework for accurate and hardware-efficient gradient boosting machines |
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 |