CN105389614B - A kind of implementation method of neutral net self refresh process - Google Patents
A kind of implementation method of neutral net self refresh process Download PDFInfo
- Publication number
- CN105389614B CN105389614B CN201510905762.0A CN201510905762A CN105389614B CN 105389614 B CN105389614 B CN 105389614B CN 201510905762 A CN201510905762 A CN 201510905762A CN 105389614 B CN105389614 B CN 105389614B
- Authority
- CN
- China
- Prior art keywords
- neutral net
- training sample
- design
- temperature
- self refresh
- 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.)
- Expired - Fee Related
Links
Abstract
The invention discloses a kind of implementation method of neutral net self refresh process, design variable and design object are determined according to design object；The initial training sample of neutral net is obtained by sampling, realizes the training to neutral net；Utilize the cross and variation process of genetic algorithm, search meets the individual of design requirement, using neutral net and Fluid Mechanics Computation method, calculates the design object value of new individual, and Fluid Mechanics Computation method result of calculation is added in initial training sample, obtain updating training sample；Neutral net is trained again using renewal training sample, realizes the renewal of neutral net；Then selected, produce new population, if new population meets convergence, design process terminates, and otherwise return to step (3), are continued executing with.Compared with the prior art, self refresh neural net method proposed by the present invention makes the amount of calculation of reverse design reduce 27.4%.
Description
Technical field
The present invention relates to field of artificial intelligence, more particularly to a kind of realization side of neutral net self refresh process
Method.
Background technology
In artificial intelligence field, genetic algorithm is often used in reversely with Fluid Mechanics Computation (CFD), nerual network technique
Design process.Genetic algorithm calculates new when Fluid Mechanics Computation (CFD) is combined progress reverse design using only CFD approach
The design object value of individual, there may be more new individual in design process, therefore more CFD calculating process can be produced, caused
Reverse design method it is computationally intensive.Genetic algorithm is when neutral net is combined progress reverse design, using only artificial neuron
The design object value of network calculations new individual, the prediction error of neutral net may cause design process not restrain.Therefore, design
During, need to be simultaneously using CFD and the design object value of neural computing new individual.Reverse design is carried out using the method
When, amount of calculation is influenceed by train samples amount.But artificial empirically determined training sample in the conventional method, can only be passed through
This amount.If training sample amount is too small, neural network prediction error is big, causes to produce more unnecessary CFD meters in design process
Calculation process.If training sample amount is excessive, needs to pay many extra CFD calculating times, obtain the training sample of surplus.Cause
This, it is very necessary to build a kind of method that can automatically determine train samples amount.
The content of the invention
In order to overcome the problem of DC system fault isolation is difficult existing for above-mentioned prior art, the present invention proposes a kind of god
Implementation method through network self refresh process, train samples amount is automatically determined, reduce reverse design amount of calculation.
The present invention proposes a kind of implementation method of neutral net self refresh process, and this method comprises the following steps：
Step 1, entrance velocity and temperature and monitoring spot speed and temperature are determined according to Blay models；
Step 2, the initial training sample by acquisition neutral net of sampling, realize the training to neutral net；
Step 3, using genetic algorithm cross and variation process simultaneously obtain multigroup entrance velocity for meeting design requirement with
Temperature
The value of degree, and input parameter is used as, using neutral net and Fluid Mechanics Computation method, calculate the monitoring of new individual
Spot speed and temperature, i.e., CFD calculating is carried out to Blay models using these input parameters as boundary condition, obtains every group of input
Output parameter corresponding to parameter, i.e., speed and temperature value at monitoring point；And Fluid Mechanics Computation method result of calculation is added
Into training sample, obtain updating training sample；
Step 4, using renewal training sample neutral net is trained again, realize the renewal of neutral net；
Step 5 and then the individual in population is ranked up by non-dominated ranking method, and entered using tournament algorithm
Row selection, produces new population, if new population meets convergence, design process terminates, and otherwise return to step (3), are continued executing with.
Compared with the prior art, self refresh neural net method proposed by the present invention reduces the amount of calculation of reverse design
27.4%.
Brief description of the drawings
Fig. 1 is the reverse design method flow chart that the genetic algorithm of the present invention is combined with self refresh neutral net；
Fig. 2 is the Blay model schematics of the specific embodiment of the invention；A) geometry of Blay models, (b) grid are drawn
Point；
Fig. 3 be neutral net without self refresh when amount of calculation change curve；
Fig. 4 is amount of calculation change curve when neutral net has self refresh.
Embodiment
Below in conjunction with the drawings and the specific embodiments, technical scheme is described in further detail.
Fig. 2 is the geometry and boundary condition of model.The object of reverse design is Blay models, and its external structure is just
Square, size is 1.04 × 1.04m, is comprising an entrance and one outlet, entrance opening dimension 0.018m, outlet size
0.022m.Entrance velocity is 0.57m/s, and inlet temperature is 15 DEG C, and wind direction is horizontal direction, and surrounding uses the wall without sliding
Face boundary condition, wall surface temperature are set to constant temperature, and monitoring site is near exit.
(1) result of calculation when, neutral net is without self refresh：
When carrying out reverse design to Blay models, design variable is entrance velocity and temperature, and design object is at monitoring point
Speed and temperature.According to numerical result, it is 0.1325m/s to obtain the velocity magnitude at monitoring point, temperature level 17.67
DEG C, and then obtained the object function of Blay model reverse designs：
The crossover probability of genetic algorithm is 0.8, mutation probability 0.1.System of selection uses father and son's brocade in composite mode
Mark match algorithm, the common choice new population in parent and filial generation, the number of individuals taken part in game every time is 5, and maximum genetic algebra is
100 generations.The condition of convergence of reverse design is FBlay=0, speed, temperature calculations and desired value at and if only if monitoring point
When (0.1325m/s, 17.67 DEG C) is completely the same, this condition of convergence can be only achieved.
Using LHS method, the combination of 40 groups of entrance velocities and temperature is produced as input parameter, is made
CFD calculating is carried out to Blay models by the use of these input parameters as boundary condition, obtains output ginseng corresponding to every group of input parameter
Number, i.e., speed and temperature value at monitoring point, every group of input parameter and its corresponding output parameter form a complete training
Sample, neutral net is trained using all training samples, the neutral net trained is combined with genetic algorithm, should
Reverse design for Blay models.Calculate the design object value of Blay models with CFD using neutral net simultaneously.First by
Neutral net is predicted to the desired value of new individual, and according to the predictor calculation F of design objectBlayValue, if this value is less than
0.1, the individual is calculated using CFD, obtains the real F of the individualBlayValue or CFD calculated values.It is as shown in figure 3, refreshing
Amount of calculation growth curve during through network without self refresh：During reverse design, when genetic algorithm was calculated to 86 generation, reversely
Design process restrains, and has calculated 164 CFD examples altogether.
(2) result of calculation when, neutral net has a self refresh：
The method being combined using genetic algorithm with self refresh neutral net, reverse design is carried out to Blay models.Heredity
Algorithm calculating parameter is identical with (one).Nerve net obtained from reverse design during using the neutral net of above-mentioned (one) without self refresh
(neutral net for training to obtain when the neutral net of above-mentioned (one) is without self refresh is as initial nerve as initial neutral net for network
Network), calculate the design object values of Blay models in design process with CFD using neutral net simultaneously, and by CFD result of calculations
Be added in train samples, and retraining carried out to neutral net, realize training sample and neutral net from more
Newly.
As illustrated, amount of calculation change curve when having self refresh for neutral net：During reverse design, work as heredity
When algorithm was calculated to 40 generation, the convergence of reverse design process, 119 CFD examples have been calculated altogether, during with neutral net without self refresh
Compare, reverse design the amount of calculation have dropped 27.4%.
Claims (1)
1. a kind of implementation method of neutral net self refresh process, it is characterised in that this method comprises the following steps：
Step (1), entrance velocity and temperature and measuring point speed and temperature determined according to Blay models；
Step (2), the initial training sample by acquisition neutral net of sampling, realize the training to neutral net；
Step (3), using genetic algorithm cross and variation process simultaneously obtain multigroup entrance velocity for meeting design requirement with temperature
The value of degree, and be used as input parameter, using neutral net and Fluid Mechanics Computation method, calculate the monitoring spot speed of new individual with
Temperature, i.e., CFD calculating is carried out to Blay models using these input parameters as boundary condition, it is corresponding to obtain every group of input parameter
Output parameter, i.e., speed and temperature value at monitoring point；And Fluid Mechanics Computation method result of calculation is added to training sample
In this, obtain updating training sample；
Step (4), using renewal training sample neutral net is trained again, realize the renewal of neutral net；Step
(5) and then by non-dominated ranking method to the individual in population it is ranked up, and is selected using tournament algorithm, is produced
Raw new population, if new population meets convergence, design process terminates, and otherwise return to step (3), are continued executing with.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510905762.0A CN105389614B (en) | 2015-12-09 | 2015-12-09 | A kind of implementation method of neutral net self refresh process |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510905762.0A CN105389614B (en) | 2015-12-09 | 2015-12-09 | A kind of implementation method of neutral net self refresh process |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105389614A CN105389614A (en) | 2016-03-09 |
| CN105389614B true CN105389614B (en) | 2018-04-03 |
Family
ID=55421881
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510905762.0A Expired - Fee Related CN105389614B (en) | 2015-12-09 | 2015-12-09 | A kind of implementation method of neutral net self refresh process |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105389614B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921359B (en) * | 2018-07-26 | 2022-03-11 | 安徽大学 | Distributed gas concentration prediction method and device |
| CN109299742A (en) * | 2018-10-17 | 2019-02-01 | 深圳信息职业技术学院 | Method, apparatus, equipment and the storage medium of automatic discovery unknown network stream |
| CN113128556B (en) * | 2021-03-10 | 2022-10-28 | 天津大学 | Deep learning test case sequencing method based on mutation analysis |
| CN115993097A (en) * | 2023-03-23 | 2023-04-21 | 长安大学 | Monitoring and early warning method and system for cable-stayed bridge stay cable broken wire |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6470261B1 (en) * | 1998-07-31 | 2002-10-22 | Cet Technologies Pte Ltd | Automatic freeway incident detection system and method using artificial neural network and genetic algorithms |
| CN102521456A (en) * | 2011-12-15 | 2012-06-27 | 河海大学常州校区 | Optimal design method of arched girder based on neural network and genetic algorithm |
| CN102999664A (en) * | 2012-11-20 | 2013-03-27 | 天津大学 | Method for identifying position of indoor pollution source |
| CN103105246A (en) * | 2012-12-31 | 2013-05-15 | 北京京鹏环球科技股份有限公司 | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm |
-
2015
- 2015-12-09 CN CN201510905762.0A patent/CN105389614B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6470261B1 (en) * | 1998-07-31 | 2002-10-22 | Cet Technologies Pte Ltd | Automatic freeway incident detection system and method using artificial neural network and genetic algorithms |
| CN102521456A (en) * | 2011-12-15 | 2012-06-27 | 河海大学常州校区 | Optimal design method of arched girder based on neural network and genetic algorithm |
| CN102999664A (en) * | 2012-11-20 | 2013-03-27 | 天津大学 | Method for identifying position of indoor pollution source |
| CN103105246A (en) * | 2012-12-31 | 2013-05-15 | 北京京鹏环球科技股份有限公司 | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm |
Non-Patent Citations (2)
| Title |
|---|
| 基于计算流体力 学的虹吸式流道形状优化设计;彭志威;《中国博士学位论文全文数据库 工程科技Ⅱ辑》;20110115(第01期);全文 * |
| 海域潮流场数值模型开边界条件补遗预测的应用数据驱动模型;李明昌等;《水资源与水工程学报》;20110228;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105389614A (en) | 2016-03-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105389614B (en) | A kind of implementation method of neutral net self refresh process | |
| CN106779148B (en) | A kind of method for forecasting wind speed of high speed railway line of multi-model multiple features fusion | |
| CN106875033B (en) | Wind power cluster power prediction method based on dynamic self-adaption | |
| CN105574615B (en) | wavelet-BP neural network wind power prediction method based on spatial correlation and GA | |
| CN103092074B (en) | The parameter optimization control method of semiconductor Advanced process control | |
| CN106779151A (en) | A kind of line of high-speed railway wind speed multi-point multi-layer coupling prediction method | |
| CN105160444A (en) | Electrical equipment failure rate determining method and system | |
| CN104747368B (en) | A kind of method and system of Wind turbines power optimization | |
| CN105045091B (en) | Dredging technique intelligent decision analysis method based on Fuzzy Neural Control system | |
| CN107168292B (en) | Submarine navigation device circuit failure diagnosis method based on ELM algorithm | |
| CN107273693A (en) | A kind of hydrocarbon fuel mechanism simplification method | |
| CN109426672B (en) | Oil reservoir injection-production parameter optimization method based on uncertain geological model | |
| CN103778482A (en) | Aquaculture dissolved oxygen short-term prediction method based on multi-scale analysis | |
| CN108204944A (en) | The Buried Pipeline rate prediction method of LSSVM based on APSO optimizations | |
| CN113255206A (en) | Hydrological prediction model parameter calibration method based on deep reinforcement learning | |
| CN105389442A (en) | Reverse design method for coupling genetic algorithm, neural network and numerical simulation | |
| CN112113146B (en) | Synchronous self-adaptive check method for roughness coefficient and node water demand of water supply pipe network pipeline | |
| CN106991493A (en) | Sewage disposal water outlet parameter prediction method based on Grey production fuction | |
| CN110097929A (en) | A kind of blast furnace molten iron silicon content on-line prediction method | |
| CN111539842B (en) | Overhead power transmission line icing prediction method based on meteorological and geographic environments | |
| CN115755219A (en) | Flood forecast error real-time correction method and system based on STGCN | |
| CN106200381A (en) | A kind of according to processing the method that water yield control by stages water factory runs | |
| KR20110034288A (en) | A method for predicting diametral creep of candu reactor pressure tubes | |
| CN105550437A (en) | Reverse design method for indoor environment based on genetic algorithm | |
| CN105372995A (en) | Measurement and control method for sewage disposal system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180403 Termination date: 20211209 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |