CN112052621B - Supercritical carbon dioxide impeller machinery supercritical region prediction and control method based on CNN - Google Patents
Supercritical carbon dioxide impeller machinery supercritical region prediction and control method based on CNN Download PDFInfo
- Publication number
- CN112052621B CN112052621B CN202010711945.XA CN202010711945A CN112052621B CN 112052621 B CN112052621 B CN 112052621B CN 202010711945 A CN202010711945 A CN 202010711945A CN 112052621 B CN112052621 B CN 112052621B
- Authority
- CN
- China
- Prior art keywords
- cnn
- blade
- carbon dioxide
- working condition
- supercritical
- 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.)
- Active
Links
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 title claims abstract description 84
- 229910002092 carbon dioxide Inorganic materials 0.000 title claims abstract description 42
- 239000001569 carbon dioxide Substances 0.000 title claims abstract description 42
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 30
- 238000005070 sampling Methods 0.000 claims abstract description 15
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 9
- 238000012795 verification Methods 0.000 claims abstract description 9
- 238000012805 post-processing Methods 0.000 claims abstract description 6
- 238000004088 simulation Methods 0.000 claims abstract description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 39
- 239000003795 chemical substances by application Substances 0.000 claims description 16
- 238000005457 optimization Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 abstract description 6
- 238000012423 maintenance Methods 0.000 abstract 1
- 230000008859 change Effects 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 239000012530 fluid Substances 0.000 description 2
- 230000000704 physical effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000000819 phase cycle Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000002922 simulated annealing Methods 0.000 description 1
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
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Software Systems (AREA)
- Strategic Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Operations Research (AREA)
- Biomedical Technology (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
Abstract
The invention discloses a supercritical carbon dioxide impeller machinery supercritical region prediction and control method based on CNN, which comprises the following steps: 1. performing line parameterization, and performing numerical simulation pulling Ding Chao cube sampling by considering each boundary condition; 2. post-processing to obtain the power, efficiency, blade surface pressure, temperature and relative speed of each working condition; 3. normalizing data, dividing a training set and a verification set, and training CNN to establish a proxy model; 4. predicting a cross-critical zone position, an adjustment molded line or an operation condition parameter by adopting a proxy model; 5. and (5) algorithm maintenance. The invention can rapidly predict and control the design working condition and the trans-critical range under the variable working condition, improves the pneumatic efficiency and the operation safety of the supercritical carbon dioxide turbine machinery, and has important engineering significance and wide application prospect.
Description
Technical Field
The invention belongs to the field of prediction of a physical field of an impeller machine, and particularly relates to a supercritical carbon dioxide impeller machine supercritical region prediction and control method based on CNN.
Background
In recent years, research on the application of supercritical carbon dioxide working substances to the brayton cycle has received a great deal of attention, and the excellent characteristics of the supercritical carbon dioxide working substances are mainly represented by: the power density is high, and the volume is small; low operating temperature (compared to gas turbines), wide material selection range; when used in compressors, small compression factors like liquids can reduce compression power consumption; the supercritical carbon dioxide brayton cycle belongs to a single-phase cycle, has no phase change process, does not need a condenser, and uses only one tenth of the number of valves of the rankine cycle. In the research of supercritical carbon dioxide turbomachinery, a region in which the physical property state of the working medium is lower than a critical point is called a supercritical region. In actual operation, the turbine expands to near the critical pressure at the trailing edge of the movable vane, the transcritical phenomenon of working medium is generated in the low pressure area, and the compressor has transcritical flow which is difficult to eradicate at the leading edges of the main vane and the splitter vane of the impeller, the leading edge of the diffuser and the clearance area of the vane top. In a supercritical region, the carbon dioxide working medium has larger physical property change, which is unfavorable for the stable operation of the turbine machinery, so that the supercritical carbon dioxide impeller machinery is predicted and controlled, and has important significance.
In the traditional supercritical carbon dioxide impeller machinery design, the performance of different molded lines and under different operation conditions must be evaluated by using a Computational Fluid Dynamics (CFD) method, and the position and the area of a critical zone of the supercritical carbon dioxide impeller machinery must be judged one by one through a complex post-processing flow, so that the time is extremely long. In the optimization of turbomachinery, proxy models such as a response surface method, kerling interpolation and the like can reduce calculation numbers of simulation working conditions to realize acceleration of optimization. To further improve accuracy and speed up optimization, machine learning based proxy models and dimension reduction strategies have become hot spots, with Convolutional Neural Networks (CNNs) being a widely used proxy model.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a supercritical carbon dioxide impeller machine transcritical region prediction and control method based on CNN, and the method for establishing a CNN proxy model through a large number of samples can predict and control the transcritical range under the design working condition and the variable working condition, thereby improving the pneumatic efficiency and the operation safety of the supercritical carbon dioxide turbine machine, and having important engineering significance and wide application prospect.
The invention is realized by adopting the following technical scheme:
the supercritical carbon dioxide impeller machinery cross-critical region prediction and control method based on CNN, wherein the CNN is a convolutional neural network, and the prediction and control method comprises the following steps:
firstly, establishing a numerical simulation sampling flow, and parameterizing molded lines by adopting the definition of the thickness of a camber line aiming at the designed supercritical carbon dioxide impeller machine, and simultaneously considering the rotating speed, the inlet state parameters and the variable working condition of the outlet flow;
secondly, post-processing is carried out on the sampling result to obtain power and efficiency under the current working condition, and parameters of blade surface pressure, temperature and relative speed are obtained;
thirdly, normalizing the data obtained in the first step and the second step, dividing a training set and a verification set according to the proportion of 9:1, randomly disturbing the training set data, and establishing a proxy model by taking the training set data as the original training input of CNN;
fourth, aiming at different variable working conditions and molded line conditions, the position and the area of a cross-critical zone can be directly predicted through a proxy model; meanwhile, aiming at the determined operation working condition or the supercritical carbon dioxide impeller machine which is designed in the circulation system, a molded line adjustment scheme or an operation working condition parameter adjustment scheme is respectively and rapidly obtained through a proxy model, so that the purpose of controlling a cross-critical region is achieved.
The invention is further improved in that the method further comprises the following steps:
and fifthly, maintaining an algorithm, wherein in the actual application process, if the range of the agent model needs to be enlarged, preprocessing is carried out according to the first step and the second step, the trained CNN network parameters are adopted as a pre-training model, and the whole neural network is retrained on the basis of the pre-training model to obtain a more accurate agent model.
The invention is further improved in that in the first step:
for the designed supercritical carbon dioxide impeller machine, the camber line of the blade is defined by a Bezier curve, and 4 control points are recorded as [ x ] m1 ,x m2 ,x m3 ,x m4 ]The method comprises the steps of carrying out a first treatment on the surface of the The thickness distribution of the blade is also defined by Bezier curve, and 4 control points are selected and marked as [ x ] t5 ,x t6 ,x t7 ,x t8 ]The method comprises the steps of carrying out a first treatment on the surface of the For the variable working condition, the independent variable of the rotating speed variable working condition is recorded as x r9 The total inlet temperature is recorded as x T10 The total inlet pressure is denoted as x P11 The outlet flow is noted as x m12 ；
Selecting a parameter range of 12 independent variables, sampling n blade molded lines and operation variable working conditions by a Latin hypercube method, and recording as [ X ] n ] 12 ＝[x m1 ,x m2 ,x m3 ,x m4 ,x t5 ,x t6 ,x t7 ,x t8 ,x r9 ,x T10 ,x P11 ,x m12 ]Wherein n is not less than 1000.
The invention is further improved in that in the second step:
in the sampling grid dividing module, the number of nodes on the surface of the blade is set to be i×j, wherein i is the number of nodes on a blade profile line, j is the number of nodes in the blade height direction, and the calculation parameters under each working condition are obtained by marking as follows:
[Y n ]＝{W n ,η n ,[P n ] i×j ,[T n ] i×j ,[V n ] i×j }
where W is power, η is efficiency, P, T, V is pressure, temperature and relative velocity at a node on the blade surface, respectively.
The invention is further improved in that in the third step:
and respectively carrying out normalization operation on the n groups of data by adopting the following formulas:
wherein Max and Min respectively represent a maximum value and a minimum value of a current parameter, and the dimension of X is 12 multiplied by n; k represents Y n The data properties of the blade, namely power, efficiency, pressure, temperature and relative speed of the blade surface are taken respectively; when k is power and efficiency, the dimension of Y is 1×n, and when k is pressure, temperature and relative speed, the dimension of Y is i×j×n; after normalization is completed, dividing all data into a training set and a verification set according to the proportion of 9:1, wherein the training set and the verification set are respectively as follows:
establishing CNN, which comprises an input layer, a convolution layer, a pooling layer, a full connection layer and an output layer, wherein the input is n groups of X, the dimension is 12 multiplied by 1, the corresponding n groups of Y are 2 multiplied by 1 when Y is power and efficiency, and the Y is pressure, temperature and relative speed are i multiplied by j; in the training process of CNN, firstly setting an optimizer as Adam, setting the initial learning rate as 0.003, and training for 100 steps; and then setting the optimizer as an SGD gradient descent algorithm, and adopting an equidistant adjustment mechanism in PyTorch to adjust the learning rate, namely attenuating to be one tenth of the original learning rate every 50 steps later.
The invention is further improved in that in the fourth step:
reconstructing and predicting a pressure field and a temperature field on the surface of a blade by a trained CNN agent model according to a certain variable working condition running condition and a corresponding blade molded line to obtain a position of a transcritical region with the pressure being 7.38MPa lower than the critical pressure of carbon dioxide or the temperature being 31.1 ℃ lower than the critical temperature;
screening a leaf pattern scheme with minimum transcritical nodes through a CNN agent model which is completed by training aiming at a determined operation working condition in a certain circulation system, and obtaining 8 Bessel control points corresponding to a mean camber line and a thickness of the leaf pattern scheme to complete leaf pattern optimization for reducing the transcritical region; aiming at the designed blade molded line, the agent model can obtain the adjustment scheme of the running condition parameters such as the rotating speed, the inlet temperature, the pressure, the flow and the like, reduce the range of the critical zone and realize the control of the critical zone in the supercritical carbon dioxide impeller machine.
Compared with the prior art, the invention has the following beneficial technical effects:
aiming at the supercritical carbon dioxide impeller machine transcritical operation problem, the invention provides a supercritical carbon dioxide impeller machine transcritical region prediction and control method based on CNN. In the traditional supercritical carbon dioxide impeller machine design, for different molded line designs and variable working condition parameters, a Computational Fluid Dynamics (CFD) method is required to evaluate the performance of the supercritical carbon dioxide impeller machine, and because the molded line designs and the variable working condition parameters are numerous, the position and the area of a supercritical region of the supercritical carbon dioxide impeller machine are required to be judged one by one through a complex post-processing flow after each CFD calculation is completed, and the time is extremely long. If conventional optimization algorithms such as a genetic algorithm, a simulated annealing algorithm and the like are adopted, the time is long, the search space for the multivariate model is very limited, and a local optimal solution can be possibly obtained. According to the method, full-automatic sampling calculation is only needed once, the trained CNN serving as a proxy model has high prediction precision of the critical area, CFD calculation is not needed, and a large-scale variable working condition curve and the critical area position under each working condition can be obtained in an order of seconds.
Meanwhile, the agent model can rapidly and accurately complete the optimization of the leaf profile for reducing the cross-critical area according to the determined operation working condition in a certain circulatory system, and screening the leaf profile scheme with the least cross-critical nodes; aiming at the designed blade molded line, an adjustment scheme of operating condition parameters is obtained rapidly, and the control of a critical zone in the supercritical carbon dioxide impeller machine is realized. In conclusion, the invention has important engineering significance and wide application prospect.
Drawings
Fig. 1 is a general flow chart of a supercritical carbon dioxide impeller machinery supercritical region prediction and control method based on CNN of the present invention.
FIG. 2 is an example of four control points for profile thickness definition.
FIG. 3 is an example of pressure distribution of the i x j nodes of the supercritical carbon dioxide impeller machine blade surface.
FIG. 4 is an example of a prediction of pressure at the intersection of the blade surface and the mid-section by a proxy model established by CNN.
Fig. 5 is an example of the prediction of efficiency by the proxy model established by CNN.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
Referring to fig. 1, the supercritical carbon dioxide impeller machinery supercritical region prediction and control method based on CNN provided by the invention comprises the following steps:
1. and establishing a numerical simulation sampling flow, parameterizing molded lines by adopting the definition of the thickness of the camber line aiming at the designed supercritical carbon dioxide impeller machine, and simultaneously considering variable working conditions such as rotating speed, inlet state parameters, outlet flow and the like. For the designed supercritical carbon dioxide impeller machine, the camber line of the blade is defined by a Bezier curve, and 4 control points are recorded as [ x ] m1 ,x m2 ,x m3 ,x m4 ]The method comprises the steps of carrying out a first treatment on the surface of the The thickness distribution of the blade is also defined by Bezier curve, and 4 control points are selected and marked as [ x ] t5 ,x t6 ,x t7 ,x t8 ]. For the variable working condition, the independent variable of the rotating speed variable working condition is recorded as x r9 The total inlet temperature is recorded as x T10 The total inlet pressure is denoted as x P11 The outlet flow is noted as x m12 . As shown in FIG. 2, four control points [ x ] are defined for the profile thickness t5 ,x t6 ,x t7 ,x t8 ]Examples are shown.
Selecting a parameter range of 12 independent variables, sampling n blade molded lines and operation variable working conditions by a Latin hypercube method, and recording as [ X ] n ] 12 ＝[x m1 ,x m2 ,x m3 ,x m4 ,x t5 ,x t6 ,x t7 ,x t8 ,x r9 ,x T10 ,x P11 ,x m12 ]Wherein n is not less than 1000.
2. And carrying out post-processing on the sampling result to obtain power and efficiency under the current working condition, and obtaining parameters of blade surface pressure, temperature and relative speed. In the sampling grid dividing module, the number of nodes on the surface of the blade is set to be i×j, wherein i is the number of nodes on a blade profile line, and j is the number of nodes in the blade height direction. The calculation parameters under each working condition are obtained as follows:
[Y n ]＝{W n ,η n ,[P n ] i×j ,[T n ] i×j ,[V n ] i×j }
where W is power, η is efficiency, P, T, V is pressure, temperature and relative velocity at a node on the blade surface, respectively. As shown in fig. 3, an example of the pressure distribution of the i×j nodes on the surface of the supercritical carbon dioxide impeller mechanical blade is shown.
3. And respectively carrying out normalization operation on the n groups of data by adopting the following formulas:
wherein Max and Min respectively represent the maximum value and the minimum value of the current parameter, and the dimension of X is 12 multiplied by n. k represents Y n I.e. power, efficiency, pressure, temperature and relative speed of the blade surface, respectively. When k is power, efficiency, the dimension of Y is 1×n, and when k is pressure, temperature, and relative speed, the dimension of Y is i×j×n. After normalization is completed, dividing all data into a training set and a verification set according to the proportion of 9:1, wherein the training set and the verification set are respectively as follows:
the CNN is established and comprises an input layer, a convolution layer, a pooling layer, a full connection layer and an output layer, wherein the input is n groups of X, the dimension is 12 multiplied by 1, the corresponding n groups of Y are respectively 2 multiplied by 1 when the Y is power and efficiency, and the Y is pressure, temperature and relative speed. In the training process of CNN, firstly setting an optimizer as Adam, setting the initial learning rate as 0.003, and training for 100 steps; and then setting the optimizer as an SGD gradient descent algorithm, and adopting an equidistant adjustment mechanism in PyTorch to adjust the learning rate, namely attenuating to be one tenth of the original learning rate every 50 steps later.
4. And (3) reconstructing and predicting the pressure field and the temperature field of the blade surface by a trained CNN agent model according to a certain variable working condition running condition and a corresponding blade molded line to obtain a position of a transcritical region with the pressure lower than the critical pressure of carbon dioxide by 7.38MPa or the temperature lower than the critical temperature by 31.1 ℃. Fig. 4 is an example of prediction of pressure at the intersection of the blade surface and the mid-section by a proxy model established by CNN, with the region below the solid line being the transcritical region, wherein the true pressure distribution is substantially the same as the predicted pressure distribution.
Screening a leaf pattern scheme with minimum transcritical nodes through a CNN agent model which is completed by training aiming at a determined operation working condition in a certain circulation system, and obtaining 8 Bessel control points corresponding to a mean camber line and a thickness of the leaf pattern scheme to complete leaf pattern optimization for reducing the transcritical region; aiming at the designed blade molded line, the agent model can obtain the adjustment scheme of the running condition parameters such as the rotating speed, the inlet temperature, the pressure, the flow and the like, reduce the range of the critical zone and realize the control of the critical zone in the supercritical carbon dioxide impeller machine. Fig. 5 is an example of the prediction of efficiency by the proxy model established by CNN, where the predicted operating points fall within 5% confidence intervals, approaching the real operating conditions.
5. In the process of actual application, if the range of the agent model needs to be enlarged, for example, a new numerical simulation sampling working condition is added or experimental data of certain working condition points are obtained, preprocessing is carried out according to the first step and the second step, the trained CNN network parameters are adopted as a pre-training model, and the whole neural network is retrained on the basis to obtain a more accurate agent model.
Claims (3)
1. The supercritical carbon dioxide impeller machinery supercritical region prediction and control method based on the CNN is characterized in that the CNN is a convolutional neural network, and the prediction and control method comprises the following steps:
firstly, establishing a numerical simulation sampling flow, and parameterizing molded lines by adopting the definition of the thickness of a camber line aiming at the designed supercritical carbon dioxide impeller machine, and simultaneously considering the rotating speed, the inlet state parameters and the variable working condition of the outlet flow; for designed supercritical twoCarbon oxide impeller machinery, wherein the camber line of the blade is defined by Bezier curve, the number of control points is 4, and the control points are marked as [ x ] m1 ,x m2 ,x m3 ,x m4 ]The method comprises the steps of carrying out a first treatment on the surface of the The thickness distribution of the blade is also defined by Bezier curve, and 4 control points are selected and marked as [ x ] t5 ,x t6 ,x t7 ,x t8 ]The method comprises the steps of carrying out a first treatment on the surface of the For the variable working condition, the independent variable of the rotating speed variable working condition is recorded as x r9 The total inlet temperature is recorded as x T10 The total inlet pressure is denoted as x P11 The outlet flow is noted as x m12 ；
Selecting a parameter range of 12 independent variables, sampling n blade molded lines and operation variable working conditions by a Latin hypercube method, and recording as [ X ] n ] 12 ＝[x m1 ,x m2 ,x m3 ,x m4 ,x t5 ,x t6 ,x t7 ,x t8 ,x r9 ,x T10 ,x P11 ,x m12 ]Wherein n is not less than 1000;
secondly, post-processing is carried out on the sampling result to obtain power and efficiency under the current working condition, and parameters of blade surface pressure, temperature and relative speed are obtained; in the sampling grid dividing module, the number of nodes on the surface of the blade is set to be i×j, wherein i is the number of nodes on a blade profile line, j is the number of nodes in the blade height direction, and the calculation parameters under each working condition are obtained by marking as follows:
[Y n ]＝{W n ,η n ,[P n ] i×j ,[T n ] i×j ,[V n ] i×j }
wherein W is power, eta is efficiency, P, T, V is pressure, temperature and relative speed of a certain node on the surface of the blade respectively;
thirdly, normalizing the data obtained in the first step and the second step, dividing a training set and a verification set according to the proportion of 9:1, randomly disturbing the training set data, and establishing a proxy model by taking the training set data as the original training input of CNN; and respectively carrying out normalization operation on the n groups of data by adopting the following formulas:
wherein Max and Min respectively represent a maximum value and a minimum value of a current parameter, and the dimension of X is 12 multiplied by n; k represents Y n The data properties of the blade, namely power, efficiency, pressure, temperature and relative speed of the blade surface are taken respectively; when k is power and efficiency, the dimension of Y is 1×n, and when k is pressure, temperature and relative speed, the dimension of Y is i×j×n; after normalization is completed, dividing all data into a training set and a verification set according to the proportion of 9:1, wherein the training set and the verification set are respectively as follows:
establishing CNN, which comprises an input layer, a convolution layer, a pooling layer, a full connection layer and an output layer, wherein the input is n groups of X, the dimension is 12 multiplied by 1, the corresponding n groups of Y are 2 multiplied by 1 when Y is power and efficiency, and the Y is pressure, temperature and relative speed are i multiplied by j; in the training process of CNN, firstly setting an optimizer as Adam, setting the initial learning rate as 0.003, and training for 100 steps; setting an optimizer as an SGD gradient descent algorithm, and adopting an equidistant adjustment mechanism in PyTorch to adjust the learning rate, namely attenuating the learning rate to be one tenth of the original learning rate after every 50 steps;
fourth, aiming at different variable working conditions and molded line conditions, the position and the area of a cross-critical zone can be directly predicted through a proxy model; meanwhile, aiming at the determined operation working condition or the supercritical carbon dioxide impeller machine which is designed in the circulation system, a molded line adjustment scheme or an operation working condition parameter adjustment scheme is respectively and rapidly obtained through a proxy model, so that the purpose of controlling a cross-critical region is achieved.
2. The CNN-based supercritical carbon dioxide impeller machine transcritical region prediction and control method of claim 1, further comprising the steps of:
and fifthly, maintaining an algorithm, wherein in the actual application process, if the range of the agent model needs to be enlarged, preprocessing is carried out according to the first step and the second step, the trained CNN network parameters are adopted as a pre-training model, and the whole neural network is retrained on the basis of the pre-training model to obtain a more accurate agent model.
3. The CNN-based supercritical carbon dioxide impeller machine supercritical region prediction and control method according to claim 1, wherein in step four:
reconstructing and predicting a pressure field and a temperature field on the surface of a blade by a trained CNN agent model according to a certain variable working condition running condition and a corresponding blade molded line to obtain a position of a transcritical region with the pressure being 7.38MPa lower than the critical pressure of carbon dioxide or the temperature being 31.1 ℃ lower than the critical temperature;
screening a leaf pattern scheme with minimum transcritical nodes through a CNN agent model which is completed by training aiming at a determined operation working condition in a certain circulation system, and obtaining 8 Bessel control points corresponding to a mean camber line and a thickness of the leaf pattern scheme to complete leaf pattern optimization for reducing the transcritical region; aiming at the designed blade molded line, the agent model can obtain the adjustment scheme of the rotation speed, inlet temperature, pressure and flow operation condition parameters, reduce the range of a supercritical region and realize the control of the supercritical region in the supercritical carbon dioxide impeller machine.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010711945.XA CN112052621B (en) | 2020-07-22 | 2020-07-22 | Supercritical carbon dioxide impeller machinery supercritical region prediction and control method based on CNN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010711945.XA CN112052621B (en) | 2020-07-22 | 2020-07-22 | Supercritical carbon dioxide impeller machinery supercritical region prediction and control method based on CNN |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112052621A CN112052621A (en) | 2020-12-08 |
| CN112052621B true CN112052621B (en) | 2024-04-16 |
Family
ID=73601875
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010711945.XA Active CN112052621B (en) | 2020-07-22 | 2020-07-22 | Supercritical carbon dioxide impeller machinery supercritical region prediction and control method based on CNN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112052621B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113051820B (en) * | 2021-03-24 | 2023-05-16 | 中国空气动力研究与发展中心超高速空气动力研究所 | Cross-drainage-basin pneumatic parameter simulation method based on convolutional neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109359379A (en) * | 2018-10-15 | 2019-02-19 | 西安交通大学 | A kind of adaptive supercritical carbon dioxide turbine design method |
| CN110008653A (en) * | 2019-02-26 | 2019-07-12 | 西北工业大学 | A kind of aviation centrifugal pump blade profile optimum design method |
| WO2020000248A1 (en) * | 2018-06-27 | 2020-01-02 | 大连理工大学 | Space reconstruction based method for predicting key performance parameters of transition state acceleration process of aircraft engine |
-
2020
- 2020-07-22 CN CN202010711945.XA patent/CN112052621B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020000248A1 (en) * | 2018-06-27 | 2020-01-02 | 大连理工大学 | Space reconstruction based method for predicting key performance parameters of transition state acceleration process of aircraft engine |
| CN109359379A (en) * | 2018-10-15 | 2019-02-19 | 西安交通大学 | A kind of adaptive supercritical carbon dioxide turbine design method |
| CN110008653A (en) * | 2019-02-26 | 2019-07-12 | 西北工业大学 | A kind of aviation centrifugal pump blade profile optimum design method |
Non-Patent Citations (1)
| Title |
|---|
| 张新华,鲁志康,谢建伟,赵建跃.叶轮故障监测系统的数据通信.机械设计与制造工程.2000,(02),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112052621A (en) | 2020-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Aulich et al. | High-dimensional constrained multiobjective optimization of a fan stage | |
| CN110173309B (en) | Optimization design method for sequence valve of four-high governing valve steam turbine | |
| CN111898212B (en) | Impeller mechanical profile design optimization method based on BezierGAN and Bayesian optimization | |
| CN111027148B (en) | Automatic calibration and industrial axial flow compressor performance calculation method for loss lag angle model | |
| CN110647052B (en) | Variable cycle engine mode switching self-adaptive identity card model construction method | |
| CN107203364B (en) | Prediction and identification method for full-working-condition characteristics of gas compressor | |
| CN112685968B (en) | Axial flow compressor pneumatic design method based on space load customization thought | |
| CN112052621B (en) | Supercritical carbon dioxide impeller machinery supercritical region prediction and control method based on CNN | |
| CN108197390A (en) | A kind of optimum design method of two-phase cryogenic liquid expanding machine anti-cavitation | |
| CN105298734A (en) | Parameter identification method for water turbine adjusting system | |
| US20210271789A1 (en) | Design and production of a turbomachine vane | |
| CN114254460A (en) | Turbo-machine pneumatic robustness optimization method based on graph convolution neural network | |
| Avdieieva et al. | Development of the typical design of the high-pressure stage of a steam turbine | |
| Popov et al. | Multicriteria optimization of axial low pressure compressor of gas turbine power plant | |
| CN113158355B (en) | All-condition optimization design method for low-temperature liquid expander | |
| Wang et al. | Integrated optimization for design and operation of turbomachinery in a solar-based Brayton cycle based on deep learning techniques | |
| CN113158356A (en) | Collaborative optimization design method for anti-cavitation rectification cone of low-temperature liquid expansion machine | |
| CN117150672A (en) | Automatic optimizing method for thickness of centrifugal pump blade | |
| Song et al. | Enhancing refrigeration capacity of turbo expander by means of multipoint design optimization | |
| CN116595874A (en) | Impeller mechanical performance prediction model parameter optimization method and device and storage medium | |
| Wang et al. | Aerodynamic prediction on the off-design performance of a S-CO2 turbine based on deep learning | |
| Pakatchian et al. | Applications of machine learning approaches in aerodynamic aspects of axial flow compressors: A review | |
| CN113591419A (en) | One-dimensional design evaluation and optimization method for radial-flow type turbine machinery | |
| Shao et al. | Aerodynamic optimization of the radial inflow turbine for a 100kW-class micro gas turbine based on metamodel-semi-assisted method | |
| Belz et al. | Goal-oriented active learning with local model networks |
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 |