CN114169605A - Physical field and performance collaborative prediction method suitable for pump and hydraulic turbine - Google Patents

Physical field and performance collaborative prediction method suitable for pump and hydraulic turbine Download PDF

Info

Publication number
CN114169605A
CN114169605A CN202111479710.3A CN202111479710A CN114169605A CN 114169605 A CN114169605 A CN 114169605A CN 202111479710 A CN202111479710 A CN 202111479710A CN 114169605 A CN114169605 A CN 114169605A
Authority
CN
China
Prior art keywords
pump
hydraulic turbine
performance
physical field
parameters
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111479710.3A
Other languages
Chinese (zh)
Inventor
谢永慧
徐涛
李云珠
张荻
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.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong University
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 Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN202111479710.3A priority Critical patent/CN114169605A/en
Publication of CN114169605A publication Critical patent/CN114169605A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • 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]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
    • 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
    • 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
    • G06N3/048—Activation functions
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/08—Fluids
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Development Economics (AREA)
  • Evolutionary Biology (AREA)
  • Game Theory and Decision Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Algebra (AREA)
  • Fluid Mechanics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computer Hardware Design (AREA)

Abstract

The invention discloses a physical field and performance collaborative prediction method suitable for a pump and a hydraulic turbine, which comprises the following steps: 1) acquiring geometric parameters of a pump model and a hydraulic turbine model and coding the geometric parameters; 2) respectively acquiring working condition parameters under two modes of a pump and a hydraulic turbine and acquiring a numerical simulation result; 3) normalizing input and output parameters of the pump and the hydraulic turbine; 4) dividing a training set and a verification set aiming at data of a pump model and a hydraulic turbine model; 5) respectively constructing two groups of physical fields and performance cooperative prediction point cloud convolution networks of a pump and a hydraulic turbine; 6) training two groups of point cloud convolution neural networks of a pump and a hydraulic turbine; 7) and (5) post-processing the performance prediction information. The method realizes the direct mapping from the related performance parameters under the pump mode to the related performance parameters under the hydraulic turbine mode, overcomes the defects of large calculation data amount and long calculation time of the traditional numerical simulation method, and has the advantages of easy data change and timely acquisition of the required performance parameters.

Description

Physical field and performance collaborative prediction method suitable for pump and hydraulic turbine
Technical Field
The invention belongs to the technical field of pumps and hydraulic turbines, and particularly relates to a physical field and performance collaborative prediction method suitable for the pumps and the hydraulic turbines.
Background
At present, a large amount of clean, green, pollution-free and surplus pressure energy is always generated in the industrial production and transmission process in the fields of petroleum, chemical engineering and the like, and the pressure energy can be fully utilized as a hydraulic turbine through pump reversal. Currently, a numerical simulation method is mostly adopted for research on hydraulic turbines for pump reversal, and the method is used for carrying out numerical simulation calculation on an internal flow field of a pump during forward and reverse rotation to obtain the flow field of the pump during forward and reverse rotation under different working conditions, so that the performance of the pump under different working conditions is obtained. The three-dimensional numerical simulation calculation consumes a large amount of resources and time, corresponding performance parameters cannot be obtained in time when relevant working condition parameters and geometric parameters change, and great resource waste and time lag exist.
In recent years, with the development of science and technology, the computing power of a computer is remarkably improved, the progress of artificial intelligence enables traditional problems to have a plurality of new solutions, and the deep learning method is popularized and propagated in various engineering fields.
Disclosure of Invention
The invention aims to provide a physical field and performance collaborative prediction method suitable for a pump and a hydraulic turbine. First P _ Net in first group1The network implements the prediction of the pump physical field from the pump's operating and geometric parameters, the second P _ Net2Network entityThe performance of the pump is now predicted by the pump physical field; first TNet of the second group1The network realizes the prediction of the physical field of the hydraulic turbine from the working condition parameters and the geometric parameters of the hydraulic turbine, and the second T _ Net2The network realizes the prediction of the performance of the hydraulic turbine by the physical field of the hydraulic turbine, and finally obtains a mapping curve from the performance of the pump to the performance of the hydraulic turbine under the variable working condition through fitting two groups of data, thereby completing the task of the multi-physical field and performance collaborative prediction of the hydraulic turbine under the variable working condition.
The invention is realized by adopting the following technical scheme:
a physical field and performance collaborative prediction method suitable for a pump and a hydraulic turbine comprises the following steps:
1) geometric parameters of pump and hydraulic turbine models are obtained and encoded
The obtained geometric parameters of the blades are the installation angle alpha of the control blade, the number a of the blades and the coordinates (x, y, z) of B points of a Bezier curve of the control blade, each pump corresponds to one blade, and because the same blade is adopted in the two modes of the pump and the hydraulic turbine, the initial geometric parameter sets in the two modes are obtained by the method
Figure BDA0003394498920000021
Modeling is carried out according to the obtained geometric parameters of the I pumps, and for each grid model, grid node information G of each grid model is obtainedi,p,qMesh node information Gi,p,qInitial set of geometric parameters associated with the blade
Figure BDA0003394498920000022
Integrating to obtain a geometric parameter set Ai,jAnd encoding the geometric parameters by using a coder-decoder combination, wherein the coder Encoder encodes the geometric parameter set A of the bladei,jMapping into a fixed-length vector form, decoding the vector obtained by encoding into a geometric parameter set A by a Decoder Decoderi,jCombining the vectors of I blades obtained by the encoder to form an indirect geometric parameter set A of the bladesi,f(ii) a Wherein, I1, 2,3,.. and I represent different pumps, I is the total number of pumps, and J1, 2, 3.. and J representDifferent geometric parameters, J is the total number of the geometric parameters, P is 1,2, 3., P represents different grid nodes, P is the total number of the grid nodes, q is the coordinate of each grid node, and comprises the coordinates in the x direction, the y direction and the z direction, if a two-dimensional model, the coordinates in the x direction and the y direction are comprised, F is 1,2, 3., F represents different indirect geometric parameter variables, and F is the total number of the indirect geometric parameter variables;
2) respectively obtaining working condition parameters of the pump and the hydraulic turbine under two modes and obtaining a numerical simulation result
The working condition parameters of the pump and the hydraulic turbine comprise the mechanical rotating speed omega when the pump and the hydraulic turbine operate in two modesRInlet pressure P in two modes of pump and hydraulic turbineinAnd inlet temperature Tin(ii) a For each pump, under two modes, respectively adopting Latin hypercube sampling method to randomly sample D working conditions in respective parameter space, so that the working condition parameter set under the pump mode is Bi,dThe operating condition parameter in the hydraulic turbine mode is Ci,dAnd respectively carrying out CFD calculation on D working condition points of each pump model in two modes to obtain a three-dimensional physical field data set of each blade in the pump mode
Figure BDA0003394498920000023
Combining with three-dimensional physical field data in hydraulic turbine mode
Figure BDA0003394498920000031
Calculating to obtain a set of performance parameters of different pumps under the pump mode according to the physical field parameters
Figure BDA0003394498920000032
And performance parameter set in hydraulic turbine mode
Figure BDA0003394498920000033
D is 1,2,3, and D represents different operating parameters, D is the total number of operating points that need to be calculated in each mode, N is 1,2,3, and N represents different physical fields including speed, temperature, pressure, and vorticity, and N is the total recorded flow field parametersNumber, m ═ 1,2, representing efficiency and head, respectively;
3) pump and hydraulic turbine input and output parameter normalization
Normalizing the simplified set and other data to obtain normalized input data set { A }i,f,Bi,d,Ci,dThe physical field is collected as
Figure BDA0003394498920000034
The set of performance parameters is
Figure BDA0003394498920000035
The input data set after normalization is
Figure BDA0003394498920000036
Set of physical fields as
Figure BDA0003394498920000037
The set of performance parameters is
Figure BDA0003394498920000038
4) Partitioning training and validation sets for pump and hydraulic turbine model data
For neural networks such as point clouds, the input data set is
Figure BDA0003394498920000039
Collecting input data
Figure BDA00033944989200000310
Randomly scrambling, and setting the input set as 2: 2: 1 into training set
Figure BDA00033944989200000311
Verification set
Figure BDA00033944989200000312
And test set
Figure BDA00033944989200000313
Similarly, the set of physical fields is divided into equal parts
Figure BDA00033944989200000314
And
Figure BDA00033944989200000315
the performance parameter set is equally divided into
Figure BDA00033944989200000316
And
Figure BDA00033944989200000317
5) cooperative prediction point cloud convolution network for respectively constructing two groups of physical fields and performances of pump and hydraulic turbine
Constructing two groups of networks of pump physical field and performance prediction and hydraulic turbine physical field prediction and performance prediction, wherein each group comprises two point cloud convolution neural networks; for the first set of pump physical fields and performance prediction networks, the first P _ Net1Point cloud convolution neural network aggregation according to input in pump mode
Figure BDA00033944989200000318
Predicting a set of physical fields in a pump mode
Figure BDA00033944989200000319
Realizing the mapping of the design variables, i.e. the geometric parameters and the working condition parameters, to the physical field of the pump in the pump mode, the second P _ Net2The point cloud convolution neural network input is P _ Net1Network predicted set of pump physical fields
Figure BDA00033944989200000320
Predicting a set of performance parameters for a pump mode
Figure BDA00033944989200000321
Mapping from the pump physical field to the pump performance parameter is realized; for the second group of hydraulic turbine physical field and performance prediction network, the first T _ Net1Point cloud convolution neural network according to input set in hydraulic turbine mode
Figure BDA00033944989200000322
Predicting to obtain physical field in hydraulic turbine mode
Figure BDA0003394498920000041
Implementing a mapping of design variables to a physical field of the hydraulic turbine in the hydraulic turbine mode, a second T _ Net2The point cloud convolution neural network input is T _ Net1Physical field predicted by network
Figure BDA0003394498920000042
Predicting to obtain a set of performance parameters in a hydraulic turbine mode
Figure BDA0003394498920000043
Mapping from a hydraulic turbine physical field to hydraulic turbine performance parameters is realized;
6) convolutional neural network for training two groups of point clouds of pump and hydraulic turbine
7) Performance prediction information post-processing
Output aggregation from two parts
Figure BDA0003394498920000044
And
Figure BDA0003394498920000045
establishing direct relation between the performance parameters in the pump mode and the performance parameters in the hydraulic turbine mode, realizing direct mapping from the efficiency in the pump mode to the efficiency in the hydraulic turbine mode, and simultaneously obtaining the related physical field analysis of the pump and the hydraulic turbine for further analysis.
The invention is further improved in that the method also comprises the following steps:
8) algorithm maintenance
In the actual application process, if the information needing to be input is increased or reduced, an encoder and a decoder are adopted, the original input information is mapped into a vector with a fixed size to serve as a new input, four groups of point cloud convolution networks including a trained pump, a trained hydraulic turbine physical field and performance prediction are adopted to serve as pre-training models, and training is carried out again on the basis.
The further improvement of the invention is that in the step 1), the Encoder Encoder and the Decoder Decoder both adopt full connection layer to construct network, and the Encoder Encoder constructs encoding mapping function to realize the geometric parameter set Ai,jEncoding mapping to a fixed length vector, Decoder constructing decoding mapping function to realize the mapping from the fixed length vector to the geometric parameter set Ai,jDecoding the map of (1); the loss function of the process is a set of geometric parameters Ai,jThe result of the predicted geometric parameters after encoding and decoding and the original real geometric parameter set Ai,jThe difference between:
Figure BDA0003394498920000046
the further improvement of the present invention is that, in step 3), the normalization operation is Sigmoid function conversion, and the specific operation is as follows:
Figure BDA0003394498920000047
wherein, the expression of the Sigmoid function is as follows:
Figure BDA0003394498920000051
the normalization operation for the remaining data is similar:
Figure BDA0003394498920000052
Figure BDA0003394498920000053
Figure BDA0003394498920000054
Figure BDA0003394498920000055
Figure BDA0003394498920000056
Figure BDA0003394498920000057
the invention further improves the point cloud convolution network used for the cooperative prediction of the physical fields and the performance of the pump and the hydraulic turbine in the step 4).
The further improvement of the invention is that the point cloud convolution Network adopts PointNet, PointNet + + or Kd-Network.
The invention further improves the method, in the step 5), for two modes of the pump and the hydraulic turbine, two point cloud convolution neural networks in each mode comprise two parts of loss functions, wherein one part is the loss function F _ loss predicted by the physical field and is used for reducing the difference between the predicted physical field and the real physical field; the other part is a loss function psi _ loss for performance prediction, which is used for reducing the gap between the predicted performance and the real performance;
Figure BDA0003394498920000058
Figure BDA0003394498920000059
Figure BDA00033944989200000510
Figure BDA00033944989200000511
wherein, wqPredicting weights of the squared differences of the physical field data and the true physical field data for each network node,
Figure BDA0003394498920000061
are respectively P _ Net1、T_Net1Two physical field prediction point cloud convolutional neural networks generated predicted physical field data,
Figure BDA0003394498920000062
real physical field data for the pump and hydraulic turbine respectively,
Figure BDA0003394498920000063
are respectively P _ Net2、T_Net2Two performance prediction point cloud convolutional neural network generated predicted performance data,
Figure BDA0003394498920000064
actual performance data for the pump and hydraulic turbine, respectively.
The further improvement of the invention is that in the step 6), in the process of training the network, firstly, an optimizer is set as Adam, the initial learning rate is set to be 0.01, and the training is carried out for 50 steps; the optimizer is then set to SGD, after which the learning rate is reduced to 1/10 per 100 steps of training, with a minimum learning rate of 0.0001.
The invention has at least the following beneficial technical effects:
the invention provides a physical field and performance collaborative prediction method suitable for a pump and a hydraulic turbine. First P _ Net in first group1The network implements the prediction of the pump physical field from the pump's operating and geometric parameters, the second P _ Net2The network realizes that the performance of the pump is predicted by the physical field of the pump; first TNet of the second group1The network realizes the prediction of the physical field of the hydraulic turbine from the working condition parameters and the geometric parameters of the hydraulic turbine, and the second T _ Net2The network realizes the prediction of the hydraulic turbine performance by the hydraulic turbine physical field, and finally obtains a mapping curve from the pump performance to the hydraulic turbine performance under variable working conditions through fitting of two groups of data. The method can realize direct mapping from the related performance parameters under the pump mode to the related performance parameters under the hydraulic turbine mode, overcomes the defects of large calculation data amount and long calculation time of the traditional numerical simulation method, has the advantages of easy data change and timely acquisition of the required performance parameters, can replace the traditional numerical simulation calculation, and has the advantages of low cost, high accuracy and the like compared with an experimental method.
Drawings
FIG. 1 is a flow chart of a data-driven variable-condition hydraulic turbine based multi-physical field and performance collaborative prediction method of the present invention;
FIG. 2 is a diagram illustrating the data processing of the present invention;
FIG. 3 is a schematic diagram of a flow field prediction point cloud convolutional neural network constructed by the present invention;
FIG. 4 is a schematic diagram of a performance prediction point cloud convolutional neural network constructed by the present invention.
Detailed Description
The present invention will be described in further detail with reference to examples in accordance with the summary of the invention. The following is an application of the present invention, but is not limited thereto, and the implementer may modify the parameters according to specific situations, the specific implementation method is shown in fig. 1, and the data processing process is shown in fig. 2.
As shown in FIG. 1, the invention provides a physical field and performance cooperative prediction method suitable for a pump and a hydraulic turbine, which comprises the following steps:
firstly, acquiring geometric parameters of a pump model and a hydraulic turbine model and coding the geometric parameters;
obtaining lamps such as a mounting angle alpha of a control blade, the number a of the blades and coordinates (x, y, z) of B points of a Bezier curve of the control bladeThe geometric parameters of the sheet are that each pump corresponds to one blade, and because the same blade is adopted in the two modes of the pump and the hydraulic turbine, the initial geometric parameter sets in the two modes are obtained
Figure BDA0003394498920000071
Modeling is carried out according to the obtained geometric parameters of the I pumps, and for each grid model, grid node information G of each grid model is obtainedi,p,qMesh node information Gi,p,qInitial set of geometric parameters associated with the blade
Figure BDA0003394498920000072
Integrating to obtain a geometric parameter set Ai,j. In order to prevent dimension disaster caused by more geometric parameters of the blade, the geometric parameters are coded by adopting a coder-decoder combination, and the coder can be used for coding a geometric parameter set A of the bladei,jMapping into a fixed-length vector, and decoding the vector into geometric parameter set Ai,jThe loss function of the process is a set of geometric parameters Ai,jThe result of the predicted geometric parameters after encoding and decoding and the original real geometric parameter set Ai,jThe difference between:
Figure BDA0003394498920000073
combining vectors of I blades obtained by an encoder to form an indirect geometric parameter set A of the bladesi,f. Wherein, I is 1,2, 3., I represents different pumps, I is the total number of pumps, J is 1,2, 3., J represents different geometric parameters, J is the total number of geometric parameters, P is 1,2, 3., P represents different grid nodes, P is the total number of grid nodes, q is the coordinate of each grid node, including the coordinates in the x direction, the y direction and the z direction, if a two-dimensional model, only includes the coordinates in the x direction and the y direction, F is 1,2, 3., F represents different indirect geometric parameter variables,
secondly, respectively acquiring working condition parameters under two modes of a pump and a hydraulic turbine and acquiring a numerical simulation result;
obtaining the mechanical rotation speed omega when the pump and the hydraulic turbine operate in two modesRInlet pressure P in two modes of pump and hydraulic turbineinAnd inlet temperature TinAnd (5) waiting for working condition parameters. For each pump, under two modes, respectively adopting Latin hypercube sampling method to randomly sample D working conditions in respective parameter space, so that the working condition parameter set under the pump mode is Bi,dThe operating condition parameter in the hydraulic turbine mode is Ci,dAnd respectively carrying out CFD calculation on D working condition points of each pump model in two modes to obtain a physical field data set of each blade in the pump mode
Figure BDA0003394498920000081
Combining with physical field data in hydraulic turbine mode
Figure BDA0003394498920000082
According to the physical field parameters, the performance parameter set of different pumps under the pump mode can be calculated
Figure BDA0003394498920000083
And performance parameter set in hydraulic turbine mode
Figure BDA0003394498920000084
D is 1,2,3, the., D represents different operating parameters, D is the total number of operating points required to be calculated in each mode, N is 1,2,3, the., N represents different physical fields including speed, temperature, pressure, vorticity and the like, N is the total number of recorded flow field parameters, and m is 1 and 2 respectively represent efficiency and a water head.
Thirdly, normalizing input and output parameters of the pump and the hydraulic turbine;
normalizing the simplified set and other data to obtain normalized input data set { A }i,f,Bi,d,Ci,dThe physical field is collected as
Figure BDA0003394498920000085
The set of performance parameters is
Figure BDA0003394498920000086
The input data set after normalization is
Figure BDA0003394498920000087
Set of physical fields as
Figure BDA0003394498920000088
The set of performance parameters is
Figure BDA0003394498920000089
The normalization operation is Sigmoid function conversion, and the specific operation is as follows:
Figure BDA00033944989200000810
wherein, the expression of the Sigmoid function is as follows:
Figure BDA00033944989200000811
the normalization operation for the remaining data is similar:
Figure BDA0003394498920000091
Figure BDA0003394498920000092
Figure BDA0003394498920000093
Figure BDA0003394498920000094
Figure BDA0003394498920000095
Figure BDA0003394498920000096
the input data set after normalization is
Figure BDA0003394498920000097
Set of physical fields as
Figure BDA0003394498920000098
The set of performance parameters is
Figure BDA0003394498920000099
Fourthly, dividing a training set and a verification set aiming at the data of the pump and the hydraulic turbine model;
for the PointNet point cloud convolution neural network, the input data set is
Figure BDA00033944989200000910
Collecting input data
Figure BDA00033944989200000911
Randomly scrambling, and setting the input set as 2: 2: 1 into training set
Figure BDA00033944989200000912
Verification set
Figure BDA00033944989200000913
And test set
Figure BDA00033944989200000914
Similarly, the set of physical fields is divided into equal parts
Figure BDA00033944989200000915
And
Figure BDA00033944989200000916
the performance parameter sets are also classified into
Figure BDA00033944989200000917
And
Figure BDA00033944989200000918
fifthly, constructing two groups of physical fields and performance cooperative prediction point cloud convolution networks of the pump and the hydraulic turbine respectively;
two sets of networks for pump physical field and performance prediction and hydraulic turbine physical field prediction and performance prediction are constructed, each set comprising a point cloud convolutional neural network as shown in fig. 3 and 4. For the first set of pump physical fields and performance prediction networks, the first P _ Net1Point cloud convolution neural network aggregation according to input in pump mode
Figure BDA00033944989200000919
Predicting a set of physical fields in a pump mode
Figure BDA00033944989200000920
The mapping from design variables (geometric parameters and working condition parameters) to the physical field of the pump in the pump mode is realized, and the loss function of the network is as follows:
Figure BDA00033944989200000921
second P _ Net2Point cloud convolution neural network composed of P _ Net1Network predicted set of pump physical fields
Figure BDA0003394498920000101
Predicting a set of performance parameters for a pump mode
Figure BDA0003394498920000102
Implementing a mapping from the physical field of the pump to the performance parameters of the pump, of the networkThe loss function is as follows:
Figure BDA0003394498920000103
for the second group of hydraulic turbine physical field and performance prediction network, the first T _ Net1Point cloud convolution neural network according to input set in hydraulic turbine mode
Figure BDA0003394498920000104
Predicting to obtain physical field in hydraulic turbine mode
Figure BDA0003394498920000105
The method realizes the mapping from design variables (geometric parameters and working condition parameters) to a physical field of the hydraulic turbine in a hydraulic turbine mode, and the loss function of the network is as follows:
Figure BDA0003394498920000106
second T _ Net2Point cloud convolution neural network composed of T _ Net1Physical field predicted by network
Figure BDA0003394498920000107
Predicting to obtain a set of performance parameters in a hydraulic turbine mode
Figure BDA0003394498920000108
Mapping from the physical field of the hydraulic turbine to the performance parameter of the hydraulic turbine is realized, and the loss function of the network is as follows:
Figure BDA0003394498920000109
sixthly, training four groups of point cloud convolution neural networks of two parts of a pump and a hydraulic turbine;
in the process of training the network, firstly setting an optimizer Adam, setting an initial learning rate to be 0.01, and training for 50 steps; the optimizer is then set to SGD, after which the learning rate is reduced to 1/10 per 100 steps of training, with a minimum learning rate of 0.0001.
Seventhly, post-processing the performance prediction information;
output aggregation from two parts
Figure BDA00033944989200001010
And
Figure BDA00033944989200001011
establishing direct relation between the performance parameters in the pump mode and the performance parameters in the hydraulic turbine mode, realizing direct mapping from the efficiency in the pump mode to the efficiency in the hydraulic turbine mode, and simultaneously obtaining the related physical field analysis of the pump and the hydraulic turbine for further analysis.
8) Algorithm maintenance
In the actual application process, if the information needing to be input is increased or reduced, an encoder and a decoder are adopted, the original input information is mapped into a vector with a fixed size to serve as a new input, four groups of point cloud convolution networks including a trained pump, a trained hydraulic turbine physical field and performance prediction are adopted to serve as pre-training models, and training is carried out again on the basis.
In the step 2), the working condition is divided into Latin hypercube sampling, the point taking of the method in the working condition parameter space is multidimensional layered point taking, the method ensures that the obtained working condition points can well cover the whole parameter space, and the method is closest to random sampling.
Although the invention has been described in detail hereinabove with respect to a general description and specific embodiments thereof, it will be apparent to those skilled in the art that modifications or improvements may be made thereto based on the invention. Accordingly, such modifications and improvements are intended to be within the scope of the invention as claimed.

Claims (8)

1. A physical field and performance cooperative prediction method suitable for a pump and a hydraulic turbine is characterized by comprising the following steps:
1) geometric parameters of pump and hydraulic turbine models are obtained and encoded
The obtained geometric parameters of the blades are the installation angle alpha of the control blade, the number a of the blades and the coordinates (x, y, z) of B points of a Bezier curve of the control blade, each pump corresponds to one blade, and because the same blade is adopted in the two modes of the pump and the hydraulic turbine, the initial geometric parameter sets in the two modes are obtained by the method
Figure FDA0003394498910000011
Modeling is carried out according to the obtained geometric parameters of the I pumps, and for each grid model, grid node information G of each grid model is obtainedi，p，qMesh node information Gi，p，qInitial set of geometric parameters associated with the blade
Figure FDA0003394498910000012
Integrating to obtain a geometric parameter set Ai，jAnd encoding the geometric parameters by using a coder-decoder combination, wherein the coder Encoder encodes the geometric parameter set A of the bladei，jMapping into a fixed-length vector form, decoding the vector obtained by encoding into a geometric parameter set A by a Decoder Decoderi，jCombining the vectors of I blades obtained by the encoder to form an indirect geometric parameter set A of the bladesi，f(ii) a Wherein, I is 1,2, 3., I represents different pumps, I is the total number of pumps, J is 1,2, 3., J represents different geometric parameters, J is the total number of geometric parameters, P is 1,2, 3., P represents different mesh nodes, P is the total number of mesh nodes, q is the coordinate of each mesh node, including the coordinates in the x direction, the y direction and the z direction, if a two-dimensional model, including the coordinates in the x direction and the y direction, F is 1,2, 3., F represents different indirect geometric parameter variables, and F is the total number of indirect geometric parameter variables;
2) respectively obtaining working condition parameters of the pump and the hydraulic turbine under two modes and obtaining a numerical simulation result
Pump and water powerThe working condition parameters of the turbine comprise the mechanical rotating speed omega when the pump and the hydraulic turbine operate in two modesRInlet pressure P in two modes of pump and hydraulic turbineinAnd inlet temperature Tin(ii) a For each pump, under two modes, respectively adopting Latin hypercube sampling method to randomly sample D working conditions in respective parameter space, so that the working condition parameter set under the pump mode is Bi，dThe operating condition parameter in the hydraulic turbine mode is Ci，dAnd respectively carrying out CFD calculation on D working condition points of each pump model in two modes to obtain a three-dimensional physical field data set of each blade in the pump mode
Figure FDA0003394498910000013
Combining with three-dimensional physical field data in hydraulic turbine mode
Figure FDA0003394498910000014
Calculating to obtain a set of performance parameters of different pumps under the pump mode according to the physical field parameters
Figure FDA0003394498910000015
And performance parameter set in hydraulic turbine mode
Figure FDA0003394498910000016
D is 1,2,3, and D represents different operating parameters, D is the total number of operating points to be calculated in each mode, N is 1,2,3, and N represents different physical fields including speed, temperature, pressure, and vorticity, N is the total number of recorded flow field parameters, and m is 1 and 2, which represent efficiency and head, respectively;
3) pump and hydraulic turbine input and output parameter normalization
Normalizing the simplified set and other data to obtain normalized input data set { A }i,f,Bi,d,Ci,dThe physical field is collected as
Figure FDA0003394498910000021
The set of performance parameters is
Figure FDA0003394498910000022
The input data set after normalization is
Figure FDA0003394498910000023
Set of physical fields as
Figure FDA0003394498910000024
The set of performance parameters is
Figure FDA0003394498910000025
4) Partitioning training and validation sets for pump and hydraulic turbine model data
For neural networks such as point clouds, the input data set is
Figure FDA0003394498910000026
Collecting input data
Figure FDA0003394498910000027
Randomly scrambling, and setting the input set as 2: 2: 1 into training set
Figure FDA0003394498910000028
Verification set
Figure FDA0003394498910000029
And test set
Figure FDA00033944989100000210
Similarly, the set of physical fields is divided into equal parts
Figure FDA00033944989100000211
And
Figure FDA00033944989100000212
the performance parameter set is equally divided into
Figure FDA00033944989100000213
And
Figure FDA00033944989100000214
5) cooperative prediction point cloud convolution network for respectively constructing two groups of physical fields and performances of pump and hydraulic turbine
Constructing two groups of networks of pump physical field and performance prediction and hydraulic turbine physical field prediction and performance prediction, wherein each group comprises two point cloud convolution neural networks; for the first set of pump physical fields and performance prediction networks, the first P _ Net1Point cloud convolution neural network aggregation according to input in pump mode
Figure FDA00033944989100000215
Predicting a set of physical fields in a pump mode
Figure FDA00033944989100000216
Realizing the mapping of the design variables, i.e. the geometric parameters and the working condition parameters, to the physical field of the pump in the pump mode, the second P _ Net2The point cloud convolution neural network input is P _ Net1Network predicted set of pump physical fields
Figure FDA00033944989100000217
Predicting a set of performance parameters for a pump mode
Figure FDA00033944989100000218
Mapping from the pump physical field to the pump performance parameter is realized; for the second group of hydraulic turbine physical field and performance prediction network, the first T _ Net1Point cloud convolution neural network according to input set in hydraulic turbine mode
Figure FDA00033944989100000219
Predicting to obtain hydraulic turbine modePhysical field of
Figure FDA00033944989100000220
Implementing a mapping of design variables to a physical field of the hydraulic turbine in the hydraulic turbine mode, a second T _ Net2The point cloud convolution neural network input is T _ Net1Physical field predicted by network
Figure FDA00033944989100000221
Predicting to obtain a set of performance parameters in a hydraulic turbine mode
Figure FDA00033944989100000222
Mapping from a hydraulic turbine physical field to hydraulic turbine performance parameters is realized;
6) convolutional neural network for training two groups of point clouds of pump and hydraulic turbine
7) Performance prediction information post-processing
Output aggregation from two parts
Figure FDA0003394498910000031
And
Figure FDA0003394498910000032
establishing direct relation between the performance parameters in the pump mode and the performance parameters in the hydraulic turbine mode, realizing direct mapping from the efficiency in the pump mode to the efficiency in the hydraulic turbine mode, and simultaneously obtaining the related physical field analysis of the pump and the hydraulic turbine for further analysis.
2. The method of claim 1, further comprising the steps of:
8) algorithm maintenance
In the actual application process, if the information needing to be input is increased or reduced, an encoder and a decoder are adopted, the original input information is mapped into a vector with a fixed size to serve as a new input, four groups of point cloud convolution networks including a trained pump, a trained hydraulic turbine physical field and performance prediction are adopted to serve as pre-training models, and training is carried out again on the basis.
3. The physical field and performance collaborative prediction method applicable to the pump and the hydraulic turbine as claimed in claim 1, wherein in the step 1), the Encoder Encoder and the Decoder Decoder both adopt full connection layer to construct the network, and the Encoder Encoder constructs the coding mapping function to realize the collaborative prediction from the geometric parameter set Ai,jEncoding mapping to a fixed length vector, Decoder constructing decoding mapping function to realize the mapping from the fixed length vector to the geometric parameter set Ai,jDecoding the map of (1); the loss function of the process is a set of geometric parameters Ai,jThe result of the predicted geometric parameters after encoding and decoding and the original real geometric parameter set Ai,jThe difference between:
Figure FDA0003394498910000033
4. the physical field and performance collaborative prediction method applicable to the pump and the hydraulic turbine as claimed in claim 1, wherein in the step 3), the normalization operation is Sigmoid function conversion, and the specific operation is as follows:
Figure FDA0003394498910000034
wherein, the expression of the Sigmoid function is as follows:
Figure FDA0003394498910000041
the normalization operation for the remaining data is similar:
Figure FDA0003394498910000042
Figure FDA0003394498910000043
Figure FDA0003394498910000044
Figure FDA0003394498910000045
Figure FDA0003394498910000046
Figure FDA0003394498910000047
5. the physical field and performance collaborative prediction method applicable to the pump and the hydraulic turbine is characterized in that in the step 4), a point cloud convolution network is used for the physical field and performance collaborative prediction of the pump and the hydraulic turbine.
6. The physical field and performance collaborative prediction method for pumps and hydraulic turbines according to claim 5, wherein the point cloud convolution Network employs PointNet, PointNet + +, or Kd-Network.
7. The physical field and performance collaborative prediction method for the pump and the hydraulic turbine is characterized in that, in the step 5), for two modes of the pump and the hydraulic turbine, two point cloud convolution neural networks in each mode comprise two parts of loss functions, one part is a predicted loss function F _ loss of the physical field, and the loss function F _ loss is used for reducing the difference between the predicted physical field and the real physical field; the other part is a loss function psi _ loss for performance prediction, which is used for reducing the gap between the predicted performance and the real performance;
Figure FDA0003394498910000048
Figure FDA0003394498910000049
Figure FDA00033944989100000410
Figure FDA0003394498910000051
wherein, wqPredicting weights of the squared differences of the physical field data and the true physical field data for each network node,
Figure FDA0003394498910000052
are respectively P _ Net1、T_Net1Two physical field prediction point cloud convolutional neural networks generated predicted physical field data,
Figure FDA0003394498910000053
real physical field data for the pump and hydraulic turbine respectively,
Figure FDA0003394498910000054
are respectively P _ Net2、T_Net2Two performance prediction point cloud convolutional neural network generated predicted performance data,
Figure FDA0003394498910000055
respectively a pump andtrue performance data for hydraulic turbines.
8. The physical field and performance collaborative prediction method suitable for the pump and the hydraulic turbine is characterized in that in the step 6), in the process of training the network, firstly, an optimizer is set to Adam, the initial learning rate is set to 0.01, and 50 steps of training are carried out; the optimizer is then set to SGD, after which the learning rate is reduced to 1/10 per 100 steps of training, with a minimum learning rate of 0.0001.
CN202111479710.3A 2021-12-06 2021-12-06 Physical field and performance collaborative prediction method suitable for pump and hydraulic turbine Pending CN114169605A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111479710.3A CN114169605A (en) 2021-12-06 2021-12-06 Physical field and performance collaborative prediction method suitable for pump and hydraulic turbine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111479710.3A CN114169605A (en) 2021-12-06 2021-12-06 Physical field and performance collaborative prediction method suitable for pump and hydraulic turbine

Publications (1)

Publication Number Publication Date
CN114169605A true CN114169605A (en) 2022-03-11

Family

ID=80483609

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111479710.3A Pending CN114169605A (en) 2021-12-06 2021-12-06 Physical field and performance collaborative prediction method suitable for pump and hydraulic turbine

Country Status (1)

Country Link
CN (1) CN114169605A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114547987A (en) * 2022-04-22 2022-05-27 中国计量大学 Centrifugal pump turbine performance prediction method based on improved artificial neural network

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114547987A (en) * 2022-04-22 2022-05-27 中国计量大学 Centrifugal pump turbine performance prediction method based on improved artificial neural network
CN114547987B (en) * 2022-04-22 2022-07-26 中国计量大学 Centrifugal pump turbine performance prediction method based on improved artificial neural network

Similar Documents

Publication Publication Date Title
Shamshirband et al. A survey of deep learning techniques: application in wind and solar energy resources
CN110348059B (en) Channel internal flow field reconstruction method based on structured grid
CN110705029B (en) Flow field prediction method of oscillating flapping wing energy acquisition system based on transfer learning
CN110264012B (en) Renewable energy power combination prediction method and system based on empirical mode decomposition
CN114580152A (en) Floating wind power structure foundation local stress time domain analysis method based on multi-body coupling analysis
CN111401603A (en) Multi-type energy demand prediction method
CN114547987B (en) Centrifugal pump turbine performance prediction method based on improved artificial neural network
CN110222453A (en) A kind of compressor outlet parameter prediction modeling method based on width learning system
CN117096867A (en) Short-term power load prediction method, device, system and storage medium
CN114169605A (en) Physical field and performance collaborative prediction method suitable for pump and hydraulic turbine
CN115758891A (en) Wing profile flow field prediction method based on Transformer decoder network
He et al. Advanced ocean wave energy harvesting: current progress and future trends
CN114021836A (en) Multivariable reservoir water inflow amount prediction system based on different-angle fusion, training method and application
CN116187508A (en) Wind turbine generator system fault prediction method integrating knowledge graph and deep learning
Yan et al. Hybrid physical and data driven modeling for dynamic operation characteristic simulation of wind turbine
CN113361047A (en) Impeller high-dimensional optimization and model selection method for multilayer sensing structure
CN117273208A (en) Combined power prediction method, device, equipment and storage medium
CN115310209B (en) Pneumatic shape migration optimization method and related device based on VAE
Zhang et al. Data-model-interactive enhancement-based Francis turbine unit health condition assessment using graph driven health benchmark model
CN112632728B (en) Turbine mechanical blade profile design and performance prediction method based on deep learning
Stergiopoulou et al. Computational Fluid Dynamics study on a 3D graphic solid model of archimedean screw turbines
CN115455819A (en) Turbine blade multi-working-condition transient flow field migration reduced-order intelligent modeling method
CN113792919A (en) Wind power prediction method based on combination of transfer learning and deep learning
CN113962357A (en) GWO-WNN-based distributed photovoltaic power data virtual acquisition method
CN111368408A (en) Seq2 seq-based data center energy efficiency optimization continuous decision method

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