CN117993303B - Pneumatic load checking method for deformation of offshore wind power blade - Google Patents
Pneumatic load checking method for deformation of offshore wind power blade Download PDFInfo
- Publication number
- CN117993303B CN117993303B CN202410389837.3A CN202410389837A CN117993303B CN 117993303 B CN117993303 B CN 117993303B CN 202410389837 A CN202410389837 A CN 202410389837A CN 117993303 B CN117993303 B CN 117993303B
- Authority
- CN
- China
- Prior art keywords
- blade
- neural network
- loss function
- data
- physical
- 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
- 238000000034 method Methods 0.000 title claims abstract description 90
- 238000013528 artificial neural network Methods 0.000 claims abstract description 112
- 238000004088 simulation Methods 0.000 claims abstract description 70
- 238000012549 training Methods 0.000 claims abstract description 35
- 239000012530 fluid Substances 0.000 claims abstract description 31
- 238000012544 monitoring process Methods 0.000 claims abstract description 27
- 238000005259 measurement Methods 0.000 claims abstract description 25
- 238000013461 design Methods 0.000 claims abstract description 23
- 238000012360 testing method Methods 0.000 claims abstract description 11
- 230000001419 dependent effect Effects 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 79
- 238000004364 calculation method Methods 0.000 claims description 56
- 230000004913 activation Effects 0.000 claims description 8
- 238000013459 approach Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 239000000203 mixture Substances 0.000 claims description 3
- 238000005293 physical law Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 description 14
- 230000008901 benefit Effects 0.000 description 7
- 230000005856 abnormality Effects 0.000 description 6
- 239000010410 layer Substances 0.000 description 6
- 238000010801 machine learning Methods 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000012937 correction Methods 0.000 description 5
- 238000001514 detection method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 230000003993 interaction Effects 0.000 description 5
- 238000005457 optimization Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- KFLQGJQSLUYUBF-WOJBJXKFSA-N Phyllanthin Chemical compound C([C@H](COC)[C@@H](COC)CC=1C=C(OC)C(OC)=CC=1)C1=CC=C(OC)C(OC)=C1 KFLQGJQSLUYUBF-WOJBJXKFSA-N 0.000 description 4
- KFLQGJQSLUYUBF-PMACEKPBSA-N Phyllanthin Natural products C([C@@H](COC)[C@H](COC)CC=1C=C(OC)C(OC)=CC=1)C1=CC=C(OC)C(OC)=C1 KFLQGJQSLUYUBF-PMACEKPBSA-N 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 4
- 238000005452 bending Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 239000002131 composite material Substances 0.000 description 2
- 238000005094 computer simulation Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000013433 optimization analysis Methods 0.000 description 2
- 238000010206 sensitivity analysis Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 241001130943 Phyllanthus <Aves> Species 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000001808 coupling effect Effects 0.000 description 1
- 238000012938 design process Methods 0.000 description 1
- 238000012942 design verification Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000012407 engineering method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 239000011229 interlayer Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000012795 verification 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
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
-
- 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
- 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
- G06N3/0499—Feedforward 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
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- 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)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Geometry (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Analysis (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Computational Mathematics (AREA)
- Computer Hardware Design (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Algebra (AREA)
- Molecular Biology (AREA)
- Databases & Information Systems (AREA)
- Operations Research (AREA)
- Fluid Mechanics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Wind Motors (AREA)
Abstract
A pneumatic load checking method for deformation of an offshore wind turbine blade comprises the following steps: s1: the selected airfoil interface is used for constructing a feedforward neural network for reconstructing the fluid state of the airfoil surface of the fan according to the independent variable and the dependent variable number of the two-dimensional incompressible fluid equation set; s2: defining a total loss function comprising data items and physical items for quantifying differences between the neural network predicted values and the actual values; the source of the data item is fan field monitoring data, and the source of the physical item is Navier-Stokes equation set and continuity equation added in; the data item is used as a supervision point of neural network training and is used for improving simulation precision; s3: constructing a physical information neural network PINN, executing a grid-free solving Navier-Stokes equation set, and carrying out simulation reproduction on the running state of the two-dimensional airfoil; s4, training a physical information neural network PINN by using test data or actual measurement data samples; s5, obtaining a new integral blade load through the trained physical information neural network PINN, and using the new integral blade load for blade design.
Description
Technical Field
The invention belongs to the field of load checking of offshore wind power deformed blades, and particularly relates to a pneumatic load checking method for offshore wind power blade deformation.
Background
The development of a high-power fan is always a main means for reducing the cost and enhancing the efficiency of offshore wind power, wherein the enlargement of blades is a key for improving the capacity of a single machine and is also an important path for improving the power generation efficiency. The research of aerodynamic characteristics of wind turbine blades is still a research focus in the field of offshore wind power, and an accurate blade aerodynamic performance calculation result is used as an index of blade design evaluation and is used for iteratively optimizing a blade design scheme, so that the blade starting performance is more ideal. The current method for calculating the aerodynamic performance of the wind turbine blade is mainly a phyllotoxin momentum theory, namely a BEM method, and is corrected through an empirical formula.
BEM method of the phyllin momentum theory, engineering method for calculating wind turbine blade forces and loads. The method is based on aerodynamic theory on the surface of the blade, and aerodynamic force and torque of each position on the blade are calculated by analyzing the flow of air on the surface of the blade;
the basic principle of BEM method of phyllin momentum theory is briefly summarized as follows:
blade section analysis: dividing the wind turbine blade into a series of fine slices, known as phyllins;
aerodynamic analysis: for each phyllanthin, analyzing the aerodynamic force it is subjected to according to the fundamental principles of hydrodynamics; this includes pressure distribution and shear forces between the airflow and the blade surface;
Blade torque calculation: calculating the torque born by the blade at different positions according to aerodynamic forces at each blade element on the blade; this involves integrating the aerodynamic forces, taking into account the geometry of the blades and the speed of the airflow;
Overall effect considerations: taking into account aerodynamic interactions and superposition effects at individual phylloxnes on the blade, as well as interactions with the entire wind turbine; performance evaluation and optimization: based on the calculation of the blade stress and torque, the performance of the wind turbine is evaluated, and the design of the blade may be optimized to improve its energy conversion efficiency and stability.
The BEM method uses the concept of leaf element, wherein the English of leaf element is 'blade element', the blade is divided into a series of tiny slices, then aerodynamic force at each leaf element is analyzed, the force and torque at the leaf element are integrated on the whole blade by considering the speed, density and geometric shape of air flow, and the stress and torque condition of the whole blade are obtained, so that the BEM belongs to two-dimensional calculation and check;
However, the conventional BEM method is not accurate in the simulation calculation of the ultra-long flexible blade, especially when the simulated airfoil interface is in high-speed motion, has a large tip speed ratio and deforms greatly; the conventional BEM is a two-dimensional computational check.
Two-dimensional assumption: the phyllanthus momentum theory is generally based on the assumption of two dimensions, i.e. ignoring the three-dimensional effects of the blade; in practice, bending and twisting of the blade can cause three-dimensional effects, which can lead to differences between theoretical calculations and actual conditions;
static assumption: the theory of phyllin momentum generally assumes that the fluid is stationary, i.e., ignoring the unsteady and turbulent effects of the airflow; however, in practical situations, the dynamics and turbulence of the air flow can have an effect on the aerodynamic forces of the blades, thereby affecting the accuracy of the theoretical calculation;
Stall effect: the theory of blade momentum is often difficult to accurately model the aerodynamic characteristics of a blade when it stalls; stall can cause sudden changes in aerodynamic forces, which can occur frequently for some high performance wind turbines;
Coupling effect: the theory of phyllin momentum generally assumes that the interaction between the blade and the surrounding environment is linear; however, in practical situations there is a complex nonlinear interaction between the blades and the tower, the ground and other blades, which may lead to deviations between theoretical calculations and the actual situation;
The empirical parameters are required: the phyllanthin momentum theory requires some empirical parameters, such as aerodynamic coefficients and correction factors, the choice of which may affect the accuracy of the calculation.
According to related data, the uncertainty of tip loss correction in the BEM method can not catch axial and radial movement of tip vortex, and the description of free vortex wake and flow around phenomenon is insufficient, so that the blade load calculation of the area beyond the impeller radius of 70% is 10-40% higher than the actual load.
Application publication number JP2022168865A discloses a method for diagnosing a wind power blade by measuring the deformation of the wind power blade in real time in an existing wind power generator, comprising the additional step of installing a sensor for detecting the deformation of the wind turbine comprising: a step of acquiring a detection signal from a sensor; a step of calculating a load acting on the wind turbine blade based on the detection signal; and detecting whether an abnormality exists in the wind turbine blade based on the calculation result. Detecting signals and preparing. The additionally mounted sensor includes an abnormality detection sensor that detects deformation of the wind turbine blade to detect whether an abnormality exists. In the step of detecting the presence or absence of an abnormality, the presence or absence of an abnormality is detected based on a detection signal from an abnormality detection sensor.
Application publication number CN115577625A discloses a method and a device for predicting limit pneumatic load of a large flexible wind power blade under turbulent wind conditions, comprising the following steps: a sweepback and bending torsion coupling design method adopted for reducing load in the design process of the large flexible wind power blade is considered, and a blade elastic torsion angle fitting curve is established; considering the vibration speed generated by a large flexible blade under the turbulent wind condition, and establishing a blade vibration speed fitting curve according to a modal superposition method; establishing a momentum phyllotoxin method load calculation flow considering an elastic torsion angle and a vibration speed caused by turbulent wind; and the limit pneumatic load prediction is defined as an optimizing process under a certain constraint condition, and the limit pneumatic load of the large flexible blade is predicted by adopting the limit pneumatic load prediction method.
However, aiming at the load of the wind power deformed blade at sea, how to improve the simulation calculation value of the load of the blade to approach the actual load, reduce errors, improve calculation accuracy, and lack effective technical measures, so that improvement is needed.
Disclosure of Invention
In view of the defects in the prior art, the application aims to provide a pneumatic load checking method for offshore wind power blade deformation, which reasonably utilizes big data and combines a machine learning method to perform design calibration calculation by utilizing the characteristics of an offshore wind power multi-sensor, and provides a checking mode for engineering design.
To achieve the above and other related objects, the present application provides a pneumatic load checking method for deformation of an offshore wind turbine blade, comprising the steps of:
s1: the selected airfoil interface is used for constructing a feedforward neural network for reconstructing the fluid state of the airfoil surface of the fan according to the independent variable and the dependent variable number of the two-dimensional incompressible fluid equation set;
s2: defining a total loss function comprising data items and physical items for quantifying differences between the neural network predicted values and the actual values;
The source of the data item is fan field monitoring data, and the source of the physical item is Navier-Stokes equation set and continuity equation added in; the data item is used as a supervision point of neural network training and is used for improving simulation precision;
s3: constructing a physical information neural network PINN, and fitting and approximating a smooth function to the physical information neural network PINN;
executing a grid-free solving Navier-Stokes equation set, and carrying out simulation reproduction on the running state of the two-dimensional airfoil;
Navier-Stokes equation set eq.1, eq.2 and continuity equation eq.3 are as follows:
The formula: eq.1
The formula: eq.2
The formula: eq.3
The formula: eq.1, eq.2, eq.3 are constraint equations;
the total loss function is expressed as the formula: eq.4
Wherein the method comprises the steps ofIs the total loss function of the physical information neural network PINN,Is a function of the loss of the data item,Is a physical law term loss function; the loss function of the physical rule term is obtained by a partial differential equation set loss function term in computational fluid dynamicsBoundary condition loss function termAnd an initial condition loss function termThe composition of the composite material comprises the components,、For adjusting weight coefficients、At the total loss functionThe ratio of (3);
The initial conditions and boundary conditions are described as follows:
The initial conditions are usually the values of the horizontal speed u of the two-dimensional airfoil surface, the vertical speed v of the two-dimensional airfoil surface and the pressure p of the two-dimensional airfoil surface at the t moment;
Boundary conditions refer to conditions that should be satisfied by a fluid physical quantity on the boundary of a solution domain of a fluid mechanics equation set, such as: the fluid should not have a velocity component across the airfoil interface.
The formula: eq.5
The formula: eq.6
The formula: eq.7
Wherein the method comprises the steps ofTo calculate the total number of data points for the fluid partial differential equation loss function,To calculate the number of data points for the fluid boundary condition loss function,Calculating a number of data points for an initial condition loss function of the fluid system; g is an implicit expression of a Navier-Stokes equation;、 The values of the measurement time t, the coordinates x and y, the blade surface gas speed U and the blade surface pressure p of the ith sensor are obtained by measurement; the blade surface gas velocity U includes a horizontal component velocity U and a vertical component velocity v; 、 to predict the values of the obtained measurement time t, coordinates x, coordinates y, blade surface gas velocity U, blade surface pressure p of the ith sensor, 、The method is obtained through simulation calculation;
after the total loss function is developed, the formula is expressed as: eq.8
；
Wherein:、、、 For adjusting weight coefficients 、、、At the total loss functionThe ratio of (3);
s4, using test data or actual measurement data samples, wherein the output result in S3 meets the physical rule, and training the physical information neural network PINN is completed; physical information neural network PINN allows for predicted speed by minimizing the total loss function And pressureAfter training the physical information neural network PINN, the input and corresponding output values, the measurement time t, the coordinate x, the coordinate y, the blade surface gas speed U and the blade surface pressure p are substituted into constraint equations eq.1, eq.2 and eq.3, and the result approaches zero, namely the given physical formula is satisfied;
S5: according to the obtained simulation calculation result of the two-dimensional blade section through the trained physical information neural network PINN, replacing the corresponding section in the BEM, calculating the total section number of the two-dimensional blade used by the BEM for N, calculating the 1 st to N two-dimensional blade sections by the BEM, and calculating the physical information of the fan blade integral load and deformation by adopting the physical information neural network PINN for the N < N, n+1 to N two-dimensional blade sections to obtain new blade integral load for blade design.
The technical scheme provided by the application also has the following technical characteristics:
Preferably, in an embodiment of the present application, values of measurement time t, coordinates x, coordinates y, blade surface gas velocity U, and blade surface pressure p obtained by measurement of four monitoring sensors distributed on the airfoil section are used as boundary conditions for simulation in S3, and are used for simulation reproduction of a two-dimensional airfoil running state;
And S4, taking four monitoring sensor positions distributed on the section of the airfoil profile as learning supervision points.
Preferably, in an embodiment of the present application, in S2 and S3, the weight coefficient in the total loss function、In the process of adjustment、After the duty ratio in the loss function, the output result meets the physical rule.
Preferably, in an embodiment of the present application, the feedforward neural network construction in S1 includes the following steps, S101: and constructing a feedforward neural network without physical information, and setting a hidden layer, an input variable, an output variable and an activation function.
Preferably, in an embodiment of the present application, the feedforward neural network construction in S1 includes the following steps, S102: taking a coordinate x, a coordinate y and a time t as inputs, taking a horizontal component speed u, a vertical component speed v and a pressure p under the two-dimensional plane coordinate of the airfoil profile as outputs, and taking tanh as an activation function of the feedforward neural network;
The activation function is a function transformation between the input and the output of the interlayer of the neural network, and aims to add nonlinear factors and enhance the expression capacity of the neural network, and common activation functions are as follows: sigmoid, tanh, relu.
Preferably, in an embodiment of the present application, in S2: the number of the supervision points is directly proportional to the simulation precision; the same airfoil section is provided with 4 supervision points.
Preferably, in an embodiment of the present application, in S2 and S3: the total loss function comprises physical information constraint, so that the data layer of the neural network is learned and trained on the physical rule layer.
Preferably, in an embodiment of the present application, the wind farm where the blade of the airfoil interface selected in S1 is located obtains actual offshore wind farm monitoring data, including global wind speed and wind shear parameters, and specific position monitoring data of the airfoil surface at a position of a certain distance from the front end of the blade.
Preferably, in an embodiment of the present application, the simulated section aerodynamic load is obtained by the BEM method by setting a simulated working condition.
Preferably, in an embodiment of the present application, the simulated condition includes wind speed.
Preferably, in an embodiment of the present application, in S2 and S3, the weight coefficient in the total loss function、、、In the process of adjustment、、、After the duty ratio in the total loss function, the output result meets the physical rule; in S4, the physical information neural network PINN is trained using the test data and the measured data samples.
The application has the beneficial effects that:
the advantages of using the physical information neural network PINN of the present application compared to computational fluid dynamics CFD methods are:
1. the data items are added, and the neural network is trained by utilizing big data and real data, so that the neural network is close to a real working condition;
2. high-quality grids are not needed to simulate fluid flow, and a grid-free scheme is more beneficial to solving engineering problems.
3. The physical information neural network PINN of the application realizes the known partial data to infer the flow field distribution condition of the whole calculation area, and the traditional CFD calculation cannot incorporate the actual measurement data.
Drawings
FIG. 1 is a schematic drawing of a sectional view of a fan blade element in a fan blade element theory of an aerodynamic load checking method for offshore wind power blade deformation;
FIG. 2 is a schematic diagram of fan airfoil interface monitoring point distribution for a pneumatic load check method for offshore wind blade deformation according to the present invention;
FIG. 3 is a physical information neural network PINN of an aerodynamic load check method for offshore wind blade deformation according to the present invention;
FIG. 4 is a conventional phyllanthin momentum theory iteration schematic diagram of an aerodynamic load checking method for offshore wind turbine blade deformation;
FIG. 5 is a schematic diagram of an iteration of the principle of momentum theory of the coupled PINN post-leaf elements of the aerodynamic load checking method for the deformation of the offshore wind turbine blade;
FIG. 6 is a graph comparing two-dimensional BEM method with three-dimensional CFD method and experimental data;
FIG. 7 is a schematic diagram of an airfoil surface sensor distribution.
Detailed Description
The following describes the embodiments of the present application in further detail with reference to the accompanying drawings. These embodiments are merely illustrative of the present application and are not intended to be limiting.
In the description of the present invention, it should be noted that the terms "center", "longitudinal", "lateral", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, merely to facilitate description of the present invention and simplify the description, and do not indicate or imply that the apparatus or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In the description of the present invention, it should be noted that, unless explicitly specified and limited otherwise, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be either fixedly connected, detachably connected, or integrally connected, for example; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
Furthermore, in the description of the present invention, unless otherwise indicated, the meaning of "a plurality" is two or more.
1-5, A pneumatic load checking method for the deformation of the offshore wind power blade utilizes a monitoring sensor arranged at the front end of a long flexible blade under the actual large deformation of the offshore wind power to couple with the traditional phyllin momentum theory so as to obtain the integral load and deformation simulation data of the blade under the more practical large deformation; the physical information-based neural network is adopted to obtain the physical information neural network PINN based on the actual measurement data of the blade operation, the aerodynamic load of the blade under the influence of the boundary layer is considered, and the blade section aerodynamic load simulation calculation neural network is established. The physical formula is added into the total loss function of the neural network, so that the trained neural network accords with the physical rule, the trained result is visualized, the axial and radial movement and turbulence phenomena of blade wake flow and blade tip vortex are accurately presented, the blade section load time course result is obtained through surface pressure integration, the blade front end wing profile calculation data in the traditional blade element momentum theory is replaced, the blade load and aerodynamic characteristic result under large deformation is obtained, and the follow-up blade research and design are facilitated.
Specifically, in an embodiment of the present application, as shown in fig. 1-5, a method for checking aerodynamic load of deformation of an offshore wind turbine blade includes the following steps:
s1: the selected airfoil interface is used for constructing a feedforward neural network for reconstructing the fluid state of the airfoil surface of the fan according to the independent variable and the dependent variable number of the two-dimensional incompressible fluid equation set;
s2: defining a total loss function comprising data items and physical items for quantifying differences between the neural network predicted values and the actual values;
The source of the data item is fan field monitoring data, and the source of the physical item is Navier-Stokes equation set and continuity equation added in; the data item is used as a supervision point of neural network training and is used for improving simulation precision;
s3: constructing a physical information neural network PINN, and fitting and approximating a smooth function to the physical information neural network PINN;
executing a grid-free solving Navier-Stokes equation set, and carrying out simulation reproduction on the running state of the two-dimensional airfoil;
Navier-Stokes equation set eq.1, eq.2 and continuity equation eq.3 are as follows:
The formula: eq.1
The formula: eq.2
The formula: eq.3
The formula: eq.1, eq.2, eq.3 are constraint equations;
the total loss function is expressed as the formula: eq.4
Wherein: Is the total loss function of the physical information neural network PINN, Is a function of the loss of the data item,Is a physical law term loss function; the loss function of the physical rule term is obtained by a partial differential equation set loss function term in computational fluid dynamicsBoundary condition loss function termAnd an initial condition loss function termThe composition of the composite material comprises the components,、For adjusting weight coefficients、At the total loss functionThe ratio of (3);
the formula: eq.5
The formula: eq.6
The formula: eq.7
Wherein: to calculate the total number of data points for the fluid partial differential equation loss function, To calculate the number of data points for the fluid boundary condition loss function,Calculating a number of data points for an initial condition loss function of the fluid system; g is an implicit expression of a Navier-Stokes equation;、 The values of the measurement time t, the coordinates x and y, the blade surface gas speed U and the blade surface pressure p of the ith sensor are obtained by measurement; the blade surface gas velocity U includes a horizontal component velocity U and a vertical component velocity v; 、 to predict the values of the obtained measurement time t, coordinates x, coordinates y, blade surface gas velocity U, blade surface pressure p of the ith sensor, 、The method is obtained through simulation calculation;
The total loss function expansion is expressed as the formula: eq.8
、、、For adjusting weight coefficients、、、At the total loss functionThe ratio of (3);
In S4, training the physical information neural network PINN by using test data and actually measured data samples, wherein the output result in S3 meets the physical rule, and training the physical information neural network PINN is completed; physical information neural network PINN allows for predicted speed by minimizing the total loss function And pressureAfter training the physical information neural network PINN, the input and corresponding output values, the measurement time t, the coordinate x, the coordinate y, the blade surface gas speed U and the blade surface pressure p are substituted into constraint equations eq.1, eq.2 and eq.3, and the result approaches zero, namely the given physical formula is satisfied;
S5: the two-dimensional blade section simulation calculation result is obtained through the trained physical information neural network PINN; replacing corresponding sections in the BEM, calculating the total section number for the two-dimensional blades used by the BEM by N, calculating the 1 st to N th two-dimensional blade sections by using the BEM, calculating the N < N, and calculating the physical information of the whole load and deformation of the fan blade again by adopting a physical information neural network PINN for the N < N, wherein the n+1 th to N th two-dimensional blade sections are used for blade design to obtain new whole load of the blade;
It should be noted that: the physical information neural network PINN is irrelevant to BEM, only the results of the last leaves in the BEM algorithm are replaced, and the physical information neural network PINN can use CFD data or monitoring data as a training supervision point; and if conditions allow, existing calculation data of the BEM and its cross-section division can be used; CFD data can also be used, but the CFD itself requires much time to calculate and requires a large sample size for training; meanwhile, the accuracy error after training is still approximate to CFD, so the BEM adopted by the application rapidly obtains a calculation result, and the CFD data or the monitoring data are used as training supervision points in combination, so that a rapid, efficient, low-cost and high-accuracy wind turbine blade checking calculation method is obtained, and further, after the wind turbine blade checking calculation method is realized, the accuracy is improved by self-iteration and further combining the monitoring data;
Therefore, the calculation of samples required by BEM training can be completed in the application, or the existing BEM calculation samples can be used as training samples in advance; similarly, for training samples used in CFD, the calculation of samples required for BEM training is completed in the present application, or existing BEM calculation samples are used as training samples in advance.
FIG. 3, where loss represents the loss value of the physical information neural network PINN during the simulation process; epsilon takingWhen the loss value is smaller thanIndicating that training is complete.
As in fig. 1, the schematic BEM two-dimensional section division, R is the total blade length, R is the airfoil section-to-blade root distance, dr is the step size;
S1, selecting actual offshore wind farm monitoring data, wherein the actual offshore wind farm monitoring data comprises global wind speed and wind shear parameters and specific positions (front 20%) of the airfoil surface at a certain distance from the front end of a blade, and monitoring the data by using position points of monitoring sensors shown in FIG. 2; four points of distribution 1,2, 3,4 on the airfoil section of fig. 2, at which the data are monitored.
In S1, selecting an airfoil interface, setting simulation working conditions including wind speed, and obtaining the aerodynamic load of the simulation section through a BEM method.
S1, constructing a feedforward neural network suitable for reconstructing the fluid state of the wing surface of a fan according to the independent variable and the dependent variable number of a two-dimensional incompressible fluid equation set aiming at a selected wing interface; firstly, building a feedforward neural network, setting a hidden layer, an input variable, an output variable and an activation function without physical information;
The coordinates x, y and time t are taken as inputs, the horizontal component speed u, the vertical component speed v and the pressure p under the two-dimensional plane coordinates of the wing profile are taken as outputs, and the tanh is taken as an activation function of the feedforward neural network.
S2 and S3, by defining a total loss function, the total loss function comprises a data item and a physical item, and is used for quantifying the difference between the predicted value and the actual value of the neural network;
The data item sources are fan field monitoring data, and the physical item sources are Navier-Stokes equation sets and continuity equations added into the fan field monitoring data. The data item is used as a supervision point for neural network training, so that the simulation precision is increased, and the number of the supervision points can be adjusted according to the actual situation; the more supervision points, the more closely the simulation accuracy is to the data, and the more data-dependent the trained neural network will be.
"Simulation accuracy" refers to the degree of accuracy or precision of the results obtained when performing a simulation or simulation; in the fields of science, engineering, computer graphics, simulation is often used by people to simulate various conditions in the real world in order to study, analyze, or predict the behavior of a particular system; the simulation precision is to evaluate the coincidence degree between the simulation results and the actual conditions;
Evaluation of simulation accuracy typically involves comparing simulation results to actual observed data, theoretical models, or results of other reliable references; if the simulation result accords with the reference results, the simulation precision is higher; conversely, if the simulation result has larger deviation or error with the reference result, the simulation model, parameters or input data need to be adjusted to improve the simulation precision;
In practical application, various factors are generally required to be considered for improving the simulation precision, including the complexity of a model, the quality of input data and the accuracy of a simulation algorithm; therefore, evaluating and improving simulation accuracy is a complex and important task, particularly when accurate modeling and analysis of complex systems is required;
The performance and effectiveness of a machine learning neural network depends largely on the data it trains; in other words, the characteristics, patterns and laws learned by the neural network are all based on the data it contacts; if the training data is representative and covers diversity, then the neural network is likely to have better generalization ability, i.e., perform well on unseen data; conversely, if the training data is of poor quality or not sufficiently diversified, the neural network may suffer from over-fitting or under-fitting, resulting in poor performance in practical applications; the quality and diversity of data are critical to the training of machine learning neural networks;
In view of the training impact of data quality and diversity on machine learning neural networks, training physical information neural networks PINN using test data or measured data samples;
In S3, the values of measurement time t, coordinate x, coordinate y, blade surface gas speed U and blade surface pressure p obtained by measurement of four monitoring sensors distributed on the section of the airfoil are used as boundary conditions for simulation in S3 and are used for simulation reproduction of the two-dimensional airfoil running state;
S2, taking four monitoring sensor positions distributed on the section of the wing profile as learning supervision points;
In general, 4 supervision points are arranged on the same airfoil section, so that the dependence of the neural network on data can be reduced, and the calculation speed of the neural network can be improved; physical information constraint is introduced into the loss function, so that the neural network can learn at the data level and train at the physical rule level; by constructing a neural network PINN based on physical information, the capability of the neural network to approach any smooth function is utilized to solve a Navier-Stokes equation set without grids, so that the two-dimensional airfoil running state is simulated and reproduced.
The ability to approximate arbitrary smooth functions using neural networks: it means that the neural network has enough flexibility and capacity, and can be fitted and approximated by any complex smooth function through appropriate parameter adjustment; this is based on the general approximation theorem of neural networks, i.e., neural networks have sufficient expressive power to represent a variety of complex functional relationships;
Smooth functions refer to functions with continuous higher derivatives whose curves appear to have no sharp points or discontinuities in the image; the design and training process of neural networks enables them to learn and represent this type of function; by adjusting the architecture of the network (e.g., the number of layers, the number of neural networks, and the manner of connection) and the training algorithm (e.g., back propagation), the neural networks can learn complex mappings from input to output;
The importance of this property is that neural networks can be applied to many different fields of problem, such as image processing, natural language processing, machine learning, as they have enough flexibility to handle various types of data and complex relationships.
In S4, the training physical information neural network PINN using the test data or the actually measured data sample has the following characteristics:
As shown in fig. 2, four points 1,2, 3 and 4 are distributed on the airfoil section, and the positions of four monitoring sensors are learning supervision points, so as to be used as boundary conditions of simulation. The ratio of the physical equation in the total loss function is increased by adding the parameter factors in the total loss function, so that the output result meets the physical rule more; the physical information neural network PINN achieves two goals by minimizing the total loss function:
1. so that the predicted speed And pressureThe error with the actual monitoring data is small enough;
2. the neural network trained by the physical information neural network PINN substitutes the input and corresponding output values, time t, coordinate x, coordinate y, blade surface gas speed U and blade surface pressure p into a constraint equation, and the result approaches zero, namely, a given physical formula is satisfied, so that a two-dimensional blade section simulation result is obtained.
As shown in fig. 5, S5, replacing the corresponding section in the BEM with the calculated result, where N is the calculated number of sections of the BEM, and represents the calculated number calculated by the physical information neural network PINN method after the n+1th blade; the physical information of the integral load and deformation of the fan blade is recalculated, so that the more accurate integral load of the blade is obtained, and the reference effect is achieved for the subsequent blade design;
aiming at the defect of the phyllanthin momentum theory, on the computer simulation of aerodynamic load of the deformation of the offshore wind power blade, the method can be optimally designed by the following measures:
three-dimensional effect simulation is introduced: by adopting the CFD three-dimensional simulation method based on computational fluid dynamics, the influence of bending and torsion of the blade on pneumatic load can be more accurately considered; the method can capture the non-uniform flow and turbulence effect on the surface of the blade, and improve the simulation accuracy;
Consider the unsteady effect: by adopting a simulation method considering an unsteady effect, the dynamic property of airflow and the pneumatic load change when the blade stalls can be accurately simulated; this can be achieved by using time domain CFD simulation or an unsteady simulation method based on a pneumatic simulation model;
coupling structure-pneumatic simulation: the structural mechanics is coupled with the pneumatic simulation, so that the interaction between the blade and the surrounding environment can be considered more comprehensively; the method can help optimize the design of the blade and improve the structural strength and durability of the blade;
And (3) verifying experimental data: experimental verification of simulation results is an important step for ensuring accuracy; the accuracy of the simulation model can be evaluated and necessary correction and optimization can be performed by performing aerodynamic test in a laboratory or a wind tunnel and comparing experimental data with simulation results;
Parameter optimization and sensitivity analysis: parameter optimization and sensitivity analysis are carried out, so that key parameters affecting the aerodynamic load of the blade can be determined, and design improvement is guided; by using an optimization algorithm and a design space searching technology, an optimal design scheme can be found so as to reduce the aerodynamic load of the blade to the greatest extent;
model refinement and grid optimization: the simulation model is refined and grid optimized, so that the simulation precision and accuracy can be improved; by adopting proper grid division and solver setting, the numerical error can be reduced, and the reliability of the simulation result can be ensured;
By adopting the specific technical measures, the aerodynamic load computer simulation of the deformation of the offshore wind power blade can be optimized, and the accuracy and reliability of the simulation are improved, so that the improvement and optimization of the blade design are guided.
FIG. 4 is a flow chart illustrating a conventional method for calculating airfoil two-dimensional interface aerodynamic loads at a BEM, where N represents the total number of two-dimensional airfoil sections used for simulation;
As shown in fig. 4, for the algorithm flow of BEM calculation, calculation is performed in the divided N two-dimensional sections, all N two-dimensional sections are calculated, and then the result is output;
As shown in fig. 5, of the above-mentioned divided N two-dimensional sections, the two-dimensional sections n+1 to N are calculated, and the result of the two-dimensional sections n+1 to N, N < N, is obtained by adopting the calculation of the physical information neural network PINN;
As shown in fig. 5, after the calculation of N two-dimensional sections of the BEM is completed, 1 to N data in the divided N two-dimensional sections are not replaced, and n+1 to N data in the two-dimensional sections are replaced by the calculation result of the physical information neural network PINN, and the physical information of the overall load and deformation of the fan blade is recalculated, so as to obtain a new overall load of the blade for blade design;
FIG. 5 sets the monitored airfoil section to be the nth airfoil section, N < N; through addition judgment, the nth and previous airfoil sections are calculated by adopting a BEM method, and n+1 and later are calculated by adopting a physical information neural network PINN, which is equivalent to replacing part of BEM results with physical information neural network PINN calculation results;
based on the physical information neural network PINN calculation, the integral load solving of the blade can be completed through a geometric accurate beam theory method.
The traditional blade integral load calculation usually uses a BEM method to calculate pneumatic loads of all two-dimensional airfoil sections, and then the blade integral load solving is completed through a geometric accurate beam theory method;
geometrically accurate beam theory is an accurate description of the geometrical nonlinearity of a beam. The theory introduces cross-section rotation into the deformation space of the beam and measures the deformation of the beam in the corresponding cross-section reference system, and the three-dimensional limited rotation of the unit can be described in conjunction with a rotation quaternion. Because it will not produce the accumulated error along with the iteration, so it is suitable for the large rotation and large displacement analysis of the space rod system structure.
Specifically, in an embodiment of the present application, the BEM translates to the phyllin momentum theory, and the BEM calculates the airfoil lift and drag through the lift drag coefficient, so that the flow field condition of the airfoil section in real operation cannot be simulated. The calculation accuracy of the method is seriously dependent on experimental data of the existing airfoil-shaped resistance-increasing coefficient, and under the actual deformation, such as torsion and section deformation, the resistance-increasing coefficient often deviates from a laboratory result, and the accuracy requirement of the correction neural network under the complex working condition is difficult to meet. And the BEM method is to simplify the flow problem of the three-dimensional blade into the two-dimensional flow problem of the flow around the airfoil. The application utilizes the physical information neural network PINN to simulate the actual flow field condition around the wing section, simulate the two-dimensional flow field distribution under the actual condition, and utilizes the surface pressure to integrate to obtain the rising resistance.
Under the conditions of high-speed motion and larger tip speed ratio, the classical BEM calculation has lower predicted value for normal distribution force and larger predicted value for tangential force; this is because BEM uses two-dimensional results, ignoring the spin enhancement caused by three-dimensional turbulence effects;
FIG. 6 shows a graph of comparison between BEM method and three-dimensional CFD method and experimental data, EXP is an experimental data curve, so in comparison, the data calculated by adopting the three-dimensional CFD method is close to EXP data, the data of the three-dimensional CFD is used as training sample and error comparison data, the distortion is small, and the acquisition cost is lower than EXP;
The application aims to provide a checking method approaching to three-dimensional CFD precision, the speed is extremely faster than that of the three-dimensional CFD method, the efficiency speed is improved by 28 times compared with that of the three-dimensional CFD method, and the advantages are obvious when the blade with a large size is 90m or more; that is, as the blade size increases, the application still maintains the advantages of small error and short calculation time.
The application utilizes a physical information neural network PINN to infer flow field and pressure distribution of the whole calculation area according to partial known data; similar to CFD, the fluid mechanics control equation is solved; the application utilizes the physical information neural network PINN to simulate the two-dimensional flow field distribution, and is different from CFD, the application utilizes the actual data as a supervision point, introduces the actual data, considers the actual flow velocity and pressure distribution after the turbulence of the bypass flow, and can more embody the actual flow field distribution condition; two-dimensional flow field simulation is also shorter and more efficient than the three-dimensional CFD method.
Table 1 below shows the errors between the present application, BEM method and three-dimensional CFD method, showing the relative errors without considering the effect of bypass flow; when the blade is changed into an ultra-long flexible blade, for example, the length is about 140m, the calculation relative error of the wing section lifting resistance at the same position is larger, and the error is further amplified after the integral along the whole blade; compared with the three-dimensional CFD method, the BEM method is two-dimensional simulation calculation, and has short calculation time and large error;
Table 1: the error between the BEM method and the three-dimensional CFD method;
table 2: the BEM method and the three-dimensional CFD method require time, and check and calculate the same operation hardware grade platform and the same blade;
According to the application, the BEM is used for realizing rapid checking calculation, N two-dimensional sections are calculated, and the cross section calculation results from n+1 to N in the N two-dimensional sections are replaced by the method, so that a new integral load of the blade is obtained, wherein N is less than N;
As can be seen from Table 2, the application greatly shortens the operation time, the precision approximates to three-dimensional CFD, the advantages are more obvious in large-size blade calculation, the error is small, the calculation time has great advantages, in the checking design, the checking result of the blade can be quickly responded and acquired to feed back the design, and a quick iteration design scheme is realized;
as shown in FIG. 6, EXP is the data obtained by actual measurement of the test, so that the three-dimensional CFD can be used as the real data for comparison in comparison, so that the training cost of the application is reduced, and the training efficiency can be improved;
however, the three-dimensional CFD has the defects of high cost and low efficiency, and cannot be matched with design verification in real time and high efficiency, so that data obtained by actual measurement are supplemented in training of the physical information neural network PINN, and the real situation is more approximated; meanwhile, by combining the test data and the measured data of the three-dimensional CFD, the method has the following advantages in blade checking design calculation:
1. The physical information neural network PINN calculation method realized by the application has the following advantages compared with two-dimensional BEM:
the BEM corrects the blade tip loss and wake flow model according to an empirical formula, and is not suitable for wind power blades with large diameters and large deformations in the future; according to the application, a two-dimensional section is selected at the key part of the blade tip for simulation, and the BEM method is optimized, so that the problem of inadaptation in correction is effectively avoided;
Based on the existing 100-meter blade data, a proper amount of blade surface sensors are added, so that airfoil surface supervision points are added for learning, a schematic diagram is shown in fig. 7, and more sensors are arranged;
For longer length blades, introducing turbulence models into the physical terms based on the physical information neural network PINN will result in more accurate results of the present application.
2. Comparing the physical information neural network PINN calculation method with the three-dimensional CFD method result, and simulating the three-dimensional CFD method through the integral blade model to obtain the most accurate result; however, more time is required for completing the simulation, and the requirement on the computational power of the computer is higher;
For example:
Compared with the traditional computational fluid dynamics CFD: the traditional CFD method solves a hydrodynamic control equation through a computer and a numerical method, but only simulation can be performed, and accuracy cannot be improved based on actual data; according to the application, the measured data is used as a machine learning supervision point, and a physical formula is added at the same time, which is equivalent to adding the measured data on the CFD method; and solving a hydrodynamic control equation by using the neural network.
Specifically, in an embodiment of the present application, a wind power generator is obtained by adopting the pneumatic load checking method for deformation of an offshore wind power blade, and the wind power blade is used for equipping the wind power generator.
The foregoing is merely a preferred embodiment of the present invention, and it should be noted that modifications and substitutions can be made by those skilled in the art without departing from the technical principles of the present invention, and these modifications and substitutions should also be considered as being within the scope of the present invention.
Claims (10)
1. The pneumatic load checking method for the deformation of the offshore wind power blade is characterized by comprising the following steps of:
s1: the selected airfoil interface is used for constructing a feedforward neural network for reconstructing the fluid state of the airfoil surface of the fan according to the independent variable and the dependent variable number of the two-dimensional incompressible fluid equation set;
s2: defining a total loss function comprising data items and physical items for quantifying differences between the neural network predicted values and the actual values;
The source of the data item is fan field monitoring data, and the source of the physical item is Navier-Stokes equation set and continuity equation added in; the data item is used as a supervision point of neural network training and is used for improving simulation precision;
s3: constructing a physical information neural network PINN, and fitting and approximating a smooth function to the physical information neural network PINN;
executing a grid-free solving Navier-Stokes equation set, and carrying out simulation reproduction on the running state of the two-dimensional airfoil;
Navier-Stokes equation set eq.1, eq.2 and continuity equation eq.3 are as follows:
The formula: eq.1
The formula: eq.2
The formula: eq.3
The formula: eq.1, eq.2, eq.3 are constraint equations;
the total loss function is expressed as the formula: eq.4
Wherein: is the total loss function of the physical information neural network PINN,/> Is a data item loss function,/>Is a physical law term loss function; the loss function of the physical rule term is obtained by the partial differential equation set loss function term/>, in computational fluid dynamicsBoundary condition loss function term/>And initial conditional loss function term/>Composition,/>、/>For adjusting/>, as weight coefficient、/>In the total loss function/>The ratio of (3);
the formula: eq.5
The formula: eq.6
The formula: eq.7
Wherein: To calculate the total number of data points of the partial differential equation loss function of the fluid,/> To calculate the number of data points of the fluid boundary condition loss function,/>Calculating a number of data points for an initial condition loss function of the fluid system; g is an implicit expression of a Navier-Stokes equation; /(I)、/>The values of the measurement time t, the coordinates x and y, the blade surface gas speed U and the blade surface pressure p of the ith sensor are obtained by measurement; the blade surface gas velocity U includes a horizontal component velocity U and a vertical component velocity v; /(I)、/>To predict the values of the measurement time t, the coordinates x, the coordinates y, the blade surface gas velocity U, the blade surface pressure p of the obtained ith sensor,/>、/>The method is obtained through simulation calculation;
The total loss function expansion is expressed as the formula: eq.8
、/>、/>、/>For adjusting/>, as weight coefficient、/>、/>、/>In the total loss function/>The ratio of (3);
s4, using test data or actual measurement data samples, wherein the output result in S3 meets the physical rule, and training the physical information neural network PINN is completed; physical information neural network PINN allows for predicted speed by minimizing the total loss function And pressure/>After training the physical information neural network PINN, the input and corresponding output values, the measurement time t, the coordinate x, the coordinate y, the blade surface gas speed U and the blade surface pressure p are substituted into constraint equations eq.1, eq.2 and eq.3, and the result approaches zero, namely the given physical formula is satisfied;
S5: the two-dimensional blade section simulation calculation result is obtained through the trained physical information neural network PINN; the corresponding sections in the BEM are replaced, N is the total number of sections calculated by the two-dimensional blades used by the BEM, the 1 st to N two-dimensional blade sections are calculated by the BEM, and N < N, n+1 to N two-dimensional blade sections adopt a physical information neural network PINN, and the physical information of the whole load and deformation of the fan blade is recalculated to obtain the new whole load of the blade for blade design.
2. The aerodynamic load checking method for offshore wind turbine blade deformation according to claim 1, wherein values of measurement time t, coordinates x, coordinates y, blade surface gas velocity U and blade surface pressure p obtained by measurement of four monitoring sensors distributed on an airfoil section are used as boundary conditions simulated in S3 for simulation reproduction of a two-dimensional airfoil running state;
And S4, taking four monitoring sensor positions distributed on the section of the airfoil profile as learning supervision points.
3. An aerodynamic load checking method for offshore wind blade deformation according to claim 2, wherein the weight coefficient in the total loss function、/>In adjusting/>、/>After the duty ratio in the total loss function, the output result meets the physical rule.
4. The method for checking aerodynamic load of marine wind blade deformation according to claim 1, wherein the construction of the feedforward neural network in S1 comprises the following steps of S101: and constructing a feedforward neural network without physical information, and setting a hidden layer, an input variable, an output variable and an activation function.
5. The method for checking aerodynamic load of blade deformation of offshore wind turbine of claim 4, wherein the construction of the feedforward neural network in S1 comprises the following steps of S102: the coordinates x, y and time t are taken as inputs, the horizontal component speed u, the vertical component speed v and the pressure p under the two-dimensional plane coordinates of the wing profile are taken as outputs, and the tanh is taken as an activation function of the feedforward neural network.
6. The aerodynamic load checking method for offshore wind blade deformation according to claim 2, wherein in S2: the number of the supervision points is directly proportional to the simulation precision; the same airfoil section is provided with 4 supervision points.
7. The method for checking aerodynamic load of marine wind blade deformation according to claim 1, wherein the total loss function comprises physical information constraint, so that a neural network data layer is learned and trained on a physical rule layer.
8. The aerodynamic load checking method for offshore wind power blade deformation according to claim 1, wherein the wind power field where the blade of the airfoil interface selected in S1 is located acquires actual offshore wind power field monitoring data including global wind speed and wind shear parameters and airfoil surface specific position monitoring data under a certain distance position of the front end of the blade.
9. The method for checking aerodynamic load of marine wind power blade deformation according to claim 8, wherein the simulated section aerodynamic load is obtained by a BEM method by setting a simulated working condition; the simulation conditions include wind speed.
10. The aerodynamic load checking method for offshore wind blade deformation of claim 1, wherein the weight coefficient in the total loss function、/>、/>、/>In adjusting/>、/>、/>、/>After the duty ratio in the total loss function, the output result meets the physical rule; in S4, the physical information neural network PINN is trained using the test data and the measured data samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410389837.3A CN117993303B (en) | 2024-04-02 | 2024-04-02 | Pneumatic load checking method for deformation of offshore wind power blade |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410389837.3A CN117993303B (en) | 2024-04-02 | 2024-04-02 | Pneumatic load checking method for deformation of offshore wind power blade |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117993303A CN117993303A (en) | 2024-05-07 |
| CN117993303B true CN117993303B (en) | 2024-06-11 |
Family
ID=90892676
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410389837.3A Active CN117993303B (en) | 2024-04-02 | 2024-04-02 | Pneumatic load checking method for deformation of offshore wind power blade |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117993303B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104612892A (en) * | 2014-12-30 | 2015-05-13 | 中国科学院工程热物理研究所 | Multi-disciplinary optimization design method for airfoil profile of wind turbine |
| WO2020176841A1 (en) * | 2019-02-28 | 2020-09-03 | Georgia Tech Research Corporation | Systems and methods for predicting the geometry and internal structure of turbine blades |
| JP2022168865A (en) * | 2021-04-26 | 2022-11-08 | 三菱重工業株式会社 | Method for diagnosing wind turbine blade |
| CN115544883A (en) * | 2022-10-08 | 2022-12-30 | 浙江大学 | Online measurement method and system for load and platform deformation of floating type offshore wind turbine generator |
| CN115563879A (en) * | 2022-10-25 | 2023-01-03 | 华北电力大学 | Method and system for delay modeling of airfoil shape stall of wind turbine blade |
| CN115577625A (en) * | 2022-09-29 | 2023-01-06 | 中国航空研究院 | Method and device for predicting ultimate aerodynamic load of large flexible wind power blade under turbulent wind condition |
| CN116432556A (en) * | 2023-04-21 | 2023-07-14 | 中国航空工业集团公司沈阳空气动力研究所 | Wing surface pressure reconstruction method, electronic equipment and storage medium |
| CN117108445A (en) * | 2023-07-25 | 2023-11-24 | 华北电力大学 | Digital twin simulation method for tandem double-wind-wheel wind turbine generator |
-
2024
- 2024-04-02 CN CN202410389837.3A patent/CN117993303B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104612892A (en) * | 2014-12-30 | 2015-05-13 | 中国科学院工程热物理研究所 | Multi-disciplinary optimization design method for airfoil profile of wind turbine |
| WO2020176841A1 (en) * | 2019-02-28 | 2020-09-03 | Georgia Tech Research Corporation | Systems and methods for predicting the geometry and internal structure of turbine blades |
| JP2022168865A (en) * | 2021-04-26 | 2022-11-08 | 三菱重工業株式会社 | Method for diagnosing wind turbine blade |
| CN115577625A (en) * | 2022-09-29 | 2023-01-06 | 中国航空研究院 | Method and device for predicting ultimate aerodynamic load of large flexible wind power blade under turbulent wind condition |
| CN115544883A (en) * | 2022-10-08 | 2022-12-30 | 浙江大学 | Online measurement method and system for load and platform deformation of floating type offshore wind turbine generator |
| CN115563879A (en) * | 2022-10-25 | 2023-01-03 | 华北电力大学 | Method and system for delay modeling of airfoil shape stall of wind turbine blade |
| CN116432556A (en) * | 2023-04-21 | 2023-07-14 | 中国航空工业集团公司沈阳空气动力研究所 | Wing surface pressure reconstruction method, electronic equipment and storage medium |
| CN117108445A (en) * | 2023-07-25 | 2023-11-24 | 华北电力大学 | Digital twin simulation method for tandem double-wind-wheel wind turbine generator |
Non-Patent Citations (5)
| Title |
|---|
| A FRAMEWORK OF DATA ASSIMILATION FOR WIND FLOW FIELDS BY PHYSICS-INFORMED NEURAL NETWORKS;Chang Yan 等;arXive;20240130;全文 * |
| Numerical modeling and machine learning for wind turbine aerodynamics and condition monitoring;Purohit, Shantanu;https://hdl.handle.net/10356/159299;20221231;全文 * |
| 大型风力机翼型气动性能优化研究;徐浩然;中国博士学位论文全文数据库 工程科技Ⅱ辑;20160515;全文 * |
| 海上漂浮式风机气动阻尼效应全耦合数值分;赵仕伦 等;中国海上油气;20231231;第35卷(第6期);全文 * |
| 风力机三维旋转叶片非定常气动特性数值模拟研究;胡国玉;孙文磊;曹莉;;可再生能源;20160620(第06期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117993303A (en) | 2024-05-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shittu et al. | Comparative study between SN and fracture mechanics approach on reliability assessment of offshore wind turbine jacket foundations | |
| Cappugi et al. | Machine learning-enabled prediction of wind turbine energy yield losses due to general blade leading edge erosion | |
| CN103983453A (en) | Differentiating method of fault diagnosis of executing mechanism of aeroengine and sensor | |
| Liu et al. | Explicit polynomial regression models of wind characteristics and structural effects on a long‐span bridge utilizing onsite monitoring data | |
| CN113642068A (en) | Bridge vortex-induced vibration amplitude prediction method based on decision tree and recurrent neural network | |
| Wang et al. | Damage identification of offshore jacket platforms in a digital twin framework considering optimal sensor placement | |
| CN113221237A (en) | Large attack angle flutter analysis method based on reduced order modeling | |
| CN110162895A (en) | A kind of two stage high energy efficiency ship form optimization design method | |
| CN115293069B (en) | System for intelligent optimization of simulation control parameters of aircraft external flow field | |
| Pena et al. | A surrogate method based on the enhancement of low fidelity computational fluid dynamics approximations by artificial neural networks | |
| CN117993303B (en) | Pneumatic load checking method for deformation of offshore wind power blade | |
| Sainmont et al. | Boundary layer behaviour on a morphing airfoil: Simulation and wind tunnel tests | |
| CN115481461A (en) | Structural wind pressure statistic prediction method based on deep learning | |
| Venkatachari et al. | Assessment of transition modeling capability in OVERFLOW with emphasis on swept-wing configurations | |
| Díaz-Casás et al. | Wind turbine design through evolutionary algorithms based on surrogate CFD methods | |
| Pahlke | The GOAHEAD project | |
| Marten et al. | QBlade documentation | |
| Shakya et al. | Fluid-structure interaction and life prediction of small-scale damaged horizontal axis wind turbine blades | |
| JP4015004B2 (en) | Helicopter airfoil simulation method | |
| CN107784143B (en) | Method and device for determining aeroelastic stability of wind generating set blade | |
| Chen et al. | Sparse pressure-based machine learning approach for aerodynamic loads estimation during gust encounters | |
| Sun et al. | Research on balancing method of single-stage rotor blade mass moment based on pointer network | |
| Xu et al. | Deep learning enhanced fluid-structure interaction analysis for composite tidal turbine blades | |
| Huang et al. | Aeroelastic analysis of wind turbine under diverse inflow conditions | |
| Venkatachari et al. | Assessment and Improvement of RANS-based Transition Models based on Experimental Data of the Common Research Model with Natural Laminar Flow |
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 |