CN110207871B - Method, device, storage medium and system for stress prediction of wind turbine generator - Google Patents
Method, device, storage medium and system for stress prediction of wind turbine generator Download PDFInfo
- Publication number
- CN110207871B CN110207871B CN201810167169.4A CN201810167169A CN110207871B CN 110207871 B CN110207871 B CN 110207871B CN 201810167169 A CN201810167169 A CN 201810167169A CN 110207871 B CN110207871 B CN 110207871B
- Authority
- CN
- China
- Prior art keywords
- stress
- current
- neural network
- current load
- network model
- 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
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01L—MEASURING FORCE, STRESS, TORQUE, WORK, MECHANICAL POWER, MECHANICAL EFFICIENCY, OR FLUID PRESSURE
- G01L5/00—Apparatus for, or methods of, measuring force, work, mechanical power, or torque, specially adapted for specific purposes
- G01L5/0047—Apparatus for, or methods of, measuring force, work, mechanical power, or torque, specially adapted for specific purposes measuring forces due to residual stresses
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Wind Motors (AREA)
Abstract
The invention discloses a method, a device, a storage medium and a system for stress prediction of a wind turbine generator. Wherein, the method comprises the following steps: collecting a plurality of current load components of the unit components of the wind turbine generator, which are coupled with each other under the current load working condition; taking a plurality of current load components which are coupled with each other under the current load working condition as input, and predicting the current stress extreme position of the unit component by using a position neural network model; selecting an extremum neural network model corresponding to the current stress extremum position according to the current stress extremum position; and taking a plurality of current load components coupled with each other under the current load working condition as input, and predicting the current stress extreme value of the current stress extreme value position by using the extreme value neural network model. According to the embodiment of the invention, the stress prediction precision of the wind turbine generator can be improved.
Description
Technical Field
The invention relates to the technical field of wind power generation, in particular to a method, a device, a computer readable and writable storage medium and a system for stress prediction of a wind turbine generator.
Background
In recent years, with the growing problem of energy, wind energy has been attracting attention as one of renewable resources, and wind power generation has been favored as the most efficient means of utilizing wind energy. In order to respond to market demands, the design of the wind energy utilization rate is high, and the environment-friendly fan becomes an industrial hotspot problem. However, as the demand of the fan increases, the performance reliability and the structural safety of the fan face serious tests. Therefore, in order to improve the reliability of the performance and the structural safety of the wind turbine generator system, the structural strength of the wind turbine generator system needs to be evaluated adaptively.
The existing structure adaptability evaluation method mainly adopts a linear interpolation method, and evaluates the structure adaptability according to a stress interpolation result. During evaluation, according to a specific load working condition, a plurality of load components influencing a stress calculation result need to be analyzed, one load component is selected as a dominant load component, and therefore a stress predicted value is extrapolated according to the dominant load component to carry out structural adaptability evaluation. However, this approach ignores the coupling effect between the different load components, and the stress is affected not only by the dominant load component but also by other load components. Therefore, there is a large error in the result of stress prediction based on a single load component.
How to accurately predict the stress of the wind generating set becomes a technical problem to be solved urgently.
Disclosure of Invention
In order to solve the problem of large error in stress prediction of a wind turbine generator system, embodiments of the present invention provide a method, an apparatus, a system, and a computer-readable storage medium for stress prediction of a wind turbine generator system.
In a first aspect, a method for stress prediction of a wind turbine is provided. The method comprises the following steps:
collecting a plurality of current load components of the unit components of the wind turbine generator, which are coupled with each other under the current load working condition;
taking a plurality of current load components which are coupled with each other under the current load working condition as input, and predicting the current stress extreme position of the unit component by using a position neural network model;
selecting an extremum neural network model corresponding to the current stress extremum position according to the current stress extremum position;
and taking a plurality of current load components coupled with each other under the current load working condition as input, and predicting the current stress extreme value of the current stress extreme value position by using the extreme value neural network model.
In a second aspect, an apparatus for stress prediction of a wind turbine is provided. The device includes:
the component acquisition module is used for acquiring a plurality of current load components which are mutually coupled under the current load working condition of the unit components of the wind turbine generator;
the position prediction module is used for taking a plurality of current load components which are coupled with each other under the current load working condition as input and predicting the current stress extreme position of the unit component by using the position neural network model;
the model selection module is used for selecting an extremum neural network model corresponding to the current stress extremum position according to the current stress extremum position;
and the extreme value prediction module is used for taking a plurality of current load components which are mutually coupled under the current load working condition as input and predicting the current stress extreme value of the current stress extreme value position by using the extreme value neural network model.
In a third aspect, an apparatus for stress prediction of a wind turbine is provided. The device includes:
a memory for storing a computer program;
a processor for executing the computer program stored by the memory, the program causing the processor to perform the method of the aspects described above.
In a fourth aspect, a computer-readable storage medium is provided. The computer-readable storage medium has stored therein a computer program which, when run on a computer, causes the computer to perform the method of the above aspects.
In a fifth aspect, a system for stress prediction for a wind turbine is provided. The system comprises the stress prediction device of the wind turbine generator.
On one hand, the embodiment of the invention can enable the relation between the input load variables and the output stress to be not only in accordance with the linear relation of a single load component any more by acquiring the current load components of the wind turbine generator which are mutually coupled under the current load working condition, but also can construct the corresponding relation in accordance with the complex nonlinearity between the load components and the stress through the neural network, fully considers the coupling among all the components and the influence of all the components on the stress calculation, and can greatly improve the precision of stress prediction.
On the other hand, the embodiment of the invention can not only accurately predict the stress extreme value, but also provide the position of the stress extreme value of the unit component of the wind turbine unit, thereby providing better and richer data support for the structural adaptability evaluation.
In another aspect, the extreme neural network model corresponding to the current stress extreme position can be selected according to the current stress extreme position, so that different positions can correspond to different prediction methods, and the precision of stress prediction can be further improved.
On the other hand, in the embodiment of the present invention, the corrected parameter (for example, the difference between the predicted data and the calculated data) may be obtained through data calculation predicted by the neural network model, and the corrected neural network model is continuously trained, so that the neural network model may continuously learn, and as the number of times of the cyclic training increases and the number of training samples increases, the accuracy of the prediction result continuously becomes higher.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments of the present invention will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a schematic flow diagram of a method for stress prediction of a wind turbine generator according to an embodiment of the invention;
FIG. 2 is a stress cloud diagram of a hub of the wind turbine generator of FIG. 1;
FIG. 3 is a schematic view of a bolted flange connection of a wind turbine generator according to an embodiment of the invention;
FIG. 4 is a schematic diagram of a model of a location neural network according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of an extremum neural network model according to an embodiment of the present invention;
FIG. 6 is a graph showing a variation relationship of stress extreme values under the action of a single load component;
FIG. 7 is a flow chart illustrating a method of stress prediction for a wind turbine generator according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of a device for predicting stress of a wind turbine generator according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The stress-predicted wind turbine component may be one or more parts or components of a wind turbine, such as: a bolted flange connection, a hub, etc., without limitation.
The following embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.
Fig. 1 is a schematic flow chart of a method for predicting stress of a wind turbine generator according to an embodiment of the present invention.
As shown in fig. 1, the method comprises the steps of: s110, collecting a plurality of current load components which are coupled with each other in the load working condition of a unit component of the wind turbine; s120, predicting the current stress extreme position of the stress extreme value of the unit component by using the position neural network model according to the current load components; and S130, predicting the current stress extreme value of the current stress extreme value position by using the extreme value neural network model according to the current stress extreme value position and the plurality of current load components.
In the present exemplary embodiment, the wind turbine component for the stress to be predicted is a hub.
FIG. 2 is a stress cloud diagram of a hub of the wind turbine of FIG. 1.
As shown in fig. 2, the geometric model of the hub is complex. The geometric model of the hub has a plurality of hot spot stress areas such as through holes, rounded corners and the like, such as local positions in a circle in fig. 1. These positions may represent the positions of different stress extremes for different load conditions.
In the present embodiment, three blade root loading points (not labeled in the figure), a plurality of load conditions and a plurality of load components under the current load condition of the hub can be considered when predicting the stress intensity of the hub. Referring to the stress cloud diagram of fig. 2, when the load condition changes, the position of the stress extreme value will change, and the stress extreme value at the position of the stress extreme value will also change. If the corresponding predicted stress is derived based on only a single load component, it will result in a discontinuous relationship between the input load (e.g., Mx) and the output stress (e.g., Fx). If one were to force the discontinuity problem to be reduced to a linear one, it could lead to a large error in the results of predicting the stress. Therefore, in step S110, in order to reduce an error of predicting stress and improve the accuracy of stress prediction, the embodiment of the present invention predicts the stress of the wind turbine by a method of collecting a plurality of current load components of the wind turbine components coupled to each other under the current load condition.
In step S110, the plurality of current load components may include: the current bending moment of each direction and the current stress value of each direction in the multidimensional space. For example, the load component may be a bending moment in an x-axis direction (e.g., Mx), a bending moment in a y-axis direction (e.g., My), a bending moment in a z-axis direction (e.g., Mz), a force in an x-axis direction (e.g., Fx), a force in a y-axis direction (e.g., Fy), a force in a z-axis direction (e.g., Fz), and the like, and conventionally, the 6-dimensional load component and the structural stress have a nonlinear complex relationship, and a coupling relationship exists between these components, which all have different degrees of influence on the structural stress calculation.
The load components under specific different load conditions can be shown in the following table (1):
watch (1)
In step S120, this step may include the following substeps:
s121, carrying out nonlinear transformation on the extreme value neural network model based on a plurality of input current load components (such as Mx, My, Mz, Fx, Fy and Fz) which are coupled with one another;
s122, performing linear interpolation calculation on the plurality of current load components coupled to each other through the nonlinear transformation to obtain a current stress extreme position (e.g., a local position in a circle in fig. 1).
In step S130, this step may include the following sub-steps:
s131, the position neural network model carries out nonlinear transformation on the basis of a plurality of input current load components (such as Mx, My, Mz, Fx, Fy and Fz) which are coupled with one another;
the nonlinear transformation is carried out on the load component to convert the 6-dimensional load component into the 12-dimensional load component, so that the nonlinear complex relation between the load component and the stress is simplified, and the subsequent linear interpolation calculation is facilitated.
S132, performing linear interpolation calculation on the plurality of current load components coupled to each other through the nonlinear transformation to obtain a current stress extreme value at the current stress extreme value position (e.g., a stress extreme value at a certain circle position in fig. 1).
In steps S120 and S130, the position neural network model or the extremum neural network model may each include: radial Basis Function (RBF) neural network models. It will be appreciated by those skilled in the art that the position neural network model or the extremum neural network model may be implemented using other types of neural network models in addition to the radial basis function neural network model, such as: a recurrent neural network or a BP neural network, etc., which are not limited in any way by the present invention.
In some embodiments, the location neural network model or the extremum neural network model may each include: the device comprises a data input layer, a data hiding layer and a data output layer. At the data input layer, a non-linear transformation may be performed based on the input mutually coupled plurality of current load components. For example, the input mutually coupled plurality of current load components may be nonlinearly transformed based on a Guass function or a Sigmoidal function as a basis function. In the data hiding layer, carrying out linear interpolation calculation on a plurality of current load components which are subjected to nonlinear transformation and mutually coupled to obtain a calculation result; and outputting the calculation result in the data output layer.
Fig. 3 is a schematic view of a bolt flange connection mechanism of a wind turbine generator according to an embodiment of the present invention.
As shown in fig. 3, the wind turbine component to be subjected to stress prediction according to the present embodiment may be a bolt flange connection mechanism. The flange may have 20 holes numbered 1-20, which may be used as locations for mounting bolts, such as locations 1-20. The current stress extreme positions of the bolt flange connecting mechanism, such as the positions 1-20, can be predicted by using a position neural network model by taking a plurality of current load components coupled with each other under the current load working condition as input.
FIG. 4 is a schematic diagram of a model of a neural network for a location, according to an embodiment of the present invention.
As shown in FIG. 4, the positional neural network model 100 may be used to predict the positions of stress extrema in FIG. 3, such as position 1-position 20. The location neural network model 100 may include: a data input layer 101, a data hiding layer 102 and a data output layer 103.
At the data input layer 101, the collected data (e.g. 6 load components in the current load condition) may be input, and then the collected data is subjected to a nonlinear transformation.
Specifically, at the data input layer 101, the collected neural network training samples may be input. These samples may be measured data or finite element data. In this embodiment, finite element data may be used as a training sample, and multiple sets of finite element data may be input. For example, the input data in each training sample set is 6 load components (such as Mx, My, Mz, Fx, Fy and Fz), and the output data is the axial force of the bolt. The load components of the first 10 sets of inputs may be as shown in table (2) below:
watch (2)
At the data input layer 101, the input data may be subjected to normalization processing based on the following formula 1.
In the above formula 1, X may be input sample data, XSMay be normalized sample data, XmeanMay be the mean of the sample data, XstdMay be the sample data variance.
During data normalization, the bending moment load, the force load in the input data and the axial force in the output data can be grouped for normalization, and the mean value, the variance parameter and the like of each group of data are reserved. Normalization processing can be used to convert the prediction data set to ensure the correspondence between training data and test data, and thus, the reliability of data processing can be improved. The normalized input load component can be shown in table (3) below:
watch (3)
| Load condition | Mx | My | Mz | Fx | Fy | Fz |
| 1 | -0.04 | -1.00 | 1.19 | -0.16 | -0.95 | 1.35 |
| 2 | 1.15 | -0.33 | 0.25 | -1.57 | 0.62 | 0.07 |
| 3 | 0.02 | -1.37 | 0.72 | -0.35 | -0.28 | 0.71 |
| 4 | -0.32 | 0.03 | -0.80 | 0.11 | 1.13 | -0.56 |
| 5 | 0.63 | -1.72 | 0.33 | 1.22 | -0.16 | -0.98 |
| 6 | 0.86 | -0.63 | 1.48 | 1.32 | -0.38 | 1.37 |
| 7 | -1.32 | 0.83 | 1.53 | 0.49 | -1.10 | -0.92 |
| 8 | 0.93 | 0.90 | -0.07 | -0.68 | -1.26 | 0.75 |
| 9 | -0.21 | -0.74 | 1.31 | -0.96 | 1.07 | 0.56 |
| 10 | -1.09 | -0.25 | -0.97 | -1.34 | -1.41 | 0.26 |
In the data hiding layer 102, the center X of the basis function can be determined based on the K-Means algorithmjAnd then carrying out linear difference calculation on the plurality of current load components which are subjected to nonlinear transformation and mutually coupled to obtain the position of the stress extreme value. In order to continuously improve the accuracy of the model predicted stress, the corrected stress extreme position (such as the difference value between the predicted stress extreme position and the calculated stress extreme position) can be obtained through the predicted stress extreme position calculation, and the corrected stress extreme position is returned to the basis function center again to carry out multiple cycles, so that the supervised feedback learning is carried out, and the parameters are adaptive. It will be appreciated by those skilled in the art that other clustering algorithms may be used to determine the basis function center X in addition to the K-Means algorithmjFor example: SOM clustering or FCM clustering, etc., which are not intended to limit the scope of the present invention.
The number p of the centers of the basis functions is 6(N +1), and N is the number of the load points (e.g., 3). When all the sample data are at the minimum Euclidean distance from the center of the sample, namely the following formula 2 is satisfied, the calculation is completed.
In the above formula 2, XjMay be the center of the basis function, P may represent the number of the centers of the basis function, and equation 2 may represent the minimum euclidean distance between all sample data and the center of the sample.
The basis function may be a gaussian function, as shown in the following equation 3:
in the above equation 3, σ may be a base function bandwidth, and an initial value of σ may take 1.
Input data XiAnd the output prediction data F (X)i) Can be expressed as the following equation 4:
in the formula 4, wjIt can be a linear interpolation weight, and the initial value is 1. XiMay be input data, F (X)i) May be the output prediction data.
When the neural network is trained, 48 groups of load working conditions and 20 groups of bolt axial forces corresponding to the load working conditions can be read, all data are grouped according to the maximum bolt number of the axial force, and then the RBF neural network is trained by predicting the position of the stress extreme value. The training method can be a gradient descent method, and an objective function e is trainediCan be as shown in equation 5 below:
in the above formula 5, XiMay be the i-th set of payload data, yiMay be payload data XiCorresponding axial force maximum bolt number, F (X)i) The predicted maximum axial force bolt may be numbered.
To minimize the objective function, the correction amount of each parameter should be proportional to its negative gradient, as shown in the following equations 6 and 7:
in the above equations 6 and 7, E may be a sample lumped error, δ may be a width of a center of a basis function, and η may be a learning rate. During training, a variable learning rate algorithm is adopted, after an initial learning rate eta is given, eta is multiplied by a random number rho during each training, and rho is more than or equal to 0.1 and less than or equal to 1. The neural network training may end when the training error is less than 1e-4 or the maximum number of iterations is reached.
At data output layer 103, the locations of stress extremes, such as location 1-location 20 in FIG. 3, may be output. Different locations may correspond to different models of the extremal neural network.
FIG. 5 is a schematic diagram of an extremum neural network model according to an embodiment of the invention.
After the stress extremum position predicting RBF neural network (position neural network model 100) is constructed, it is necessary to continue to construct the stress extremum predicting neural network corresponding to the stress extremum position predicting RBF neural network, that is, each stress extremum position has one stress extremum predicting RBF neural network corresponding to the stress extremum position, for example, 20 bolt positions (position 1-position 20) in fig. 3 may correspond to 20 extremum neural network models 200.
As shown in FIG. 5, the extreme neural network model 200 may be used to predict stress extremes for the stress extreme positions in FIG. 3. Such as 20 bolt positions in fig. 3, there are 20 extremum neural network models 200. The extremum neural network model 200 can include: a data input layer 201, a data hiding layer 202 and a data output layer 203.
At the data input layer 201, the collected data (e.g. 6 load components in the current load condition) may be input, and then the collected data is subjected to a nonlinear transformation. In the data input layer 201, normalization processing may be performed on input data.
In the data hiding layer 202, the center X of the basis function can be determined based on the K-Means algorithmjAnd then carrying out linear difference calculation on the plurality of current load components which are coupled with each other and subjected to nonlinear transformation to obtain a stress extreme value. In order to continuously improve the accuracy of the model predicted stress, a corrected stress extreme value (such as a predicted stress extreme value and a calculated stress extreme value) can be obtained by calculating the position of the predicted stress extreme valueThe difference of the force extreme values) and returning the force extreme values to the center of the basis function again to perform multiple cycles, thereby performing supervised feedback learning and enabling the parameters to be self-adaptive.
At the data output layer 203, a stress extremum at the stress extremum location may be output.
Specifically, a stress extremum predicting RBF neural network may be constructed for a bolted flange connection structure including 20 bolts. For example, after the stress extreme position prediction neural network is constructed, a stress prediction neural network corresponding to each stress extreme position is constructed for each stress extreme position. At this time, the input data in the training sample is normalized 6 load components, the output data is normalized maximum bolt axial force, and the input and output data of the stress extremum neural network can be shown in the following table (4):
watch (4)
| Load condition | Mx | My | Mz | Fx | Fy | Fz | Fs |
| 1 | -1.38 | 0.10 | -0.22 | -0.55 | 1.43 | 1.70 | 2.12 |
| 2 | 0.21 | 0.39 | -0.37 | 1.70 | -1.69 | -1.02 | -0.45 |
| 3 | 0.40 | -0.09 | 1.11 | 0.20 | 0.08 | 0.38 | 0.48 |
| 4 | -0.92 | -1.18 | -1.28 | -0.78 | -0.31 | 1.50 | 1.86 |
| 5 | -0.40 | 1.40 | 0.90 | -1.69 | -0.13 | 1.19 | 2.03 |
The construction and training method of the extremum prediction neural network is the same as the construction and training method of the stress extremum position prediction neural network, and the content of the part is not repeated. And when the two parts of neural networks are completely constructed, constructing the whole stress prediction platform. After the stress prediction platform is built, the platform can be used for carrying out data test on the accuracy of the stress prediction value so as to verify the accuracy of the predicted stress.
After 6 load components of the load working condition are input (the number of the components can be flexibly changed according to the number of actual loading points), firstly, a stress extreme value position prediction RBF neural network is called to predict the stress extreme value position, and the predicted position is output. Then, the stress extremum prediction RBF neural network corresponding to the stress extremum prediction RBF neural network can be called to output a predicted stress extremum, and the prediction result is shown in table (5):
watch (5)
As can be seen from table (5) above, when 6 load components (e.g., Mx, My, Mz, Fx, Fy, and Fz) are input to the position neural network model 100, the result of prediction of the stress extreme value position (axial force maximum bolt number) is 15, which is the same as the result of calculation 15, and the error is 0. The predicted result 570.76 of the stress extremum for the predicted stress extremum location for the positional neural network model 200 is a 570.02, with an error of only 0.75, which is small.
Therefore, the wind turbine generator system has high stress prediction accuracy, and the prediction accuracy is gradually increased along with supervised feedback learning and parameter self-adaption.
Besides the above-mentioned testing precision by means of stress calculation, the following curve fitting method can be used to test the stress testing precision of the wind turbine generator.
FIG. 6 is a graph showing the variation of stress extreme under the action of a single load component.
In the present exemplary embodiment, the wind turbine installation component to be predicted can be the bolted flange connection in fig. 3.
As shown in fig. 6, the curve is a graph showing the variation of the stress extreme value of the bolted flange connection mechanism under the action of a single load component. In this curve, the abscissa may represent the value of a single load component (dominant load) Mx in KNm. The ordinate may represent the bolt axial force in KN. The curve may represent the variation of the bolt axial force with the external load dominant load component Mx. The accuracy of the curve can be verified with a number of small black blocks (distributed on both sides of the curve of fig. 6) actually measured. And each small black square represents the bolt axial force under the action of different abscissa dominant loads. As can be seen from fig. 6, the fitting degree of the small black blocks and the curve is not very high, and a certain error exists. Specifically, the bolt axial force tends to increase with the increase of Mx, but certain fluctuation still exists relative to the overall trend. From this, it can be seen that the bolt axial force is affected not only by the Mx load component but also by the coupling action of a plurality of other load components. Therefore, the prediction method for simplifying a plurality of load components into a single dominant load component has certain errors from the actual situation.
Thus, the relationship between the different load components and the output stress is discontinuous and cannot be reduced to a linear problem. Reducing the non-linearity problem directly to a linearity problem loses precision. Specifically, the relationship between input load and structural response is non-linear due to the existence of non-linear factors such as bearing connection, bolt contact and the like, and the local stress concentration phenomenon caused by the complex structure of the fan component. If the stress of the wind turbine components is predicted by a single load component, a large error may exist in the prediction result. Therefore, a single load component is omitted, and the stress prediction is carried out on the wind turbine generator by adopting a plurality of load components which are mutually coupled under the current working condition, so that the precision of the stress prediction can be greatly improved.
Fig. 7 is a flowchart illustrating a method for predicting stress of a wind turbine according to an embodiment of the present invention.
As shown in fig. 7, the method may include:
s701, collecting a plurality of current load components of the wind turbine generator system which are coupled with each other under the current load working condition.
And carrying out data homogenization treatment on the positions corresponding to the plurality of load working conditions and the stress extreme values to obtain first input data to be input.
S702, taking a plurality of current load components coupled with each other under the current load working condition as input, and predicting to obtain current stress extreme positions of the unit component by using the position neural network model 100, such as extreme position 1 and extreme position 2 … … extreme position N.
And inputting the first input data into a stress extreme position prediction RBF neural network, processing the first input data by a data hiding layer of a position neural network model, and predicting to obtain stress extreme position prediction results, such as extreme position 1 and extreme position 2 … … extreme position N.
And S703, selecting the extremum neural network model corresponding to the current stress extremum position according to the current stress extremum position.
If extremum position 1 selects extremum neural network model 1, extremum position 2 selects extremum neural network model 2 … … extremum position N selects extremum neural network model N.
And S704, taking a plurality of current load components which are coupled with each other under the current load working condition as input data.
And carrying out data homogenization treatment on the plurality of load working conditions and stress extreme values to obtain second input data to be input.
S705, when the extreme value position 1 selects the extreme value neural network model 1, inputting the second input data into the stress extreme value prediction RBF neural network 1, processing the stress extreme value prediction RBF neural network 1 by the data hidden layer of the extreme value neural network model 1, and predicting to obtain a stress extreme value prediction result 1.
When the extreme value position 2 selects the extreme value neural network model 2, the second input data is input into the stress extreme value prediction RBF neural network 2, the stress extreme value prediction RBF neural network is processed by the data hidden layer of the extreme value neural network model 2, and the stress extreme value prediction result 2 is obtained through prediction.
And when the extreme value position N selects the extreme value neural network model N, inputting the second input data into the stress extreme value prediction RBF neural network N, processing the stress extreme value prediction RBF neural network N by the data hidden layer of the extreme value neural network model N, and predicting to obtain a stress extreme value prediction result N.
Specifically, in this embodiment, a stress extreme position prediction module may first construct and train an RBF neural network (position neural network model) according to load data in training sample data and a stress extreme position corresponding to the load data, and predict the stress extreme position when the load data is input.
Secondly, the embodiment may further include a plurality of (N) stress extreme value prediction modules, where each module corresponds to a stress extreme value position. And constructing the RBF neural network according to the load data in the training sample data and the stress extreme value corresponding to the load data, and predicting the stress extreme value when the load data is input. And finally, connecting all the modules in the two parts, constructing a structural stress prediction platform, and simultaneously giving the value and the position of the maximum stress according to the input load working condition.
Compared with the existing stress prediction method, the stress extreme value output method has the advantages that the position corresponding to the stress extreme value can be given while the stress extreme value is output, and the function is stronger.
Compared with a univariate and linear interpolation method, the embodiment considers the influence of a plurality of load components on the stress result, performs nonlinear transformation and linear interpolation on the plurality of load components, and simultaneously each position has a unique stress extreme value prediction RBF neural network corresponding to the position, so that the reliability and the accuracy of the calculated result are higher.
In addition, in the case of no conflict, those skilled in the art can flexibly adjust the order of the above operation steps or flexibly combine the above steps according to actual needs. Various implementations are not described again for the sake of brevity. In addition, the contents of the various embodiments may be mutually incorporated by reference.
Fig. 8 is a schematic structural diagram of a device for predicting stress of a wind turbine generator according to an embodiment of the present invention.
As shown in fig. 8, the apparatus may include: a component acquisition module 801, a location prediction module 802, a model selection module 803, and an extremum prediction module 804.
The component acquisition module 801 may be configured to acquire a plurality of current load components of the wind turbine generator set components coupled to each other under a current load condition; the position prediction module 802 may be configured to predict a current stress extreme position of the unit component using a position neural network model, with a plurality of current load components coupled to each other under a current load condition as inputs; the model selection module 803 may be configured to select an extremum neural network model corresponding to the current stress extremum position according to the current stress extremum position; the extremum predicting module 804 may be configured to predict a current stress extremum at a current stress extremum location using an extremum neural network model, using a plurality of current load components coupled to each other under a current load condition as inputs.
In some embodiments, extremum predicting module 804 may include: a first transformation unit and a first calculation unit. Wherein the first transformation unit is operable to perform a non-linear transformation by the extremal neural network model based on the input mutually coupled plurality of current load components; the first calculating unit may be configured to perform linear difference calculation on the multiple current load components coupled to each other through the nonlinear transformation, so as to obtain a current stress extreme value at the current stress extreme value position.
In some embodiments, the location prediction module 802 may include: a second transformation unit and a second calculation unit. Wherein the second transformation unit may be configured to perform a non-linear transformation by the position neural network model based on the input mutually coupled plurality of current load components; the second calculating unit may be configured to perform linear difference calculation on the multiple current load components coupled to each other through the nonlinear transformation, so as to obtain a current stress extreme value position.
In some embodiments, the location neural network model or the extremum neural network model may include: the device comprises a data input layer, a data hiding layer and a data output layer. Wherein: at a data input layer, carrying out nonlinear transformation based on a plurality of current load components which are input and coupled with each other; in the data hiding layer, linear difference value calculation is carried out on a plurality of current load components which are subjected to nonlinear transformation and mutually coupled to obtain a calculation result; and outputting the calculation result in the data output layer.
In some embodiments, the location neural network model or the extremum neural network model may include: RBF neural network model.
In some embodiments, the data entry layer may be used to: and carrying out nonlinear transformation on the input mutually coupled current load components based on a Guass function or a Sigmoidal function as a basis function.
In some embodiments, the plurality of current load components includes two or more of the following components: the current bending moment of each direction and the current stress value of each direction in the multidimensional space.
In some embodiments, the means for stress prediction of a wind turbine may comprise: a memory and a processor. The memory may be used to store computer programs; the processor may be configured to execute a program stored in the memory, the program causing the processor to perform the methods of fig. 1 and 7.
In some embodiments, a computer readable storage medium may include a computer program that, when run on a computer, causes the computer to perform the methods of fig. 1 and 7.
In some embodiments, a stress prediction system for a wind turbine may include: and the stress prediction device of the extreme value neural network model base and the wind generating set. A plurality of extremal neural network models can be included in the extremal neural network model library. And predicting the current stress extreme value of the current stress extreme value position by using each extreme value neural network model.
Therefore, the embodiment of the invention can increase the position of the stress extreme value as a prediction output variable, and give out the corresponding position while predicting the stress extreme value; the prediction of the stress extreme value can be associated with the position, and different positions can correspond to different prediction methods; and when stress is predicted, a plurality of load components coupled with each other are considered at the same time, the input load variable and the output stress variable are not in a linear interpolation relation any more, and a nonlinear relation is constructed through an RBF neural network.
It should be noted that the apparatuses in the foregoing embodiments can be used as the execution main body in the methods in the foregoing embodiments, and can implement corresponding processes in the methods to achieve the same technical effects, and for brevity, the contents of this aspect are not described again.
In the above embodiments, all or part may be implemented by software, hardware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device to perform the methods described in the embodiments or some portions of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (19)
1. The method for predicting the stress of the wind turbine is characterized by comprising the following steps of:
collecting a plurality of current load components of the unit components of the wind turbine generator, which are coupled with each other under the current load working condition;
taking the plurality of current load components coupled with each other under the current load working condition as input, and predicting the current stress extreme value position of the unit component by using a position neural network model;
selecting an extremum neural network model corresponding to the current stress extremum position according to the current stress extremum position;
and taking the plurality of current load components which are coupled with each other under the current load working condition as input, and predicting a current stress extreme value of the current stress extreme value position by using an extreme value neural network model.
2. The method of claim 1, wherein using the plurality of current load components coupled to each other under the current load condition as inputs, predicting the current stress extremum locations of the crew member using a positional neural network model comprises:
the position neural network model performs a nonlinear transformation based on the input plurality of current load components coupled to each other;
and performing linear interpolation calculation on the plurality of mutually coupled current load components subjected to the nonlinear transformation to obtain the current stress extreme value position.
3. The method of claim 1, wherein using the plurality of current load components coupled to each other under the current load condition as inputs, predicting a current stress extremum for the current stress extremum location using an extremal neural network model comprises:
the extremum neural network model performs a nonlinear transformation based on the input plurality of intercoupled current load components;
and performing linear interpolation calculation on the plurality of mutually coupled current load components subjected to the nonlinear transformation to obtain a current stress extreme value of the current stress extreme value position.
4. The method of any one of claims 1-3, wherein the location neural network model or the extremum neural network model comprises:
the device comprises a data input layer, a data hiding layer and a data output layer.
5. The method of claim 4, wherein:
at the data input layer, performing a non-linear transformation based on the input mutually coupled plurality of current load components;
in the data hiding layer, carrying out linear interpolation calculation on the mutually coupled current load components subjected to the nonlinear transformation to obtain a calculation result;
and outputting the calculation result at the data output layer.
6. The method of claim 5, wherein the location neural network model or the extremum neural network model comprises:
RBF neural network model.
7. The method of claim 6, wherein performing a non-linear transformation at the data input layer based on the input of the mutually coupled plurality of current payload components comprises:
and carrying out nonlinear transformation on the input mutually coupled current load components based on a Guass function or a Sigmoidal function as a basis function.
8. The method of claim 1, wherein the plurality of current load components includes two or more of the following components:
the current bending moment of each direction and the current stress value of each direction in the multidimensional space.
9. An apparatus for stress prediction of a wind turbine, comprising:
the component acquisition module is used for acquiring a plurality of current load components which are mutually coupled under the current load working condition of the unit components of the wind turbine generator;
the position prediction module is used for taking the current load components which are coupled with each other under the current load working condition as input and predicting the current stress extreme position of the unit component by using a position neural network model;
the model selection module is used for selecting an extremum neural network model corresponding to the current stress extremum position according to the current stress extremum position;
and the extreme value prediction module is used for taking the current load components which are mutually coupled under the current load working condition as input and predicting the current stress extreme value of the current stress extreme value position by using an extreme value neural network model.
10. The apparatus of claim 9, wherein the location prediction module comprises:
a second transformation unit for performing a nonlinear transformation by the position neural network model based on the input current load components coupled to each other;
and the second calculation unit is used for performing linear interpolation calculation on the plurality of current load components which are coupled with each other and subjected to the nonlinear transformation to obtain the current stress extreme value position.
11. The apparatus of claim 9, wherein the extremum predicting module comprises:
a first transformation unit, configured to perform a nonlinear transformation on the basis of the input mutually coupled current load components by the extremum neural network model;
and the first calculation unit is used for performing linear interpolation calculation on the plurality of current load components which are coupled with each other and subjected to the nonlinear transformation to obtain a current stress extreme value of the current stress extreme value position.
12. The apparatus of any one of claims 9-11, wherein the location neural network model or the extremum neural network model comprises:
the device comprises a data input layer, a data hiding layer and a data output layer.
13. The apparatus of claim 12, wherein:
at the data input layer, performing a non-linear transformation based on the input mutually coupled current load components;
in the data hiding layer, carrying out linear interpolation calculation on a plurality of current load components which are subjected to nonlinear transformation and mutually coupled to obtain a calculation result;
and outputting the calculation result at the data output layer.
14. The apparatus of claim 13, wherein the location neural network model or the extremum neural network model comprises:
RBF neural network model.
15. The apparatus of claim 14, wherein the data input layer is configured to:
and carrying out nonlinear transformation on the input mutually coupled current load components based on a Guass function or a Sigmoidal function as a basis function.
16. The apparatus of claim 9, wherein the plurality of current load components comprises two or more of:
the current bending moment of each direction and the current stress value of each direction in the multidimensional space.
17. An apparatus for stress prediction of a wind turbine, comprising:
a memory for storing a computer program;
a processor for executing a computer program stored by the memory, the computer program causing the processor to perform the method of any of claims 1-8.
18. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, causes the processor to carry out the method according to any one of claims 1-8.
19. A system for stress prediction for a wind turbine, comprising:
the apparatus of any one of claims 9-16.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810167169.4A CN110207871B (en) | 2018-02-28 | 2018-02-28 | Method, device, storage medium and system for stress prediction of wind turbine generator |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810167169.4A CN110207871B (en) | 2018-02-28 | 2018-02-28 | Method, device, storage medium and system for stress prediction of wind turbine generator |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110207871A CN110207871A (en) | 2019-09-06 |
| CN110207871B true CN110207871B (en) | 2021-04-06 |
Family
ID=67778931
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810167169.4A Active CN110207871B (en) | 2018-02-28 | 2018-02-28 | Method, device, storage medium and system for stress prediction of wind turbine generator |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110207871B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113252218B (en) * | 2021-05-12 | 2023-11-17 | 国网山西省电力公司电力科学研究院 | Insulator surface stress prediction method and prediction device |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6925361B1 (en) * | 1999-11-30 | 2005-08-02 | Orion Engineering Corp. | Distributed energy neural network integration system |
| CN101814160A (en) * | 2010-03-08 | 2010-08-25 | 清华大学 | RBF neural network modeling method based on feature clustering |
| CN103374931A (en) * | 2012-04-25 | 2013-10-30 | 同济大学 | Test device for simulating wind power base affected by three-way coupling loads |
| CN104537424A (en) * | 2014-10-28 | 2015-04-22 | 北京天源科创风电技术有限责任公司 | Method for establishing predicated response system based on wind turbine generator load database |
| CN105243253A (en) * | 2014-12-05 | 2016-01-13 | 宁波大学 | Detection method and device of climbing frame state |
| CN105604807A (en) * | 2015-12-31 | 2016-05-25 | 北京金风科创风电设备有限公司 | Wind turbine generator monitoring method and device |
| CN105673325A (en) * | 2016-01-13 | 2016-06-15 | 湖南世优电气股份有限公司 | Individual pitch control method of wind driven generator set based on RBF neural network PID |
| CN107578453A (en) * | 2017-10-18 | 2018-01-12 | 北京旷视科技有限公司 | Compressed image processing method, apparatus, electronic equipment and computer-readable medium |
-
2018
- 2018-02-28 CN CN201810167169.4A patent/CN110207871B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6925361B1 (en) * | 1999-11-30 | 2005-08-02 | Orion Engineering Corp. | Distributed energy neural network integration system |
| CN101814160A (en) * | 2010-03-08 | 2010-08-25 | 清华大学 | RBF neural network modeling method based on feature clustering |
| CN103374931A (en) * | 2012-04-25 | 2013-10-30 | 同济大学 | Test device for simulating wind power base affected by three-way coupling loads |
| CN104537424A (en) * | 2014-10-28 | 2015-04-22 | 北京天源科创风电技术有限责任公司 | Method for establishing predicated response system based on wind turbine generator load database |
| CN105243253A (en) * | 2014-12-05 | 2016-01-13 | 宁波大学 | Detection method and device of climbing frame state |
| CN105604807A (en) * | 2015-12-31 | 2016-05-25 | 北京金风科创风电设备有限公司 | Wind turbine generator monitoring method and device |
| CN105673325A (en) * | 2016-01-13 | 2016-06-15 | 湖南世优电气股份有限公司 | Individual pitch control method of wind driven generator set based on RBF neural network PID |
| CN107578453A (en) * | 2017-10-18 | 2018-01-12 | 北京旷视科技有限公司 | Compressed image processing method, apparatus, electronic equipment and computer-readable medium |
Non-Patent Citations (1)
| Title |
|---|
| 基于神经网络的模锻压机载荷在线预测模型;蔺永诚，梁英杰，谌东东;《锻压技术》;20161018;第41卷(第10期);98-102 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110207871A (en) | 2019-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107231436B (en) | Method and device for scheduling service | |
| Barlas et al. | Model predictive control for wind turbines with distributed active flaps: incorporating inflow signals and actuator constraints | |
| EP3364324B1 (en) | Method and device for detecting equivalent load of wind turbine generator system | |
| CN113553755B (en) | Power system state estimation method, device and equipment | |
| Jamshidi et al. | Neuro-fuzzy system identification for remaining useful life of electrolytic capacitors | |
| CN112414668A (en) | Wind tunnel test data static bomb correction method, device, equipment and medium | |
| CN110207871B (en) | Method, device, storage medium and system for stress prediction of wind turbine generator | |
| CN115455793A (en) | High-rise structure complex component stress analysis method based on multi-scale model correction | |
| CN113919221A (en) | Fan load prediction and analysis method and device based on BP neural network and storage medium | |
| CN117170980B (en) | Early warning method, device, equipment and storage medium for server hardware abnormality | |
| CN108460177B (en) | Reliability approximate calculation method for large-scale multi-state series-parallel system | |
| CN112287605B (en) | Power flow checking method based on graph convolution network acceleration | |
| CN112131753A (en) | Method, system and device for evaluating fatigue life of fan and readable medium | |
| Cevallos et al. | The extended Kalman filter in the dynamic state estimation of electrical power systems | |
| CN110991741A (en) | Section constraint probability early warning method and system based on deep learning | |
| CN111177855B (en) | Pneumatic structure solving method and system in global aeroelasticity optimization | |
| CN111651843B (en) | Design method and system of main frame of generator and electronic equipment | |
| Bin et al. | Simple and effective fault diagnosis method of power lithium-ion battery based on GWA-DBN | |
| CN112052604A (en) | Method, system, equipment and readable medium for predicting equivalent fatigue load of fan | |
| Sefriti et al. | Fuzzy observer-based fault detection for wind turbines using the finite-frequency approach | |
| Ulloa et al. | The extended kalman filter in the dynamic state estimation of electrical power systems | |
| CN116736173B (en) | Energy storage battery model construction and energy storage battery state judgment method and device | |
| Tian et al. | Fusion modeling for wind power forecasting based on redundant method elimination | |
| Devie et al. | Transient stability prediction in multimachine system using data mining techniques | |
| Liu et al. | A fusion prognostic approach based on multi-kernel relevance vector machine and Bayesian model averaging |
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 | ||
| CP01 | Change in the name or title of a patent holder |
Address after: No.107 Shanghai Road, Urumqi Economic and Technological Development Zone, Urumqi City, Xinjiang Uygur Autonomous Region Patentee after: Jinfeng Technology Co.,Ltd. Address before: No.107 Shanghai Road, Urumqi Economic and Technological Development Zone, Urumqi City, Xinjiang Uygur Autonomous Region Patentee before: XINJIANG GOLDWIND SCIENCE & TECHNOLOGY Co.,Ltd. |
|
| CP01 | Change in the name or title of a patent holder |