CN115660206A - Training method and device for wind power plant power prediction actor model - Google Patents
Training method and device for wind power plant power prediction actor model Download PDFInfo
- Publication number
- CN115660206A CN115660206A CN202211398184.2A CN202211398184A CN115660206A CN 115660206 A CN115660206 A CN 115660206A CN 202211398184 A CN202211398184 A CN 202211398184A CN 115660206 A CN115660206 A CN 115660206A
- Authority
- CN
- China
- Prior art keywords
- training
- matrix
- wind
- model
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Wind Motors (AREA)
Abstract
The disclosure provides a training method and a device for a power prediction actor model of a wind power plant, which relate to the technical field of power systems, and the method comprises the following steps: acquiring historical wind power plant data and an informar model to be trained corresponding to different wind condition grades; carrying out standardization processing on historical wind power plant data to obtain a historical data matrix; intercepting the historical data matrix to obtain a plurality of training matrices; and determining the sample wind condition grade of the training matrix, and inputting the training matrix into an informar model matched with the sample wind condition grade for training to obtain the trained informar model. Compared with the wind power plant power prediction model in the prior art, the wind power plant power prediction method has the advantages that the prediction speed of inference calculation is improved by establishing the wind power plant power prediction actor model and training, better effects are achieved on prediction accuracy and calculation speed, and the accuracy and prediction effect of the actor model prediction can be improved by establishing the actor models with different wind condition grades.
Description
Technical Field
The disclosure relates to the technical field of power systems, and in particular to a training method and device for a wind power plant power prediction actor model, electronic equipment and a storage medium.
Background
Because wind energy belongs to unstable energy which fluctuates randomly, large-scale wind power is merged into a system, and new challenges are brought to the stability of the system. For a wind power plant, wind power prediction is beneficial to reasonably arranging maintenance plans and improving the profitability of enterprises, and the accuracy of the wind power plant power prediction has very important significance for wind power operation management. The more accurate the power prediction of the wind power plant is, the power limit of the wind power plant can be reduced by the power grid, and the wind power generation capacity of the power grid is greatly improved, so that the generated energy of the wind power plant is improved.
There are some related patents that use Transformer to predict wind power generation power. The wind power prediction method based on the Transformer model refers to the wind power prediction by using the Transformer model. A wind power prediction method, a system and equipment based on a convolution Transformer framework input meteorological data and operation data into a convolution Transformer network to predict wind power generation power. The wind power prediction method and system for optimizing the depth Transformer network combine the transform network prediction model with the whale swarm optimization algorithm, and the accuracy of wind power prediction is greatly improved. Although the prediction accuracy of the Transformer model can obtain good performance, the model has high time and space complexity, and the timeliness is poor for the wind power plant power prediction problem needing real-time calculation, particularly ultra-short-term prediction.
BRIEF SUMMARY OF THE PRESENT DISCLOSURE
The present disclosure is directed to solving, at least in part, one of the technical problems in the related art.
To this end, an object of the present disclosure is to provide a training method for a wind farm power prediction former model.
The second purpose of the present disclosure is to provide a training device for a wind farm power prediction actor model.
A third object of the present disclosure is to provide an electronic device.
A fourth object of the present disclosure is to propose a non-transitory computer-readable storage medium.
In order to achieve the above purpose, an embodiment of the first aspect of the present disclosure provides a training method for a wind farm power prediction former model, including: acquiring historical wind power plant data and an informar model to be trained corresponding to different wind condition grades; carrying out standardization processing on historical wind power plant data to obtain a historical data matrix; intercepting the historical data matrix to obtain a plurality of training matrices; and determining the sample wind condition grade of the training matrix, and inputting the training matrix into an informar model matched with the sample wind condition grade for training to obtain the trained informar model.
According to one embodiment of the present disclosure, the method for obtaining a historical data matrix by normalizing historical wind farm data, which includes: z-score normalizing each matrix element to obtain normalized data; a historical data matrix is generated based on the normalized data for all matrix elements.
According to one embodiment of the present disclosure, intercepting a historical data matrix to obtain a plurality of training matrices includes: acquiring the interception width of a training matrix to be generated, wherein the interception width is smaller than the width of a historical data matrix; and selecting adjacent multi-column data in the intercepted width historical data matrix to obtain a training matrix.
According to one embodiment of the present disclosure, the matrix element contains a sampling time stamp, and the determining of the sample wind condition level of the training matrix comprises: acquiring an average wind speed value and a wind condition grade mapping table of a training matrix; and performing table lookup based on the average wind speed value and the wind condition grade mapping table to determine the sample wind condition grade of the training matrix.
According to one embodiment of the present disclosure, obtaining an average wind speed value of a training matrix comprises: acquiring wind speed data of all matrix elements in a training matrix; an average wind speed value of the training matrix is determined based on the wind speed data.
According to one embodiment of the present disclosure, an nformer model includes an Encoder module and a Decoder module, and inputs a training matrix into an nformer model matched with a sample wind condition grade for training to obtain a trained nformer model, including: acquiring input historical wind power corresponding to all matrix elements of a training matrix; inputting the training matrix into an Encoder module to obtain a characteristic matrix; inputting the extracted feature matrix and the input historical wind power into a Decoder module to generate predicted wind power; training the inner model based on the predicted wind power.
According to one embodiment of the present disclosure, training an actor model based on predicted wind power includes: acquiring real future wind power corresponding to the training matrix; and determining a loss value based on the real power of the future wind and the predicted wind power, and optimizing the informar model based on the loss value.
In order to achieve the above object, an embodiment of a second aspect of the present disclosure provides a training apparatus for a wind farm power prediction actor model, including: the acquisition module is used for acquiring historical wind power plant data and an informar model to be trained corresponding to different wind condition grades; the standardization module is used for carrying out standardization processing on historical wind power plant data to obtain a historical data matrix; the intercepting module is used for intercepting the historical data matrix to obtain a plurality of training matrices; and the training module is used for determining the sample wind condition grade of the training matrix, inputting the training matrix into an informar model matched with the sample wind condition grade for training so as to obtain the trained informar model.
To achieve the above object, an embodiment of a third aspect of the present disclosure provides an electronic device, including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to implement a method for training a wind farm power prediction actor model as defined in embodiments of the first aspect of the disclosure.
To achieve the above object, a fourth aspect of the present disclosure provides a non-transitory computer-readable storage medium storing computer instructions for implementing a training method of a wind farm power prediction former model according to the first aspect of the present disclosure.
Compared with the wind power plant power prediction model in the prior art, the wind power plant power prediction method has the advantages that the prediction speed of inference calculation is improved by establishing the wind power plant power prediction actor model and training, better effects are achieved on prediction accuracy and calculation speed, and the accuracy and prediction effect of the actor model prediction can be improved by establishing the actor models with different wind condition grades.
Drawings
FIG. 1 is a schematic diagram of a training method of a wind farm power prediction actor model according to an embodiment of the disclosure;
FIG. 2 is a schematic diagram of another method for training a wind farm power prediction, in accordance with an embodiment of the present disclosure;
FIG. 3 is a schematic diagram of another method for training a wind farm power prediction actor model according to an embodiment of the present disclosure;
FIG. 4 is a schematic diagram of another method for training a wind farm power prediction actor model according to an embodiment of the present disclosure;
FIG. 5 is a schematic diagram of an equation model according to an embodiment of the present disclosure;
FIG. 6 is a schematic diagram of a training device for a wind farm power prediction actor model according to an embodiment of the present disclosure;
fig. 7 is a schematic diagram of an electronic device according to an embodiment of the present disclosure.
Detailed Description
Reference will now be made in detail to the embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and intended to be illustrative of the present disclosure, and should not be construed as limiting the present disclosure.
Fig. 1 is a schematic diagram of an exemplary embodiment of a training method for a wind farm power prediction actor model according to the present disclosure, and as shown in fig. 1, the training method for the wind farm power prediction actor model includes the following steps:
s101, acquiring historical wind power plant data and an informar model to be trained corresponding to different wind condition grades.
In order to solve the problem of the Transformer model in wind power plant power prediction, the invention applies the inform model to wind power plant power prediction. The Informer model improves the self-attention structure (temporal and spatial complexity O (N2)) in the Transformer model to a probabilistic sparse self-attention structure, reduces temporal and spatial complexity improvement to O (NlogN), and reduces spatial complexity to O ((2-e) NlogN) using self-attention extraction operations. Compared with a transform wind power plant power prediction model, the Informer model greatly improves the prediction speed of inference calculation, and achieves better effects on the prediction accuracy and the calculation speed.
Because the result of the power prediction of the wind farm is not regularly changed in the equation model at different wind speeds, separate modeling and prediction need to be performed for different wind speeds. It should be noted that, in the embodiment of the present disclosure, different wind condition classes may be divided, and an informar model corresponding to the wind condition class is established for training, so that the method for establishing the informar model based on the wind condition classes may make the prediction result more accurate.
In the embodiment of the disclosure, the historical wind farm data includes historical data of operation of a wind measuring tower and a wind farm, the wind measuring tower data includes meteorological environment data such as wind speed, wind direction, air temperature, air humidity, air density, air pressure and the like, and the wind farm operation data includes grid-connected point three-phase voltage, active power, reactive power, grid frequency and the like, which are not limited at all and can be specifically limited according to actual design requirements.
It should be noted that, when building the model, the time length to be predicted also needs to be determined, and the information models corresponding to different time lengths to be predicted may be different, and specifically may be set according to actual prediction needs. For example, the time period may be 10min, 20min, etc.
S102, carrying out standardization processing on historical wind power plant data to obtain a historical data matrix.
It should be noted that the normalization of the data is to convert the raw data into a dimensionless index value by converting the raw data into a mathematical transform, i.e. each index value is in the same numerical level, so as to perform comprehensive analysis and comparison.
In the embodiment of the disclosure, after the historical wind power plant data is obtained, the historical wind power plant data needs to be standardized, so that unified data can be obtained, and an actor model can conveniently process and train standardized data.
In the embodiment of the present disclosure, the data normalization process may be performed in various ways, for example, the data z-score normalization, centralization, etc. are included, which are not limited herein, and may be specifically defined according to the actual design requirement.
In the disclosed embodiment, the historical data matrix may be an m × n matrix, where m represents the number of variables and n represents the time sampling point.
S103, intercepting the historical data matrix to obtain a plurality of training matrices.
Since the wind farm power prediction problem is a long time series prediction problem, the data of the current period of time needs to be used for predicting the power data of a future period of time. In one implementation of the present disclosure, if the number of time sampling points in a current period of time is d, and the number of time prediction points in a future period of time is t, the problem needs to be solved by inputting a data matrix of m × d to output a predicted power vector of t.
In the embodiment of the present disclosure, after an mxn historical data matrix is obtained, the mxn historical data matrix needs to be decomposed into a plurality of mxd data matrices, so as to train an informar model. It should be noted that windowing interception may be performed on the mxn historical data matrix, and a plurality of training matrices with the same size are intercepted by setting a certain window width.
And S104, determining the sample wind condition grade of the training matrix, and inputting the training matrix into an informar model matched with the sample wind condition grade for training to obtain the trained informar model.
In the embodiment of the present disclosure, the method for determining the sample wind condition level of the training matrix may be various, and is not limited herein. Alternatively, the wind speed of each matrix element in the training matrix may be averaged, and the sample wind condition level of the training matrix may be determined based on the average wind speed.
Optionally, the wind speed of the last matrix element in the training matrix may also be determined as the target wind speed, for example, the last matrix element may be the matrix element ordered as the last one, and may also be the matrix element with the largest sampling time. After the target wind speed is obtained, a current wind condition rating may be determined based on the target wind speed.
The wind condition level may be set in advance, and may be changed according to actual design requirements, which is not limited herein.
In the embodiment of the disclosure, historical wind power plant data and an informar model to be trained corresponding to different wind speed grades are firstly obtained, then the historical wind power plant data are subjected to standardization processing to obtain a historical data matrix, then the historical data matrix is intercepted to obtain a plurality of training matrices, finally the sample wind speed grade of the training matrix is determined, and the training matrices are input into the informar model matched with the sample wind speed grade to be trained to obtain the trained informar model. Therefore, by establishing the wind power plant power prediction informar model and training, compared with a wind power plant power prediction model in the prior art, the prediction speed of inference calculation is improved, better effects on prediction accuracy and calculation speed are achieved, and the accuracy and the prediction effect of the informar model prediction can be improved by establishing the informar models of different wind condition grades.
In the above embodiment, the historical wind farm data includes a plurality of matrix elements, and the historical wind farm data is normalized to obtain a historical data matrix, which can be further explained by using fig. 2, where the method includes:
s201, performing z-score standardization on each matrix element to obtain standardized data.
In the disclosed embodiment, the mean and standard deviation of n time sampling points of each variable are first calculated, and then the historical data D is z-score normalized, with the normalized data as normalized data X, which is an m × n matrix. The Z-score normalization procedure was calculated according to the following formula:
wherein
The value of the nth time sample point of the mth variable in normalized data X is shown,
the value of the nth time sample point of the mth variable in the history data D, mean (D) (m) ) Denotes the mean value of the mth variable in the history data D, std (x) (m) ) Indicating the standard deviation of the mth variable in the historical data D.
S202, generating a historical data matrix based on the standardized data of all matrix elements.
After the normalized data for each matrix element is obtained, a historical data matrix may be generated from the normalized data based on the collected timestamps of the matrix elements. It should be noted that the element sequence in the historical data matrix is the same as the element sequence of the historical wind farm data, and the element data are arranged according to the time sequence.
In the disclosed embodiment, each matrix element is first z-score normalized to obtain normalized data, and then a historical data matrix is generated based on the normalized data for all matrix elements. Therefore, the matrix elements are subjected to z-score standardization, so that the matrix elements can be conveniently input into the model for training and processing, and the training effect of the model is improved.
In the above embodiment, the historical data matrix is intercepted to obtain a plurality of training matrices, the interception width of the training matrix to be generated may be first obtained, the interception width is smaller than the width of the historical data matrix, and then adjacent columns of data in the historical data matrix with the interception width are selected to obtain the training matrix. It should be noted that the truncation width is generally much smaller than the width of the history matrix, so as to ensure that a sufficient training matrix is generated. For example, normalized data X of m × n size may be collated into training data T of m × d × (n-d + 1) size. T comprises (n-d + 1) training matrixes with the size of m multiplied by d as input data.
In the above embodiment, the matrix elements include sampling time stamps, and the determination of the sample wind speed levels of the training matrix is further explained by fig. 3, and the method includes:
s301, obtaining an average wind speed value and a wind condition grade mapping table of the training matrix.
In the embodiment of the disclosure, the wind speed data of all matrix elements in the training matrix can be obtained by obtaining the historical wind power plant data, and then the average wind speed value of the training matrix is determined based on the wind speed data.
In the disclosed embodiment, the wind condition level mapping table may be generated by collecting wind data of a current wind farm. Specifically, the wind speed data in the time interval may be collected, the average value of the wind speed data in the time interval may be determined, the average value may be used as a basic wind condition level, and then other wind condition levels may be determined according to a certain division interval, so as to generate a wind condition level mapping table. For example, the wind condition level mapping table may be generated by obtaining an average wind speed of the first 10min of the last sampling point in the historical data as a classification variable of the wind condition division, and then determining wind speeds corresponding to other wind condition levels according to the division interval of 0.5 m/s.
S302, based on the average wind speed value and the wind condition level mapping table, determining the wind speed level corresponding to the matrix element with the maximum sampling time stamp as a sample wind speed level.
After the average wind speed value and the wind condition level mapping table are obtained, the current wind speed level can be determined through table lookup based on the current average wind speed value. For example, the wind conditions are divided at 1m/s intervals, for example, the wind speed zone of 0.5m/s represents 0-1 m/s, the division is 0.5m/s, 1.5m/s, 2.5m/s, 3.5m/s … …, and so on. And according to the wind condition division, dividing the training data T into training matrixes under different wind conditions.
In the embodiment of the disclosure, an average wind speed value and a wind condition level mapping table of a training matrix are obtained first, and then a wind speed level corresponding to a matrix element with a maximum sampling time stamp is determined as a sample wind speed level based on the average wind speed value and the wind condition level mapping table. Therefore, the wind speed grade of the training matrix is determined by the average wind speed value of the matrix element with the largest time stamp, and a basis can be provided for subsequent training of determining different models and models based on different wind speed grades.
In the above embodiment, the inner model includes an Encoder module and a Decoder module, and the training matrix is input into the inner model matched with the sample wind speed level for training, so as to obtain the trained inner model, which can be further explained by using fig. 4, where the method includes:
s401, obtaining input historical wind power corresponding to all matrix elements of the training matrix.
It should be noted that, in the embodiment of the present disclosure, the input historical wind power is data that has not been subjected to the standardization process.
In embodiments of the present disclosure, the corresponding input historical wind power may be selected from historical wind farm data based on a timestamp in the matrix element.
S402, inputting the training matrix into an Encoder module to obtain a characteristic matrix.
In an embodiment of the disclosure, the Encoder may generate a feature matrix based on an input training matrix. As shown in FIG. 5, the Encoder includes K modules stacked for feature extraction, and each sub-module includes a Multi-head ProbSparse Self-attribute and a Self-attribute distinguishing. In the Self-orientation structure, the matrix comprises WQ, WK and WV, and if the input data is I, the matrix is
Q＝W Q I
K＝W K I
V＝W V I
Q, K, V denotes query, key, value, respectively. Features are then extracted using the following formula:
in Multi-head ProbSparse Self-attention, a sparse matrix Q is utilized - And carrying out feature extraction calculation. First, setting the sampling factor c, u = clnL, for each vector Q _ i in Q, the sparsity metric M (Q _ i, K) is calculated:
selecting u q with maximum sparse measure i Recombined to form
Then theAnd performing feature extraction calculation. Computing only the selected u q i And q is others i Directly taking mean (V), and then adding S 1 And S 0 The results of (1) are combined to obtain the final output S.
In the Self-orientation distinguishing operation, a higher weight is given to the dominant feature having the main feature, and a focused Self-orientation feature matrix is generated at the next layer. The Self-anchoring partitioning includes a one-dimensional convolution (Conv 1 d), an ELU activation function, and a max-pooling (Maxpool) operation.
I j+1 ＝MaxPool(ELU(Conv1d[S j ]))
And S403, inputting the feature matrix and the input historical wind power into a Decoder module to generate predicted wind power.
The Decoder is used for outputting a wind power prediction result of the wind power plant for a period of time in the future by using the extracted feature matrix. Besides the feature matrix output by the Encoder, the input historical wind power of a previous period of time is input and output to the model.
It should be noted that, in the embodiment of the present disclosure, in order to predict the wind farm power at the future prediction time, the training matrix is also required to be filled. For example, the historical time length is d, and the predicted time length is t. The Decoder input is "generative inference", the predicted future power sequence is filled with a 0 value of length t, and the input vector is (d + t). As shown in FIG. 5, the input vector of the Decoder prevents each position from focusing on future positions by a Masked Multi-head ProbSparse Self-attribute, thereby avoiding autoregressive. The output of mask Multi-head ProbSparse Self-attribute is used as query, the output characteristic diagram S of Encoder is used as key and value, and a Multi-head Attention is input. And (4) outputting the power prediction result of the wind power plant with the future time length of t through a full connection layer by the Decoder.
S404, training an informar model based on the predicted wind power.
After the predicted wind power is obtained, the real future wind power corresponding to the training matrix is obtained, a loss value is determined based on the real future wind power and the predicted wind power, and the informar model is optimized based on the loss value. In an embodiment of the present disclosure, the real future wind power is wind power data of a preset time length in the future of the largest timestamp in the training matrix. In the disclosed embodiment, the loss value may be obtained by first determining a loss function of the informar model and inputting the predicted wind power and the actual future wind power into the loss function. The loss function may be various and is not limited herein. For example, RMSE losses can be used for training based on a loss function, model parameters are updated by using an Adam optimization algorithm, and errors between actual wind power plant power and predicted wind power are reduced through continuous iterative training.
It can be understood that the training of the model is a repeated iterative process, and the training is performed by continuously adjusting the network parameters of the model until the overall loss function value of the model is smaller than a preset value, or the overall loss function value of the model is not changed or the change amplitude is slow, and the model converges to obtain the trained model. In the embodiment of the present disclosure, the preset value may be changed according to actual design requirements, and is not limited herein.
In the embodiment of the disclosure, input historical wind power corresponding to all matrix elements of a training matrix is firstly obtained, then the training matrix is input into an Encoder module to obtain a characteristic matrix, then the characteristic matrix and the input historical wind power are input into a Decoder module to generate predicted wind power, and finally an informar model is trained based on the predicted wind power. Therefore, the historical wind power and the training matrix are input into the equation model, and the equation model capable of accurately predicting the power of the wind power plant can be obtained through continuous iteration and optimization.
After the trained informar model is obtained, when real-time data are predicted, wind speed data in a preset time interval before the current moment are obtained, average wind speed is obtained, the current wind condition grade is obtained through table lookup of the average wind speed in a wind condition grade mapping table, and the corresponding trained informar model is selected for wind power plant power prediction.
Corresponding to the training methods for the wind farm power prediction actor model provided in the foregoing embodiments, an embodiment of the present disclosure further provides a training device for the wind farm power prediction actor model, and since the training device for the wind farm power prediction actor model provided in the embodiment of the present disclosure corresponds to the training methods for the wind farm power prediction actor model provided in the foregoing embodiments, the implementation of the training method for the wind farm power prediction actor model provided in the embodiment of the present disclosure is also applicable to the training device for the wind farm power prediction actor model provided in the embodiment of the present disclosure, and will not be described in detail in the following embodiments.
Fig. 6 is a schematic diagram of a training device for a wind farm power prediction actor model according to the present disclosure, and as shown in fig. 6, the training device 600 for the wind farm power prediction actor model includes: an acquisition module 610, a normalization module 620, an interception module 630, and a training module 640.
The obtaining module 610 is configured to obtain historical wind farm data and an informar model to be trained corresponding to different wind condition grades.
And the normalization module 620 is used for performing normalization processing on the historical wind farm data to obtain a historical data matrix.
And an intercepting module 630, configured to intercept the historical data matrix to obtain a plurality of training matrices.
The training module 640 is configured to determine a sample wind condition level of a training matrix, and input the training matrix into an inner model matched with the sample wind condition level for training to obtain the trained inner model.
In an embodiment of the present disclosure, the normalization module 620 is further configured to: z-score normalizing each matrix element to obtain normalized data; a historical data matrix is generated based on the normalized data for all matrix elements.
In an embodiment of the present disclosure, the normalization module 620 is further configured to: acquiring the interception width of a training matrix to be generated, wherein the interception width is smaller than the width of a historical data matrix; and selecting adjacent multi-column data in the intercepted width historical data matrix to obtain a training matrix.
In an embodiment of the present disclosure, the training module 640 is further configured to: acquiring an average wind speed value and a wind condition grade mapping table of a training matrix; and performing table lookup based on the average wind speed value and the wind condition grade mapping table to determine the wind condition grade of the training matrix sample.
In an embodiment of the present disclosure, the training module 640 is further configured to: acquiring wind speed data of all matrix elements in the training matrix; an average wind speed value for the training matrix is determined based on the wind speed data.
In an embodiment of the present disclosure, the inner model includes an Encoder module and a Decoder module, and the training module 640 is further configured to: acquiring input historical wind power corresponding to all matrix elements of a training matrix; inputting the training matrix into the Encoder module to obtain a characteristic matrix; inputting the characteristic matrix and the input historical wind power into a Decoder module to generate predicted wind power; training the inner model based on the predicted wind power.
In an embodiment of the present disclosure, the training module 640 is further configured to: acquiring real future wind power corresponding to the training matrix; determining a loss value based on the real power of the coming wind and the predicted wind power, and optimizing the inner model based on the loss value.
In order to implement the above embodiments, an embodiment of the present disclosure further provides an electronic device 700, as shown in fig. 7, where the electronic device 700 includes: a processor 701 and a memory 702 communicatively connected to the processor, the memory 702 storing instructions executable by the at least one processor, the instructions being executable by the at least one processor 701 to implement a method for training a wind farm power prediction actor model according to an embodiment of the first aspect of the disclosure.
To achieve the above embodiments, the present disclosure also proposes a non-transitory computer-readable storage medium storing computer instructions, where the computer instructions are used to enable a computer to implement the training method of the wind farm power prediction former model according to the first aspect of the present disclosure.
In order to implement the foregoing embodiments, the present disclosure further provides a computer program product, which includes a computer program and when executed by a processor, implements the method for training the wind farm power prediction indicator model according to the first aspect of the present disclosure.
In the description of the present disclosure, it is to be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," "circumferential," and the like are used in the orientations and positional relationships indicated in the drawings for convenience in describing the present disclosure and to simplify the description, but are not intended to indicate or imply that the referenced devices or elements must have a particular orientation, be constructed and operated in a particular orientation, and are therefore not to be considered limiting of the present disclosure.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present disclosure, "a plurality" means two or more unless specifically limited otherwise.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present disclosure have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present disclosure, and that changes, modifications, substitutions and alterations may be made to the above embodiments by those of ordinary skill in the art within the scope of the present disclosure.
Claims (10)
1. A training method for a wind power plant power prediction actor model is characterized by comprising the following steps:
acquiring historical wind power plant data and an informar model to be trained corresponding to different wind condition grades;
carrying out standardization processing on the historical wind power plant data to obtain a historical data matrix;
intercepting the historical data matrix to obtain a plurality of training matrices;
and determining the sample wind condition grade of the training matrix, and inputting the training matrix into an informar model matched with the sample wind condition grade for training to obtain the trained informar model.
2. The method of claim 1, wherein the historical wind farm data comprises a plurality of matrix elements, and wherein normalizing the historical wind farm data to obtain a historical data matrix comprises:
z-score normalizing each of the matrix elements to obtain normalized data;
generating the historical data matrix based on the normalized data of all the matrix elements.
3. The method of claim 1 or 2, wherein the truncating the historical data matrix to obtain a number of training matrices comprises:
acquiring the interception width of the training matrix to be generated, wherein the interception width is smaller than the width of the historical data matrix;
and selecting adjacent multi-column data in the intercepted width historical data matrix to obtain the training matrix.
4. The method of claim 1, wherein the matrix elements contain sampling time stamps, and wherein determining the sample wind condition class of the training matrix comprises:
acquiring an average wind speed value and a wind condition grade mapping table of the training matrix;
and performing table look-up based on the average wind speed value and the wind condition grade mapping table, and determining the sample wind condition grade of the training matrix.
5. The method of claim 4, wherein the obtaining of the average wind speed value of the training matrix comprises:
acquiring wind speed data of all matrix elements in the training matrix;
an average wind speed value for the training matrix is determined based on the wind speed data.
6. The method of claim 4, wherein the informar model comprises an Encoder module and a Decoder module, and the inputting the training matrix into the informar model matched with the sample wind condition levels for training to obtain the trained informar model comprises:
acquiring input historical wind power corresponding to all matrix elements of the training matrix;
inputting the training matrix into the Encoder module to obtain a feature matrix;
inputting the feature matrix and the input historical wind power into the Decoder module to generate a predicted wind power;
training the informar model based on the predicted wind power.
7. The method of claim 6, wherein training the informar model based on the predicted wind power comprises:
acquiring real future wind power corresponding to the training matrix;
determining a loss value based on the real oncoming wind power and the predicted wind power, and optimizing the inner model based on the loss value.
8. A training device for a wind power plant power prediction actor model is characterized by comprising the following components:
the acquisition module is used for acquiring historical wind power plant data and an informar model to be trained corresponding to different wind condition grades;
the standardization module is used for carrying out standardization processing on the historical wind power plant data to obtain a historical data matrix;
the intercepting module is used for intercepting the historical data matrix to obtain a plurality of training matrices;
and the training module is used for determining the sample wind condition grade of the training matrix, inputting the training matrix into an inner model matched with the sample wind condition grade for training so as to obtain the trained inner model.
9. An electronic device comprising a memory, a processor;
wherein the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory for implementing the method according to any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium having stored thereon computer instructions for causing the computer to perform the method for recognizing a sensitive word according to any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211398184.2A CN115660206A (en) | 2022-11-09 | 2022-11-09 | Training method and device for wind power plant power prediction actor model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211398184.2A CN115660206A (en) | 2022-11-09 | 2022-11-09 | Training method and device for wind power plant power prediction actor model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115660206A true CN115660206A (en) | 2023-01-31 |
Family
ID=85016634
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211398184.2A Pending CN115660206A (en) | 2022-11-09 | 2022-11-09 | Training method and device for wind power plant power prediction actor model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115660206A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117093873A (en) * | 2023-10-19 | 2023-11-21 | 国网浙江省电力有限公司丽水供电公司 | Hydropower station storage capacity assessment method and system based on natural water inflow prediction |
-
2022
- 2022-11-09 CN CN202211398184.2A patent/CN115660206A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117093873A (en) * | 2023-10-19 | 2023-11-21 | 国网浙江省电力有限公司丽水供电公司 | Hydropower station storage capacity assessment method and system based on natural water inflow prediction |
| CN117093873B (en) * | 2023-10-19 | 2024-01-30 | 国网浙江省电力有限公司丽水供电公司 | Hydropower station storage capacity assessment method and system based on natural water inflow prediction |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113935562A (en) | Intelligent rating and automatic early warning method for health condition of power equipment | |
| CN109523084A (en) | A kind of ultrashort-term wind power prediction method based on pivot analysis and machine learning | |
| CN110601250B (en) | Method for evaluating stability of static voltage of access point based on new energy power station | |
| CN115660206A (en) | Training method and device for wind power plant power prediction actor model | |
| CN112653198A (en) | Wind power output scene generation method and system based on prediction box | |
| Dokur et al. | Hybrid model for short term wind speed forecasting using empirical mode decomposition and artificial neural network | |
| CN103971175A (en) | Short-term load prediction method of multistage substations | |
| Guo et al. | Appling an improved method based on ARIMA model to predict the short-term electricity consumption transmitted by the Internet of Things (IoT) | |
| CN111242355A (en) | Photovoltaic probability prediction method and system based on Bayesian neural network | |
| CN116307291B (en) | Distributed photovoltaic power generation prediction method and prediction terminal based on wavelet decomposition | |
| CN114021483A (en) | Ultra-short-term wind power prediction method based on time domain characteristics and XGboost | |
| CN114548498A (en) | Wind speed prediction method and system for local area of overhead transmission line | |
| Buonanno et al. | Comprehensive method for modeling uncertainties of solar irradiance for PV power generation in smart grids | |
| CN113991711A (en) | Capacity configuration method for energy storage system of photovoltaic power station | |
| CN112540159A (en) | Nuclear power plant atmospheric diffusion prediction correction method, system, medium and electronic equipment | |
| CN115936238B (en) | Method, system, equipment and medium for predicting medium-long-term output of global wind power | |
| Niu et al. | Knowledge mining based on environmental simulation applied to wind farm power forecasting | |
| CN116703114A (en) | Classified storage method, system, device and medium for power dispatching data | |
| Naoui et al. | Integrating iot devices and deep learning for renewable energy in big data system | |
| CN116565840A (en) | High-precision wind speed soft measurement method for wind power prediction of wind power plant | |
| CN117033923A (en) | Method and system for predicting crime quantity based on interpretable machine learning | |
| Wenting et al. | Short-term wind power forecasting model based on random forest algorithm and TCN | |
| CN113779892B (en) | Method for predicting wind speed and wind direction | |
| CN115526429A (en) | Decoupling analysis method for wind power prediction error, processor and storage medium | |
| CN114692937A (en) | Main transformer load prediction method based on main transformer load influence factor prediction |
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 |