An Offshore Wind Power Prediction Method and System for Eliminating Boundary Effects

Abstract

The present invention relates to the technical field of wind power prediction, and more specifically, to an offshore wind power prediction method and system for eliminating boundary effects, including: S1, obtaining power characteristic data and meteorological characteristic data of a wind farm and performing preprocessing; S2, constructing a power characteristic matrix and a wind speed characteristic matrix; S3, respectively inputting the power characteristic matrix and the wind speed characteristic matrix into a gated recurrent unit network to output a power prediction sequence and a wind speed prediction sequence; S4, constructing a boundary effect interval identification function, inputting the wind speed prediction sequence into the boundary effect interval identification function to output a boundary effect interval; S5, constructing a three-dimensional piecewise threshold function, inputting the wind speed prediction sequence, the power prediction sequence and the boundary effect interval into the three-dimensional piecewise threshold function to output a future power sequence. The present invention can eliminate the influence of boundary effects on the wind power prediction result, thereby improving the prediction accuracy of wind power.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and more specifically, to an offshore wind power prediction method and system for eliminating boundary effects. Background Art

[0002] With the continuous development of wind power generation technology, the grid connection of large-capacity wind power has brought challenges to the safe and stable operation of large power grids. Therefore, accurate prediction of the output power of wind power generation can schedule the grid connection capacity of wind power in advance, reduce the volatility of wind power grid connection power, and is of great significance to the safe and stable operation of the power system. Currently, wind power generation is mainly divided into offshore wind power and onshore wind power. In the prediction of wind power, the meteorological differences between offshore wind power and onshore wind power should be considered, and adaptive adjustments should be made to obtain accurate prediction results.

[0003] In the prior art, for example, Chinese Patent CN117893361A discloses an offshore wind power prediction method based on a multi-scale analyzer, as follows: initially process the power, wind speed, and temperature data of the obtained offshore wind farm to obtain a wind power time series, a wind speed time series, and a temperature time series; input the wind power time series into the multi-scale analyzer to obtain trend characteristics, long-period characteristics, and short-period characteristics; recombine the trend characteristics, long-period characteristics, short-period characteristics, wind power time series, wind speed time series, and temperature time series into a feature vector matrix; input the feature vector matrix into the offshore wind power prediction model for offshore wind power prediction. However, this prediction method does not consider the problem that due to the relatively fast offshore wind speed, there is a boundary effect in the offshore wind power sequence, and the boundary effect will cause the gradient disappearance phenomenon during the training process of the prediction model, reducing the accuracy of the wind power prediction result. Summary of the Invention

[0004] An object of the present invention is to overcome the deficiency in the prior art that the accuracy of wind power prediction results is reduced due to boundary effects, and to provide an offshore wind power prediction method and system for eliminating boundary effects, which eliminate the boundary effect phenomenon during the wind power prediction process, thereby improving the prediction accuracy of offshore wind power.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is:

[0006] Provide an offshore wind power prediction method for eliminating boundary effects, including the following steps:

[0007] S1. Obtain the power characteristic data and meteorological characteristic data of the target wind farm, preprocess the power characteristic data and the meteorological characteristic data to obtain the historical power sequence P, historical wind speed sequence WS, historical temperature sequence Tem, and historical air pressure sequence Pre;

[0008] S2. Construct a power characteristic matrix X using the historical power sequence P, historical wind speed sequence WS, historical temperature sequence Tem, and historical air pressure sequence Pre power and a wind speed characteristic matrix X wind ;

[0009] S3. Input the power characteristic matrix X power and the wind speed characteristic matrix X wind into the gated recurrent unit network respectively. The gated recurrent unit network outputs a power prediction sequence P power and a wind speed prediction sequence P wind ;

[0010] S4. Construct a boundary effect interval recognition function ST according to the fan operation state curve, input the wind speed prediction sequence P wind into the boundary effect interval recognition function ST, and the boundary effect interval recognition function ST identifies the boundary effect interval;

[0011] S5. Construct a three-dimensional piecewise threshold function 3DST according to the fan operation state curve and the boundary effect interval, input the wind speed prediction sequence P wind , the power prediction sequence P power and the boundary effect interval recognition function ST into the three-dimensional piecewise threshold function 3DST, and the three-dimensional piecewise threshold function 3DST outputs the future power sequence.

[0012] The offshore wind power prediction method for eliminating boundary effects of the present invention first collects the power characteristic data and meteorological characteristic data of the wind farm to be predicted. The meteorological characteristic data includes wind speed, temperature, and air pressure. After preprocessing the power characteristic data and meteorological characteristic data, the historical power sequence P, historical wind speed sequence WS, historical temperature sequence Tem, and historical air pressure sequence Pre are obtained; subsequently, a power characteristic matrix X power and a wind speed characteristic matrix X wind are constructed according to the historical power sequence P, historical wind speed sequence WS, historical temperature sequence Tem, and historical air pressure sequence Pre; then the obtained power characteristic matrix X power and the wind speed characteristic matrix X wind are input into the gated recurrent unit network respectively. The gated recurrent unit network makes predictions based on the input data and finally outputs two prediction sequences, namely the power prediction sequence P power and the wind speed prediction sequence Pwind ; Then, according to the fan operation state curve, a boundary effect interval recognition function ST is constructed, and the predicted wind speed prediction sequence P wind is input into the boundary effect discrimination function ST for discrimination to obtain the boundary effect interval of wind power in the future moment; finally, the power prediction sequence P power , the wind speed prediction sequence P wind and the boundary effect interval recognition function ST are jointly input into the three-dimensional piecewise threshold function 3DST. The three-dimensional piecewise threshold function 3DST divides the power prediction sequence into those within the boundary effect interval and those outside the boundary effect interval, and independently processes the power prediction sequences in the two different intervals, so as to eliminate the influence of the boundary effect on the wind power prediction result in the wind power prediction process, avoid the inconsistent coupling relationship between the power sequence and the meteorological feature sequence, prevent the prediction model from being underfitted, and thus improve the prediction accuracy of wind power.

[0013] Preferably, in step S1, the meteorological feature data includes wind speed, temperature, and air pressure data at different heights, and the preprocessing includes performing min-max normalization on the power feature data and the wind speed, temperature, and air pressure data at different heights. Min-max normalization is a data standardization method that linearly maps data to a specific range. During the data collection process, the numerical ranges of different meteorological features may vary greatly. After min-max normalization, the numerical ranges of all power feature data and meteorological feature data are unified, avoiding certain features from dominating the prediction result due to large numerical values.

[0014] Preferably, in step S2, the power feature matrix X power and the wind speed feature matrix X wind are as follows:

[0015]

[0016]

[0017] where represents the matrix composed of power features from time t - 1 to t - n at the m-th height, represents the matrix composed of wind speed features from time t - 1 to t - n at the m-th height. The power feature matrix X power mainly provides historical power information, and the wind speed feature matrix X wind mainly provides meteorological information, reflecting the influence of wind speed and meteorological conditions on power. The two work together to comprehensively describe the operation state of the wind farm and improve the prediction accuracy.

[0018] Preferably, the power feature matrix X power and the wind speed feature matrix Xwind The mathematical expression is:

[0019]

[0020]

[0021] Wherein, , , , respectively represent the power, wind speed, temperature and air pressure at the m th height at the (t - n)th moment. Using parameters in multiple dimensions such as power, wind speed, temperature and air pressure at different moments and different heights to construct a feature matrix can make the collected data more comprehensive, thereby increasing the accuracy of the prediction result.

[0022] Preferably, in step S3, the gated recurrent unit network is a two - layer gated recurrent unit network, and the number of neurons in the two - layer gated recurrent unit network is different, and an activation function is provided. Setting a two - layer gated recurrent unit network can enhance the modeling ability and expression ability of the power prediction model, and at the same time achieve a good balance between computational efficiency and performance cost, and it is easier to maintain stability in deep learning training.

[0023] Preferably, the mathematical expression of the two - layer gated recurrent unit network is:

[0024]

[0025] Wherein, , , are weight matrices, corresponding to the input parts of the update gate , the reset gate and the candidate activation respectively, , , are weight matrices corresponding to the current input , and are used to calculate the hidden states of the update gate , the reset gate and the candidate activation respectively, , , are bias parameter matrices, corresponding to the weight biases of the update gate , the reset gate and the candidate activation respectively, is matrix multiplication, is the Sigmod function, is the reset gate, For the update gate, is the candidate state of the hidden layer at the current moment, is the current hidden state, is the hidden state at the previous moment, is the input state at the current moment. In the gated recurrent unit network, the inputs of the reset gate and the update gate are both the input at the current time step and the hidden state at the previous time step, and the output is calculated by a fully connected layer with the sigmoid function as the activation function. The reset gate helps to capture short-term dependencies in the time series, while the update gate helps to capture long-term dependencies in the time series. Therefore, the gated recurrent unit network shows better computational ability than other algorithms in the wind power prediction model.

[0026] Preferably, the number of neurons in the two-layer gated recurrent unit network is 4 and 8 respectively. When selecting the number of neurons, the complexity of the prediction task and the amount of data should be considered. A smaller number of neurons may cause the gated recurrent unit network to be unable to capture dependencies, while a larger number of neurons will increase the computational cost and may also result in overfitting, causing a decrease in prediction accuracy. Selecting 4 and 8 neurons for the two-layer gated recurrent unit network can better adapt to the amount of data in the wind power prediction process, and at the same time will not cause overfitting, and can balance the computational efficiency and prediction accuracy.

[0027] Preferably, in step S4, the mathematical expression of the boundary effect interval recognition function ST is:

[0028]

[0029] In the formula, is the wind speed prediction sequence P wind , is the cut-in wind speed set in the fan operation state curve, is the cut-out wind speed set in the fan operation state curve. The fan operation state curve of the wind turbine corresponds to the operation state of the fan in different wind speed intervals. When the wind speed V is less than the cut-in wind speed V in , the fan is in the shutdown state, and the output power is 0 at this time; when the input wind speed V is between the cut-in wind speed V in and the cut-out wind speed V out , the fan operates normally; when the input wind speed V is greater than the cut-out wind speed V out , the fan protection system starts to operate, and the fan is in the shutdown state, and the output power is 0 at this time. The boundary effect interval recognition function ST divides the boundary effect interval according to different input wind speeds, and different input wind speeds correspond to different function output values, so as to reduce the influence of the boundary effect on the prediction result.

[0030] Preferably, in step S5, the mathematical expression of the three-dimensional piecewise threshold function 3DST is as follows:

[0031]

[0032] In the formula, is the maximum rated output power of the wind turbine, is the wind speed when the wind turbine operates at the maximum rated power. When the input wind speed V is less than the cut-in wind speed V in or greater than the cut-out wind speed V out , the wind turbine remains in the shutdown state and the output power is 0; when the input wind speed V is greater than the cut-in wind speed V in and less than the wind speed V max when the wind turbine operates at the maximum rated power, the wind turbine operates normally according to the corresponding relationship between power and wind speed in the wind turbine operation state curve; when the input wind speed V is at the wind speed V max when the wind turbine operates at the maximum rated power and the cut-out wind speed V out When between, the maximum power tracking system inside the wind turbine starts to operate, and the wind turbine is in the maximum power tracking state. At this time, the output power is always the maximum rated power P max . The three-dimensional piecewise threshold function 3DST can perform collaborative processing on the power prediction sequence P power and the wind speed prediction sequence P wind output by the gated recurrent unit network, so as to eliminate the influence of the offshore wind power boundary effect on a single power prediction model.

[0033] The present invention also provides a wind power prediction system for implementing the wind power prediction method, including:

[0034] Data acquisition module: used to acquire the power characteristic data and meteorological characteristic data of the target wind farm, and transmit the power characteristic data and the meteorological characteristic data to the data processing module;

[0035] Data processing module: used to preprocess the power characteristic data and the meteorological characteristic data to obtain the power characteristic matrix X power and the wind speed characteristic matrix X wind ;

[0036] Gated recurrent unit network prediction module: based on the input power characteristic matrix X power and the wind speed characteristic matrix X wind , output the power prediction sequence P power and the wind speed prediction sequence P wind ;

[0037] Boundary effect interval identification module: used to construct the boundary effect interval identification function ST, and use the boundary effect interval identification function ST to process the wind speed prediction sequence P wind for identification, and obtain the boundary effect interval;

[0038] Three-dimensional piecewise prediction module: used to construct the three-dimensional piecewise threshold function 3DST, and input the wind speed prediction sequence P wind , power prediction sequence P power and the boundary effect interval identification function ST into the three-dimensional piecewise threshold function 3DST, and finally output the future power sequence.

[0039] The wind power prediction system of the present invention collects the power characteristic data and meteorological characteristic data of the wind farm through the data acquisition module, and transmits the data to the data processing module, where preprocessing is performed, and a power characteristic matrix X power and a wind speed characteristic matrix X wind are constructed. In the gated recurrent unit network prediction module, the power characteristic matrix X power and the wind speed characteristic matrix X wind are input into the gated recurrent network unit for prediction, and the obtained power prediction sequence P power and the wind speed prediction sequence P wind are input into the boundary effect interval identification module. The boundary effect interval identification module ST identifies the boundary effect interval. Finally, the three-dimensional piecewise prediction module further couples and corrects the power prediction sequence P power according to the result of the boundary effect interval identification function ST, and outputs the final future power sequence.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] 1. It can eliminate the influence of boundary effects on the wind power prediction results, keep the coupling relationship between the power sequence and the meteorological characteristic sequence consistent, and thus improve the prediction accuracy of wind power;

[0042] 2. It realizes the separate processing of the power prediction sequences in the boundary effect interval and the non-boundary effect interval, and prevents the prediction model from being underfitted. Description of the Drawings

[0043] Figure 1 is a flowchart of the offshore wind power prediction method for eliminating boundary effects;

[0044] Figure 2 is a comparison diagram of the prediction effects of the offshore wind power prediction method for eliminating boundary effects;

[0045] Figure 3 is a schematic diagram of the wind power prediction system for eliminating boundary effects. Specific embodiments

[0046] The present invention will be further described below in conjunction with specific embodiments.

[0047] Embodiment 1

[0048] This embodiment is the first embodiment of a method for predicting the power of offshore wind farms to eliminate boundary effects, including: S1. Obtain the power characteristic data and meteorological characteristic data of the target wind farm, and preprocess the power characteristic data and meteorological characteristic data to obtain the historical power sequence P, historical wind speed sequence WS, historical temperature sequence Tem, and historical air pressure sequence Pre;

[0049] S2. Construct a power feature matrix X using the historical power sequence P, historical wind speed sequence WS, historical temperature sequence Tem, and historical air pressure sequence Pre power and a wind speed feature matrix X wind ;

[0050] S3. Input the power feature matrix X power and the wind speed feature matrix X wind into the gated recurrent unit network respectively, and the gated recurrent unit network outputs the power prediction sequence P power and the wind speed prediction sequence P wind ;

[0051] S4. According to the fan operation state curve, construct a boundary effect interval identification function ST, input the wind speed prediction sequence P wind into the boundary effect interval identification function ST, and the boundary effect interval identification function ST identifies the boundary effect interval;

[0052] S5. According to the fan operation state curve and the boundary effect interval, construct a three-dimensional piecewise threshold function 3DST, input the wind speed prediction sequence P wind , the power prediction sequence P power and the boundary effect interval identification function ST into the three-dimensional piecewise threshold function 3DST, and the three-dimensional piecewise threshold function 3DST outputs the future power sequence.

[0053] The method for predicting the power of offshore wind farms to eliminate boundary effects in this embodiment, as Figure 1 shown, first collects the power characteristic data and meteorological characteristic data of the wind farm to be predicted, where the meteorological characteristic data includes wind speed, temperature, and air pressure. After preprocessing the power characteristic data and meteorological characteristic data, the historical power sequence P, historical wind speed sequence WS, historical temperature sequence Tem, and historical air pressure sequence Pre are obtained; subsequently, a power feature matrix X is constructed according to the historical power sequence P, historical wind speed sequence WS, historical temperature sequence Tem, and historical air pressure sequence Pre powerand the wind speed feature matrix X wind ; Then the obtained power feature matrix X power and the wind speed feature matrix X wind are input into the gated recurrent unit network. The gated recurrent unit network makes predictions based on the input data and finally outputs two prediction sequences, namely the power prediction sequence P power and the wind speed prediction sequence P wind ; Then, according to the fan operation state curve, a boundary effect interval recognition function ST is constructed. The predicted wind speed prediction sequence P wind is input into the boundary effect discrimination function ST for discrimination to obtain the wind power boundary effect interval in the future moment; Finally, the power prediction sequence P power , the wind speed prediction sequence P wind and the obtained boundary effect interval recognition function ST are jointly input into the three-dimensional piecewise threshold function 3DST. The three-dimensional piecewise threshold function 3DST divides the power prediction sequence into those within the boundary effect interval and those outside the boundary effect interval, and independently processes the power prediction sequences in the two different intervals, so as to eliminate the influence of the boundary effect on the wind power prediction result during the wind power prediction process.

[0054] In step S1, the meteorological feature data includes wind speed, temperature, and air pressure data at different heights. The preprocessing includes performing min-max normalization on the power feature data and the wind speed, temperature, and air pressure data at different heights. Min-max normalization is a simple and effective standardization method. In wind farm power prediction, normalization can accelerate model training, improve prediction accuracy, and ensure that different features (such as wind speed, temperature, air pressure) have the same weight in the prediction model.

[0055] In step S2, the power feature matrix X power and the wind speed feature matrix X wind are as follows:

[0056]

[0057]

[0058] where represents the matrix composed of power features from time t - 1 to t - n at the m-th height, represents the matrix composed of wind speed features from time t - 1 to t - n at the m-th height. The power feature matrix X power mainly provides historical power information, and the wind speed feature matrix X wind mainly provides meteorological information, reflecting the influence of wind speed and meteorological conditions on power. The two work together to comprehensively describe the operation state of the wind farm and improve prediction accuracy.

[0059] Power characteristic matrix X power and wind speed characteristic matrix X wind The mathematical expressions are as follows:

[0060]

[0061]

[0062] Wherein, , , , respectively represent the power, wind speed, temperature and air pressure at the m th height at the (t - n)th moment. Using parameters in multiple dimensions such as power, wind speed, temperature and air pressure at different moments and different heights to construct the characteristic matrix can make the collected data more comprehensive, thereby increasing the accuracy of the prediction result.

[0063] In step S4 of this embodiment, the mathematical expression of the boundary effect interval identification function ST is:

[0064]

[0065] Wherein, is the wind speed prediction sequence P wind , is the cut-in wind speed set in the fan operation state curve, is the cut-out wind speed set in the fan operation state curve. The fan operation state curve of the wind turbine corresponds to the operation state of the fan in different wind speed intervals. When the wind speed V is less than the cut-in wind speed V in , the fan is in the shutdown state, and the output power is 0 at this time; when the input wind speed V is between the cut-in wind speed V in and the cut-out wind speed V out , the fan operates normally; when the input wind speed V is greater than the cut-out wind speed V out , the fan protection system starts to operate, and the fan is in the shutdown state, and the output power is 0 at this time. The boundary effect interval identification function ST divides the boundary effect interval according to different input wind speeds, and different input wind speeds correspond to different function output values, thereby reducing the influence of the boundary effect on the prediction result.

[0066] In step S5 of this embodiment, the mathematical expression of the three-dimensional piecewise threshold function 3DST is:

[0067]

[0068] Wherein, is the maximum rated output power of the fan, is the wind speed when the fan operates at the maximum rated power. When the input wind speed V is less than the cut-in wind speed Vin or greater than the cut-out wind speed V out When this occurs, the wind turbine remains in the shutdown state and the output power is 0; when the input wind speed V is greater than the cut-in wind speed V in and less than the wind speed V when the wind turbine operates at the maximum rated power max the wind turbine operates normally according to the corresponding relationship between power and wind speed in the wind turbine operation state curve; when the input wind speed V is in the wind speed V when the wind turbine operates at the maximum rated power max and the cut-out wind speed V out When in between, the maximum power tracking system inside the wind turbine starts to operate, and the wind turbine is in the maximum power tracking state. At this time, the output power is always the maximum rated power P max . The three-dimensional piecewise threshold function 3DST can perform collaborative processing on the power prediction sequence P power output by the gated recurrent unit network and the wind speed prediction sequence P wind to eliminate the influence of the offshore wind power boundary effect on a single power prediction model.

[0069] The working principle of the offshore wind power prediction method for eliminating boundary effects in this embodiment is as follows: First, the power characteristic data and meteorological characteristic data are preliminarily processed to obtain a power characteristic matrix and a wind speed characteristic matrix. Then, the wind speed characteristic matrix sequence is input into the gated recurrent unit network to predict the future wind speed sequence. Next, the predicted wind speed sequence is input into the boundary effect discrimination function for discrimination to obtain the boundary effect interval and non-boundary effect interval of the wind power at future moments. Finally, the power prediction sequence and the wind speed prediction sequence are input into the three-dimensional piecewise threshold activation function together to realize the separate processing of the power prediction sequence in the boundary effect interval and non-boundary effect interval.

[0070] In this embodiment, according to the wind turbine operation state curve, the cut-in wind speed is set to 3 m / s and the cut-out wind speed is set to 25 m / s. Substituting into the above wind power prediction method for prediction, the prediction results are as Figure 2 shown, where the solid line represents the true value of the wind power, and the two dashed lines respectively represent the predicted values using the ordinary prediction method without considering the boundary effect of offshore wind power and the predicted values obtained by using the method in this embodiment considering the boundary effect. When not considering the boundary effect, the mean absolute error MAE is 11.0062, and when considering the boundary effect and using the method in this embodiment for prediction, the mean absolute error MAE is reduced to 7.2616. Therefore, the wind power prediction method in this embodiment can significantly eliminate the influence of the boundary effect during the offshore wind power prediction process, thereby effectively improving the accuracy of short-term wind power prediction.

[0071] Embodiment 2

[0072] This embodiment is the second embodiment of the offshore wind power prediction method for eliminating boundary effects. This embodiment is similar to the first embodiment, except that the gated recurrent unit network is a two-layer gated recurrent unit network, the number of neurons in the two-layer gated recurrent unit network is different, and an activation function is provided. Setting up a two-layer gated recurrent unit network can enhance the modeling ability and expression ability of the power prediction model, while achieving a good balance between computational efficiency and performance cost, and it is easier to maintain stability in deep learning training.

[0073] The mathematical expression of the two-layer gated recurrent unit network is:

[0074]

[0075] In the formula, , , are weight matrices, corresponding to the update gate , the reset gate and the candidate activation input parts, , , are weight matrices corresponding to the current input , respectively used to calculate the hidden states of the update gate , the reset gate and the candidate activation , , , are bias parameter matrices, corresponding to the update gate , the reset gate and the candidate activation weight biases, is matrix multiplication, is the Sigmod function, is the reset gate, is the update gate, is the candidate state of the hidden layer at the current time, is the current hidden state, is the hidden state at the previous time, is the input state at the current time step. In the gated recurrent unit network, the inputs of the reset gate and the update gate are both the input at the current time step and the hidden state at the previous time step, and the outputs are calculated by a fully connected layer with the activation function being the sigmoid function. The reset gate helps to capture short-term dependencies in the time series, and the update gate helps to capture long-term dependencies in the time series. Therefore, the gated recurrent unit network shows better computational ability than other algorithms in the wind power prediction model.

[0076] The number of neurons in the two - layer gated recurrent unit network is 4 and 8 respectively. When selecting the number of neurons, the complexity of the prediction task and the amount of data should be taken into account. A smaller number of neurons may cause the gated recurrent unit network to fail to capture dependencies, while a larger number of neurons will increase the computational cost and may lead to overfitting, resulting in a decrease in prediction accuracy. Selecting 4 and 8 neurons for the two - layer gated recurrent unit network can better adapt to the amount of data in the wind power prediction process, without causing overfitting, and can balance the computational efficiency and prediction accuracy.

[0077] Embodiment III

[0078] This embodiment is the first embodiment of the wind power prediction system, including: a data acquisition module: used to acquire the power characteristic data and meteorological characteristic data of the target wind farm, and transfer the power characteristic data and meteorological characteristic data to the data processing module;

[0079] A data processing module: used to pre - process the power characteristic data and meteorological characteristic data to obtain the power characteristic matrix X power and the wind speed characteristic matrix X wind ;

[0080] A gated recurrent unit network prediction module: based on the input power characteristic matrix X power and the wind speed characteristic matrix X wind , output the power prediction sequence P power and the wind speed prediction sequence P wind ;

[0081] A boundary effect interval identification module: used to construct the boundary effect interval identification function ST, and use the boundary effect interval identification function ST to identify the wind speed prediction sequence P wind to identify the boundary effect interval;

[0082] A three - dimensional piece - wise prediction module: used to construct the three - dimensional piece - wise threshold function 3DST, and input the wind speed prediction sequence P wind , the power prediction sequence P power and the boundary effect interval identification function ST into the three - dimensional piece - wise threshold function 3DST, and finally output the future power sequence.

[0083] The wind power prediction system in this embodiment, as Figure 3 shown, collects the power characteristic data and meteorological characteristic data of the wind farm through the data acquisition module, and transfers the data to the data processing module, where pre - processing is performed to obtain the historical power sequence P, historical wind speed sequence WS, historical temperature sequence Tem, and historical pressure sequence Pre; then, according to the historical power sequence P, historical wind speed sequence WS, historical temperature sequence Tem, and historical pressure sequence Pre, construct the power characteristic matrix Xpower and the wind speed feature matrix X wind , in the gated recurrent unit network prediction module, the power feature matrix X power and the wind speed feature matrix X wind are input into the gated recurrent network unit for prediction, and the obtained power prediction sequence P power and the wind speed prediction sequence P wind are input into the boundary effect interval recognition module. The boundary effect interval recognition module ST identifies the boundary effect interval. Finally, the three-dimensional piecewise prediction module further couples and corrects the power prediction sequence P power according to the result of the boundary effect interval recognition function ST, and outputs the final future power sequence.

[0084] In the specific content of the above specific implementation manner, each technical feature can be combined arbitrarily without contradiction. For the sake of concise description, not all possible combinations of the above technical features are described. However, as long as the combinations of these technical features do not exist in contradiction, they should be considered as the scope described in this specification.

[0085] Obviously, the above embodiments of the present invention are merely examples for clearly explaining the present invention, rather than limiting the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made on the basis of the above description. It is not necessary and impossible to enumerate all the implementation manners here. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. A method for predicting offshore wind power to eliminate boundary effects, characterized in that: The following steps are involved: S1. Acquire power characteristic data and meteorological characteristic data of a target wind farm, pre-process the power characteristic data and the meteorological characteristic data to obtain a historical power sequence P, a historical wind speed sequence WS, a historical temperature sequence Tem, and a historical air pressure sequence Pre; S2, using the historical power sequence P, the historical wind speed sequence WS, the historical temperature sequence Tem and the historical air pressure sequence Pre to construct a power feature matrix X power And wind speed feature matrix X wind ; S3, the power characteristic matrix X power And the wind speed characteristic matrix X wind are respectively input into the gated recurrent unit network, and the gated recurrent unit network outputs a power prediction sequence P power And the wind speed prediction sequence P wind ; S4, constructing a boundary effect interval identification function ST according to the wind turbine operation status curve, and converting the wind speed prediction sequence P wind The boundary effect interval identification function ST is input, and the boundary effect interval identification function ST identifies the boundary effect interval. The mathematical expression of the boundary effect interval identification function ST is: In the formula, is the wind speed prediction sequence P wind , is the cut-in wind speed set in the fan operation status curve, It is the cut-out wind speed set in the fan operation status curve; S5, constructing a three-dimensional segmented threshold function 3DST according to the wind turbine operating state curve and the boundary effect interval, and converting the wind speed prediction sequence P wind , power prediction sequence P power And the boundary effect interval identification function ST inputs the three-dimensional segmented threshold function 3DST, and the three-dimensional segmented threshold function 3DST outputs the future power sequence. The mathematical expression of the three-dimensional segmented threshold function 3DST is: In the formula, is the maximum rated output power of the fan, The wind speed is maintained at the maximum rated power of the fan.

2. The offshore wind power prediction method for eliminating boundary effects according to claim 1, characterized in that: In step S1, the meteorological characteristic data includes wind speed, temperature and air pressure data at different heights, and the preprocessing includes performing min-max normalization processing on the power characteristic data and the wind speed, temperature and air pressure data at different heights.

3. The offshore wind power prediction method for eliminating boundary effects according to claim 1, characterized in that: In step S2, the power characteristic matrix X power And wind speed feature matrix X wind for: In the formula, Indicates The matrix composed of the power characteristics from time t-1 to time tn at each height, Indicates The matrix consists of the wind speed characteristics at each height from time t-1 to time tn.

4. The offshore wind power prediction method for eliminating boundary effects according to claim 3, characterized in that: The power characteristic matrix X power And wind speed feature matrix X wind The mathematical expression is: In the formula, , , , Respectively represent Time Power, wind speed, temperature and air pressure at each altitude.

5. The offshore wind power prediction method for eliminating boundary effects according to any one of claims 1 to 4, characterized in that: In step S3, the gated recurrent unit network is a two-layer gated recurrent unit network, the two-layer gated recurrent unit network has different numbers of neurons and is provided with an activation function .

6. The offshore wind power prediction method for eliminating boundary effects according to claim 5, characterized in that: The mathematical expression of the two-layer gated recurrent unit network is: In the formula, , , is the weight matrix, corresponding to the update gate , Reset Gate and candidate activation The input part, , , For the current input The corresponding weight matrices are used to calculate the update gate , reset gate and candidate activation The hidden state of , , is the bias parameter matrix, corresponding to the update gate , reset gate and candidate activation The weight bias of is matrix multiplication, is the Sigmod function, To reset the gate, To update the gate, is the candidate state of the hidden layer at the current moment, is the current implicit state, is the implicit state at the previous moment, The input status at the current moment.

7. The offshore wind power prediction method for eliminating boundary effects according to claim 5, characterized in that: The number of neurons in the two-layer gated recurrent unit network is 4 and 8 respectively.

8. A wind power prediction system, used to implement the wind power prediction method according to any one of claims 1 to 7, characterized in that: include: Data acquisition module: used to acquire power characteristic data and meteorological characteristic data of the target wind farm, and transmit the power characteristic data and the meteorological characteristic data to the data processing module; Data processing module: used to pre-process the power characteristic data and the meteorological characteristic data to obtain the power characteristic matrix X power And wind speed feature matrix X wind ; Gated recurrent unit network prediction module: based on the input power feature matrix X power And wind speed feature matrix X wind , output power prediction sequence P power And the wind speed prediction sequence P wind ; Boundary effect interval identification module: used to construct a boundary effect interval identification function ST, and use the boundary effect interval identification function ST to identify the wind speed prediction sequence P wind Identify and obtain the boundary effect interval; Three-dimensional segmented prediction module: used to construct a three-dimensional segmented threshold function 3DST and convert the wind speed prediction sequence P wind , power prediction sequence P power And the boundary effect interval identification function ST inputs the three-dimensional segmented threshold function 3DST, and finally outputs the future power sequence.

Images