A wind power generation prediction method for a wind farm

Abstract

The application relates to the wind power generation technology field, in particular to a wind power generation capacity prediction method for a wind farm, which comprises the following steps: clustering generators based on environment data to obtain multiple environment clusters, wherein the environment data comprises a terrain category, an altitude and an effective power generation matrix, the effective power generation matrix is used for representing the power generation capacity of the generators under various wind farm information; a prediction model of each environment cluster is constructed; the prediction power generation capacity of each generator is obtained according to the prediction model of the environment cluster to which each generator in the wind farm belongs; and the sum of all the prediction power generation capacities is taken as the power generation capacity prediction result of the wind farm at the next adjacent time. Through the technical scheme, the accurate wind power generation capacity prediction result can be obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power generation, in particular to a wind power generation capacity prediction method for a wind farm. BACKGROUND

[0002] With the increasing attention to energy shortage and environmental problems, wind energy as a clean energy is continuously developed and utilized, and wind power generation technology is also continuously developed. However, due to the volatility and randomness of wind, the power generation capacity of wind power generation also has great randomness and uncertainty, which causes certain influence on the stable operation of the power grid. In order to improve the utilization rate of wind energy and improve the consumption level of wind energy, it is necessary to accurately predict the power generation capacity of the wind power station.

[0003] At present, the patent application file with the publication number CN111311022A discloses a power generation capacity prediction method, device, equipment and computer readable storage medium, wherein the method comprises: acquiring a to-be-predicted satellite image and a first shooting parameter corresponding to the to-be-predicted satellite image, and acquiring first outdoor wind power information corresponding to the to-be-predicted satellite image; the power generation capacity prediction model comprises a deep learning model and a power generation capacity regression model; the to-be-predicted satellite image is input into the deep learning model to obtain a plurality of first target rectangular frame information corresponding to the to-be-predicted satellite image and a probability that a first target frame corresponding to the first target rectangular frame information is a wind turbine; the first shooting parameter, the first outdoor wind power information, the first target rectangular frame information and the probability that the first target frame corresponding to the first target rectangular frame information is a wind turbine are input into the power generation capacity regression model to obtain the power generation capacity of the wind power station corresponding to the to-be-predicted satellite image.

[0004] The above method first acquires the position coordinates of the wind turbine in the to-be-predicted satellite image, and then predicts the power generation capacity of the wind power station corresponding to the to-be-predicted satellite image according to the first outdoor wind power information, the first shooting parameter and the position coordinates of the wind turbine. However, the above method only acquires the position coordinates of the wind turbine, ignores the influence of the orientation of the wind turbine and other information on wind power generation, and thus the prediction result of the wind power generation capacity is inaccurate. SUMMARY

[0005] In order to solve the technical problem that the prediction result of the wind power generation capacity is inaccurate, the present application provides a wind power generation capacity prediction method for a wind farm, which can acquire an accurate prediction result of the wind power generation capacity.

[0006] In a first aspect, the present application provides a wind farm wind power generation capacity prediction method, the prediction method comprising: clustering generators based on environmental data to obtain a plurality of environmental clusters, the environmental data comprising terrain category, altitude and effective power generation matrix, the effective power generation matrix being used to represent the power generation capacity of the generators under various wind farm information; constructing a prediction model for each environmental cluster, the input of the prediction model being the time series of wind farm information of each grid region within the neighborhood range of any generator, and the output being the predicted power generation capacity of the generator, the wind farm information comprising wind speed and the included angle between wind direction and generator yaw angle; obtaining the predicted power generation capacity of each generator according to the prediction model of the environmental cluster to which each generator in the wind farm belongs, and taking the sum of all predicted power generation capacities as the power generation capacity prediction result of the wind farm at the next adjacent time; wherein the method for obtaining the effective power generation matrix comprises: constructing an initial matrix with numerical values of 0, taking wind speed as rows and the included angle between wind direction and yaw angle as columns; statistically calculating the average power generation under each combination of wind speed and included angle within a historical time, and adding the average power generation to the numerical value at the corresponding position in the initial matrix; and normalizing all numerical values in the initial matrix to obtain the effective power generation matrix.

[0007] Considering that wind power generation is greatly affected by the environment, the power generation capacity of wind power generation under different environments is different, therefore, the generators are divided into a plurality of environmental clusters according to environmental data, the power generation capacity of each generator in an environmental cluster is basically consistent; further, a prediction model for each environmental cluster is constructed, the prediction model is used to predict the power generation capacity of the generators in the environmental cluster, the influence of environmental data on the power generation capacity is excluded, and the accuracy of power generation capacity prediction is improved; the predicted power generation capacity of each generator is predicted according to the prediction model of the environmental cluster to which each generator in the wind farm belongs, and the sum of all predicted power generation capacities is taken as the power generation capacity prediction result of the wind farm, so as to obtain an accurate prediction result of wind power generation capacity.

[0008] Preferably, obtaining the plurality of environmental clusters comprises: defining a clustering distance and an initial clustering number, and clustering the generators by using a K-prototypes algorithm to obtain a clustering result, the clustering distance being the sum of the distances of each environmental data between any two generators; adjusting the initial clustering number multiple times, and determining a target clustering number by using an elbow method, the clustering result corresponding to the target clustering number corresponding to the plurality of environmental clusters.

[0009] Preferably, the prediction model comprises a convolutional sub-model, a time series sub-model and a regression sub-model; the convolutional sub-model is used to extract features of the wind farm information of each grid region within the neighborhood range at any time to obtain a wind farm feature vector at each time; the time series sub-model is used to map the wind farm feature vector at each time into a wind farm time series vector; and the wind farm time series vector is input into the regression sub-model to obtain the predicted power generation capacity.

[0010] The convolution sub-model is used to extract the wind field information of each grid area within the neighborhood range (spatial dimension) at any time, and then the time series sub-model is used to extract the time series features of the wind field information (time dimension), obtaining a wind field time series vector that can characterize the changes in wind field information. The predicted power generation of the generator is output by combining the spatial and time dimensions.

[0011] Preferably, the convolution sub-model is a convolutional neural network, the timing sub-model is a recurrent neural network, and the regression sub-model is a fully connected neural network.

[0012] Preferably, the construction of the prediction model for each environmental cluster includes: collecting the historical time series of wind field information of each grid area within the neighborhood of any generator in the environmental cluster as an input sample, and taking the actual power generation at the next adjacent moment of the historical time series as the label of the input sample; inputting the input sample into the prediction model to obtain the output result, and iteratively training the prediction model according to the loss function, where the loss function is the mean square error function between the output result and the label; in response to the loss function being less than the preset loss or the number of iterations being greater than the preset number, completing the construction of the environmental cluster prediction model.

[0013] To ensure that the prediction model of each environmental cluster can accurately predict the power generation of the generator, the historical time series of wind field information of each grid area within the neighborhood of any generator in the environmental cluster is used to train the prediction model of the environmental cluster.

[0014] Preferably, the loss function also includes: calculating the Euclidean distance between the environmental data of each generator in the environmental cluster and the cluster center, and using the normalized Euclidean distance as the attention of the corresponding generator; and using the product of the mean square error function between the output result and the label and the attention as the loss function.

[0015] Since the environmental data of each generator in an environmental cluster may still be different, during the training process, higher attention is given to the generators that deviate from the cluster center in the environmental cluster to enhance the robustness of the prediction model within the environmental cluster, that is, to ensure that the prediction model can accurately predict the power generation of all generators in the environmental cluster.

[0016] Preferably, the loss function for: ; Environmental Cluster Internal generator The Euclidean distance between the environmental data and the cluster center, Environmental Cluster The sum of the Euclidean distances between the environmental data of all generators and the cluster center, Environmental Cluster Internal generator The mean square error function between the output result and the label corresponding to the input sample.

[0017] Preferably, the method for determining the environment cluster to which the generator belongs comprises: collecting environment data of any generator, calculating the similarity between the environment data and the clustering center of each environment cluster, and taking the environment cluster corresponding to the maximum similarity as the environment cluster to which the generator belongs.

[0018] Preferably, the similarity is negatively correlated with the Euclidean distance between the environment data and the clustering center of each environment cluster.

[0019] Preferably, the neighborhood range is a rectangular region centered on the generator, the rectangular region includes a plurality of grid regions, and the wind field information of each grid region in the neighborhood range is a two-channel image data.

[0020] The technical solution of the present application has the following beneficial technical effects:

[0021] Firstly, considering that wind power generation is greatly affected by the environment, the power generation capacity of wind power generation is different under different environments, therefore, the generators are divided into a plurality of environment clusters according to the environment data, and the power generation capacity of the generators in one environment cluster is basically consistent; further, a prediction model of each environment cluster is constructed, the prediction model is used to predict the power generation of the generators in the environment cluster, the influence of the environment data on the power generation capacity is excluded, and the accuracy of the power generation prediction is improved; the predicted power generation of each generator is predicted according to the prediction model of the environment cluster to which each generator belongs in the wind farm, the sum of all the predicted power generations is taken as the power generation prediction result of the wind farm, and an accurate wind power generation prediction result is obtained.

[0022] The environment data includes terrain category, altitude and effective power generation matrix, the terrain category and the altitude distinguish the power generation capacity of the generators from the perspective of geographical environment, the effective power generation matrix distinguishes the power generation capacity of the generators from the perspective of wind field information, the accurate division of the environment cluster is realized, and the power generation capacity of the generators in one environment cluster is basically consistent. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 is a flowchart of a wind power generation prediction method of a wind farm according to an embodiment of the present application.

[0024] Figure 2 is a structure diagram of a prediction model according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0026] According to a first aspect of the present application, the present application provides a wind farm wind power generation capacity prediction method. Figure 1 is a flow chart of a wind farm wind power generation capacity prediction method according to an embodiment of the present application. As shown in the figure, the wind farm wind power generation capacity prediction method comprises steps S101 to S103, which are described in detail below. Figure 1

[0027] S101, clustering the generators based on environmental data to obtain a plurality of environmental clusters, the environmental data comprising terrain category, altitude and effective power generation matrix, the effective power generation matrix being used to represent the power generation capacity of the generators under each wind farm information.

[0028] In one embodiment, since wind power generation is greatly affected by the environment, the power generation capacity of wind power generation under different environments is different, in order to ensure the accuracy of the power generation capacity prediction result, it is necessary to cluster the generators according to the environmental data to obtain a plurality of environmental clusters, if two generators belong to the same environmental cluster, the power generation capacity of the two generators is the same, a prediction model will be constructed for each environmental cluster to exclude the influence of environmental data on the power generation capacity and improve the accuracy of power generation capacity prediction.

[0029] The environmental data comprises terrain category, altitude and effective power generation matrix, wherein the terrain category comprises plain, hilly and gentle hill, the terrain category and altitude distinguish the power generation capacity of the generators from the perspective of geographical environment, and the effective power generation matrix distinguishes the power generation capacity of the generators from the perspective of wind farm information, in other words, the effective power generation matrix can represent the power generation capacity of the generators under different wind farm information.

[0030] Specifically, the method for obtaining the effective power generation matrix comprises: constructing an initial matrix with numerical values of 0, with wind speed as rows and the included angle between wind direction and yaw angle as columns; calculating the average power generation under each combination of wind speed and included angle in the historical time, and adding the average power generation to the numerical value in the corresponding position of the initial matrix; and normalizing all numerical values in the initial matrix to obtain the effective power generation matrix. Wherein, the normalization can be realized by calculating the ratio of any numerical value to the sum of the numerical value in the initial matrix, which plays a role in eliminating the dimension.

[0031] Wherein, the wind speed is divided into equal intervals with 1 meter per second as the interval interval, and the wind speed interval from 0 meters per second to 30 meters per second is divided into 30 interval segments, i.e. the number of rows of the initial matrix is 30 rows; further, the included angle between wind direction and yaw angle ranges from 0 degrees to 180 degrees, with 5 degrees as the interval interval, and the included angle value range is divided into 36 interval segments, i.e. the number of columns of the initial matrix is 36 columns, i.e. the initial matrix is a 30 row 36 column 0 matrix. The yaw angle represents the orientation information of the generator, which can be directly obtained by using the angle sensor deployed on the generator.​

[0032] It can be understood that one value in the initial matrix corresponds to one wind field information, for example, the 5th row and the 10th column corresponds to the wind field information that the wind speed is 5 meters per second and the included angle between the wind direction and the yaw angle is 45 degrees to 50 degrees.

[0033] In one embodiment, the obtaining the plurality of environment clusters comprises: defining a clustering distance and an initial clustering number, and clustering each generator by using a K-prototypes algorithm to obtain a clustering result, the clustering distance being the sum of distances between each environment data of any two generators; adjusting the initial clustering number multiple times, and determining a target clustering number by using an elbow method, the clustering result of the target clustering number corresponding to the plurality of environment clusters.

[0034] In the embodiment, the K-prototypes algorithm is a clustering algorithm for processing mixed data, and can be used to process numerical data and category data at the same time. In the embodiment, the terrain category is category data, and the altitude and the effective power generation matrix are numerical data, so the K-prototypes algorithm is used for clustering. The initial clustering number is 2. The number of environment clusters is equal to the target clustering number, and each environment cluster includes a plurality of generators. The average value of the environment data of the plurality of generators in the environment cluster is taken as the clustering center of the environment cluster. The clustering center is used to determine the environment cluster to which each generator in the wind farm belongs.

[0035] In this way, the generators are divided into a plurality of environment clusters according to the environment data, and the power generation capacities of the generators in one environment cluster are basically consistent.

[0036] S102, constructing a prediction model of each environment cluster, the input of the prediction model being a time sequence of wind field information of each grid region in a neighborhood range of any generator, and the output being a predicted power generation of the generator, the wind field information including a wind speed and an included angle between a wind direction and a yaw angle of the generator.

[0037] In one embodiment, since the power generation capacities of the generators in one environment cluster are basically consistent, the prediction model of each environment cluster is constructed, the difference in the power generation capacities of the generators caused by the environment clusters is excluded, and the accuracy of the power generation prediction result is improved.

[0038] The input of the prediction model is a time sequence of wind field information of each grid region in a neighborhood range of any generator, and the output is a predicted power generation of the generator. The wind field information of one grid region includes a wind speed and an included angle between a wind direction and a yaw angle of the generator in the grid region. The neighborhood range is a rectangular region with the generator as the center. One rectangular region includes a plurality of grid regions. The wind field information of each grid region in the neighborhood range can be regarded as a two-channel image data.

[0039] For example, the grid area is 10m*10m, one wind speed sensor and one wind direction sensor are deployed in one grid area to collect the wind speed and wind direction in the grid area, and further combined with the yaw angle of the generator, the wind speed and the included angle between the wind direction and the yaw angle of the generator in the grid area can be obtained. If a rectangular area includes 9 grid areas, the wind field information of each grid area in the neighborhood range is a 3*3 two-channel image data, wherein one channel image includes the wind speed data of the grid area where the generator is located and the other 8 grid areas in the neighborhood range; and the other channel image includes the included angle between the wind direction and the yaw angle of the generator in the grid area where the generator is located and the other 8 grid areas.

[0040] According to the same method, the wind field information of each grid area in the neighborhood range at each moment can be obtained, and the wind field information of each grid area in the neighborhood range is arranged in chronological order, so that the time sequence of the wind field information of each grid area in the neighborhood range is obtained. The time sequence can represent the change of the wind field information in the neighborhood range of the generator, and the time sequence includes the current moment and a plurality of historical moments before the current moment.

[0041] Figure 2 is a structural diagram of a prediction model according to an embodiment of the present application. The prediction model includes a convolutional sub-model, a time sequence sub-model, and a regression sub-model. The convolutional sub-model is used to extract features of the wind field information of each grid area in the neighborhood range at any moment to obtain a wind field feature vector at each moment. The time sequence sub-model is used to map the wind field feature vector at each moment to a wind field time sequence vector. The wind field time sequence vector is input into the regression sub-model to obtain a predicted power generation.

[0042] The convolutional sub-model can use existing convolutional neural networks such as ResNet, VGGNet, or ShuffleNet, the time sequence sub-model can use existing recurrent neural networks such as LSTM or Transformer, and the regression sub-model uses a fully connected neural network.

[0043] In one embodiment, in order to ensure that the prediction model of each environment cluster can accurately predict the power generation of the generator, the prediction model of each environment cluster needs to be trained. The model structures of the prediction models of each environment cluster are the same, but the training processes are different. Specifically, the construction of the prediction model of each environment cluster includes: collecting a historical time sequence of the wind field information of each grid area in the neighborhood range of any generator in the environment cluster as an input sample, and collecting the actual power generation at the next adjacent moment of the historical time sequence as a label of the input sample; inputting the input sample into the prediction model to obtain an output result, and iteratively training the prediction model according to a loss function, wherein the loss function is a mean square error function between the output result and the label; and in response to the loss function being less than a preset loss or the number of iterations being greater than a preset number, the construction of the environment cluster prediction model is completed.

[0044] wherein the preset loss is 0.01 and the preset number of times is 500.

[0045] In another embodiment, in one environment cluster, the environment data of each generator still has differences. In order to enhance the robustness of the prediction model in the environment cluster, that is, the prediction model can accurately predict the power generation of all generators in the environment cluster, in the training process, the generator deviating from the cluster center in the environment cluster is given higher attention. The loss function further comprises: calculating the Euclidean distance between the environment data of each generator in the environment cluster and the cluster center, and taking the normalized Euclidean distance as the attention degree of the corresponding generator; and taking the product of the mean square error function between the output result and the label and the attention degree as the loss function.

[0046] Specifically, the loss function is:

[0047] ; The Euclidean distance between the environment data of the generator in the environment cluster and the cluster center, The sum of the Euclidean distances between the environment data of all generators in the environment cluster and the cluster center, The mean square error function between the output result corresponding to the input sample of the generator in the environment cluster and the label.

[0048] In this way, the prediction model of each environment cluster is obtained; the prediction model can accurately predict the power generation of all generators in the corresponding environment cluster.

[0049] S103, obtaining the predicted power generation of each generator according to the prediction model of the environment cluster to which each generator in the wind farm belongs, and taking the sum of all predicted power generations as the power generation prediction result of the wind farm at the next adjacent time.

[0050] In one embodiment, the method for determining the environment cluster to which the generator belongs comprises: collecting the environment data of any generator, calculating the similarity between the environment data and the cluster center of each environment cluster, and taking the environment cluster corresponding to the maximum similarity as the environment cluster to which the generator belongs.

[0051] The similarity is a similarity calculation method based on the Euclidean distance, and the similarity between the environment data and the cluster center of any environment cluster is negatively correlated with the Euclidean distance between the environment data and the cluster center of the environment cluster.

[0052] ​​The wind power plant comprises a plurality of generators, and each generator belongs to an environment cluster according to environment data of the generator, and then a predicted power generation of each generator at a next adjacent time is accurately obtained according to a prediction model of the environment cluster to which the generator belongs; and a sum of all the predicted power generations is taken as a power generation prediction result of the wind power plant at the next adjacent time, so that accurate prediction of wind power generation of the wind power plant is realized.

[0053] It should be noted that, for those skilled in the art of the present neighborhood, without departing from the concept of the present application, a number of modifications and improvements can be made, which all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A wind farm wind power generation amount prediction method characterized by comprising: The prediction method comprises: clustering each generator based on environment data to obtain a plurality of environment clusters, the environment data comprising a terrain category, an altitude, and an effective power generation matrix, the effective power generation matrix being used to represent the power generation capacity of the generator under each wind farm information; constructing a prediction model of each environment cluster, the input of the prediction model being a time series of wind farm information of each grid area in a neighborhood range of any generator, the output being a predicted power generation of the generator, the wind farm information comprising a wind speed and an included angle of a wind direction and a yaw angle of the generator; obtaining the predicted power generation of each generator according to the prediction model of the environment cluster to which each generator belongs in the wind farm, and taking the sum of all the predicted power generations as a prediction result of the power generation of the wind farm at a next adjacent time; wherein the method for obtaining the effective power generation matrix comprises: constructing an initial matrix with a value of 0, taking the wind speed as a row and the included angle of the wind direction and the yaw angle as a column; statistically calculating the average power generation under each combination of the wind speed and the included angle in a historical time, and adding the average power generation to the value in the corresponding position of the initial matrix; and normalizing all the values in the initial matrix to obtain the effective power generation matrix; the method for constructing the prediction model of each environment cluster comprises: collecting a historical time series of wind farm information of each grid area in a neighborhood range of any generator in the environment cluster as an input sample, taking the actual power generation at a next adjacent time of the historical time series as a label of the input sample; inputting the input sample into the prediction model to obtain an output result, iteratively training the prediction model according to a loss function, the loss function being a mean square error function between the output result and the label; and in response to the loss function being less than a preset loss or the number of iterations being greater than a preset number, completing the construction of the prediction model of the environment cluster; the loss function further comprises: calculating the Euclidean distance between the environment data of each generator in the environment cluster and a clustering center, taking the normalized Euclidean distance as an attention degree of the corresponding generator; and taking the product of the mean square error function between the output result and the label and the attention degree as the loss function; Loss Function for: ; Environmental Cluster Internal generator The Euclidean distance between the environmental data and the cluster center, Environmental Cluster The sum of the Euclidean distances between the environmental data of all generators and the cluster center, Environmental Cluster Internal generator The mean square error function between the output result and the label corresponding to the input sample.

2. A method for predicting wind power generation in a wind farm according to claim 1, characterized in that: the method for obtaining the plurality of environment clusters comprises: defining a clustering distance and an initial clustering number, and clustering each generator by using a K-prototypes algorithm to obtain a clustering result, the clustering distance being the sum of the distances of each environment data between any two generators; adjusting the initial clustering number multiple times, and determining a target clustering number by using an elbow method, the clustering result corresponding to the target clustering number corresponding to the plurality of environment clusters.

3. The wind power generation amount prediction method of a wind farm according to claim 1, characterized in that, The prediction model comprises a convolutional sub-model, a time series sub-model, and a regression sub-model; the convolutional sub-model is used for feature extraction of the wind farm information of each grid area in a neighborhood range at any time to obtain a wind farm feature vector at each time; the time series sub-model is used for mapping the wind farm feature vector at each time into a wind farm time series vector; the wind farm time series vector is input into the regression sub-model to obtain the predicted power generation.

4. The wind power generation amount prediction method of a wind farm according to claim 3, characterized in that, The convolutional sub-model is a convolutional neural network, the time series sub-model is a recurrent neural network, and the regression sub-model is a fully connected neural network.

5. The wind power generation amount prediction method of a wind farm according to claim 1, characterized in that, The method for determining the environment cluster to which the generator belongs comprises: The environment data of any generator is collected, the similarity between the environment data and the clustering center of each environment cluster is calculated, and the environment cluster corresponding to the maximum similarity is taken as the environment cluster to which the generator belongs.

6. The wind farm wind power generation amount prediction method according to claim 5, characterized in that, The similarity is negatively correlated with the Euclidean distance between the environment data and the clustering center of each environment cluster.

7. The wind power generation amount prediction method of a wind farm according to claim 1, characterized in that, The neighborhood range is a rectangular region centered on the generator, the rectangular region includes a plurality of grid regions, and the wind field information of each grid region in the neighborhood range is a two-channel image data.

Images