CN117175535A - Wind power group power prediction method, system, equipment and medium - Google Patents
Wind power group power prediction method, system, equipment and medium Download PDFInfo
- Publication number
- CN117175535A CN117175535A CN202210567499.9A CN202210567499A CN117175535A CN 117175535 A CN117175535 A CN 117175535A CN 202210567499 A CN202210567499 A CN 202210567499A CN 117175535 A CN117175535 A CN 117175535A
- Authority
- CN
- China
- Prior art keywords
- data
- wind power
- weather forecast
- weather
- wind
- 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
- 238000000034 method Methods 0.000 title claims abstract description 87
- 238000012549 training Methods 0.000 claims abstract description 41
- 238000000547 structure data Methods 0.000 claims description 149
- 238000013528 artificial neural network Methods 0.000 claims description 28
- 238000010586 diagram Methods 0.000 claims description 25
- 230000035945 sensitivity Effects 0.000 claims description 16
- 238000005070 sampling Methods 0.000 claims description 15
- 238000000605 extraction Methods 0.000 claims description 14
- 238000012512 characterization method Methods 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 12
- 230000008859 change Effects 0.000 claims description 10
- 230000007246 mechanism Effects 0.000 claims description 10
- 238000013527 convolutional neural network Methods 0.000 claims description 9
- 238000002790 cross-validation Methods 0.000 claims description 8
- 238000012546 transfer Methods 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 2
- 230000001131 transforming effect Effects 0.000 claims description 2
- 230000003044 adaptive effect Effects 0.000 claims 1
- 238000013135 deep learning Methods 0.000 abstract description 5
- 238000013136 deep learning model Methods 0.000 abstract description 3
- 238000013507 mapping Methods 0.000 description 11
- 238000012360 testing method Methods 0.000 description 11
- 238000010276 construction Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 238000010248 power generation Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 230000006872 improvement Effects 0.000 description 3
- 230000001788 irregular Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a wind power group power prediction method, a system, equipment and a medium, which comprise the steps of obtaining grid numerical weather forecast of time to be predicted covering the space range of a wind power group and single-point numerical weather forecast of time to be predicted of each wind power place in the wind power group at a geographic position; inputting the grid numerical weather forecast and the single-point numerical weather forecast into an weather map structure model of the wind power group to obtain a weather map structure of the wind power group; and inputting the weather map structure into a pre-trained deep learning model, and outputting the predicted power of the wind power group to be predicted. According to the method, the multisource numerical weather forecast is taken as a basic data source, the position distribution of wind power stations in the area is considered, an weather map structure is constructed, the characteristic abstract capacity of high-dimensional data is achieved through deep learning, the power of the multi-wind power stations in the area is finally predicted through model training, and the modeling efficiency is improved on the premise that the prediction accuracy of the power of the wind power clusters is guaranteed.
Description
Technical Field
The invention relates to the technical field of new energy power generation of power systems, in particular to a wind power group power prediction method, a system, equipment and a medium.
Background
The wind power generation has strong volatility and randomness, brings great challenges to the operation of a power system, and predicts the output power of a wind power plant, so that the wind power generation is an important means for relieving the peak regulation and frequency modulation pressure of the power system and promoting the wind power absorption capacity. The wind power prediction system is established to obviously promote the wind power consumption. Meanwhile, as the penetration rate of wind power is continuously increased, under the constraint of uncertainty bearable by a power system, continuous improvement requirements are provided for wind power prediction accuracy.
At present, an superposition method is mainly adopted for power prediction of regional wind power stations, and the main thought is that each wind power station in a cluster is respectively established with a meteorological-power correlation model, and then the predicted power of each wind power station is summed to obtain the total power of the cluster. For a reference wind farm for wind power cluster power prediction, i.e. a single wind farm, its prediction model is usually constructed by using a physical method, a statistical method and a combination method. Firstly, preprocessing numerical weather forecast data, actual measurement data and historical power for the wind power plant, namely carrying out processes such as missing value and abnormal value removal on the data; secondly, taking a numerical weather forecast or a combination of the numerical weather forecast and measured data as input and historical power generation as output, and establishing a weather-power mapping model by using the method; and finally, preprocessing meteorological data at the moment to be predicted, and inputting the preprocessed meteorological data into the mapping model to obtain the predicted power of the wind power station. And establishing a prediction model for each wind power plant in the cluster according to the operation, and further adding the prediction power to obtain the cluster output.
Firstly, the existing wind power cluster power prediction technology generally establishes a weather-power mapping model by utilizing a mathematical algorithm for each wind power plant in a cluster, and then carries out simple accumulation summation on prediction results so as to obtain the total power of the wind power cluster. The premise of implementing the method is that a power prediction model is required to be built for each wind farm in the cluster, the modeling efficiency is low, and a great deal of manpower, material resources and time cost are required for links such as model building, deployment and maintenance. Secondly, a prediction model established for each wind farm in the cluster is usually based on a shallow network such as an artificial neural network and a support vector machine, the dimension of processed data is extremely limited, the knowledge of field experts is lacked, and a mapping model of weather forecast and power with high dimension values is difficult to establish, namely, the deep mapping relation between weather forecast information and power with available numerical values is not fully excavated. Finally, when the weather-power correlation model is established, the capturing capability of space-time characteristics of high-dimensional complex data by deep learning is not fully utilized, namely, the space resource correlation characteristic and the wind power output time sequence correlation are not fully considered, so that the prediction effect is sometimes poor.
When the weather-power depth mapping model is constructed for wind power cluster prediction, the organization form of weather data of a regular grid is the same as the image format in the field of computer vision, and the airspace feature extraction can be performed by using a convolutional neural network. However, strictly speaking, the general meteorological data should be defined on an irregular grid of non-European geometric space, influenced by the shape of the earth itself and the terrain of the ground. The graph structure may be used to describe an irregular grid, defining spatial domain feature extraction similar to a normal convolution with the convolution operator on the graph structure. The irregular grid defines vertex information in the graph structure, i.e. each wind farm in the area is considered as a vertex of the graph, and the similarity between any two points is considered as the weight of the edge connecting the two vertices. For connection relation research among vertexes, a K nearest neighbor graph is mostly adopted to construct a similarity graph at present, and similarity among wind power plants is evaluated based on Euclidean distance similarity measurement improved by a density sensitive item.
Disclosure of Invention
In order to solve the problems of low modeling efficiency and large deviation of the prediction effect of the existing wind power prediction method, the invention provides a method for establishing a wind power plant graph structure model, which comprises the following steps:
taking each wind power plant in a wind power group as a node, and converting single-point numerical weather forecast data of each wind power plant into graph structure data by adopting a density self-adaptive algorithm based on shared nearest neighbors;
extracting airspace characteristic data of grid numerical weather forecast data covering the spatial range of the wind power group and airspace characteristic data of the map structure data by adopting a convolution network;
and characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data on the space with the same meteorological semantic information to obtain the wind power plant map structure model.
Preferably, the converting the single-point numerical weather forecast data of each wind farm into the graph structure data by using each wind farm in the wind farm group as a node and adopting a density self-adaptive algorithm based on shared nearest neighbor includes:
selecting an optimal weather sensitivity factor affecting the power prediction performance of the wind power group from single-point numerical weather forecast data of each wind power plant by using a K-fold cross validation method;
And taking each wind power plant as a node, taking the optimal weather sensitivity factor data corresponding to the single-point numerical weather forecast data as the characteristics of the node, adopting a density self-adaptive algorithm based on shared nearest neighbors to determine the connection weight between the nodes, forming a data set of the graph structure by the geographic position data of each wind power plant, the characteristics of the nodes and the connection weight between the nodes, and converting the single-point numerical weather forecast data into the graph structure data.
Preferably, the extracting airspace feature data of the grid-type numerical weather forecast data covering the spatial range of the wind power group and airspace feature data of the graph structure data by adopting a convolution network includes:
taking the union of the optimal weather sensitive factor data of each wind power plant as the optimal weather element of the grid numerical weather forecast data, and extracting airspace characteristic data of the optimal weather element by adopting a convolution self-coding decoding network; and extracting airspace characteristic data of the graph structure data by adopting a graph convolution neural network.
Preferably, the extracting the spatial domain feature data of the graph structure data by using the graph convolution neural network includes: and constructing a filter in a Fourier domain by adopting the graph convolution neural network, sampling the graph structural data by using the filter to obtain sampling data, and extracting spatial domain characteristic data of the graph structural data from the sampling data.
Preferably, the characterizing the airspace feature data of the gridding numerical weather forecast data and the airspace feature data of the map structure data in the space with the same weather semantic information to obtain the wind power plant map structure model includes:
characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data in a space with the same weather semantic information by adopting an embedded mode to obtain weather tensor data;
and encoding the meteorological tensor data and the geographic position data of each wind power plant to obtain the wind power plant graph structure model.
Based on the same thought, the invention also provides a system for establishing the wind power plant graph structural model, which comprises the following steps:
the graph structure module is used for converting single-point numerical weather forecast data of each wind power plant into graph structure data by taking each wind power plant in a wind power group as a node and adopting a density self-adaptive algorithm based on shared nearest neighbors;
the extraction module is used for extracting gridding numerical weather forecast data covering the spatial range of the wind power group and airspace characteristic data of the map structure data by adopting a convolution network;
And the characterization module is used for characterizing the airspace characteristic data of the gridding numerical weather forecast data and the airspace characteristic data of the graph structure data on the space with the same meteorological semantic information to obtain a wind power plant graph structure model of the wind power group.
Preferably, the graph structure module includes:
the selecting unit is used for selecting the optimal weather sensitivity factor affecting the power prediction performance of the wind power group from the single-point numerical weather forecast data of each wind power plant by using a K-fold cross validation method;
the transformation unit is used for taking each wind power plant as a node, taking the optimal weather sensitivity factor data corresponding to the single-point numerical weather forecast data as the characteristics of the node, and transforming the single-point numerical weather forecast data into the graph structure data by adopting a density self-adaptive algorithm based on the shared nearest neighbor.
Preferably, the extraction module includes:
the convolution self-coding decoding network unit is used for extracting the airspace characteristic data of the meshing numerical weather forecast data by adopting a convolution self-coding decoding network;
and the graph convolution neural network unit is used for extracting airspace characteristic data of the graph structure data by adopting the graph convolution neural network.
Preferably, the characterization module includes:
the embedding unit is used for representing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data in a space with the same weather semantic information in an embedding mode to obtain weather tensor data;
and the encoding unit is used for encoding the meteorological tensor data and the geographical position data of each wind power plant to obtain the wind power plant diagram structure model.
Based on the same thought, the invention also provides a training method of the wind power group power prediction model, which comprises the following steps:
acquiring historical meteorological data of a wind power group and power data corresponding to the historical meteorological data, and acquiring historical meteorological graph structure data of the wind power group by utilizing a pre-constructed wind power plant graph structure model;
taking the historical weather map structure data as input, and taking power data corresponding to the historical weather map structure data as output to train the wind power group power prediction model to obtain a trained wind power group power prediction model;
the wind power group prediction model comprises a graph convolutional neural network and a gating circulation network;
The pre-constructed wind power plant diagram structure model is constructed by adopting the method for constructing the wind power plant diagram structure model.
Preferably, the training the wind power group power prediction model by taking the historical weather map structure data as input and taking the power data corresponding to the historical weather map structure data as output, and obtaining the trained wind power group power prediction model includes:
extracting airspace characteristic time sequence data of the historical weather map structure data by adopting the map convolution neural network, and recalculating connection weights among nodes of the historical weather map structure by adopting a space-time attention mechanism;
and adopting the gating circulation network to transmit internal information of the airspace characteristic time sequence data, capturing dynamic changes of the airspace characteristic time sequence data, forming a corresponding relation between the power data and the dynamic changes, and obtaining the trained wind power group power prediction model.
Based on the same thought, the invention also provides a training system of the wind power group power prediction model, which comprises the following steps:
the acquisition module is used for acquiring historical meteorological data of a wind power group and power data corresponding to the historical meteorological data, and acquiring historical meteorological graph structure data of the wind power group by utilizing a pre-constructed wind power plant graph structure model;
The training module is used for taking the historical weather pattern structure data as input, taking the power data corresponding to the historical weather pattern structure data as output and training the wind power group power prediction model to obtain a trained wind power group power prediction model;
the wind power plant graph structure model is built by adopting the building method provided by the invention, and the wind power group prediction model comprises a graph convolutional neural network and a gating circulation network.
Preferably, the training module includes:
an attention mechanism unit for recalculating connection weights between nodes of the historical weather pattern structure using a spatiotemporal attention mechanism;
and the corresponding relation acquisition unit is used for carrying out internal information transfer on the historical weather map structure data by adopting the gating circulation network, capturing the dynamic change of the historical weather map structure data and forming the corresponding relation between the power data and the dynamic change.
Based on the same thought, the invention also provides a wind power group power prediction method, which comprises the following steps:
acquiring grid numerical weather forecast data of time to be predicted covering a space range of a wind power group and single-point numerical weather forecast data of time to be predicted of each wind power plant in the wind power group;
Acquiring weather map structure data of the time to be predicted of the wind power group by utilizing a pre-constructed wind power plant map structure model based on the grid numerical weather forecast data and the single-point numerical weather forecast data;
based on the weather map structure data, predicting the power of the wind power group to be predicted time by using a pre-trained wind power group power prediction model;
the method for building the wind power plant diagram structure model comprises the steps of building a wind power plant diagram structure model, wherein the pre-built wind power plant diagram structure model is built by adopting the building method of the wind power plant diagram structure model provided by the invention; the wind power group power prediction model is obtained by adopting the training method of the wind power group power prediction model.
Preferably, the obtaining, based on the gridding numerical weather forecast data and the single-point numerical weather forecast data, weather map structure data of the time to be predicted of the wind power group by using a pre-constructed wind power plant map structure model includes:
converting the single-point numerical weather forecast data into graph structure data by the pre-constructed wind power plant graph structure model by adopting a density self-adaptive algorithm based on shared nearest neighbor, extracting airspace characteristic data of the gridding numerical weather forecast data and the graph structure data by adopting the pre-constructed wind power plant graph structure model, and representing the airspace characteristic data of the gridding numerical weather forecast data and the airspace characteristic data of the graph structure data on a space with the same meteorological semantic information to obtain the weather graph structure data.
Preferably, the predicting the power of the wind power group to be predicted time by using a pre-trained wind power group power prediction model based on the weather map structure data includes:
inputting the weather pattern structure data into the pre-trained wind power group power prediction model, extracting airspace feature time sequence data of the weather pattern structure data by using the pre-trained wind power group prediction model, carrying out internal information transfer on the airspace feature time sequence data, capturing dynamic changes of the airspace feature time sequence data, and outputting power of the wind power group to be predicted time based on the dynamic changes.
Based on the same thought, the invention also provides a wind power group power prediction system, which comprises:
the acquisition module is used for acquiring grid numerical weather forecast data of the time to be predicted covering the space range of the wind power group and single-point numerical weather forecast data of the time to be predicted of each wind power plant in the wind power group;
the weather map structure module is used for obtaining weather map structure data of the time to be predicted of the wind power group by utilizing a pre-constructed wind power plant map structure model based on the grid numerical weather forecast data and the single-point numerical weather forecast data;
And the prediction module is used for predicting the power of the time to be predicted of the wind power group by using a pre-trained wind power group power prediction model based on the weather map structure data.
The method for building the wind power plant diagram structure model comprises the steps of building a wind power plant diagram structure model, wherein the pre-built wind power plant diagram structure model is built by adopting the building method of the wind power plant diagram structure model provided by the invention; the pre-trained wind power group power prediction model is obtained by adopting the training method of the wind power group power prediction model.
Preferably, the aerial image structure module comprises an image structure unit, an extraction unit and a characterization unit;
the graph structure unit is used for converting single-point numerical weather forecast data of the time to be predicted of each wind power plant into graph structure data by taking each wind power plant in a wind power group as a node and adopting a density self-adaptive algorithm based on shared nearest neighbor;
the extraction unit is used for extracting airspace feature data of grid-type numerical weather forecast data covering the time to be predicted of the wind power group space range and airspace feature data of the map structure data by adopting a convolution network;
the characterization unit is used for characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the graph structure data in the space with the same weather semantic information to obtain the weather graph structure data of the time to be predicted of the wind power group.
Preferably, the prediction module comprises a capturing unit and a prediction unit;
the capturing unit is used for capturing dynamic changes of airspace characteristic time sequence data of the weather image structure data; and the prediction unit is used for predicting the power of the wind power group to-be-predicted time based on the dynamic change.
Based on the same inventive thought, the invention also provides a computer device, comprising: one or more processors;
the processor is used for storing one or more programs;
when the one or more programs are executed by the one or more processors, the wind power group power prediction method provided by the invention is realized.
Based on the same thought, the invention also provides a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and when the computer program is executed, the wind power group power prediction method provided by the invention is realized.
Compared with the prior art, the invention has the beneficial effects that:
according to the wind power group power prediction method provided by the invention, the grid numerical weather forecast and the single-point numerical weather forecast of the wind power group are taken as basic data sources, the weather image structure model of the wind power group is constructed, the weather image structure data is taken as input, the pre-trained wind power group power prediction model is utilized to capture the dynamic change of time sequence data, the mapping relation between the weather data and the wind power group power is utilized to output the predicted power, the purpose of predicting the power of a plurality of wind power plants in the wind power group by utilizing a single model is realized, and the modeling efficiency is improved on the premise of ensuring the prediction precision.
Drawings
FIG. 1 is a schematic diagram of steps of a wind power group power prediction method provided by the invention;
FIG. 2 is a schematic diagram of the structure data format of the weather map provided by the present invention.
Detailed Description
The invention discloses a wind power group power prediction method, which is characterized in that an meteorological graph structure model of a wind power group is constructed, the meteorological graph structure model is used as the input of a power prediction model, and the aim of predicting the power of a plurality of wind power plants of the wind power group by a single model is fulfilled by means of the characteristic abstract capacity of high-dimensional data and the capturing capacity of continuous information in time sequence data of a deep learning mapping model.
Example 1:
as shown in fig. 1, the wind power group power prediction method provided by the invention comprises the following steps:
step 1: acquiring grid numerical weather forecast data of time to be predicted covering a space range of a wind power group and single-point numerical weather forecast data of time to be predicted of each wind power field in the wind power group;
step 2: acquiring weather map structure data of the time to be predicted of a wind power group by utilizing a pre-constructed wind power plant map structure model based on grid numerical weather forecast data and single-point numerical weather forecast data;
step 3: based on the weather map structure data, predicting the power of the wind power group to be predicted time by using a pre-trained wind power group power prediction model;
Before the step 1, the method further comprises the steps of constructing an meteorological graph structure model of the wind power group and constructing and training a wind power group power prediction model.
The method for establishing the meteorological graph structure model of the wind power group specifically comprises the following steps:
1) Converting single-point numerical weather forecast data of each wind power plant into graph structure data
And taking grid numerical weather forecast covering the space range of the wind power group and single-point numerical weather forecast of the corresponding geographic position of each wind power plant in the group as basic weather parameters, wherein the two numerical weather forecast comprise the same weather elements, and the K-fold cross validation method is utilized to optimize the optimal weather sensitivity factor influencing the power prediction performance of each wind power plant.
The data similarity of each wind power plant in the same cluster is very high, and the data similarity of each wind power plant in different clusters is low, and constructing a wind power cluster graph structure based on a KNN algorithm generally causes redundant connection of wind power plants in different clusters, namely, incorrect connection of nodes in the graph structure, so that a data space cannot be reflected well. For each wind farm in the same cluster, the similarity is measured in terms of euclidean distance, and such a similarity measure sometimes has difficulty reflecting the true degree of similarity between data points. Two wind farms are generally considered to have more shared neighboring wind farms, and are considered to have higher pairwise similarities between them, but merely considering the number of shared neighboring wind farms may ignore the compactness of the two wind farms themselves. The weight of the edge between two points is determined by taking into account not only the number of shared adjacent nodes of the two nodes but also whether the two nodes are closely connected or not.
As shown in FIG. 2, each wind power plant is taken as a node, and the optimal weather sensitivity factor data corresponding to single-point numerical weather forecast data is taken as the characteristics or attributes of the node, wherein the characteristics or attributes of the node can be wind speed, wind direction, temperature, humidity, air pressure and the like. Determining connection relations among nodes by adopting a neighborhood construction algorithm based on density self-adaption, determining weights among the nodes by adopting a method based on shared nearest neighbor, forming a data set of a graph structure by geographic position data of each wind power plant, characteristics of the nodes and connection weights among the nodes, and converting single-point numerical weather forecast data into graph structure data. The construction method of the graph structure data can more accurately reflect the spatial structure of the wind power plant data.
2) Extracting airspace characteristic data of gridding numerical weather forecast data and airspace characteristic data of graph structure data;
aiming at the gridding numerical weather forecast, taking the union of the optimal weather sensitive factor data of each wind power plant as the optimal weather element of the gridding numerical weather forecast data, and extracting the airspace characteristic data of the optimal weather element by adopting a convolution self-coding decoding network; and extracting airspace characteristic data of the graph structure data by adopting a graph convolution neural network.
For single-point numerical weather forecast, drawing convolution neural network is adopted to extract airspace characteristic data of drawing structure data, and the method specifically comprises the following steps: and constructing a filter in a Fourier domain by adopting a graph convolution neural network, sampling from graph structure data by using the filter to obtain sampling data, and extracting airspace characteristic data of the graph structure data from the sampling data.
3) Construction of wind farm graph structure model
Characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data in a space with the same weather semantic information by adopting an embedded mode to obtain weather tensor data; and encoding the meteorological tensor data and the geographic position data of each wind power plant to obtain a wind power plant diagram structure model.
The meteorological graph structure model is established based on multi-source meteorological data and geographic position data of each wind farm, and can more accurately reflect the spatial structure of the meteorological data of the wind farms.
The wind power group power prediction model construction and training method specifically comprises the following steps:
the wind power group power prediction model is constructed by the following ideas: the short-term wind power prediction, such as power prediction for 24 to 72 hours in the future, is essentially to construct a mapping model between a numerical weather prediction result and actual power, and the prediction accuracy can be improved through two links of the numerical weather prediction and a prediction model. The improvement of the accuracy of the numerical weather forecast is a slow process, and based on the latest achievements in the expert cognition and deep learning fields, the deep mapping relation between the available numerical weather forecast weather information and power is fully excavated, so that the method is the only way for improving the prediction accuracy of wind power in a short period.
The wind power group power prediction model constructed based on the thought comprises a graph convolution neural network and a gating circulation network, wherein the prediction model takes the weather image structure data as input, and the weather space characteristics of the weather image structure data are excavated by using the graph convolution neural network through internal operations such as convolution, updating, resetting, pooling, activating and the like. The graph convolutional neural network utilizes the filtering object gas graph structural data of the Fourier domain to sample the central node and the neighborhood thereof, and applies a multi-level space-time attention mechanism to realize the attention of each wind power plant in the space region and the adjacent sites, and the connection weight of the distributed nodes is recalculated in the subsequent training process. And then taking the meteorological space characteristic time sequence data mined by the graph convolution neural network as the input of the gating circulation network, capturing dynamic changes through the transmission of internal information in the gating circulation network, thereby extracting time characteristics and outputting wind power group prediction power mined by the space-time characteristics. The power prediction model takes the weather pattern structure data of a wind power group as input, takes the weather pattern structure data-the graph convolution neural network data-the gate control circulation network data-the predicted power as data flow direction, and the output power can be the power prediction result of each wind power station or the total sum of the predicted powers of a plurality of wind power stations, thereby obtaining the cluster output power.
The power prediction model builds a weather-power depth mapping model by means of the capturing capability of deep learning on space-time characteristics of high-dimensional complex data, so that the space-time characteristics of a wind power cluster can be captured better, meanwhile, the attention of each wind power plant in a space region and adjacent sites is realized by applying a multi-stage space-time attention mechanism, and therefore, the data space can be reflected accurately. According to the prediction model, the meteorological graph structure data is used as the input of the graph convolution gating circulation network, the power generation power of multiple wind power plants in the wind power group area is used as the output, and the single model predicts the power generation power of multiple wind power plants at the same time, so that the modeling efficiency is improved.
The prior art and the prediction model of the invention are respectively used for modeling the power predictions of 20 wind power plants in a certain area, and the time required for modeling is shown in table 1. Compared with the prior art, the modeling efficiency of the invention is greatly improved.
TABLE 1 modeling time comparison of prior art and the present invention
| The method adopted is | Modeling time |
| Prior Art | About 200 minutes |
| The invention is that | About 110 minutes |
The training process of the power prediction model is as follows:
1) Obtaining historical weather map structure data and historical power data
Historical weather data is selected from grid-type numerical weather forecast data and single-point numerical weather forecast data of a wind power group, historical power data of each wind power plant in the wind power group corresponding to the historical weather data is obtained, the historical weather data is input into a pre-built wind power plant structure model, historical weather map structure data corresponding to the historical weather data is obtained, one group of historical weather map structure data and the corresponding historical power data form a sample, a training sample set is formed by a plurality of samples, and the training sample set has stable and reliable output for the trained model, and the data duration of the training sample set is at least one year.
2) Training and testing wind power group power prediction model
And taking the historical gas image structure data in the training sample as input, taking the historical power data in the sample as target output, and carrying out initial training on the power prediction model. The power prediction model which is trained initially is tested by a test sample set, the test sample is also composed of historical weather map structure data of a wind power group and corresponding historical power data, and a data source which is different from the training sample needs to be selected from the historical weather data. The data duration of the test sample set can be selected according to the test requirement and the test result, and in the embodiment, the data duration of the test sample set is two months. And taking the power prediction model after passing the test as a trained wind power group power prediction model.
The specific process for predicting the wind power group power by using the trained wind power group power prediction model comprises the following steps: firstly, grid numerical weather forecast of time to be predicted covering a space range of a wind power group and single-point numerical weather forecast of time to be predicted of each wind power place in a geographic position in the wind power group are obtained; inputting a grid numerical weather forecast and a single-point numerical weather forecast of the time to be predicted into a pre-constructed weather map structure model of the wind power group to obtain a weather map structure with weather information of the time to be predicted of the wind power group; and inputting the weather image structure with the weather information of the wind power group to be predicted time into a pre-trained deep learning model, and outputting the power of the wind power group to be predicted time through the deep learning model.
The power prediction method can simultaneously predict the power of a plurality of wind power stations in the wind power group area by using a single model, improves the complicated operation of links such as deployment, training and maintenance of the traditional wind power group power prediction model, and reduces the cost of manpower, material resources and time.
Example 2:
based on the same thought, the invention also discloses a system for establishing the wind power plant graph structural model, which comprises the following steps:
the graph structure module takes each wind power plant in the wind power group as a node, and adopts a density self-adaptive algorithm based on shared nearest neighbors to convert single-point numerical weather forecast data of each wind power plant into graph structure data;
the extraction module is used for extracting meshing numerical weather forecast data and airspace characteristic data of graph structure data covering the spatial range of the wind power group by adopting a convolution network;
the characterization module is used for characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data on the space with the same meteorological semantic information to obtain a wind power plant map structure model of the wind power group.
Wherein the graph structure module comprises: the selecting unit is used for selecting an optimal weather sensitivity factor affecting the power prediction performance of the wind power group from single-point numerical weather forecast data of each wind power plant by using a K-fold cross validation method; and the conversion unit takes each wind power plant as a node, takes optimal weather sensitivity factor data corresponding to the single-point numerical weather forecast data as characteristics of the node, and converts the single-point numerical weather forecast data into graph structure data by adopting a density self-adaptive algorithm based on shared nearest neighbors.
The extraction module comprises: the convolution self-coding decoding network unit is used for extracting spatial domain characteristic data of the gridding numerical weather forecast data by adopting the convolution self-coding decoding network; and the graph convolution neural network unit is used for extracting airspace characteristic data of the graph structural data by adopting the graph convolution neural network.
The characterization module comprises: the embedding unit is used for representing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data in the space with the same weather semantic information in an embedding mode to obtain weather tensor data; and the encoding unit is used for encoding the meteorological tensor data and the geographical position data of each wind power plant to obtain a wind power plant diagram structure model.
Example 3:
based on the same thought, the invention also discloses a training system of the wind power group power prediction model, which comprises the following steps:
the acquisition module is used for acquiring historical meteorological data of the wind power group and power data corresponding to the historical meteorological data, and acquiring historical meteorological graph structure data of the wind power group by utilizing a pre-constructed wind power plant graph structure model;
the training module is used for training the wind power group power prediction model by taking the historical weather pattern structure data as input and taking the power data corresponding to the historical weather pattern structure data as output to obtain a trained wind power group power prediction model.
The construction process of the wind power plant graph structure model comprises the following steps:
1) Converting single-point numerical weather forecast data of each wind power plant into graph structure data
And taking grid numerical weather forecast covering the space range of the wind power group and single-point numerical weather forecast of the corresponding geographic position of each wind power plant in the group as basic weather parameters, wherein the two numerical weather forecast comprise the same weather elements, and the K-fold cross validation method is utilized to optimize the optimal weather sensitivity factor influencing the power prediction performance of each wind power plant. And taking each wind power plant as a node, taking the optimal weather sensitivity factor data corresponding to the single-point numerical weather forecast data as the characteristics of the node, adopting a density self-adaptive algorithm based on shared nearest neighbors to determine the connection weight between the nodes, forming a data set of the graph structure by the geographic position data of each wind power plant, the characteristics of the nodes and the connection weight between the nodes, and converting the single-point numerical weather forecast data into graph structure data.
2) Extracting airspace characteristic data of gridding numerical weather forecast data and airspace characteristic data of graph structure data;
aiming at the gridding numerical weather forecast, taking the union of the optimal weather sensitive factor data of each wind power plant as the optimal weather element of the gridding numerical weather forecast data, and extracting the airspace characteristic data of the optimal weather element by adopting a convolution self-coding decoding network; and extracting airspace characteristic data of the graph structure data by adopting a graph convolution neural network.
For single-point numerical weather forecast, drawing convolution neural network is adopted to extract airspace characteristic data of drawing structure data, and the method specifically comprises the following steps: and constructing a filter in a Fourier domain by adopting a graph convolution neural network, sampling from graph structure data by using the filter to obtain sampling data, and extracting airspace characteristic data of the graph structure data from the sampling data.
3) Construction of wind farm graph structure model
Characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data in a space with the same weather semantic information by adopting an embedded mode to obtain weather tensor data; and encoding the meteorological tensor data and the geographic position data of each wind power plant to obtain a wind power plant diagram structure model.
The wind power group prediction model in the training system comprises a graph convolutional neural network and a gating circulation network, the prediction model takes weather space characteristics of weather image structure data as input, the weather space characteristics of the weather image structure data are mined by using the graph convolutional neural network, the weather space characteristic time sequence data mined by the graph convolutional neural network are taken as input of the gating circulation network, dynamic changes are captured through transmission of internal information in the gating circulation network, so that time characteristics are extracted, and wind power group prediction power mined by space-time characteristics is output.
The training module of the training system comprises:
an attention mechanism unit for recalculating the connection weight between the nodes of the historical weather pattern structure by adopting a space-time attention mechanism;
and the corresponding relation acquisition unit is used for carrying out internal information transfer on the historical gas image structure data by adopting a gate control circulation network, capturing the dynamic change of the historical gas image structure data and forming the corresponding relation between the power data and the dynamic change.
Example 4:
based on the same thought, the invention also discloses a wind power group power prediction system, which comprises: the acquisition module is used for acquiring grid numerical weather forecast data of the time to be predicted covering the space range of the wind power group and single-point numerical weather forecast data of the time to be predicted of each wind power plant in the wind power group; the weather map structure module is used for obtaining weather map structure data of the time to be predicted of the wind power group by utilizing a pre-constructed wind power plant structure model based on the grid numerical weather forecast data and the single-point numerical weather forecast data; and the prediction module is used for predicting the power of the wind power group to be predicted time by utilizing a pre-trained wind power group power prediction model based on the weather map structure data.
The construction process of the pre-constructed wind power plant diagram structure model comprises the following steps:
1) Converting single-point numerical weather forecast data of each wind power plant into graph structure data
And taking grid numerical weather forecast covering the space range of the wind power group and single-point numerical weather forecast of the corresponding geographic position of each wind power plant in the group as basic weather parameters, wherein the two numerical weather forecast comprise the same weather elements, and the K-fold cross validation method is utilized to optimize the optimal weather sensitivity factor influencing the power prediction performance of each wind power plant. And taking each wind power plant as a node, taking the optimal weather sensitivity factor data corresponding to the single-point numerical weather forecast data as the characteristics of the node, adopting a density self-adaptive algorithm based on shared nearest neighbors to determine the connection weight between the nodes, forming a data set of the graph structure by the geographic position data of each wind power plant, the characteristics of the nodes and the connection weight between the nodes, and converting the single-point numerical weather forecast data into graph structure data.
2) Extracting airspace characteristic data of gridding numerical weather forecast data and airspace characteristic data of graph structure data;
aiming at the gridding numerical weather forecast, taking the union of the optimal weather sensitive factor data of each wind power plant as the optimal weather element of the gridding numerical weather forecast data, and extracting the airspace characteristic data of the optimal weather element by adopting a convolution self-coding decoding network; and extracting airspace characteristic data of the graph structure data by adopting a graph convolution neural network.
For single-point numerical weather forecast, drawing convolution neural network is adopted to extract airspace characteristic data of drawing structure data, and the method specifically comprises the following steps: and constructing a filter in a Fourier domain by adopting a graph convolution neural network, sampling from graph structure data by using the filter to obtain sampling data, and extracting airspace characteristic data of the graph structure data from the sampling data.
3) Construction of wind farm graph structure model
Characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data in a space with the same weather semantic information by adopting an embedded mode to obtain weather tensor data; and encoding the meteorological tensor data and the geographic position data of each wind power plant to obtain a wind power plant diagram structure model.
The training method of the wind power group power prediction model comprises the following steps:
1) Obtaining historical weather map structure data and historical power data
Historical weather data is selected from grid-type numerical weather forecast data and single-point numerical weather forecast data of a wind power group, historical power data of each wind power plant in the wind power group corresponding to the historical weather data is obtained, the historical weather data is input into a pre-constructed wind power plant structure model, historical weather map structure data corresponding to the historical weather data is obtained, a group of historical weather map structure data and corresponding historical power data form a sample, and a training sample set is formed by a plurality of samples.
2) Training and testing wind power group power prediction model
And taking the historical gas image structure data in the training sample as input, taking the historical power data in the sample as target output, and carrying out initial training on the power prediction model. The power prediction model which is trained initially is tested by a test sample set, the test sample also comprises the historical weather pattern structure data of the wind power group and the corresponding historical power data, and the power prediction model which is tested to be qualified is used as the power prediction model of the trained wind power group.
The gas image structure module comprises a graph structure unit, an extraction unit and a characterization unit; the graph structure unit is used for converting single-point numerical weather forecast data of the time to be predicted of each wind power plant into graph structure data by taking each wind power plant in a wind power group as a node and adopting a density self-adaptive algorithm based on shared nearest neighbors; the extraction unit is used for extracting airspace characteristic data of grid-type numerical weather forecast data and airspace characteristic data of graph structure data of the time to be predicted covering the space range of the wind power group by adopting a convolution network; the characterization unit is used for characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the graph structure data on the space with the same weather semantic information to obtain the weather graph structure data of the wind power group to-be-predicted time.
The prediction module comprises a capturing unit and a prediction unit; the capturing unit is used for capturing dynamic changes of airspace characteristic time sequence data of the weather image structure data; the prediction unit is used for predicting the power of the wind power group to be predicted time based on dynamic change.
Example 5:
based on the same inventive concept, the invention also provides a computer device comprising a processor and a memory for storing a computer program comprising program instructions, the processor being for executing the program instructions stored in the computer memory. The processor may be a central processing unit (Central Processing Unit, CPU), but may also be other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application SpecificIntegrated Circuit, ASIC), off-the-shelf Programmable gate arrays (FPGAs) or other Programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., which are the computational cores and control cores of the terminals adapted to implement one or more instructions, in particular to load and execute one or more instructions in a computer storage medium to implement the corresponding method flow or corresponding functions, to implement the steps of a wind farm power prediction method in embodiment 1 described above.
Example 6:
based on the same inventive concept, the present invention also provides a storage medium, in particular, a computer readable storage medium (Memory), which is a Memory device in a computer device, for storing programs and data. It is understood that the computer readable storage medium herein may include both built-in storage media in a computer device and extended storage media supported by the computer device. The computer-readable storage medium provides a storage space storing an operating system of the terminal. Also stored in the memory space are one or more instructions, which may be one or more computer programs (including program code), adapted to be loaded and executed by the processor. The computer readable storage medium herein may be a high-speed RAM memory or a non-volatile memory (non-volatile memory), such as at least one magnetic disk memory. One or more instructions stored in a computer-readable storage medium may be loaded and executed by a processor to implement the steps of a method for predicting wind farm power in embodiment 1 described above.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The foregoing is illustrative of the present invention and is not to be construed as limiting thereof, but rather as providing for the use of additional embodiments and advantages of all such modifications, equivalents, improvements and similar to the present invention are intended to be included within the scope of the present invention as defined by the appended claims.
Claims (21)
1. A method for predicting wind power population power, the method comprising: the method for establishing the wind power plant graph structure model comprises the following steps:
Taking each wind power plant in a wind power group as a node, and converting single-point numerical weather forecast data of each wind power plant into graph structure data by adopting a density self-adaptive algorithm based on shared nearest neighbors;
extracting airspace characteristic data of grid numerical weather forecast data covering the spatial range of the wind power group and airspace characteristic data of the map structure data by adopting a convolution network;
and characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data on the space with the same meteorological semantic information to obtain the wind power plant map structure model.
2. The prediction method of claim 1, wherein the converting the single-point numerical weather forecast data of each wind farm into the graph structure data by using each wind farm in the wind farm group as a node and adopting a density adaptive algorithm based on shared nearest neighbor comprises:
selecting an optimal weather sensitivity factor affecting the power prediction performance of the wind power group from single-point numerical weather forecast data of each wind power plant by using a K-fold cross validation method;
and taking each wind power plant as a node, taking the optimal weather sensitivity factor data corresponding to the single-point numerical weather forecast data as the characteristics of the node, adopting a density self-adaptive algorithm based on shared nearest neighbors to determine the connection weight between the nodes, forming a data set of the graph structure by the geographic position data of each wind power plant, the characteristics of the nodes and the connection weight between the nodes, and converting the single-point numerical weather forecast data into the graph structure data.
3. The prediction method according to claim 2, wherein the extracting airspace feature data of meshing numerical weather forecast data covering the spatial range of the wind power group and airspace feature data of the map structure data using a convolution network includes:
taking the union of the optimal weather sensitive factor data of each wind power plant as the optimal weather element of the grid numerical weather forecast data, and extracting airspace characteristic data of the optimal weather element by adopting a convolution self-coding decoding network; and extracting airspace characteristic data of the graph structure data by adopting a graph convolution neural network.
4. A method of predicting as claimed in claim 3 wherein said extracting spatial signature data of said map structural data using a map convolutional neural network comprises: and constructing a filter in a Fourier domain by adopting the graph convolution neural network, sampling the graph structural data by using the filter to obtain sampling data, and extracting spatial domain characteristic data of the graph structural data from the sampling data.
5. The method of predicting as set forth in claim 4, wherein characterizing the spatial signature data of the meshing numerical weather forecast data and the spatial signature data of the map structure data in space with the same weather semantic information to obtain the wind farm map structure model comprises:
Characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data in a space with the same weather semantic information by adopting an embedded mode to obtain weather tensor data;
and encoding the meteorological tensor data and the geographic position data of each wind power plant to obtain the wind power plant graph structure model.
6. A system for building a structural model of a wind farm, comprising:
the graph structure module is used for converting single-point numerical weather forecast data of each wind power plant into graph structure data by taking each wind power plant in a wind power group as a node and adopting a density self-adaptive algorithm based on shared nearest neighbors;
the extraction module is used for extracting gridding numerical weather forecast data covering the spatial range of the wind power group and airspace characteristic data of the map structure data by adopting a convolution network;
and the characterization module is used for characterizing the airspace characteristic data of the gridding numerical weather forecast data and the airspace characteristic data of the graph structure data on the space with the same meteorological semantic information to obtain a wind power plant graph structure model of the wind power group.
7. The build system of claim 6, wherein the graph structure module comprises:
the selecting unit is used for selecting the optimal weather sensitivity factor affecting the power prediction performance of the wind power group from the single-point numerical weather forecast data of each wind power plant by using a K-fold cross validation method;
the transformation unit is used for taking each wind power plant as a node, taking the optimal weather sensitivity factor data corresponding to the single-point numerical weather forecast data as the characteristics of the node, and transforming the single-point numerical weather forecast data into the graph structure data by adopting a density self-adaptive algorithm based on the shared nearest neighbor.
8. The setup system of claim 7, wherein the extraction module comprises:
the convolution self-coding decoding network unit is used for extracting the airspace characteristic data of the meshing numerical weather forecast data by adopting a convolution self-coding decoding network;
and the graph convolution neural network unit is used for extracting airspace characteristic data of the graph structure data by adopting the graph convolution neural network.
9. The setup system of claim 8, wherein the characterization module comprises:
the embedding unit is used for representing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the map structure data in a space with the same weather semantic information in an embedding mode to obtain weather tensor data;
And the encoding unit is used for encoding the meteorological tensor data and the geographical position data of each wind power plant to obtain the wind power plant diagram structure model.
10. A method for predicting wind power population power, the method comprising: the training method of the wind power group power prediction model comprises the following steps:
acquiring historical meteorological data of a wind power group and power data corresponding to the historical meteorological data, and acquiring historical meteorological graph structure data of the wind power group by utilizing a pre-constructed wind power plant graph structure model;
taking the historical weather map structure data as input, and taking power data corresponding to the historical weather map structure data as output to train the wind power group power prediction model to obtain a trained wind power group power prediction model;
the wind power group power prediction model comprises a graph convolution neural network and a gating circulation network;
the pre-built wind farm graph structure model is built by adopting the building method of the wind farm graph structure model in the prediction method according to any one of claims 1 to 5.
11. The prediction method of claim 10, wherein training the wind power group power prediction model with the historical weather map structure data as input and the power data corresponding to the historical weather map structure data as output to obtain a trained wind power group power prediction model comprises:
Extracting airspace characteristic time sequence data of the historical weather map structure data by adopting the map convolution neural network, and recalculating connection weights among nodes of the historical weather map structure by adopting a space-time attention mechanism;
and adopting the gating circulation network to transmit internal information of the airspace characteristic time sequence data, capturing dynamic changes of the airspace characteristic time sequence data, forming a corresponding relation between the power data and the dynamic changes, and obtaining the trained wind power group power prediction model.
12. A training system for a wind farm power prediction model, comprising:
the acquisition module is used for acquiring historical meteorological data of a wind power group and power data corresponding to the historical meteorological data, and acquiring historical meteorological graph structure data of the wind power group by utilizing a pre-constructed wind power plant graph structure model;
the training module is used for taking the historical weather pattern structure data as input, taking the power data corresponding to the historical weather pattern structure data as output and training the wind power group power prediction model to obtain a trained wind power group power prediction model;
the pre-constructed wind power plant graph structure model is constructed by adopting the method for constructing the wind power plant graph structure model in the prediction method according to any one of claims 1-5, and the wind power population prediction model comprises a graph convolutional neural network and a gating circulation network.
13. The training system of claim 12, wherein the training module comprises:
an attention mechanism unit for recalculating connection weights between nodes of the historical weather pattern structure using a spatiotemporal attention mechanism;
and the corresponding relation acquisition unit is used for carrying out internal information transfer on the historical weather map structure data by adopting the gating circulation network, capturing the dynamic change of the historical weather map structure data and forming the corresponding relation between the power data and the dynamic change.
14. A method for predicting wind power population power, comprising:
acquiring grid numerical weather forecast data of time to be predicted covering a space range of a wind power group and single-point numerical weather forecast data of time to be predicted of each wind power plant in the wind power group;
acquiring weather map structure data of the time to be predicted of the wind power group by utilizing a pre-constructed wind power plant map structure model based on the grid numerical weather forecast data and the single-point numerical weather forecast data;
based on the weather map structure data, predicting the power of the wind power group to be predicted time by using a pre-trained wind power group power prediction model;
Wherein the pre-built wind farm graph structure model is built by adopting a building method of the wind farm graph structure model in the prediction method according to any one of claims 1 to 5; the wind power group power prediction model is obtained by adopting a training method of the wind power group power prediction model in the prediction method according to claim 10 or claim 11.
15. The prediction method according to claim 14, wherein the obtaining, based on the gridded numerical weather forecast data and the single-point numerical weather forecast data, weather map structure data of the time to be predicted for the wind farm using a pre-constructed wind farm map structure model includes:
converting the single-point numerical weather forecast data into graph structure data by the pre-constructed wind power plant graph structure model by adopting a density self-adaptive algorithm based on shared nearest neighbor, extracting airspace characteristic data of the gridding numerical weather forecast data and the graph structure data by adopting the pre-constructed wind power plant graph structure model, and representing the airspace characteristic data of the gridding numerical weather forecast data and the airspace characteristic data of the graph structure data on a space with the same meteorological semantic information to obtain the weather graph structure data.
16. The method of predicting as recited in claim 15 wherein said predicting power of said wind farm to be predicted time using a pre-trained wind farm power prediction model based on said weather map structure data comprises:
inputting the weather pattern structure data into the pre-trained wind power group power prediction model, extracting airspace feature time sequence data of the weather pattern structure data by using the pre-trained wind power group prediction model, carrying out internal information transfer on the airspace feature time sequence data, capturing dynamic changes of the airspace feature time sequence data, and outputting power of the wind power group to be predicted time based on the dynamic changes.
17. A wind farm power prediction system, comprising:
the acquisition module is used for acquiring grid numerical weather forecast data of the time to be predicted covering the space range of the wind power group and single-point numerical weather forecast data of the time to be predicted of each wind power plant in the wind power group;
the weather map structure module is used for obtaining weather map structure data of the time to be predicted of the wind power group by utilizing a pre-constructed wind power plant map structure model based on the grid numerical weather forecast data and the single-point numerical weather forecast data;
And the prediction module is used for predicting the power of the time to be predicted of the wind power group by using a pre-trained wind power group power prediction model based on the weather map structure data.
Wherein the pre-built wind farm graph structure model is built by adopting a building method of the wind farm graph structure model in the prediction method according to any one of claims 1 to 5; the pre-trained wind power group power prediction model is obtained by adopting a training method of the wind power group power prediction model in the prediction method according to claim 10 or claim 11.
18. The prediction system of claim 17 wherein the aerial image structure module comprises a graph structure unit, an extraction unit, and a characterization unit;
the graph structure unit is used for converting single-point numerical weather forecast data of the time to be predicted of each wind power plant into graph structure data by taking each wind power plant in a wind power group as a node and adopting a density self-adaptive algorithm based on shared nearest neighbor;
the extraction unit is used for extracting airspace feature data of grid-type numerical weather forecast data covering the time to be predicted of the wind power group space range and airspace feature data of the map structure data by adopting a convolution network;
The characterization unit is used for characterizing the airspace characteristic data of the grid numerical weather forecast data and the airspace characteristic data of the graph structure data in the space with the same weather semantic information to obtain the weather graph structure data of the time to be predicted of the wind power group.
19. The prediction system of claim 18, wherein the prediction module comprises a capture unit and a prediction unit;
the capturing unit is used for capturing dynamic changes of airspace characteristic time sequence data of the weather image structure data; and the prediction unit is used for predicting the power of the wind power group to-be-predicted time based on the dynamic change.
20. A computer device, comprising: one or more processors;
the processor is used for storing one or more programs;
the method of wind power population power prediction of any one of claims 14-16, when the one or more programs are executed by the one or more processors.
21. A computer readable storage medium, having stored thereon a computer program which, when executed, implements a wind farm power prediction method according to any of claims 14-16.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210567499.9A CN117175535A (en) | 2022-05-23 | 2022-05-23 | Wind power group power prediction method, system, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210567499.9A CN117175535A (en) | 2022-05-23 | 2022-05-23 | Wind power group power prediction method, system, equipment and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117175535A true CN117175535A (en) | 2023-12-05 |
Family
ID=88939970
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210567499.9A Pending CN117175535A (en) | 2022-05-23 | 2022-05-23 | Wind power group power prediction method, system, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117175535A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117688975A (en) * | 2024-02-02 | 2024-03-12 | 南京信息工程大学 | Meteorological event prediction method and system based on evolution rule mining |
-
2022
- 2022-05-23 CN CN202210567499.9A patent/CN117175535A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117688975A (en) * | 2024-02-02 | 2024-03-12 | 南京信息工程大学 | Meteorological event prediction method and system based on evolution rule mining |
| CN117688975B (en) * | 2024-02-02 | 2024-05-14 | 南京信息工程大学 | Meteorological event prediction method and system based on evolution rule mining |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ghaderi et al. | Deep forecast: Deep learning-based spatio-temporal forecasting | |
| Jiao et al. | A graph neural network based deep learning predictor for spatio-temporal group solar irradiance forecasting | |
| CN115293415A (en) | Multi-wind-farm short-term power prediction method considering time evolution and space correlation | |
| CN103268366A (en) | Combined wind power prediction method suitable for distributed wind power plant | |
| CN110570030A (en) | Wind power cluster power interval prediction method and system based on deep learning | |
| Li et al. | A novel combined prediction model for monthly mean precipitation with error correction strategy | |
| CN111242351A (en) | Tropical cyclone track prediction method based on self-encoder and GRU neural network | |
| Li et al. | Deep spatio-temporal wind power forecasting | |
| CN116307291B (en) | Distributed photovoltaic power generation prediction method and prediction terminal based on wavelet decomposition | |
| Chen et al. | Research on wind power prediction method based on convolutional neural network and genetic algorithm | |
| CN113984198B (en) | Shortwave radiation prediction method and system based on convolutional neural network | |
| CN117175535A (en) | Wind power group power prediction method, system, equipment and medium | |
| CN115293424A (en) | New energy maximum power generation capacity calculation method, terminal and storage medium | |
| CN114298132A (en) | Wind power prediction method and device and electronic equipment | |
| CN116976227B (en) | Storm water increasing forecasting method and system based on LSTM machine learning | |
| CN113570132A (en) | Wind power prediction method for space-time meteorological feature extraction and deep learning | |
| Wan et al. | Turbine location wind speed forecast using convolutional neural network | |
| CN114444763A (en) | Wind power prediction method based on AFSA-GNN | |
| CN112784477A (en) | WRF-LES and BP-PSO-Bagging combined wind power prediction method | |
| Singhal et al. | Solar-Cast: Solar Power Generation Prediction from Weather Forecasts using Machine Learning | |
| Dhakal et al. | Deep Learning Model with Probability Density Function and Feature Engineering for Short Term Wind Speed Prediction | |
| Afrasiabi et al. | Nonparametric Maximum Likelihood Probabilistic Photovoltaic Power Generation Forecasting based on Spatial-Temporal Deep Learning | |
| CN116703007B (en) | Wind power cluster prediction model construction method, power prediction method and device | |
| Tanaka et al. | SVM-based multi-point prediction system of solar irradiance using DTW-based cluster analysis | |
| Su et al. | A Short-term PV Power Forecasting Method based on NWP Correction Considering Meteorological Coupling Correlation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication |