CN117934208A - Multi-channel depth network-based multi-source data offshore wind power prediction method - Google Patents
Multi-channel depth network-based multi-source data offshore wind power prediction method Download PDFInfo
- Publication number
- CN117934208A CN117934208A CN202410303887.5A CN202410303887A CN117934208A CN 117934208 A CN117934208 A CN 117934208A CN 202410303887 A CN202410303887 A CN 202410303887A CN 117934208 A CN117934208 A CN 117934208A
- Authority
- CN
- China
- Prior art keywords
- fan
- feature
- wind power
- data
- fans
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 239000013598 vector Substances 0.000 claims abstract description 24
- 238000011176 pooling Methods 0.000 claims abstract description 12
- 238000012216 screening Methods 0.000 claims abstract description 8
- 238000012545 processing Methods 0.000 claims description 42
- 238000012549 training Methods 0.000 claims description 29
- 238000012417 linear regression Methods 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 4
- 230000008878 coupling Effects 0.000 claims description 3
- 238000010168 coupling process Methods 0.000 claims description 3
- 238000005859 coupling reaction Methods 0.000 claims description 3
- 238000003475 lamination Methods 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 239000010410 layer Substances 0.000 description 27
- 238000010248 power generation Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 241000287127 Passeridae Species 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Landscapes
- Wind Motors (AREA)
Abstract
The invention relates to a multi-channel depth network-based multi-source data offshore wind power prediction method, which comprises the following steps: the method comprises the steps of firstly, acquiring SCADA data, meteorological data and geographic position information of all fans in an offshore wind farm in an area, superposing SCADA data alignment timestamps of all fans to form a three-dimensional tensor to form a fan feature input channel, screening useful feature information through a feature selection module based on L-1 norm regularization, forming a map feature input channel by the geographic position information composition map of all fans, extracting potential space-time features through a depth map attention convolution loop network, superposing meteorological data alignment timestamps in the area to form a meteorological feature input channel, screening useful feature information through L1 norm regularization regression, connecting the features output by the three feature input channels in series through a global feature pooling layer to obtain final potential feature vectors, inputting the final potential feature vectors into a prediction model, and improving the prediction precision of offshore wind power.
Description
Technical Field
The invention relates to the technical field of offshore wind power prediction, in particular to a multi-source data offshore wind power prediction method based on a multi-channel depth network.
Background
Because the wind energy has the characteristics of cleanness, richness and environmental protection, the offshore wind farm has the advantages of no consumption of land resources, high wind speed, zero dust emission and the like, and compared with the onshore wind farm, the offshore wind farm has 20-40% higher energy efficiency, and is more suitable for large-scale development. However, the offshore wind power generation is easily affected by a plurality of factors, so that the offshore wind power generation has uncertainty, and adverse effects are brought to the dispatching and operation of the power system, so that accurate offshore wind power prediction has important significance for the safe and stable operation of the power system.
Aiming at offshore wind power prediction, the prior art provides a method based onAccording to the short-term offshore wind power prediction method, a CEEMDAN algorithm is utilized to decompose a historical offshore wind power sequence to obtain different subsequences, then a BiLSTM prediction model is constructed for each subsequence, super-parameters such as the hidden layer unit number, the learning rate and the iteration number of the BiLSTM model are optimized by utilizing a sparrow optimization algorithm, the optimized super-parameters are utilized to predict different subsequences, and finally prediction results of the different subsequences are overlapped to obtain the offshore wind power prediction power. However, the method only relies on the historical power sequence of the offshore wind power to predict, the influence of the geographic position information of the fan and the complicated and changeable climate of the sea on the offshore wind power generation is not considered, the generalization of the model is poor, and the prediction precision of the offshore wind power applied to complex conditions is not ideal. The fan installation position in the wind power plant in a certain area can influence the power generation of surrounding fans, so that the geographic position information of the fans is also an important characteristic and cannot be ignored. A great deal of researches show that the characteristics collected by the SCADA system of the fan, such as wind speed and wind direction, have close relation with wind power generation, and the two important characteristics are considered to help to improve the prediction accuracy of the model. The offshore climate is complex, and the input of meteorological information such as sea waves, tides, typhoons and the like into the model is beneficial to improving the prediction precision.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a multi-source data offshore wind power prediction method and system based on a multi-channel depth network, which fully utilize information collected by a fan SCADA (supervisory control and data acquisition), geographical position information of the fan and meteorological information, can effectively reduce prediction errors caused by wind power generation uncertainty and improve the prediction precision of wind power of an offshore wind power plant.
In order to solve the technical problems, the invention adopts the following technical scheme:
the multi-channel depth network-based multi-source data offshore wind power prediction method comprises the following steps:
S1, acquiring SCADA data acquisition and monitoring control data, meteorological data and geographic position information of a fan of an offshore wind farm in an area, and performing primary processing on related data;
S2, acquiring a fan characteristic input channel and processing three-dimensional tensor characteristics;
acquiring a graph characteristic input channel and processing regional fan graph characteristics;
Acquiring a meteorological feature input channel and processing meteorological features;
S3, constructing a multi-channel depth network feature processing model, connecting the features output by the three feature channels in series through a global feature pooling layer, and processing the features to form final potential features of an input prediction model;
S4, constructing a training part of the linear regression prediction model, and training by utilizing final potential features obtained by the multi-channel depth network feature processing model;
s5, constructing a test part of the linear regression prediction model, and inputting a training result into the prediction model to predict the wind power of the target fan.
Further, the operation of step S2 is specifically as follows:
The SCADA data alignment time stamps of all fans are overlapped to form a three-dimensional tensor to form a fan characteristic input channel, and useful characteristic information is screened through a characteristic selection module based on L-1 norm regularization;
Regarding each fan in the area as a node, wherein each node corresponds to the geographic position information of the fan, determining the spatial relationship of each pair of fans according to the distance and mutual information between the fans, and establishing an area fan graph between the fans to form a graph characteristic input channel;
And overlapping the alignment time stamps of the meteorological data in the region to form a meteorological feature input channel, and screening useful feature information through an L1 norm regularized regression feature selection module.
Further, the operation of the fan feature input channel to process the three-dimensional tensor feature is as follows:
s21, superposing N characteristic alignment time stamps of P typhoon SCADA records with the history record length of T in the area to form a three-dimensional tensor And then/>Expanded to one/>A long vector of dimensions;
s22, constructing a fan characteristic selection module based on L-1 norm regularization Passing the long vector obtained in S21Selecting a set of mean-effective fan features/>, from a high-dimensional input spaceThe expression is:
。
further, the operation of the map feature input channel processing region fan map feature is as follows:
Representing geographical information of fans in an area as Wherein P represents the number of fans and M represents the features contained in the geographic information: longitude, latitude and altitude; constructing a multi-fan graph G, wherein each node corresponds to one wind turbine, and the edge reflects the spatial relationship between every two fans, and the specific process is as follows:
S23: the normalized distance ND between the fan pairs is calculated, and the normalized distance ND between the fans i, j is calculated as follows:
wherein, Representing the longitude of the ith blower,/>Represents the longitude of the jth fan,/>Represents the latitude of the ith fan,/>Represents the latitude of the jth fan,/>Representing the elevation of the ith blower,/>Representing the elevation of the jth fan,R=6371 denotes the earth radius,/>Intermediate variable representing transition,/>，/>Representing the distance between fans i, j,/>Represents a distance maximum,/>Representing a distance minimum;
s24: calculating normalized mutual information NMI between fan pairs, and enabling wind power of an ith fan to be equal to that of the ith fan The wind power/>, expressed as I, of the jth fanDenoted J, normalized mutual information NMI between fans i and J is calculated as follows:
wherein, Is the joint distribution of i, j,/>Is the edge distribution of i,/>Information entropy representing blower i,/>Information entropy representing blower j,/>Combined information entropy representing fans i and j,/>Mutual information representing fans i and j;
S25: constructing an adjacency matrix of the graph G based on normalized distance and normalized mutual information between fan pairs The expression is:
，
wherein, Represents the wind power of the ith fan,/>The wind power of the jth fan is represented,For the normalized distance between i and j fan pairs,/>Is normalized mutual information between fans i and j.
Further, when the fan graph characteristics of the graph characteristic input channel processing region are constructed, a graph characteristic selection module is constructed，The module is composed of two parts, identified/>And G, extracting low-dimensional potential features by using a space-time coupling mode; the first sub-module will/>, via a full connection layer network NNMapping to a potential eigenvector/>; The second sub-module is a deep network structure of a graph roll lamination GCN-GRU layer combined with a gating circulation unit network GRU and is based on potential eigenvectors/>And graph G, extracting potential spatiotemporal features/>。
Further, the meteorological feature input channel processes meteorological features as follows:
s26: overlapping m meteorological feature alignment time stamps with the length of the historical record T in the region to form a two-dimensional tensor And then/>Expanded to one/>A long vector of dimensions;
S27: constructing meteorological feature selection module based on L-1 norm regularization Passing the long vector obtained in S26 throughSelecting a set of mean-valid meteorological features/>, from a high-dimensional input spaceThe expression is:
。
Further, the specific operation in step S3 is:
Building global feature pooling layer module Fan feature selection module/>, in three feature channelsGraph feature selection Module/>Meteorological feature selection Module/>Separately output blower characteristics/>Graph characteristics/>Meteorological characteristics/>In series, a pooling operation based on L1 norm penalty is adopted to reduce the dimension to obtain the final potential feature/>The calculation expression is as follows:
。
further, the specific process of the training part for constructing the linear regression prediction model in the step S4 is as follows:
constructing a linear regression prediction model Utilizing final latent features/>Wind power prediction training,/>For predicting/>Wind power/>, momentThe calculation is as follows:
。
Further, the specific process of constructing the test part of the linear regression prediction model in step S5 is as follows:
obtaining a test model by using the trained test model Training/>Obtain the predicted value/>。
Further, the SCADA data in step S1 includes historical wind power data, historical wind speed data, historical wind direction data, meteorological data includes sea wave data, tide data and typhoon data, and geographic position information includes longitude and latitude and altitude of each fan; the preliminary processing of the data includes: and (3) carrying out min-max normalization processing on wind power data, wind speed data, sea wave data, tide data and typhoon data to obtain a [0,1] section, obtaining a processed wind power sequence P, a wind speed sequence S, a sea wave sequence W, a tide sequence TW and a typhoon sequence H, and carrying out sine and cosine processing on wind direction sequence data to obtain wind direction sine WS and wind direction cosine WC.
Compared with the prior art, the invention has the beneficial effects that:
According to the multi-channel depth network-based multi-source data offshore wind power prediction method, SCADA data, meteorological data and geographic position information of all fans in an offshore wind power plant in an area are firstly obtained, SCADA data alignment time stamps of all fans are overlapped into a three-dimensional tensor to form a fan characteristic input channel, and useful characteristic information is screened through a characteristic selection module based on L-1 norm regularization. And forming a map feature input channel by mapping the geographical position information of each fan, and extracting potential space-time features through a depth map attention convolution loop network. And (3) overlapping the alignment time stamps of the meteorological data in the region to form a meteorological feature input channel, and screening useful feature information through L1 norm regularization regression. And then, the features output by the three feature input channels are connected in series through the global feature pooling layer to obtain final potential feature vectors, the final potential feature vectors are input into the prediction model, the information collected by the fan SCADA, the geographical position information of the fan and the meteorological information are fully utilized, the prediction errors caused by wind power generation uncertainty are effectively reduced by combining the three feature input information, and the prediction precision of the wind power of the offshore wind farm is improved.
Drawings
FIG. 1 shows a flow diagram of a multi-channel depth network-based multi-source data offshore wind power prediction method.
FIG. 2 shows a detailed development flow diagram of a multi-channel depth network-based multi-source data offshore wind power prediction method.
Fig. 3 shows a structure diagram of a multi-channel deep network feature processing model in embodiment 1 of the present invention.
Fig. 4 shows a graph of wind power prediction effect obtained by using the multi-channel depth network-based multi-source data offshore wind power prediction method in embodiment 2 of the invention.
FIG. 5 shows a block diagram of a multi-channel depth network-based multi-source data offshore wind turbine prediction system in accordance with embodiment 3 of the present invention.
Detailed Description
The invention is further described below in connection with the following detailed description. Wherein the drawings are for illustrative purposes only and are shown in schematic, non-physical, and not intended to be limiting of the present patent; for the purpose of better illustrating embodiments of the invention, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the size of the actual product; it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings are merely for convenience in describing the present invention and simplifying the description, and do not indicate or imply that the device or element being referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present invention, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise.
In the present invention, unless explicitly specified and limited otherwise, the terms "mounted," "connected," "secured," and the like are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally formed; can be mechanically or electrically connected; either directly or indirectly, through intermediaries, or both, may be in communication with each other or in interaction with each other, unless expressly defined otherwise. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
In the present invention, unless expressly stated or limited otherwise, a first feature "up" or "down" a second feature may be the first and second features in direct contact, or the first and second features in indirect contact via an intervening medium. Moreover, a first feature being "above," "over" and "on" a second feature may be a first feature being directly above or obliquely above the second feature, or simply indicating that the first feature is level higher than the second feature. The first feature being "under", "below" and "beneath" the second feature may be the first feature being directly under or obliquely below the second feature, or simply indicating that the first feature is less level than the second feature.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
Example 1
The embodiment provides a multi-channel depth network-based multi-source data offshore wind power prediction method, a flow chart of which is shown in fig. 1 and 2, and referring to fig. 1, the method comprises the following steps:
S1, SCADA data, meteorological data and geographic position information of a fan of an offshore wind farm in an area are obtained, and related data are subjected to preliminary processing;
S2, overlapping SCADA data alignment time stamps of all fans into a three-dimensional tensor to form a fan characteristic input channel, and screening useful characteristic information through a characteristic selection module based on L-1 norm regularization;
Regarding each fan in the area as a node, wherein each node corresponds to the geographic position information of the fan, determining the spatial relationship of each pair of fans according to the distance and mutual information between the fans, and establishing an area fan graph between the fans to form a graph characteristic input channel;
overlapping the alignment time stamps of the meteorological data in the region to form a meteorological feature input channel, and screening useful feature information through an L1 norm regularized regression feature selection module;
s21, superposing N characteristic alignment time stamps of P typhoon SCADA records with the history record length of T in the area to form a three-dimensional characteristic vector And then/>Expanded to one/>A long vector of dimensions;
s22, constructing a fan characteristic selection module based on L-1 norm regularization Passing the long vector obtained in S21Selecting a set of mean-effective fan features/>, from a high-dimensional input spaceThe expression is:
；
feature extraction module for constructing graph meaning convolution recursion network The module is formed by stacking n layers of GCN-LSTM, and for the total time step T, at the moment T, for a fan j, the input/>, of the model is obtained according to whether the number of layers of GCN-LSTM is equal to 1 or notInput features/>, when the GCN-LSTM layer number l=1By features/>Obtained, when GCN-LSTM layer number l is not equal to 1, feature/>From the last hidden layer state/>Obtained, then input features/>And last state/>、/>Input gate, forget gate and output gate input to LSTM, and vectors output by the three gates are/>, respectively、/>、/>At the same time, will/>And/>Coding as an output vector/>These four sets of vectors/>、/>、/>、/>Respectively synthesized into four matrixes/>、/>、/>And/>Then matrix/>And/>Multiplying by Laplace matrix/>The matrix integrates all the characteristics of the fan, and finally, the information in the four matrixes is mixed through linear transformation to obtain the state/>, of the current step、/>. The GCN-LSTM layer I can be expressed by the equation:
wherein, ，/>；/>Representing elements in the adjacency matrix; /(I)An input representing a model; /(I)Representing a last hidden layer state of the model; /(I)、/>Representing the final hidden layer state of the model; /(I)、、/>、/>、/>、/>、/>、/>、/>Representing weight parameters of the corresponding hidden layer; /(I)、/>、/>、/>Representing deviation parameters of the corresponding hidden layer; /(I)、/>、/>Output vectors representing LSTM input gates, forget gates, output gates; /(I)The representation is composed ofAnd/>Encoding the output vector of the formed neuron; /(I)、/>、/>And/>Representation by four sets of vectors/>、/>、/>、A synthetic matrix; for each time step t, matrix/>And/>Multiplying by the laplace matrix L as a matrix of attention to extract the spatial pattern. For n GCN-LSTM layers, normalized LN operator pairs/>, are appliedNormalized and series these features to give/>Then, using a single layer neural network will/>Conversion to spatiotemporal features/>:
Representing geographical information of fans in an area asWherein P represents the number of fans and M represents the features contained in the geographic information: longitude, latitude and altitude; building a multi-fan graph G, wherein each node corresponds to one wind turbine, and the edge reflects the spatial relationship between every two fans, and the specific process is as follows:
S23: the normalized distance ND between the fan pairs is calculated, and the normalized distance ND between the fans i, j is calculated as follows:
wherein, Representing the longitude of the ith blower,/>Represents the longitude of the jth fan,/>Represents the latitude of the ith fan,/>Represents the latitude of the jth fan,/>Representing the elevation of the ith blower,/>Representing the elevation of the jth fan,R=6371 denotes the earth radius,/>Intermediate variable representing transition,/>，/>Representing the distance between fans i, j,/>Represents a distance maximum,/>Representing a distance minimum;
s24: calculating normalized mutual information NMI between fan pairs, and enabling wind power of an ith fan to be equal to that of the ith fan The wind power/>, expressed as I, of the jth fanDenoted J, normalized mutual information NMI between fans i and J is calculated as follows:
wherein, Is the joint distribution of i, j,/>Is the edge distribution of i,/>Information entropy representing blower i,/>Information entropy representing blower j,/>Combined information entropy representing fans i and j,/>Mutual information representing fans i and j;
S25: constructing an adjacency matrix of the graph G based on normalized distance and normalized mutual information between fan pairs The expression is:
，
wherein, Represents the wind power of the ith fan,/>The wind power of the jth fan is represented,For the normalized distance between i and j fan pairs,/>Normalized mutual information between fans i and j;
The expression of the adjacency matrix of G obtained is:
When the fan graph characteristics of the region are processed by the graph characteristic input channel, a graph characteristic selection module is constructed ，/>The module is composed of two parts, identified/>And G, extracting low-dimensional potential features by using a space-time coupling mode; the first sub-module will/>, via a full connection layer network NNMapping to a potential eigenvector/>; The second sub-module is a deep network structure of a graph roll lamination GCN-GRU layer combined with a gate control loop unit network GRU, and is based on potential feature vectorsAnd graph G, extracting potential spatiotemporal features/>。
S26: overlapping m meteorological feature alignment time stamps with the length of the historical record T in the region to form a two-dimensional tensorAnd then/>Expanded to one/>Long vector of dimension,/>The expression of (2) is:
s27, constructing a meteorological feature selection module based on L-1 norm regularization Passing the long vector obtained in S26 throughSelecting a set of mean-valid features/>, from a high-dimensional input spaceThe expression is:
；
S3, constructing a multi-channel depth network feature processing model, wherein as shown in FIG. 3, a fan feature input channel is used for processing three-dimensional tensor features, a map feature input channel is used for processing regional fan map features, a meteorological feature input channel is used for processing meteorological features, and finally, the features output by the three feature channels are connected in series through a global feature pooling layer to form final potential features of an input prediction model;
Building global feature pooling layer module Fan feature selection module/>, in three feature channelsGraph feature selection Module/>Meteorological feature selection Module/>Separately output blower characteristics/>Graph characteristics/>Meteorological characteristics/>In series, a pooling operation based on L1 norm penalty is adopted to reduce the dimension to obtain the final potential feature/>The calculation expression is as follows:
；
S4, constructing a training part of the linear regression prediction model, and training by utilizing final potential features obtained by the multi-channel depth network feature processing model;
constructing a linear regression prediction model Utilizing final latent features/>Wind power prediction training,/>For predicting/>Wind power/>, momentThe calculation is as follows:
；
training each sub-module of the multichannel deep network characteristic processing model and a prediction model respectively:
Wherein ① divides the data set obtained in the S1 into a training set and a testing set;
② Feature extraction module of training graph meaning convolution recursion network Dividing the data set into/>, by adopting a batch training methodBatch B instances, each batch expressed as/>Each batch/>And G pass/>Extracting to obtain potential characteristics/>Then will/>, through a single full connectivity layer network NNConversion to predicted wind powerBased on local loss function/>, using a back propagation algorithmPair/>Updating parameters of (a);
③ Training feature selection module Obtain a set of effective subset features/>Then, a linear regression layer LR is used to obtain the/>Wind power/>;
④ Training feature selection moduleObtain a set of effective subset features/>Then, a linear regression layer LR is used to obtain the/>Wind power/>。
⑤ Training global feature fusion moduleFirst three feature engineering channels/>、/>、/>Output characteristic pass/>Conversion to final latent fusion feature/>Then obtaining a predicted value/>, through a linear regression layer LR。
S5: a test part of a linear regression prediction model is constructed, and a training result is input into the prediction model to predict the wind power of a target fan;
training predictive models Utilizing the final latent fusion feature/>Training/>After that, the test set is used to obtain the predicted value/>。
Example 2
In order to verify the effectiveness of the multi-source data offshore wind power prediction method of the multi-channel depth network provided by the invention, first in step S1, SCADA data, meteorological data and geographic position information of a certain offshore wind farm 2020/1/1/0:00-2020/12/31/23:59 time period in guangdong are obtained, the data are preprocessed, then a fan characteristic input channel, a graph characteristic input channel and a meteorological characteristic input channel are formed according to the preprocessed data, the number of GCN-LSTM layers of the graph-meaning convolution recursive network is set to be 4, then a prediction effect graph of the offshore wind power of the multi-source data offshore wind power prediction method of the multi-channel depth network provided by the invention is utilized, in fig. 4, the dotted trend line represents a true offshore wind power value, the solid trend line represents the true offshore wind power value, and the wind power prediction value error is very small as can be seen from curves corresponding to the two values shown in fig. 4, and the accuracy of the offshore wind power prediction can be effectively improved by utilizing the method provided by the invention.
Example 3
As shown in fig. 5, the present application proposes a multi-channel depth network-based multi-source data offshore wind power prediction system, referring to fig. 5, the system includes:
the offshore wind farm data acquisition and processing unit is used for acquiring SCADA data, meteorological data and geographic position information of the offshore wind farm fans in the area and performing preliminary processing on related data;
the fan characteristic input unit is used for superposing SCADA data alignment time stamps of all fans into a three-dimensional tensor to form a fan characteristic input channel, and useful characteristic information is screened through the characteristic selection module based on L-1 norm regularization;
The map feature input unit is used for regarding each fan in the area as a node, each node corresponds to the geographic position information of the fan, the spatial relationship of each pair of fans is determined according to the distance and mutual information between the fans, and a map feature input channel is formed by the regional fan map between the fans;
The meteorological feature input unit is used for superposing the meteorological data alignment time stamps in the area to form a meteorological feature input channel, and screening useful feature information through the L1 norm regularized regression feature selection module;
The model construction unit is used for constructing a multi-channel depth network feature processing model and a linear regression prediction model, wherein a fan feature input channel is used for processing three-dimensional tensor features, a graph feature input channel is used for processing regional fan graph features, a meteorological feature input channel is used for processing meteorological features, and finally, the features output by the three feature channels are connected in series through the global feature pooling layer to form potential features of the final input prediction model; the linear regression prediction model predicts final potential features obtained by utilizing the multi-channel depth network feature processing model;
The model training unit is a training part of the linear regression prediction model; each sub-module and the prediction model are used for training the multichannel deep network characteristic processing model;
The prediction unit is a test part of the linear regression prediction model and is used for inputting final potential features obtained by the multi-channel depth network feature processing model into the final linear regression prediction model to predict the wind power of the target fan;
The offshore wind farm data acquisition and processing unit is respectively in communication connection with the fan characteristic input unit and the weather characteristic input unit, the fan characteristic input unit is also in communication connection with the graph characteristic input unit, the fan characteristic input unit, the graph characteristic input unit and the weather characteristic input unit are all in communication connection with the model building unit, and the model training unit is respectively in communication connection with the model building unit and the prediction unit.
It is to be understood that the above examples of the present invention are provided by way of illustration only and not by way of limitation of the embodiments of the present invention. Other variations or modifications of the above teachings will be apparent to those of ordinary skill in the art. It is not necessary here nor is it exhaustive of all embodiments. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the invention are desired to be protected by the following claims.
Claims (10)
1. A multi-channel depth network-based multi-source data offshore wind power prediction method is characterized by comprising the following steps:
S1, acquiring SCADA data acquisition and monitoring control data, meteorological data and geographic position information of a fan of an offshore wind farm in an area, and performing primary processing on related data;
S2, acquiring a fan characteristic input channel and processing three-dimensional tensor characteristics;
acquiring a graph characteristic input channel and processing regional fan graph characteristics;
Acquiring a meteorological feature input channel and processing meteorological features;
S3, constructing a multi-channel depth network feature processing model, and connecting the features output by the three feature channels in series through a global feature pooling layer and processing to form final potential features;
S4, constructing a training part of the linear regression prediction model, and training by utilizing final potential features obtained by the multi-channel depth network feature processing model;
s5: and (3) constructing a test part of the linear regression prediction model, and inputting a training result into the prediction model to predict the wind power of the target fan.
2. The multi-channel depth network-based multi-source data offshore wind power prediction method according to claim 1, wherein the operation of step S2 is specifically as follows:
The SCADA data alignment time stamps of all fans are overlapped to form a three-dimensional tensor to form a fan characteristic input channel, and useful characteristic information is screened through a characteristic selection module based on L-1 norm regularization;
Regarding each fan in the area as a node, wherein each node corresponds to the geographic position information of the fan, determining the spatial relationship of each pair of fans according to the distance and mutual information between the fans, and establishing an area fan graph between the fans to form a graph characteristic input channel;
And overlapping the alignment time stamps of the meteorological data in the region to form a meteorological feature input channel, and screening useful feature information through an L1 norm regularized regression feature selection module.
3. The multi-channel depth network-based multi-source data offshore wind power prediction method according to claim 2, wherein the operation of the fan feature input channel to process the three-dimensional tensor feature is as follows:
s21, superposing N characteristic alignment time stamps of P typhoon SCADA records with the history record length of T in the area to form a three-dimensional tensor And then/>Expanded to one/>A long vector of dimensions;
s22, constructing a fan characteristic selection module based on L-1 norm regularization Passing the long vector obtained in S21/>Selecting a set of mean-effective fan features/>, from a high-dimensional input spaceThe expression is:
。
4. The multi-channel depth network-based multi-source data offshore wind power prediction method according to claim 2, wherein the operation of the map feature input channel processing region fan map feature is as follows:
Representing geographical information of fans in an area as Wherein P represents the number of fans and M represents the features contained in the geographic information: longitude, latitude and altitude; constructing a multi-fan graph G, wherein each node corresponds to one wind turbine, and the edge reflects the spatial relationship between every two fans, and the specific process is as follows:
S23: the normalized distance ND between the fan pairs is calculated, and the normalized distance ND between the fans i, j is calculated as follows:
wherein, Representing the longitude of the ith blower,/>Represents the longitude of the jth fan,/>Represents the latitude of the ith fan,/>Represents the latitude of the jth fan,/>Representing the elevation of the ith blower,/>Representing the elevation of the jth fan,R=6371 denotes the earth radius,/>Intermediate variable representing transition,/>，/>Representing the distance between fans i, j,/>Represents a distance maximum,/>Representing a distance minimum;
s24: calculating normalized mutual information NMI between fan pairs, and enabling wind power of an ith fan to be equal to that of the ith fan Representation of
Wind power of the j-th fanDenoted J, normalized mutual information NMI between fans i and J is calculated as follows:
wherein, Is the joint distribution of i, j,/>Is the edge distribution of i,/>The information entropy of the fan i is represented,Information entropy representing blower j,/>Combined information entropy representing fans i and j,/>Mutual information representing fans i and j;
S25: constructing an adjacency matrix of the graph G based on normalized distance and normalized mutual information between fan pairs The expression is:
，
wherein, Represents the wind power of the ith fan,/>Represents the wind power of the jth fan,/>For the normalized distance between i and j fan pairs,/>Is normalized mutual information between fans i and j.
5. The multi-channel depth network-based multi-source data offshore wind power prediction method according to claim 4, wherein when the map feature input channel processing region fan map feature is constructed, a map feature selection module is constructed，/>The module is composed of two parts, identified/>And G, extracting low-dimensional potential features by using a space-time coupling mode; the first sub-module will/>, via a full connection layer network NNMapping to a potential eigenvector/>; The second sub-module is a deep network structure of a graph roll lamination GCN-GRU layer combined with a gating circulation unit network GRU and is based on potential eigenvectors/>And graph G, extracting potential spatiotemporal features/>。
6. The multi-channel depth network-based multi-source data offshore wind power prediction method according to claim 2, wherein the meteorological feature input channel processes meteorological features as follows:
s26: overlapping m meteorological feature alignment time stamps with the length of the historical record T in the region to form a two-dimensional tensor And then/>Expanded to one/>A long vector of dimensions;
S27: constructing meteorological feature selection module based on L-1 norm regularization Passing the long vector obtained in S26/>Selecting a set of mean-valid meteorological features/>, from a high-dimensional input spaceThe expression is:
。
7. The multi-channel depth network-based multi-source data offshore wind power prediction method according to claim 1, wherein the specific operations in step S3 are as follows:
Building global feature pooling layer module Fan feature selection module/>, in three feature channelsGraph feature selection Module/>Meteorological feature selection Module/>Separately output blower characteristics/>Graph characteristics/>Meteorological characteristics/>In series, a pooling operation based on L1 norm penalty is adopted to reduce the dimension to obtain the final potential feature/>The calculation expression is as follows:
。
8. the multi-channel depth network-based multi-source data offshore wind power prediction method according to claim 7, wherein the specific process of constructing the training part of the linear regression prediction model in step S4 is as follows:
constructing a linear regression prediction model Utilizing final latent features/>Wind power prediction training,/>For predicting/>Wind power/>, momentThe calculation is as follows:
。
9. The multi-channel depth network-based multi-source data offshore wind power prediction method according to claim 8, wherein the specific process of constructing the test part of the linear regression prediction model in step S5 is as follows:
obtaining a test model by using the trained test model Training/>Obtain the predicted value/>。
10. The multi-channel depth network-based multi-source data offshore wind power prediction method according to claim 1, wherein the SCADA data in the step S1 comprises historical wind power data, historical wind speed data and historical wind direction data, the meteorological data comprises sea wave data, tide data and typhoon data, and the geographic position information comprises longitude and latitude and altitude of each fan; the preliminary processing of the data includes: and (3) carrying out min-max normalization processing on wind power data, wind speed data, sea wave data, tide data and typhoon data to obtain a [0,1] section, obtaining a processed wind power sequence P, a wind speed sequence S, a sea wave sequence W, a tide sequence TW and a typhoon sequence H, and carrying out sine and cosine processing on wind direction sequence data to obtain wind direction sine WS and wind direction cosine WC.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410303887.5A CN117934208B (en) | 2024-03-18 | 2024-03-18 | Multi-channel depth network-based multi-source data offshore wind power prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410303887.5A CN117934208B (en) | 2024-03-18 | 2024-03-18 | Multi-channel depth network-based multi-source data offshore wind power prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117934208A true CN117934208A (en) | 2024-04-26 |
| CN117934208B CN117934208B (en) | 2024-06-11 |
Family
ID=90766947
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410303887.5A Active CN117934208B (en) | 2024-03-18 | 2024-03-18 | Multi-channel depth network-based multi-source data offshore wind power prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117934208B (en) |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103617462A (en) * | 2013-12-10 | 2014-03-05 | 武汉大学 | Geostatistics-based wind power station wind speed spatio-temporal data modeling method |
| KR20190118387A (en) * | 2018-04-10 | 2019-10-18 | 배재대학교 산학협력단 | Convolutional neural network based image processing system and method |
| US20220373984A1 (en) * | 2021-05-19 | 2022-11-24 | Shandong University | Hybrid photovoltaic power prediction method and system based on multi-source data fusion |
| CN115392525A (en) * | 2022-05-10 | 2022-11-25 | 云南电网有限责任公司电力科学研究院 | Distributed wind power prediction method, device and equipment |
| CN116070769A (en) * | 2023-02-17 | 2023-05-05 | 华北电力大学 | Ultra-short-term wind power plant power multi-step interval prediction modularization method and device thereof |
| CN116454875A (en) * | 2023-04-17 | 2023-07-18 | 国网新疆电力有限公司电力科学研究院 | Regional wind farm mid-term power probability prediction method and system based on cluster division |
| CN116485004A (en) * | 2023-03-14 | 2023-07-25 | 中国农业大学 | Wind power prediction method based on power heat map and convolutional neural network |
| CN116894384A (en) * | 2023-07-11 | 2023-10-17 | 湖北工业大学 | Multi-fan wind speed space-time prediction method and system |
| CN116960954A (en) * | 2023-07-19 | 2023-10-27 | 贵州电网有限责任公司 | Short-term wind power prediction method and device, storage medium and computer equipment |
| CN117151927A (en) * | 2023-08-31 | 2023-12-01 | 南京信息工程大学 | Prediction method for wind power probability interval |
| CN117252448A (en) * | 2023-11-20 | 2023-12-19 | 华东交通大学 | Wind power prediction method based on space-time feature extraction and secondary decomposition polymerization |
| CN117521497A (en) * | 2023-11-06 | 2024-02-06 | 国网浙江省电力有限公司青田县供电公司 | Renewable energy power station power prediction method and system based on deep learning |
| CN117543567A (en) * | 2023-12-01 | 2024-02-09 | 国家电投集团广西兴安风电有限公司 | Wind power prediction method and device for reactive power optimization of wind farm |
-
2024
- 2024-03-18 CN CN202410303887.5A patent/CN117934208B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103617462A (en) * | 2013-12-10 | 2014-03-05 | 武汉大学 | Geostatistics-based wind power station wind speed spatio-temporal data modeling method |
| KR20190118387A (en) * | 2018-04-10 | 2019-10-18 | 배재대학교 산학협력단 | Convolutional neural network based image processing system and method |
| US20220373984A1 (en) * | 2021-05-19 | 2022-11-24 | Shandong University | Hybrid photovoltaic power prediction method and system based on multi-source data fusion |
| CN115392525A (en) * | 2022-05-10 | 2022-11-25 | 云南电网有限责任公司电力科学研究院 | Distributed wind power prediction method, device and equipment |
| CN116070769A (en) * | 2023-02-17 | 2023-05-05 | 华北电力大学 | Ultra-short-term wind power plant power multi-step interval prediction modularization method and device thereof |
| CN116485004A (en) * | 2023-03-14 | 2023-07-25 | 中国农业大学 | Wind power prediction method based on power heat map and convolutional neural network |
| CN116454875A (en) * | 2023-04-17 | 2023-07-18 | 国网新疆电力有限公司电力科学研究院 | Regional wind farm mid-term power probability prediction method and system based on cluster division |
| CN116894384A (en) * | 2023-07-11 | 2023-10-17 | 湖北工业大学 | Multi-fan wind speed space-time prediction method and system |
| CN116960954A (en) * | 2023-07-19 | 2023-10-27 | 贵州电网有限责任公司 | Short-term wind power prediction method and device, storage medium and computer equipment |
| CN117151927A (en) * | 2023-08-31 | 2023-12-01 | 南京信息工程大学 | Prediction method for wind power probability interval |
| CN117521497A (en) * | 2023-11-06 | 2024-02-06 | 国网浙江省电力有限公司青田县供电公司 | Renewable energy power station power prediction method and system based on deep learning |
| CN117252448A (en) * | 2023-11-20 | 2023-12-19 | 华东交通大学 | Wind power prediction method based on space-time feature extraction and secondary decomposition polymerization |
| CN117543567A (en) * | 2023-12-01 | 2024-02-09 | 国家电投集团广西兴安风电有限公司 | Wind power prediction method and device for reactive power optimization of wind farm |
Non-Patent Citations (4)
| Title |
|---|
| 李飞;纪元;: "基于高层气象大数据的风电场中长期风功率预测研究", 电力大数据, vol. 23, no. 02, 21 February 2020 (2020-02-21) * |
| 殷豪;欧祖宏;陈德;孟安波;: "基于二次模式分解和级联式深度学习的超短期风电功率预测", 电网技术, vol. 44, no. 02, 23 September 2019 (2019-09-23) * |
| 高盛许;沛华;陈正洪;成驰;: "基于改进LSTM神经网络的风电功率短期预报算法", 科学技术与工程, vol. 01, no. 11, 30 January 2024 (2024-01-30) * |
| 黄玲玲;石孝华;符杨;刘阳;应飞祥;: "基于DCGCN模型的海上风电场超短期功率预测", 电力系统自动化, 9 April 2024 (2024-04-09) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117934208B (en) | 2024-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20220373984A1 (en) | Hybrid photovoltaic power prediction method and system based on multi-source data fusion | |
| JP6759966B2 (en) | How to operate the photovoltaic power generation system | |
| Chen et al. | Online 24-h solar power forecasting based on weather type classification using artificial neural network | |
| CN109902885A (en) | Typhoon prediction method based on deep learning mixed CNN-LSTM model | |
| CN103268366A (en) | Combined wind power prediction method suitable for distributed wind power plant | |
| CN114240000A (en) | Air quality prediction method based on space-time graph convolution network | |
| Labati et al. | A decision support system for wind power production | |
| CN112183625A (en) | PM based on deep learning2.5High-precision time-space prediction method | |
| CN112862274A (en) | WRF-LES and Deepar combined wind power prediction method | |
| CN105373849A (en) | Stepped temperature prediction method for photovoltaic battery assembly | |
| CN114418174A (en) | Electric vehicle charging load prediction method | |
| CN117556197A (en) | Typhoon vortex initialization method based on artificial intelligence | |
| US20240054267A1 (en) | Method for planning a layout of a renewable energy site | |
| CN114882373A (en) | Multi-feature fusion sandstorm prediction method based on deep neural network | |
| CN117934208B (en) | Multi-channel depth network-based multi-source data offshore wind power prediction method | |
| CN111488974B (en) | Ocean wind energy downscaling method based on deep learning neural network | |
| CN117493476A (en) | Runoff backtracking simulation method and system integrating physical mechanism and artificial intelligence | |
| CN117293809A (en) | Multi-time space scale new energy generation power prediction method based on large model | |
| Deng et al. | A survey of the researches on grid-connected solar power generation systems and power forecasting methods based on ground-based cloud atlas | |
| CN115860248A (en) | Wind speed numerical assimilation short-term prediction method and system | |
| Ma et al. | Multi-features fusion for short-term photovoltaic power prediction | |
| KR102411074B1 (en) | Devices and Methods for Detection and Prediction of Cold Water Mass Using Artificial Intelligence and Satellite Imaging | |
| Mohan et al. | Crop production rate estimation using parallel layer regression with deep belief network | |
| Li et al. | Deepphysinet: Bridging deep learning and atmospheric physics for accurate and continuous weather modeling | |
| CN112784477A (en) | WRF-LES and BP-PSO-Bagging combined wind power prediction method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |