CN117094426A - Station area division type photovoltaic power station group power prediction method - Google Patents
Station area division type photovoltaic power station group power prediction method Download PDFInfo
- Publication number
- CN117094426A CN117094426A CN202310727851.5A CN202310727851A CN117094426A CN 117094426 A CN117094426 A CN 117094426A CN 202310727851 A CN202310727851 A CN 202310727851A CN 117094426 A CN117094426 A CN 117094426A
- Authority
- CN
- China
- Prior art keywords
- space
- photovoltaic power
- station
- time
- low
- 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 37
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 18
- 230000003068 static effect Effects 0.000 claims description 28
- 239000011159 matrix material Substances 0.000 claims description 24
- 238000000605 extraction Methods 0.000 claims description 18
- 230000004927 fusion Effects 0.000 claims description 16
- 238000013507 mapping Methods 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 230000008034 disappearance Effects 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims 1
- 238000013527 convolutional neural network Methods 0.000 claims 1
- 238000005516 engineering process Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- General Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Quality & Reliability (AREA)
- Photovoltaic Devices (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Operations Research (AREA)
Abstract
The invention discloses a power prediction method of a distributed photovoltaic power station group in a station area, which comprises sequence decomposition, low-frequency component space-time prediction, high-frequency component space-time prediction and sequence addition.
Description
Technical Field
The invention relates to the technical field of power input prediction of distributed photovoltaic power stations, in particular to a power prediction method of a station area distributed photovoltaic power station group.
Background
The existing distributed photovoltaic prediction technology mainly comprises an interpolation method, a reference power station method and a historical output method, wherein the interpolation method is used for calculating meteorological data of a geographical position where a distributed station is located mainly by interpolating satellite meteorological data, then a physical model method or a data driving method is used for predicting power generation, the reference power station method is used for mainly taking a certain centralized photovoltaic power station adjacent to the distributed photovoltaic station as the reference station, an output related model is established, a distributed photovoltaic output predicted value is calculated by means of a power predicted value of the reference power station, and the historical output method is used for establishing a time sequence prediction model only based on the historical output data of the distributed photovoltaic under the condition that the meteorological data are not available;
however, in the existing distributed photovoltaic prediction technology, the interpolation method depends on satellite meteorological data, but the spatial resolution of the satellite meteorological data is lower, the error of the meteorological interpolation result is larger, the error of the power prediction result is larger, the reference power station method depends on the meteorological data provided by adjacent centralized photovoltaic power stations, but in practice, centralized photovoltaic power stations cannot be ensured to exist around all distributed photovoltaic power stations, so that the applicability is limited, the historical output method is often used for predicting the power of a single distributed photovoltaic power station, the consideration of the spatial correlation among the output forces of a plurality of power stations is lacked, the prediction error is larger, and the model adaptability among different power stations is poor and cannot be shared.
Disclosure of Invention
The invention provides a power prediction method for a regional distributed photovoltaic power station group, which can effectively solve the problems that in the background technology, the current existing distributed photovoltaic prediction technology is proposed, an interpolation method depends on satellite meteorological data, the spatial resolution of the satellite meteorological data is lower, the error of a meteorological interpolation result is larger, the error of the power prediction result is larger, a reference power station method depends on meteorological data provided by a nearby centralized photovoltaic power station, but in practice, centralized photovoltaic power stations cannot be ensured around all distributed photovoltaic power stations, so that applicability is limited, a historical output method often carries out power prediction on a single distributed photovoltaic power station, consideration of spatial correlation among output of a plurality of power stations is lacked, prediction error is larger, model adaptability among different power stations is poor, and the power stations cannot be shared.
In order to achieve the above purpose, the present invention provides the following technical solutions: a power prediction method of a station area type photovoltaic power station group fully considers the time-space correlation of different components of a photovoltaic power station group output sequence based on a means of sequence decomposition and gate control fusion graph convolution, and realizes simultaneous prediction of power of a plurality of photovoltaic power stations in the station area based on historical output data only under the condition of no meteorological information, and specifically comprises the following steps:
s1, sequence decomposition, namely decomposing an original photovoltaic output sequence into a low-frequency sequence component and a plurality of high-frequency sequence components by utilizing extremely-overlapped discrete wavelet transformation;
s2, low-frequency component space-time prediction, wherein the low-frequency component is predicted based on a static graph convolution network and a transducer model;
s3, predicting high-frequency components in a space-time mode based on the gating fusion graph convolution and a transducer model;
s4, adding the sequences, and adding the low-frequency predicted value and the high-frequency predicted value, so as to obtain a final predicted result.
According to the above technical solution, in S1, the sequence decomposition specifically includes the following implementation steps:
s101, inputting a model;
s102, normalizing the maximum value;
s103, wavelet decomposition.
According to the above technical solution, in S101, the model is input as the output P of the N photovoltaic power stations at T historic moments in the transformer area i,t A matrix P consisting of i=1, 2 …, N, t=1, 2 …, T;
in S102, maximum normalization refers to normalizing the output sequences of N photovoltaic power stations in the transformer area, and there are
In S103, the wavelet decomposition mainly refers to decomposing the photovoltaic output sequence into a trend component a3 and a fluctuation component d1, d2, d3 by using a wavedec function and a wrcouf function in a matlab wavelet analysis tool box, and selecting a 'db 5' wavelet, and the specific procedure is as follows:
[C,L]=wavedec(P i ,3,'db5');
d1=wrcoef('d'C,L,wname,1);
d2=wrcoef('d',C,L,wname,2);
d3=wrcoef('d',C,L,,wname,3);
a3=wrcoef('a',C,L,wname,3)。
according to the above technical solution, in S2, the low-frequency component space-time prediction specifically includes the following implementation steps:
s201, a space-time feature extraction layer;
s202, predicting a layer.
According to the above technical solution, in S201, the space-time feature extraction layer is a space-time feature for constructing a static graph convolution network to extract low-frequency components, and the specific algorithm is as follows:
firstly, solving the geographical distance of each photovoltaic power station according to a Haverine formula:
d ij =Rarccos(cosβ i cosβ j cos(α i -α j )+sinβ i sinβ j )(i,j=1,2,...,N);
constructing a static diagram structure adjacency matrix according to the geographical distance of the photovoltaic power station:
then constructing a double-layer SGCN to extract the spatial characteristics of the photovoltaic power station at each moment:
wherein,for self-connected static adjacency matrix +.>Degree matrix being adjacency matrix, P t A For the value of the low-frequency component at time t, W 1 、W 2 Is a weight parameter;
splicing T historical moment output spatial features
The spatial characteristics of each moment extracted by SGCN and the original low-frequency components are spliced and then input into an encoder structure of a transducer, and the specific algorithm is as follows:
X A =concat(H A ,P A )
x is to be A Is decomposed intoWherein->
Corresponding position coding:
wherein,is->Position codes corresponding to the jth column of the t row;
will beAnd (3) adding the position code E in a para position: there is->
From the following componentsLinear mapping is performed to obtain:
obtaining space-time characteristics through dot multiplication:
in the step S202, the prediction layer refers to, to alleviate the problems of overfitting and gradient disappearance, adopting residual connection to splice the space-time features extracted in the step S201 with the original input and then inputting the spliced space-time features into the full-connection layer to obtain the predicted value of the low-frequency component of the photovoltaic power station group:
transformer outputs low-frequency component predicted values of N power stations in parallel
According to the above technical solution, in S3, the high-frequency component space-time prediction specifically includes the following implementation steps:
s301, static space-time feature extraction layer
S302, a dynamic space-time feature extraction layer;
s303, gating a fusion layer;
s304, a full connection layer.
According to the above technical solution, in S301, the static space-time feature extraction layer refers to constructing a double-layer SGCN to extract the space features of the photovoltaic power station at each moment, as shown in the following formula:
wherein,for self-connected static adjacency matrix +.>As a degree matrix of the adjacency matrix,for the value of the low-frequency component at time t, W 1 、W 2 Is a weight parameter;
splicing T historical moment output spatial features
The spatial characteristics of each moment extracted by SGCN and the original low-frequency components are spliced and then input into an encoder structure of a transducer, and the specific algorithm is as follows:
X SD1 =concat(H SD1 ,P SD1 )
x is to be D1 Is decomposed intoWherein (1)>
Will beAnd (3) adding the position code E in a para position:
from the following componentsLinear mapping is performed to obtain:
obtaining space-time characteristics through dot multiplication:
according to the above technical solution, in S302, the dynamic space-time feature extraction layer refers to constructing a dynamic graph convolution neural network to extract the space features of each moment of the photovoltaic power station:
wherein A is t Representing dynamic dependencies between nodes;
splicing T historical moment output spatial features
The space characteristics of each moment extracted by the DGCN and the original low-frequency components are spliced and then input into the encoder structure of the transducer, and the specific algorithm is as follows:
X DD1 =concat(H DD1 ,P D1 )
x is to be DD1 Is decomposed intoWherein (1)>
Will beAnd (3) adding the position code E in a para position: there is->
From the following componentsLinear mapping is performed to obtain:
obtaining space-time characteristics through dot multiplication:
according to the above technical solution, in S303, the gating fusion layer means to construct a gating fusion mechanism to fuse the space-time features learned from the fixed and dynamic graph convolution layersAnd->The resulting high frequency component P D1 Is characterized by the space-time characteristics of:
in S404, the full-connection layer refers to, in order to alleviate the problems of overfitting and gradient disappearance, adopting residual connection to splice the space-time characteristics in steps S301-S303 with the original input, and inputting the full-connection layer to obtain the predicted value of the low-frequency component of the photovoltaic power station group:
transformer outputs low-frequency component predicted values of N power stations in parallel
The same applies to obtain a low frequency component P D2 、P D3 Predicted value of (2)
According to the above technical solution, in S4, the final prediction result obtained by the sequence addition is as follows:
compared with the prior art, the invention has the beneficial effects that:
1. according to the method, the purpose of prediction can be achieved by only using the photovoltaic output data without meteorological data, the time-space correlation of different components of the output sequence of the photovoltaic power station group is fully considered, the simultaneous prediction of a plurality of photovoltaic power stations in a station area can be realized under the condition without meteorological information, and meteorological measurement equipment is not required to be additionally installed for distributed photovoltaic, so that the economic cost is saved;
meanwhile, the output of the invention is the output predicted value of a plurality of distributed photovoltaic power stations in the platform region, the output of the plurality of distributed photovoltaic power stations can be predicted at the same time, compared with single-station prediction, the space-time correlation among the output power of the multi-station photovoltaic can be better learned, the precision of regional photovoltaic prediction is improved, the training cost is reduced, the original output sequence is decomposed into low-frequency and high-frequency components for modeling respectively, the subcomponent sequence obtained after decomposition can better represent the characteristics of the original data from different dimensionalities, and more accurate predicted results can be obtained through targeted prediction modeling.
2. According to the invention, a photovoltaic output sequence is decomposed into low-frequency and high-frequency components aiming at distributed photovoltaic prediction, a space-time prediction model is respectively established to obtain predicted values of all components in view of different influence factors of all output components, a final predicted value is obtained by adding, a more accurate predicted result can be obtained by targeted modeling prediction of all components, a static graph convolution+transform model is established for low-frequency components related to static attributes, and a space feature extracted by static graph convolution and dynamic graph convolution is weighted by a gating fusion mechanism aiming at high-frequency components related to the static attributes and influenced by atmospheric motion, and the high-frequency components are predicted by combining the transform model.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention. In the drawings:
FIG. 1 is a schematic flow chart of the prediction method of the present invention;
FIG. 2 is a schematic diagram of a sequence decomposition of the present invention;
FIG. 3 is a flow chart of the low frequency component prediction of the present invention;
fig. 4 is a flow chart of the high frequency component prediction of the present invention.
Detailed Description
The preferred embodiments of the present invention will be described below with reference to the accompanying drawings, it being understood that the preferred embodiments described herein are for illustration and explanation of the present invention only, and are not intended to limit the present invention.
Examples: as shown in fig. 1-4, the present invention provides a technical scheme, a method for predicting power of a distributed photovoltaic power station group in a station, based on a means of sequence decomposition and gate control fusion graph convolution, fully considers space-time correlation of different components of a power output sequence of the photovoltaic power station group, and under the condition of no weather information, realizes simultaneous prediction of power of a plurality of photovoltaic power stations in the station based on only historical output data, and specifically includes the following steps:
s1, sequence decomposition, namely decomposing an original photovoltaic output sequence into a low-frequency sequence component and a plurality of high-frequency sequence components by utilizing extremely-overlapped discrete wavelet transformation;
s2, low-frequency component space-time prediction, wherein the low-frequency component is predicted based on a static graph convolution network and a transducer model;
s3, predicting high-frequency components in a space-time mode based on the gating fusion graph convolution and a transducer model;
s4, adding the sequences, and adding the low-frequency predicted value and the high-frequency predicted value, so as to obtain a final predicted result.
Based on the above technical solution, in S1, the sequence decomposition specifically includes the following implementation steps:
s101, inputting a model;
s102, normalizing the maximum value;
s103, wavelet decomposition.
Based on the above technical solution, in S101, the model is input as the output P of the N photovoltaic power stations at T historic moments in the station area i,t A matrix P consisting of i=1, 2 …, N, t=1, 2 …, T;
in S102, maximum normalization refers to normalizing the output sequences of N photovoltaic power stations in the transformer area, and there is
In S103, wavelet decomposition mainly refers to decomposing the photovoltaic output sequence into a trend component a3 and a fluctuation component d1, d2, d3 by using a wavedec function and a wrcoef function in a matlab wavelet analysis tool box, and selecting a wavelet of "db5", wherein the specific procedure is as follows:
[C,L]=wavedec(P i ,3,'db5');
d1=wrcoef('d'C,L,wname,1);
d2=wrcoef('d',C,L,wname,2);
d3=wrcoef('d',C,L,,wname,3);
a3=wrcoef('a',C,L,wname,3)。
based on the above technical solution, in S2, the low-frequency component space-time prediction specifically includes the following implementation steps:
s201, a space-time feature extraction layer;
s202, predicting a layer.
Based on the above technical solution, in S201, the space-time feature extraction layer is a space-time feature for constructing a static graph convolution network to extract low-frequency components, where the static graph convolution network is Static Graph Convolution Neural Networks, and the SGCN specifically includes:
firstly, solving the geographical distance of each photovoltaic power station according to a Haverine formula:
d ij =Rarccos(cosβ i cosβ j cos(α i -α j )+sinβ i sinβ j )(i,j=1,2,...,N);
constructing a static diagram structure adjacency matrix according to the geographical distance of the photovoltaic power station:
then constructing a double-layer SGCN to extract the spatial characteristics of the photovoltaic power station at each moment:
wherein,for self-connected static adjacency matrix +.>Degree matrix being adjacency matrix, P t A For the value of the low-frequency component at time t, W 1 、W 2 Is a weight parameter;
splicing T historical moment output spatial features
The spatial characteristics of each moment extracted by SGCN and the original low-frequency components are spliced and then input into an encoder structure of a transducer, and the specific algorithm is as follows:
X A =concat(H A ,P A )
x is to be A Is decomposed intoWherein->
Corresponding position coding:
wherein,is->Position codes corresponding to the jth column of the t row;
will beAnd (3) adding the position code E in a para position: thenThere is->
From the following componentsLinear mapping is performed to obtain:
obtaining space-time characteristics through dot multiplication:
in S202, the prediction layer refers to, to alleviate the problems of overfitting and gradient disappearance, adopting residual connection to splice the space-time features extracted in step S201 with the original input, and inputting the spliced space-time features into the full-connection layer to obtain the predicted value of the low-frequency component of the photovoltaic power station group:
transformer outputs low-frequency component predicted values of N power stations in parallel
Based on the above technical solution, in S3, the high-frequency component space-time prediction specifically includes the following implementation steps:
s301, static space-time feature extraction layer
S302, a dynamic space-time feature extraction layer;
s303, gating a fusion layer;
s304, a full connection layer.
Based on the above technical solution, in S301, the static space-time feature extraction layer refers to constructing a double-layer SGCN to extract the space features of the photovoltaic power station at each moment, as shown in the following formula:
wherein,for self-connected static adjacency matrix +.>Degree matrix being adjacency matrix, P t D1 For the value of the low-frequency component at time t, W 1 、W 2 Is a weight parameter;
splicing T historical moment output spatial features
The spatial characteristics of each moment extracted by SGCN and the original low-frequency components are spliced and then input into an encoder structure of a transducer, and the specific algorithm is as follows:
X SD1 =concat(H SD1 ,P SD1 )
x is to be D1 Is decomposed intoWherein (1)>
Will beAnd (3) adding the position code E in a para position:
from the following componentsLinear mapping is performed to obtain:
obtaining space-time characteristics through dot multiplication:
based on the above technical solution, in S302, the dynamic space-time feature extraction layer is to construct a dynamic graph convolution neural network to extract the space features of each moment of the photovoltaic power station, where the dynamic graph convolution neural network is Dynamic Graph Convolution Neural Networks, and the DGCN includes:
wherein A is t Representing dynamic dependencies between nodes;
splicing T historical moment output spatial features
The space characteristics of each moment extracted by the DGCN and the original low-frequency components are spliced and then input into the encoder structure of the transducer, and the specific algorithm is as follows:
X DD1 =concat(H DD1 ,P D1 )
x is to be DD1 Is decomposed intoWherein (1)>
Will beAnd (3) adding the position code E in a para position: there is->
From the following componentsLinear mapping is performed to obtain:
obtaining space-time characteristics through dot multiplication:
based on the above technical solution, in S303, the gating fusion layer means to construct a gating fusion mechanism to fuse the space-time learned from the fixed and dynamic graph convolution layersFeatures (e.g. a character)And->The resulting high frequency component P D1 Is characterized by the space-time characteristics of:
in S404, the full connection layer refers to, to alleviate the problems of overfitting and gradient disappearance, adopting residual connection to splice the space-time features in steps S301-S303 with the original input, and inputting the full connection layer to obtain the predicted value of the low-frequency component of the photovoltaic power station group:
transformer outputs low-frequency component predicted values of N power stations in parallel
The same applies to obtain a low frequency component P D2 、P D3 Predicted value of (2)
Based on the above technical solution, in S4, the final prediction result obtained by sequence addition is as follows:
finally, it should be noted that: the foregoing is merely a preferred example of the present invention, and the present invention is not limited thereto, but it is to be understood that modifications and equivalents of some of the technical features described in the foregoing embodiments may be made by those skilled in the art, although the present invention has been described in detail with reference to the foregoing embodiments. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A station area division type photovoltaic power station group power prediction method is characterized by comprising the following steps of: based on the means of sequence decomposition and gating fusion graph convolution, the space-time correlation of different components of the photovoltaic power station group output sequence is fully considered, and under the condition of no meteorological information, the simultaneous prediction of the power of a plurality of photovoltaic power stations in a station area is realized only based on historical output data, and the method specifically comprises the following steps:
s1, sequence decomposition, namely decomposing an original photovoltaic output sequence into a low-frequency sequence component and a plurality of high-frequency sequence components by utilizing extremely-overlapped discrete wavelet transformation;
s2, low-frequency component space-time prediction, wherein the low-frequency component is predicted based on a static graph convolution network and a transducer model;
s3, predicting high-frequency components in a space-time mode based on the gating fusion graph convolution and a transducer model;
s4, adding the sequences, and adding the low-frequency predicted value and the high-frequency predicted value, so as to obtain a final predicted result.
2. A method for predicting power of a group of station-area-distributed photovoltaic power stations as set forth in claim 1, wherein: in the step S1, the sequence decomposition specifically comprises the following implementation steps:
s101, inputting a model;
s102, normalizing the maximum value;
s103, wavelet decomposition.
3. A method for predicting power of a group of station-area-distributed photovoltaic power stations as set forth in claim 2, wherein: in S101, the model input is the output P of T historical moments of N photovoltaic power stations in the transformer area i,t Matrix P of composition, where i=1, 2 …,N,t=1,2…,T;
In S102, maximum normalization refers to normalizing the output sequences of N photovoltaic power stations in the transformer area, and there are
In S103, the wavelet decomposition mainly refers to decomposing the photovoltaic output sequence into a trend component a3 and a fluctuation component d1, d2, d3 by using a wavedec function and a wrcouf function in a matlab wavelet analysis tool box, and selecting a 'db 5' wavelet, and the specific procedure is as follows:
[C,L]=wavedec(P i ,3,'db5');
d1=wrcoef('d'C,L,wname,1);
d2=wrcoef('d',C,L,wname,2);
d3=wrcoef('d',C,L,,wname,3);
a3=wrcoef('a',C,L,wname,3)。
4. a method for predicting power of a group of station-area-distributed photovoltaic power stations as set forth in claim 1, wherein: in the step S2, the low-frequency component space-time prediction specifically includes the following implementation steps:
s201, a space-time feature extraction layer;
s202, predicting a layer.
5. The method for predicting power of a group of station-area-distributed photovoltaic power stations according to claim 4, wherein: in S201, the spatio-temporal feature extraction layer is a spatio-temporal feature for constructing a static graph convolution network (Static Graph Convolution Neural Networks, SGCN) to extract low-frequency components, and the specific algorithm is as follows:
firstly, solving the geographical distance of each photovoltaic power station according to a Haverine formula:
d ij =Rarccos(cosβ i cosβ j cos(α i -α j )+sinβ i sinβ j )(i,j=1,2,...,N);
constructing a static diagram structure adjacency matrix according to the geographical distance of the photovoltaic power station:
then constructing a double-layer SGCN to extract the spatial characteristics of the photovoltaic power station at each moment:
wherein,for self-connected static adjacency matrix +.>Degree matrix being adjacency matrix, P t A For the value of the low-frequency component at time t, W 1 、W 2 Is a weight parameter;
splicing T historical moment output spatial features
The spatial characteristics of each moment extracted by SGCN and the original low-frequency components are spliced and then input into an encoder structure of a transducer, and the specific algorithm is as follows:
X A =concat(H A ,P A )
x is to be A Is decomposed intoWherein->
Corresponding position coding:
wherein,is->Position codes corresponding to the jth column of the t row;
will beAnd (3) adding the position code E in a para position: there is->
From the following componentsLinear mapping is performed to obtain:
obtaining space-time characteristics through dot multiplication:
in the step S202, the prediction layer refers to, to alleviate the problems of overfitting and gradient disappearance, adopting residual connection to splice the space-time features extracted in the step S201 with the original input and then inputting the spliced space-time features into the full-connection layer to obtain the predicted value of the low-frequency component of the photovoltaic power station group:
transformer outputs low-frequency component predicted values of N power stations in parallel
6. A method for predicting power of a group of station-area-distributed photovoltaic power stations as set forth in claim 1, wherein: in the step S3, the high-frequency component space-time prediction specifically includes the following implementation steps:
s301, static space-time feature extraction layer
S302, a dynamic space-time feature extraction layer;
s303, gating a fusion layer;
s304, a full connection layer.
7. The method for predicting power of a group of station-area-distributed photovoltaic power stations according to claim 6, wherein: in S301, the static space-time feature extraction layer refers to the construction of a double-layer SGCN to extract the space features of each moment of the photovoltaic power station, as shown in the following formula:
wherein,for self-connected static adjacency matrix +.>Degree matrix being adjacency matrix, P t D1 For the value of the low-frequency component at time t, W 1 、W 2 Is a weight parameter;
splicing T historical moment output spatial features
The spatial characteristics of each moment extracted by SGCN and the original low-frequency components are spliced and then input into an encoder structure of a transducer, and the specific algorithm is as follows:
x is to be D1 Is decomposed intoWherein (1)>
Will beAnd (3) adding the position code E in a para position:
from the following componentsLinear mapping is performed to obtain:
obtaining space-time characteristics through dot multiplication:
8. the method for predicting power of a group of station-area-distributed photovoltaic power stations according to claim 6, wherein: in S302, the dynamic space-time feature extraction layer is used for constructing a dynamic graph convolutional neural network (Dynamic Graph Convolution Neural Networks, DGCN) to extract the space features of each moment of the photovoltaic power station:
wherein A is t Representing dynamic dependencies between nodes;
splicing T historical moment output spatial features
The space characteristics of each moment extracted by the DGCN and the original low-frequency components are spliced and then input into the encoder structure of the transducer, and the specific algorithm is as follows:
X DD1 =concat(H DD1 ,P D1 )
x is to be DD1 Is decomposed intoWherein (1)>
Will beAnd (3) adding the position code E in a para position: there is->
From the following componentsLinear mapping is performed to obtain:
obtaining space-time characteristics through dot multiplication:
9. the method for predicting power of a group of station-area-distributed photovoltaic power stations according to claim 6, wherein: by a means ofIn S303, the gating fusion layer is to construct a gating fusion mechanism to fuse the space-time characteristics learned from the fixed and dynamic graph convolution layersAnd->The resulting high frequency component P D1 Is characterized by the space-time characteristics of:
in S404, the full-connection layer refers to, in order to alleviate the problems of overfitting and gradient disappearance, adopting residual connection to splice the space-time characteristics in steps S301-S303 with the original input, and inputting the full-connection layer to obtain the predicted value of the low-frequency component of the photovoltaic power station group:
transformer outputs low-frequency component predicted values of N power stations in parallel
The same applies to obtain a low frequency component P D2 、P D3 Predicted value of (2)
10. A method for predicting power of a group of station-area-distributed photovoltaic power stations as set forth in claim 1, wherein: in S4, the final prediction result obtained by the sequence addition is as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310727851.5A CN117094426A (en) | 2023-06-19 | 2023-06-19 | Station area division type photovoltaic power station group power prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310727851.5A CN117094426A (en) | 2023-06-19 | 2023-06-19 | Station area division type photovoltaic power station group power prediction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117094426A true CN117094426A (en) | 2023-11-21 |
Family
ID=88775916
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310727851.5A Pending CN117094426A (en) | 2023-06-19 | 2023-06-19 | Station area division type photovoltaic power station group power prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117094426A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117293826A (en) * | 2023-11-27 | 2023-12-26 | 山东大学 | Distributed photovoltaic missing power real-time prediction method, system, medium and equipment |
-
2023
- 2023-06-19 CN CN202310727851.5A patent/CN117094426A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117293826A (en) * | 2023-11-27 | 2023-12-26 | 山东大学 | Distributed photovoltaic missing power real-time prediction method, system, medium and equipment |
CN117293826B (en) * | 2023-11-27 | 2024-04-05 | 山东大学 | Distributed photovoltaic missing power real-time prediction method, system, medium and equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110070226B (en) | Photovoltaic power prediction method and system based on convolutional neural network and meta-learning | |
Wu et al. | A comprehensive review on deep learning approaches in wind forecasting applications | |
Qu et al. | A temporal distributed hybrid deep learning model for day-ahead distributed PV power forecasting | |
CN117094426A (en) | Station area division type photovoltaic power station group power prediction method | |
CN109840633B (en) | Photovoltaic output power prediction method, system and storage medium | |
CN112613657B (en) | Short-term wind speed prediction method for wind power plant | |
CN110619442A (en) | Vehicle berth prediction method based on reinforcement learning | |
CN116307291B (en) | Distributed photovoltaic power generation prediction method and prediction terminal based on wavelet decomposition | |
CN112508279A (en) | Regional distributed photovoltaic prediction method and system based on spatial correlation | |
CN111141879B (en) | Deep learning air quality monitoring method, device and equipment | |
CN116014722A (en) | Sub-solar photovoltaic power generation prediction method and system based on seasonal decomposition and convolution network | |
Pinto et al. | Reliability‐constrained robust expansion planning of active distribution networks | |
CN113344283B (en) | Energy internet new energy consumption capability assessment method based on edge intelligence | |
CN117151927A (en) | Prediction method for wind power probability interval | |
CN117273076A (en) | Electric vehicle charging station load prediction method and system based on attention-based time-space multi-graph convolution network | |
CN115310711B (en) | New energy photovoltaic power generation power prediction method | |
CN115696203A (en) | Transformer-based smart grid 5G mobile terminal multi-track prediction method | |
CN111327453B (en) | Communication bandwidth estimation method considering gridding dynamic and static components | |
CN111950765B (en) | Probabilistic transient stability prediction method based on stacked noise reduction self-encoder | |
CN114386666A (en) | Wind power plant short-term wind speed prediction method based on space-time correlation | |
CN110069719B (en) | Internet environment-oriented behavior prediction method and prediction system thereof | |
Hou et al. | Stochastic pre‐disaster planning and post‐disaster restoration to enhance distribution system resilience during typhoons | |
CN111310103A (en) | Densification processing method for ecological environment monitoring data | |
CN117498320B (en) | Wave power generation power prediction method and system | |
CN114363951B (en) | Inter-cell flow collaborative prediction method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication |