CN114971007A - Photovoltaic power generation prediction method and system based on multi-scale graph convolutional neural network - Google Patents
Photovoltaic power generation prediction method and system based on multi-scale graph convolutional neural network Download PDFInfo
- Publication number
- CN114971007A CN114971007A CN202210560677.5A CN202210560677A CN114971007A CN 114971007 A CN114971007 A CN 114971007A CN 202210560677 A CN202210560677 A CN 202210560677A CN 114971007 A CN114971007 A CN 114971007A
- Authority
- CN
- China
- Prior art keywords
- time
- space
- photovoltaic power
- convolution
- graph
- 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
- 238000010248 power generation Methods 0.000 title claims abstract description 49
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 16
- 230000004927 fusion Effects 0.000 claims abstract description 13
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 238000010586 diagram Methods 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 15
- 230000004913 activation Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 230000002776 aggregation Effects 0.000 claims description 6
- 238000004220 aggregation Methods 0.000 claims description 6
- 235000004257 Cordia myxa Nutrition 0.000 claims description 5
- 244000157795 Cordia myxa Species 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 238000013075 data extraction Methods 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- 150000001875 compounds Chemical class 0.000 claims description 2
- 238000013507 mapping Methods 0.000 claims description 2
- 230000002123 temporal effect Effects 0.000 claims description 2
- 230000007246 mechanism Effects 0.000 abstract description 4
- 238000010276 construction Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 230000010339 dilation Effects 0.000 description 3
- 238000000053 physical method Methods 0.000 description 3
- 238000007619 statistical method Methods 0.000 description 3
- 241000196324 Embryophyta Species 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
- 238000004064 recycling Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Images
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
-
- 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
- G06N3/048—Activation functions
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
The invention provides a photovoltaic power generation prediction method and a photovoltaic power generation prediction system based on a multi-scale graph convolutional neural network, wherein an attention space-time convolution kernel is constructed, time and space characteristics of data of a plurality of photovoltaic power stations in the same area are respectively extracted by utilizing a 1x1 convolution and a 1x1 graph convolution, an attention mechanism is introduced to model the mutual relation of all moments, and the influence degree of the data at different moments on a prediction result is measured. According to the network construction method, Res2Net is introduced, the attention space-time convolution kernel is embedded into the Res2Net network, the space-time feature fusion of multiple scales is achieved, the multi-scale space-time feature expression is further achieved, and finally the photovoltaic power generation prediction of the multi-scale dynamic graph convolution neural network is achieved.
Description
Technical Field
The invention relates to the technical field of photovoltaic power generation, in particular to a photovoltaic power generation prediction method and system based on a multi-scale graph convolutional neural network.
Background
In recent years, photovoltaic grid connection is gradually popularized in the world due to the characteristics of distributed operation, capability of reducing the pressure of a power transmission and distribution system, improvement on the reliability of a power grid and the like. The photovoltaic power station can only generate electric energy in the daytime, is a typical intermittent power supply, and has larger fluctuation and randomness because the power generation power is influenced by meteorological conditions such as solar irradiation intensity, ambient temperature and the like. Due to the characteristics of photovoltaic power generation, severe impact can be brought to the stable operation of a power grid when large-scale photovoltaic power generation is connected to the grid, and adverse effects can be caused to the whole power system. If the photovoltaic power generation power can be predicted timely and accurately, the influence of the fluctuation of the photovoltaic power generation on a power grid is greatly reduced, and the method has important significance on power grid dispatching and photovoltaic power station operation. Accurate photovoltaic power prediction helps to solve the problems faced by large-scale photovoltaic development and utilization. The prediction result can be used as a decision reference for power grid scheduling and photovoltaic power station operation and maintenance.
Currently, many researchers have studied photovoltaic power generation prediction. In the existing research, the prediction technology for photovoltaic power generation is mainly divided into two categories: physical methods, statistical methods. The physical method is mainly based on photovoltaic modules and weather conditions affecting power production: solar irradiance, temperature, etc. These techniques use mathematical models to convert weather conditions into opportunities for electrical energy. Statistical methods predict photovoltaic power generation over a future period of time by analyzing historical data. Compared with a physical method, the statistical method can be better suitable for photovoltaic power generation power prediction.
In photovoltaic power generation, a plurality of prediction models such as an artificial neural network, a support vector machine, a BP neural network and the like are well used for predicting photovoltaic power generation in a plurality of documents, but the common neural network prediction model cannot effectively capture the time and space characteristics of distributed photovoltaic stations at the same time, so that the prediction precision is influenced.
Disclosure of Invention
The invention provides a photovoltaic power generation prediction method and system based on a multi-scale graph convolutional neural network, which are used for solving the problem of low prediction precision of the conventional photovoltaic power generation prediction model.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention provides a photovoltaic power generation prediction method based on a multi-scale graph convolution neural network, which comprises the following steps:
acquiring input information, and extracting time characteristics and space characteristics of photovoltaic power station data from the input information to form a space-time characteristic diagram, wherein the time characteristics are extracted through one-dimensional expansion convolution;
dividing the space-time feature graph into a plurality of feature sub-graphs along the channel dimension, calculating attention space-time convolution kernels corresponding to each feature sub-graph, wherein the attention space-time convolution kernels corresponding to the rest feature sub-graphs comprise space-time information of all previous feature sub-graphs except the attention space-time convolution kernel corresponding to the first feature sub-graph, and the feature sub-graphs and the corresponding attention space-time convolution kernels form space-time features of the current feature sub-graphs;
and splicing the space-time characteristics of various scales along the channel dimension to form spliced space-time characteristics consistent with the dimension of the input information, and performing weighted fusion on the space-time characteristics of various scales in the spliced space-time characteristics to obtain a photovoltaic power generation prediction result.
Further, the extracting of the time feature specifically includes:
wherein Z is the extracted time characteristic, delta is the Selu activation function,
and d is a one-dimensional expansion convolution operator, and H is the original data of the time characteristics of the photovoltaic power station.
Further, the spatial features are extracted by an approximate graph convolution operator, specifically:
the frequency domain plot convolution of the computation plot signal X with the convolution kernel Θ
approximating the frequency domain graph convolution by a K-order Chebyshev polynomial
Further, the calculation of the attention space-time convolution kernel specifically includes:
performing aggregation operation on the features of the spatial feature map except time by using global average pooling to obtain global spatial features of all moments;
and mapping the correlation to a preset interval by using an activation function based on the correlation of the two full-connection layers at different moments to obtain a time weight coefficient, and distributing the time weight coefficient to the space characteristic diagram to obtain an attention space-time convolution kernel.
Further, the calculation of the global spatial feature at each time specifically includes:
in the formula, S t And the GAP represents the aggregation operation for the global space characteristic at the moment t,
is a space-time characteristic diagram at the time t,
is that
The value on node f channel at time n, N, F represents the number of nodes and the number of channels, respectively.
Further, the calculation of the weight coefficient specifically includes:
in the formula (I), the compound is shown in the specification,
for the weight coefficients, S is the global spatial feature, 2FC denotes two fully connected layers,
the parameter matrixes are related to two full connections respectively, r is a scaling rate, sigma is a sigmoid activation function, and delta is a Selu activation function.
Further, the attention spatiotemporal convolution kernel is calculated as:
in the formula, Q t 、
And
respectively representing attention space-time convolution kernels, weight coefficients and space-time feature maps at the time t.
Further, the process of splicing the spatiotemporal features of multiple scales along the channel dimension includes:
and introducing a Res2Net network structure to obtain the space-time correlation of the photovoltaic power station data in a multi-scale space-time feature fusion mode, and then utilizing a convolution layer to arrange the extracted space-time features into dimensions consistent with a prediction target.
The invention provides a photovoltaic power generation prediction system of a multi-scale dynamic graph convolution neural network, which comprises:
the data extraction unit is used for acquiring input information, extracting time characteristics and space characteristics of photovoltaic power station data from the input information to form a space-time characteristic diagram, and extracting the time characteristics through one-dimensional expansion convolution;
the data processing unit is used for dividing the spatio-temporal feature graph into a plurality of feature sub-graphs along the channel dimension, calculating attention spatio-temporal convolution kernels corresponding to each feature sub-graph, wherein the attention spatio-temporal convolution kernels corresponding to the other feature sub-graphs comprise spatio-temporal information of all previous feature sub-graphs except the attention spatio-temporal convolution kernel corresponding to the first feature sub-graph, and the feature sub-graphs and the corresponding attention spatio-temporal convolution kernels form spatio-temporal features of the current feature sub-graph;
and the power generation prediction unit is used for splicing the space-time characteristics of various scales along the channel dimension to form spliced space-time characteristics consistent with the dimension of the input information, and performing weighted fusion on the space-time characteristics of each scale in the spliced space-time characteristics to obtain a photovoltaic power generation prediction result.
A third aspect of the invention provides a computer storage medium having stored thereon computer instructions which, when run on the system, cause the system to perform the steps of the method.
The photovoltaic power generation prediction system according to the second aspect of the present invention can implement the methods according to the first aspect and the respective implementation manners of the first aspect, and achieve the same effects.
The effect provided in the summary of the invention is only the effect of the embodiment, not all the effects of the invention, and one of the above technical solutions has the following advantages or beneficial effects:
1. in the process of photovoltaic power generation prediction, the extracted space-time feature map is divided into a plurality of feature sub-maps, the space-time feature fusion of various scales is realized based on attention space-time convolution kernels, the relevance among all moments is considered in the attention space-time convolution kernels, the model expression capacity is improved, the time and space features of a photovoltaic power station can be effectively extracted, and the data characteristics are deeply mined, so that the prediction accuracy of the graph convolution neural network photovoltaic power generation prediction model is improved.
2. When time feature extraction is carried out, a one-dimensional expansion convolution is adopted to set an expansion rate, and a proper number of zero values are added among each element of a convolution kernel to control the dimensionality of the convolution kernel, so that the time dimensionality is adjusted nonlinearly, and the problem that the time feature memory is influenced by the convolution kernel with a fixed size is solved.
3. When the spatial feature is extracted, the Chebyshev approximation is carried out on the traditional graph convolution based on the frequency domain, and the complexity of formula calculation is reduced.
Drawings
In order to more clearly illustrate the embodiments or technical solutions in the prior art of the present invention, the drawings used in the description of the embodiments or prior art will be briefly described below, and it is obvious for those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.
FIG. 1 is a schematic flow diagram of an embodiment of the method of the present invention;
FIG. 2 is a schematic diagram illustrating a process of constructing an attention spatiotemporal convolution kernel according to an embodiment of the method of the present invention;
FIG. 3 is a schematic diagram illustrating a data dimension variation corresponding to a weight coefficient calculation process in an embodiment of the method of the present invention;
fig. 4 is a schematic structural diagram of an embodiment of the system of the present invention.
Detailed Description
In order to clearly explain the technical features of the present invention, the following detailed description of the present invention is provided with reference to the accompanying drawings. The following disclosure provides many different embodiments, or examples, for implementing different features of the invention. To simplify the disclosure of the present invention, the components and arrangements of specific examples are described below. Furthermore, the present invention may repeat reference numerals and/or letters in the various examples. This repetition is for the purpose of simplicity and clarity and does not in itself dictate a relationship between the various embodiments and/or configurations discussed. It should be noted that the components illustrated in the figures are not necessarily drawn to scale. Descriptions of well-known components and processing techniques and procedures are omitted so as to not unnecessarily limit the invention.
As shown in fig. 1, an embodiment of the present invention provides a photovoltaic power generation prediction method based on a multi-scale map convolutional neural network, where the method includes the following steps:
s1, acquiring input information, extracting time characteristics and space characteristics of photovoltaic power station data from the input information to form a space-time characteristic diagram, wherein the time characteristics are extracted through one-dimensional expansion convolution;
s2, dividing the space-time feature graph into a plurality of feature sub-graphs along the channel dimension, calculating attention space-time convolution kernels corresponding to each feature sub-graph, wherein the attention space-time convolution kernels corresponding to the rest feature sub-graphs comprise the space-time information of all previous feature sub-graphs except the attention space-time convolution kernel corresponding to the first feature sub-graph, and the feature sub-graphs and the corresponding attention space-time convolution kernels form the space-time features of the current feature sub-graph;
and S3, splicing the space-time characteristics of various scales along the channel dimension to form spliced space-time characteristics consistent with the dimension of the input information, and performing weighted fusion on the space-time characteristics of each scale in the spliced space-time characteristics to obtain a photovoltaic power generation prediction result.
As shown in fig. 2, in step S1, the time data is regarded as one-dimensional structured data, and the time feature is extracted by a one-dimensional convolutional neural network. Since the convolution kernel size of the convolutional neural network is fixed, the memory capacity of the convolutional neural network to the time characteristic can only be short-term. To alleviate this problem, the present invention employs one-dimensional dilation convolution to capture the temporal characteristics of photovoltaic plant data. As a special one-dimensional convolution, the dilated convolution can be performed by setting the dilation rate d at the convolution kernel
D-1 zero values are added between each element to control the dimension of the convolution kernel, thereby adjusting the time dimension non-linearly:
wherein δ is Selu activation function d Is a one-dimensional expansion convolution operator. Assume that the number of convolution kernels of the dilation convolution is C' and the size of the feature map is maintained using a padding operation. H is the raw data of the time characteristics of the photovoltaic power station.
After the time dimension of the photovoltaic power station data is subjected to feature extraction through one-dimensional expansion convolution, spatial feature extraction is performed on a plurality of photovoltaic power station data in the same area. Most of the existing photovoltaic power plant data prediction models use frequency domain based graph convolution.
The frequency domain graph convolution operation of the graph signal X with the convolution kernel Θ can be defined as:
wherein
Is a graph convolution operator, L is a graph Laplace matrix, and after normalization processing
I N Is a unit matrix which is formed by a plurality of unit matrixes,
the method is characterized in that the method is an L eigenvector matrix, D is a degree matrix, Λ is a diagonal matrix taking an eigenvalue of L as a diagonal element, U is an L eigenvector matrix, A is an adjacent matrix reflecting the internal connection condition of the graph X, and the calculation amount is large because the characteristic decomposition of Laplace of the formula is complex, and a K-order Chebyshev polynomial is adopted to approximate a graph convolution kernel:
wherein
A Chebyshev coefficient vector, a Chebyshev polynomial being defined as T k (x)＝2xT k-1 (x)-T k-2 (x)T 0 (x)＝1，T 1 (x) X. While
λ max Representing the maximum eigenvalue of L. By this approximation, the computational complexity of the above formula is greatly reduced.
In the historical data of the photovoltaic power station, the contribution of photovoltaic data at different times to the prediction result is different. Therefore, different weights are dynamically assigned to the features at different times according to the relationship between the features at the respective times, so that the model focuses on the features more favorable for the prediction result. According to the method, an attention mechanism is applied to the time dimension of photovoltaic power station data, the characteristics of each moment are integrated, the correlation among the characteristics of different moments is modeled, and further the weight is dynamically distributed to the space-time characteristics Z of each moment, so that valuable characteristics are emphasized, and the influence of invalid characteristics is weakened.
In step S3, the specific steps of calculating the attention space-time convolution kernel corresponding to each feature sub-graph are as follows:
firstly, global average pooling is utilized to align the spatio-temporal feature map
Performing aggregation operation on the features of other dimensions except time to obtain the global spatial features of all the moments
Then, the two fully-connected layers (2FC) consider the interrelation between different time instants by the following formula and map this relation to [0,1 ] using sigmoid activation function σ]Obtaining a time weight coefficient
Finally, the weight is assigned to
Deriving attention space-time convolution kernels
The whole process is formulated as follows:
in the formula
Is that
At time t n the value on node f-channel,
which are the parameter matrices involved in the two full connections, respectively, the parameter number of the module can be controlled by adjusting the scaling rate r.
Figure 3 shows a characteristic diagram
In generating the weight coefficients
The dimensions of the process vary.
In step S3, a multi-scale spatio-temporal map convolutional layer is proposed using the constructed attention spatio-temporal convolution kernel as a basic unit. As shown in FIG. 4, assume that the input to the system is
After 1x1 convolution processing, a characteristic diagram is obtained
Wherein F' is F × s. Then H is measured along the channel dimension (l) Are equally divided into s feature subgraphs, i.e.
Each sub-feature map has a corresponding attention space-time convolution kernel AST i (l) (. carries out space-time characteristic extraction on the signal and outputs Q i (l) 。H i (l) AST will be fused i-1 (l) Output Q of (c) i-1 (l) Are input to AST together i (l) In (c). Thus, each attention space-time convolution kernel AST i (l) All the previous sub-feature maps H are received j (l) J is less than or equal to iSpatio-temporal information. In order to reduce the parameter quantity of the model, the system omits the pair H 1 (l) And (4) extracting the space-time characteristics, and directly taking the space-time characteristics as the characteristics for recycling. The above process is described mathematically as:
the system then compares all the dimensions of the spatiotemporal feature Q i (l) Spliced along the dimension of the channel
Finally, for Q integrating s-order space-time characteristics (l) Polymerization is carried out to obtain (l) Dimensionally consistent
The aggregation operation aims to enable the model to dynamically adjust the importance of the space-time characteristics of different scales, so that the space-time characteristic expression of the node mixed scale is obtained, and the subsequent prediction task is completed more accurately. The specific implementation mode adopts an attention mechanism to carry out attention weighting on space-time characteristics at different positions.
Based on the cycle characteristics of photovoltaic power generation, weather conditions and illumination conditions of corresponding time periods from one day, one month or even one year before the target sequence is predicted can bring strong guiding significance to current prediction. In order to consider the periodicity of the photovoltaic power station data, the recent, monthly and yearly period sequences X of the photovoltaic power station data are respectively processed by 3 network components with the same structure r ,X m ,X y Is modeled.
Each component consists of a multi-scale space-time diagram convolutional layer and a convolutional layer, and a Res2Net network structure is introduced to obtain the space-time correlation of the photovoltaic power station data in a multi-scale space-time characteristic fusion mode. The extracted spatiotemporal features are then aligned into dimensions consistent with the predicted target using the convolutional layer. And finally, performing weighted fusion on the output of each component to obtain a final prediction result.
The invention provides a photovoltaic power generation prediction method based on a multi-scale dynamic graph convolution neural network, which constructs an attention space-time convolution kernel, respectively extracts time and space characteristics of data of a plurality of photovoltaic power stations in the same region by utilizing a 1x1 convolution and a 1x1 graph convolution, introduces an attention mechanism to model the interrelation of each moment, and measures the influence degree of the data at different moments on a prediction result. In a network construction method, Res2Net is introduced, the attention space-time convolution kernel is embedded into the Res2Net network, the space-time feature fusion of multiple scales is achieved, the multi-scale space-time feature expression is further achieved, and finally the photovoltaic power generation prediction of the multi-scale dynamic graph convolution neural network tray is achieved.
As shown in fig. 4, the invention further provides a photovoltaic power generation prediction system of a multi-scale dynamic graph convolution neural network, which comprises a data extraction unit 1, a data processing unit 2 and a power generation prediction unit 3.
The data extraction unit 1 is used for acquiring input information, extracting time characteristics and space characteristics of photovoltaic power station data from the input information to form a space-time characteristic diagram, and extracting the time characteristics through one-dimensional expansion convolution; the data processing unit 2 is configured to divide the spatio-temporal feature map into a plurality of feature sub-maps along a channel dimension, calculate an attention spatio-temporal convolution kernel corresponding to each feature sub-map, where the attention spatio-temporal convolution kernels corresponding to the remaining feature sub-maps include spatio-temporal information of all previous feature sub-maps except an attention spatio-temporal convolution kernel corresponding to a first feature sub-map, and the feature sub-maps and the corresponding attention spatio-temporal convolution kernels form spatio-temporal features of a current feature sub-map; the power generation prediction unit 3 is used for splicing the space-time characteristics of multiple scales along the channel dimension to form spliced space-time characteristics consistent with the input information dimension, and performing weighted fusion on the space-time characteristics of each scale in the spliced space-time characteristics to obtain a photovoltaic power generation prediction result.
The present invention also provides a computer storage medium having stored thereon computer instructions which, when run on the system, cause the system to perform the steps of the method.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.
Claims (10)
1. A photovoltaic power generation prediction method based on a multi-scale graph convolution neural network is characterized by comprising the following steps:
acquiring input information, and extracting time characteristics and space characteristics of photovoltaic power station data from the input information to form a space-time characteristic diagram, wherein the time characteristics are extracted through one-dimensional expansion convolution;
dividing the spatio-temporal feature graph into a plurality of feature sub-graphs along a channel dimension, calculating attention spatio-temporal convolution kernels corresponding to each feature sub-graph, wherein the attention spatio-temporal convolution kernels corresponding to other feature sub-graphs comprise spatio-temporal information of all previous feature sub-graphs except the attention spatio-temporal convolution kernel corresponding to the first feature sub-graph, and the feature sub-graphs and the corresponding attention spatio-temporal convolution kernels form spatio-temporal features of the current feature sub-graph;
and splicing the space-time characteristics of various scales along the channel dimension to form spliced space-time characteristics consistent with the dimension of the input information, and performing weighted fusion on the space-time characteristics of various scales in the spliced space-time characteristics to obtain a photovoltaic power generation prediction result.
2. The multi-scale graph convolutional neural network-based photovoltaic power generation prediction method of claim 1, wherein the extraction of the temporal features specifically comprises:
wherein Z is the extracted time characteristic, delta is the Selu activation function,
and d is a one-dimensional expansion convolution operator, and H is the original data of the time characteristics of the photovoltaic power station.
3. The method for predicting photovoltaic power generation based on the multi-scale graph convolutional neural network as claimed in claim 1, wherein the spatial features are extracted by an approximate graph convolution operator, and specifically comprise:
the frequency domain plot convolution of the computation plot signal X with the convolution kernel Θ
approximating the frequency domain plot convolution by a Chebyshev polynomial of order K
4. The method for predicting photovoltaic power generation based on the multi-scale graph convolutional neural network as claimed in claim 1, wherein the calculation of the attention space-time convolutional kernel specifically comprises:
performing aggregation operation on the features of the spatial feature map except time by using global average pooling to obtain global spatial features of all moments;
and mapping the correlation to a preset interval by using an activation function based on the correlation of the two full-connection layers at different moments to obtain a time weight coefficient, and distributing the time weight coefficient to the space characteristic diagram to obtain an attention space-time convolution kernel.
5. The multi-scale graph convolutional neural network-based photovoltaic power generation prediction method of claim 4, wherein the calculation of the global spatial features at each time is specifically as follows:
in the formula, S t And the GAP represents the aggregation operation for the global space characteristic at the moment t,
is a space-time characteristic diagram at the time t,
is that
The value on node f channel at time n, N, F represents the number of nodes and the number of channels, respectively.
6. The multi-scale graph convolutional neural network-based photovoltaic power generation prediction method of claim 5, wherein the weight coefficients are specifically calculated as follows:
in the formula (I), the compound is shown in the specification,
for the weight coefficients, S is the global spatial feature, 2FC denotes two fully connected layers,
the parameter matrixes are related to two full connections respectively, r is a scaling rate, sigma is a sigmoid activation function, and delta is a Selu activation function.
7. The method for predicting photovoltaic power generation based on the multi-scale graph convolutional neural network of claim 6, wherein the attention space-time convolutional kernel is calculated as:
in the formula, Q t 、
And
respectively representing attention space-time convolution kernels, weight coefficients and space-time feature maps at the time t.
8. The method for predicting photovoltaic power generation based on the multi-scale graph convolutional neural network as claimed in claim 1, wherein the process of splicing the multi-scale spatio-temporal features along the channel dimension comprises the following steps:
and introducing a Res2Net network structure to obtain the space-time correlation of the photovoltaic power station data in a multi-scale space-time feature fusion mode, and then utilizing a convolution layer to arrange the extracted space-time features into dimensions consistent with a prediction target.
9. A photovoltaic power generation prediction system based on a multi-scale graph convolution neural network is characterized by comprising:
the data extraction unit is used for acquiring input information, extracting time characteristics and space characteristics of photovoltaic power station data from the input information to form a space-time characteristic diagram, and extracting the time characteristics through one-dimensional expansion convolution;
the data processing unit is used for dividing the spatio-temporal feature graph into a plurality of feature sub-graphs along the channel dimension, calculating attention spatio-temporal convolution kernels corresponding to each feature sub-graph, wherein the attention spatio-temporal convolution kernels corresponding to the other feature sub-graphs comprise spatio-temporal information of all previous feature sub-graphs except the attention spatio-temporal convolution kernel corresponding to the first feature sub-graph, and the feature sub-graphs and the corresponding attention spatio-temporal convolution kernels form spatio-temporal features of the current feature sub-graph;
and the power generation prediction unit is used for splicing the space-time characteristics of various scales along the channel dimension to form spliced space-time characteristics consistent with the dimension of the input information, and performing weighted fusion on the space-time characteristics of each scale in the spliced space-time characteristics to obtain a photovoltaic power generation prediction result.
10. A computer storage medium having computer instructions stored thereon, which, when run on the system of claim 9, cause the system to perform the steps of the method of any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210560677.5A CN114971007A (en) | 2022-05-20 | 2022-05-20 | Photovoltaic power generation prediction method and system based on multi-scale graph convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210560677.5A CN114971007A (en) | 2022-05-20 | 2022-05-20 | Photovoltaic power generation prediction method and system based on multi-scale graph convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114971007A true CN114971007A (en) | 2022-08-30 |
Family
ID=82986071
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210560677.5A Pending CN114971007A (en) | 2022-05-20 | 2022-05-20 | Photovoltaic power generation prediction method and system based on multi-scale graph convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114971007A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116128130A (en) * | 2023-01-31 | 2023-05-16 | 广东电网有限责任公司 | Short-term wind energy data prediction method and device based on graphic neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111292562A (en) * | 2020-05-12 | 2020-06-16 | 北京航空航天大学 | Aviation flow prediction method |
| CN112307958A (en) * | 2020-10-30 | 2021-02-02 | 河北工业大学 | Micro-expression identification method based on spatiotemporal appearance movement attention network |
| WO2021068528A1 (en) * | 2019-10-11 | 2021-04-15 | 平安科技（深圳）有限公司 | Attention weight calculation method and apparatus based on convolutional neural network, and device |
| CN113283581A (en) * | 2021-05-14 | 2021-08-20 | 南京邮电大学 | Multi-fusion graph network collaborative multi-channel attention model and application method thereof |
| CN113852492A (en) * | 2021-09-01 | 2021-12-28 | 南京信息工程大学 | Network flow prediction method based on attention mechanism and graph convolution neural network |
-
2022
- 2022-05-20 CN CN202210560677.5A patent/CN114971007A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021068528A1 (en) * | 2019-10-11 | 2021-04-15 | 平安科技（深圳）有限公司 | Attention weight calculation method and apparatus based on convolutional neural network, and device |
| CN111292562A (en) * | 2020-05-12 | 2020-06-16 | 北京航空航天大学 | Aviation flow prediction method |
| CN112307958A (en) * | 2020-10-30 | 2021-02-02 | 河北工业大学 | Micro-expression identification method based on spatiotemporal appearance movement attention network |
| CN113283581A (en) * | 2021-05-14 | 2021-08-20 | 南京邮电大学 | Multi-fusion graph network collaborative multi-channel attention model and application method thereof |
| CN113852492A (en) * | 2021-09-01 | 2021-12-28 | 南京信息工程大学 | Network flow prediction method based on attention mechanism and graph convolution neural network |
Non-Patent Citations (1)
| Title |
|---|
| 温钧林: "基于多尺度时空图卷积网络的交通流预测算法研究" * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116128130A (en) * | 2023-01-31 | 2023-05-16 | 广东电网有限责任公司 | Short-term wind energy data prediction method and device based on graphic neural network |
| CN116128130B (en) * | 2023-01-31 | 2023-10-24 | 广东电网有限责任公司 | Short-term wind energy data prediction method and device based on graphic neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059878B (en) | Photovoltaic power generation power prediction model based on CNN LSTM and construction method thereof | |
| Wu et al. | A short-term load forecasting method based on GRU-CNN hybrid neural network model | |
| US20220373984A1 (en) | Hybrid photovoltaic power prediction method and system based on multi-source data fusion | |
| CN110909919A (en) | Photovoltaic power prediction method of depth neural network model with attention mechanism fused | |
| CN112529282A (en) | Wind power plant cluster short-term power prediction method based on space-time graph convolutional neural network | |
| CN116128170B (en) | Photovoltaic power station power ultra-short-term prediction method and device and related equipment | |
| CN110782071A (en) | Method for predicting wind power by convolutional neural network based on time-space characteristic fusion | |
| CN114792156A (en) | Photovoltaic output power prediction method and system based on curve characteristic index clustering | |
| CN111242355A (en) | Photovoltaic probability prediction method and system based on Bayesian neural network | |
| CN114676923A (en) | Method and device for predicting generated power, computer equipment and storage medium | |
| CN116629416A (en) | Photovoltaic power station power prediction method and device | |
| Dokur | Swarm decomposition technique based hybrid model for very short-term solar PV power generation forecast | |
| CN114971007A (en) | Photovoltaic power generation prediction method and system based on multi-scale graph convolutional neural network | |
| CN116014722A (en) | Sub-solar photovoltaic power generation prediction method and system based on seasonal decomposition and convolution network | |
| US20240054267A1 (en) | Method for planning a layout of a renewable energy site | |
| CN112836876A (en) | Power distribution network line load prediction method based on deep learning | |
| CN113610665B (en) | Wind power generation power prediction method based on multi-delay output echo state network | |
| CN116341728A (en) | Ultra-short-term photovoltaic output power prediction method based on data driving | |
| Hui et al. | Ultra-short-term prediction of wind power based on fuzzy clustering and RBF neural network | |
| CN113779861B (en) | Photovoltaic Power Prediction Method and Terminal Equipment | |
| CN115545256A (en) | Small-scale photovoltaic power prediction method based on multi-dimensional data feature learning | |
| CN113988395A (en) | Wind power ultra-short-term power prediction method based on SSD and dual attention mechanism BiGRU | |
| CN112215383A (en) | Distributed photovoltaic power generation power prediction method and system | |
| Zhang et al. | Multi-Input Deep Convolutional Neural Network Model for Short-Term Power Prediction of Photovoltaics | |
| CN115860281B (en) | Multi-entity load prediction method and device for energy system based on cross-entity attention |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220830 |