CN116612393A - Solar radiation prediction method, system, electronic equipment and storage medium - Google Patents
Solar radiation prediction method, system, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN116612393A CN116612393A CN202310498367.XA CN202310498367A CN116612393A CN 116612393 A CN116612393 A CN 116612393A CN 202310498367 A CN202310498367 A CN 202310498367A CN 116612393 A CN116612393 A CN 116612393A
- Authority
- CN
- China
- Prior art keywords
- sequence
- solar radiation
- radiation
- time
- long
- 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
- 230000005855 radiation Effects 0.000 title claims abstract description 244
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000000605 extraction Methods 0.000 claims abstract description 71
- 230000008859 change Effects 0.000 claims abstract description 60
- 230000007246 mechanism Effects 0.000 claims abstract description 39
- 238000004821 distillation Methods 0.000 claims abstract description 21
- 230000003068 static effect Effects 0.000 claims abstract description 15
- 238000011176 pooling Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 9
- 238000010586 diagram Methods 0.000 claims description 8
- 230000001419 dependent effect Effects 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 3
- 239000003550 marker Substances 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 description 12
- 239000011159 matrix material Substances 0.000 description 11
- 230000008901 benefit Effects 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 5
- 230000033001 locomotion Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000010248 power generation Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000000053 physical method Methods 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 230000010485 coping Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000002803 fossil fuel Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000000873 masking effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000006798 recombination Effects 0.000 description 1
- 238000005215 recombination Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000010792 warming Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/10—Devices for predicting weather conditions
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/62—Extraction of image or video features relating to a temporal dimension, e.g. time-based feature extraction; Pattern tracking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Environmental & Geological Engineering (AREA)
- Environmental Sciences (AREA)
- Atmospheric Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Remote Sensing (AREA)
- Ecology (AREA)
- Astronomy & Astrophysics (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a solar radiation prediction method, a solar radiation prediction system, electronic equipment and a storage medium, and relates to the technical field of weather prediction. The method comprises the following steps: acquiring initial data; the initial data comprises a static satellite image long sequence and a ground radiation observation long sequence; determining a 10-minute solar radiation length sequence prediction result according to the initial data and a solar radiation prediction model; the solar radiation prediction model comprises a cloud layer feature extraction module, a radiation change feature extraction module and a long-time sequence radiation prediction module; the cloud layer feature extraction module and the radiation change feature extraction module are connected with the long-time-sequence radiation prediction module; the long-time sequence radiation prediction module is constructed according to a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation type decoder mechanism. The method can improve the solar radiation prediction fineness and realize 10-minute-level ultra-short-time solar radiation prediction.
Description
Technical Field
The invention relates to the technical field of weather prediction, in particular to a solar radiation prediction method, a solar radiation prediction system, electronic equipment and a storage medium.
Background
With the continuous increase of global energy demand, the long-term development and utilization of fossil fuels exacerbate the problems of global warming, environmental pollution and the like. Under the background, the wide application of solar energy becomes an important way for coping with climate change and realizing sustainable development of global economy and society. According to the international energy agency report, the global solar photovoltaic installed capacity has increased from 40GW in 2010 to 760GW in 2020, and solar photovoltaic power generation has become one of the main sources of the global newly increased installed capacity of electricity. However, due to day-night alternation, seasonal variation, cloud cover, atmospheric optical properties, etc., solar radiation energy received by the ground has intermittent and fluctuating properties, which can limit the stability of the photovoltaic power supply, thereby affecting its large-scale application. Therefore, it is necessary to predict solar radiation accurately in real time, so as to support intelligent regulation and control of a photovoltaic power generation system, improve the running reliability of a photovoltaic power grid, and maximize economic and environmental benefits.
Among various prediction tasks in different advance periods, solar radiation ultra-short time prediction is most commonly used for real-time scheduling and evaluation of a photovoltaic power grid, and the existing ultra-short time prediction mainly adopts a statistical method, a physical method or a mixed model method. The statistical method is simple and easy to realize, but the generalization capability is insufficient; the physical method has strong universality, but the model is complex, the calculation cost is high, and the prediction effect is poor in a cloudy region. The mixed model method combines the advantages of the two methods, but only based on ground local observation, the influence of a large-scale cloud layer on solar radiation change is not considered, and a significant prediction hysteresis problem exists under a cloudy condition, so that the accuracy of solar radiation prediction needs to be improved by including large-scale cloud layer motion information in the prediction process.
With the development of remote sensing technology, a stationary satellite provides an effective means for obtaining high-resolution large-range continuous cloud field observation. At present, a few prediction methods combined with satellite images mostly adopt numerical weather forecast or traditional machine learning models, and have limited capabilities in aspects of cloud layer feature extraction, cloud layer motion-radiation change relation simulation and the like. In contrast, deep learning can be trained by a large number of pre-observation sequences and corresponding future radiation sequence samples, space-time feature extraction is adaptively performed, and a nonlinear relationship between the pre-sequences and the future radiation is fitted to obtain a more accurate prediction result. However, the existing deep learning model still has a disadvantage in obtaining high time resolution predictions, because the length of the sequence to be processed increases with increasing resolution, which requires the model to be able to learn effectively the dependency (i.e. long range dependency) between solar radiation at future time steps and cloud layer and radiation variation information before longer time paths. But is limited by the simulation capability, model complexity and calculation efficiency of the long-range dependency relationship, the existing method can only predict with the resolution of 15 minutes at most, and the fine degree still has a large improvement space.
Disclosure of Invention
The invention aims to provide a solar radiation prediction method, a solar radiation prediction system, electronic equipment and a storage medium, which can improve the solar radiation prediction fineness and realize 10-minute-level ultra-short-time solar radiation prediction.
In order to achieve the above object, the present invention provides the following solutions:
a solar radiation prediction method, comprising:
acquiring initial data; the initial data comprises a static satellite image long sequence and a ground radiation observation long sequence;
determining a 10-minute solar radiation length sequence prediction result according to the initial data and a solar radiation prediction model; the solar radiation prediction model comprises a cloud layer feature extraction module, a radiation change feature extraction module and a long-time sequence radiation prediction module; the cloud layer feature extraction module and the radiation change feature extraction module are connected with the long-time-sequence radiation prediction module;
the cloud layer feature extraction module is constructed through a plurality of three-dimensional convolution blocks; the radiation change characteristic extraction module is constructed by a plurality of one-dimensional convolution blocks; the long-time sequence radiation prediction module is constructed according to a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation type decoder mechanism; the cloud layer feature extraction module is used for determining cloud layer space features of the long sequence of the static satellite images; the radiation change feature extraction module is used for determining solar radiation time change features of the ground radiation observation long sequence; the long-time-sequence radiation prediction module is used for determining a 10-minute-level solar radiation long-sequence prediction result according to the cloud layer space characteristics and the solar radiation time change characteristics.
Optionally, each of the three-dimensional convolution blocks and each of the one-dimensional convolution blocks includes a convolution layer and a maximum pooling layer connected in sequence;
the long-time-sequence radiation prediction module comprises a first embedding sub-module, a second embedding sub-module, an encoder, a decoder and a fully-connected output layer; the first embedded submodule, the encoder and the decoder are connected in sequence; the decoder is also connected with the second embedded submodule;
the encoder comprises two multi-head probability distribution sparse self-attention layers and two self-attention distillation layers which are sequentially and alternately connected; the decoder comprises a mask multi-head probability distribution sparse self-attention layer and a multi-head self-attention layer which are connected in sequence.
Optionally, determining a 10-minute solar radiation length sequence prediction result according to the initial data and the solar radiation prediction model specifically comprises:
extracting features of the long sequence of the static satellite image by utilizing three-dimensional convolution to obtain cloud layer space features;
extracting features of the ground radiation observation long sequence by utilizing one-dimensional convolution to obtain solar radiation time change features;
and carrying out feature prediction according to the cloud layer space features and the solar radiation time change features to obtain a 10-minute solar radiation length sequence prediction result.
Optionally, performing feature prediction according to the cloud layer space feature and the solar radiation time variation feature to obtain a 10-minute solar radiation length sequence prediction result, which specifically includes:
carrying out space-time fusion pretreatment on the cloud layer space characteristics and the solar radiation time variation characteristics to obtain a front space-time characteristic sequence and a generated decoder input sequence;
performing marker embedding, position embedding and global time embedding on the pre-space-time feature sequence, and adding according to the time steps and feature dimensions of the corresponding sequence to obtain a first input sequence;
extracting the long-range dependence of the first input sequence by utilizing multi-head probability distribution sparse self-attention operation and self-attention distillation operation to obtain a long-range dependence characteristic diagram;
performing mark embedding, position embedding and global time embedding on the input sequence of the generating decoder, and adding according to the time steps and feature dimensions of the corresponding sequence to obtain a second input sequence;
and generating a 10-minute-level solar radiation long sequence prediction result according to the second input sequence and the long-range dependent feature map by using a feature mask method and multi-head probability distribution sparse self-attention operation.
Optionally, preprocessing the cloud layer space feature and the solar radiation time change feature to obtain a pre-space time feature sequence and a generated decoder input sequence, which specifically comprises the following steps:
aligning and splicing the cloud layer space characteristics and the solar radiation time change characteristics according to time steps to obtain a front space-time characteristic sequence;
dividing the preposed space-time characteristic sequence into two sections with equal length, and after the second half section sequence, splicing the 0 value sequence with the length of the predicted time steps along the time dimension to obtain the input sequence of the generating decoder.
The present invention provides a solar radiation prediction system comprising:
the data acquisition module is used for acquiring initial data; the initial data comprises a static satellite image long sequence and a ground radiation observation long sequence;
the radiation prediction module is used for determining a 10-minute-level solar radiation length sequence prediction result according to the initial data and the solar radiation prediction model; the solar radiation prediction model comprises a cloud layer feature extraction module, a radiation change feature extraction module and a long-time sequence radiation prediction module; the cloud layer feature extraction module and the radiation change feature extraction module are connected with the long-time-sequence radiation prediction module;
the cloud layer feature extraction module is constructed through a plurality of three-dimensional convolution blocks; the radiation change characteristic extraction module is constructed by a plurality of one-dimensional convolution blocks; the long-time sequence radiation prediction module is constructed according to a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation type decoder mechanism; the cloud layer feature extraction module is used for determining cloud layer space features of the long sequence of the static satellite images; the radiation change feature extraction module is used for determining solar radiation time change features of the ground radiation observation long sequence; the long-time-sequence radiation prediction module is used for determining a 10-minute-level solar radiation long-sequence prediction result according to the cloud layer space characteristics and the solar radiation time change characteristics.
The invention provides an electronic device comprising a memory for storing a computer program and a processor for running the computer program to cause the electronic device to perform a solar radiation prediction method according to the above.
The present invention provides a computer readable storage medium storing a computer program which when executed by a processor implements a solar radiation prediction method as described above.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention discloses a solar radiation prediction method, a system, electronic equipment and a storage medium, wherein the method comprises the steps of carrying out feature extraction on a static satellite image long sequence and a ground radiation observation long sequence by utilizing a solar radiation prediction model and carrying out long time sequence radiation prediction according to features to obtain a 10-minute-level ultra-short solar radiation prediction result, wherein the solar radiation prediction model comprises a cloud layer feature extraction module, a radiation change feature extraction module and a long time sequence radiation prediction module, the long time sequence radiation prediction module is constructed according to a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation type decoder mechanism, and the future ultra-short solar radiation of a region is predicted by the deep learning method, so that the fineness of solar radiation prediction in the prior art can be improved, and the 10-minute-level ultra-short solar radiation prediction is realized.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a solar radiation prediction method of the present invention;
FIG. 2 is a schematic diagram of a prediction logic flow in the present embodiment;
FIG. 3 is a schematic diagram of a feature extraction module in the present embodiment;
FIG. 4 is a schematic diagram of a long-time-series radiation prediction module according to the present embodiment;
FIG. 5 is a graph showing the comparison of the solar radiation prediction results in the present example;
FIG. 6 is a schematic diagram of the cloud deck movement and the solar radiation variation in the present embodiment;
fig. 7 is a block diagram of the solar radiation prediction system of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention aims to provide a solar radiation prediction method, a solar radiation prediction system, electronic equipment and a storage medium, which can improve the solar radiation prediction fineness and realize 10-minute-level ultra-short-time solar radiation prediction.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
As shown in fig. 1, the present invention provides a solar radiation prediction method, including:
step 100: acquiring initial data; the initial data includes a long sequence of still satellite images and a long sequence of ground radiation observations.
Step 200: determining a 10-minute solar radiation length sequence prediction result according to the initial data and a solar radiation prediction model; the solar radiation prediction model comprises a cloud layer feature extraction module, a radiation change feature extraction module and a long-time sequence radiation prediction module; and the cloud layer feature extraction module and the radiation change feature extraction module are both connected with the long-time sequence radiation prediction module.
The cloud layer feature extraction module is constructed through a plurality of three-dimensional convolution blocks; the radiation change characteristic extraction module is constructed by a plurality of one-dimensional convolution blocks; the long-time sequence radiation prediction module is constructed according to a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation type decoder mechanism; the cloud layer feature extraction module is used for determining cloud layer space features of the long sequence of the static satellite images; the radiation change feature extraction module is used for determining solar radiation time change features of the ground radiation observation long sequence; the long-time-sequence radiation prediction module is used for determining a 10-minute-level solar radiation long-sequence prediction result according to the cloud layer space characteristics and the solar radiation time change characteristics.
In the above modules, each three-dimensional convolution block and each one-dimensional convolution block comprise a convolution layer and a maximum pooling layer which are sequentially connected; the long-time-sequence radiation prediction module comprises a first embedding sub-module, a second embedding sub-module, an encoder, a decoder and a fully-connected output layer; the first embedded submodule, the encoder and the decoder are connected in sequence; the decoder is also connected with the second embedded submodule; the encoder comprises two multi-head probability distribution sparse self-attention layers and two self-attention distillation layers which are sequentially and alternately connected; the decoder comprises a mask multi-head probability distribution sparse self-attention layer and a multi-head self-attention layer which are connected in sequence.
As a specific embodiment of step 200, it includes:
extracting features of the long sequence of the static satellite image by utilizing three-dimensional convolution to obtain cloud layer space features; extracting features of the ground radiation observation long sequence by utilizing one-dimensional convolution to obtain solar radiation time change features; and carrying out feature prediction according to the cloud layer space features and the solar radiation time change features to obtain a 10-minute solar radiation length sequence prediction result.
Specifically, the step of performing feature prediction according to the cloud layer space feature and the solar radiation time change feature to obtain a 10-minute solar radiation length sequence prediction result comprises the following steps:
carrying out space-time fusion pretreatment on the cloud layer space characteristics and the solar radiation time variation characteristics to obtain a front space-time characteristic sequence and a generated decoder input sequence; performing marker embedding, position embedding and global time embedding on the pre-space-time feature sequence, and adding according to the time steps and feature dimensions of the corresponding sequence to obtain a first input sequence; extracting the long-range dependence of the first input sequence by utilizing multi-head probability distribution sparse self-attention operation and self-attention distillation operation to obtain a long-range dependence characteristic diagram; performing mark embedding, position embedding and global time embedding on the input sequence of the generating decoder, and adding according to the time steps and feature dimensions of the corresponding sequence to obtain a second input sequence; and generating a 10-minute-level solar radiation long sequence prediction result according to the second input sequence and the long-range dependent feature map by using a feature mask method and multi-head probability distribution sparse self-attention operation.
In a further scheme, the step of preprocessing the cloud layer space characteristics and the solar radiation time variation characteristics to obtain a front space-time characteristic sequence and a generated decoder input sequence comprises the following steps:
aligning and splicing the cloud layer space characteristics and the solar radiation time change characteristics according to time steps to obtain a front space-time characteristic sequence; dividing the preposed space-time characteristic sequence into two sections with equal length, and after the second half section sequence, splicing the 0 value sequence with the length of the predicted time steps along the time dimension to obtain the input sequence of the generating decoder.
On the basis of the technical scheme, the following embodiments are provided.
As shown in fig. 2, the specific operation flow in this embodiment is as follows: according to the embodiment, through a cloud layer feature extraction module, the space features of cloud layers in a satellite observation long sequence are extracted by utilizing three-dimensional (3D) convolution; secondly, extracting solar radiation time variation characteristics in a ground station observation long sequence by utilizing one-dimensional (1D) convolution through a radiation variation characteristic extraction module; finally, the long-sequence space-time characteristics are fused, and a solar radiation fine prediction result with the resolution of 0-8 hours and 10 minutes in the future is obtained by utilizing a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generating decoder mechanism through a long-sequence radiation prediction module.
The operation flow relates to two stages of feature extraction and radiation prediction, including three key modules of cloud layer feature extraction, radiation change feature extraction and long-time sequence radiation prediction. The characteristic extraction stage mainly utilizes convolution and maximum pooling operation to obtain cloud layer movement and earth surface solar radiation change information from a front satellite image long sequence and a ground observation long sequence through a cloud layer characteristic extraction module and a radiation change characteristic extraction module; the radiation prediction stage is mainly based on a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation type decoder mechanism by a long-time sequence radiation prediction module, and a solar radiation length sequence prediction result of 10 minutes in the future is obtained according to the extracted characteristics. The modules will be further described below.
1. Cloud layer feature extraction module
The module consists of three 3D convolution blocks (as shown in fig. 3) for extracting the spatial pattern of the cloud layer. Wherein each convolution block includes a 3D convolution layer and a 3D max-pooling layer. Setting the convolution kernel sizes of three 3D convolution layers to be 3 multiplied by 3, the convolution step length to be 1, the number of the convolution kernels to be 64, 128 and 256 in sequence, adopting a sigmoid activation function to introduce nonlinear factors, and adopting a filling (padding) strategy to ensure that the feature images before and after convolution are consistent in size; the convolution extracted features are context aggregated in windows set to the three 3D max-pooling layers of sizes 1 x 2, 1 x 4 and 1 x 4, respectively, to reduce the feature map size. And inputting the front satellite image sequence into the module, and finally obtaining the space characteristic sequence of the cloud layer in the front time period.
2. Radiation variation feature extraction module
The module consists of three 1D convolution blocks (as shown in fig. 3) for extracting the varying features of the surface solar radiation caused by the cloud layer during the pre-period. Each convolution block comprises a 1D convolution layer and a 1D maximum pooling layer, wherein the window length of three 1D convolutions is 3, the convolution step length is 1, the number of convolution kernels is 64, 128 and 256 in sequence, and a sigmoid activation function and a padding strategy are adopted; the window sizes of the three 1D max-pooling layers are all 2. By inputting the pre-ground radiation observation sequence into the module, a time evolution characteristic sequence of the ground radiation can be obtained.
3. Long time sequence radiation prediction module
The existing ultra-short solar radiation prediction method cannot efficiently and accurately provide a long time sequence prediction result with resolution of more than 15 minutes due to factors such as simulation capability, model complexity, calculation efficiency and the like of a long-range dependency relationship. Aiming at the defect, the technology designs a long-time sequence radiation prediction module, improves the solving capability of the model on long-range dependence problem through a probability distribution sparse self-attention mechanism and a self-attention distillation mechanism, and reduces the time complexity of the model. Meanwhile, a generating decoder mechanism is adopted, so that the generation efficiency of the prediction result is improved.
The module is mainly composed of two Embedding (Embedding) sub-modules, an encoder, a decoder and a full connection output layer (as shown in fig. 4). Firstly, aligning and splicing cloud layer space features and earth surface radiation time features obtained by a cloud layer feature extraction module and a radiation change feature extraction module according to time steps, processing the cloud layer space features and the earth surface radiation time features by a first embedding sub-module, and inputting the cloud layer space features and the earth surface radiation time features into an encoder; meanwhile, dividing the front ground observation sequence into two sections with equal length, taking a second half section sequence, splicing the 0 value sequence with the length of the predicted time step number along the time dimension, processing the 0 value sequence by a second embedding sub-module, and inputting the 0 value sequence into a decoder; secondly, extracting long-range dependent features of the pre-space-time feature sequence through an encoder, and transmitting an extraction result to a decoder; then, the decoder generates a high-dimension long sequence prediction according to the input of the decoder and the processing result of the encoder; and finally, mapping the decoder prediction into one dimension through a fully connected output layer to form a final 10-minute solar radiation length sequence prediction result. Each sub-module will be described below.
(1) Embedded sub-module
The sub-module is used to further incorporate long sequences of local and global timing features into the input of the encoder/decoder, including three parts of Token embedding, position embedding and global time embedding. Tag embedding uses a 1D convolution (window size 3, step size 1) to map the original input sequence to the embedding dimension. Position embedding represents local timestamp information by calculating a temporal position code:
where pos is the relative position of a time step in the input sequence,for inputting dimension sequence number d of sequence model For the embedding dimension (here set d model ＝512)，L x Is the input sequence length. Global time embedding with a d model The trainable full connection layer of each unit obtains global timestamp information for each time step. And finally, adding the three embedding results according to the corresponding time steps and the characteristic dimension to obtain a final input sequence embedding result.
(2) Encoder with a plurality of sensors
The submodule is mainly used for extracting long-range dependence of a front space-time characteristic sequence and is formed by superposing two secondary submodules, and each secondary submodule comprises a multi-head probability distribution sparse self-attention layer and a self-attention distillation three-level submodule (shown in figure 4). The multi-head probability distribution sparse self-attention layer considers the long tail distribution characteristic of self-attention weight, adopts K-L divergence as sparse measurement, and selects the Query (Query) vector with the highest importance to perform dot product calculation, so that the purpose of reducing the complexity of a model is achieved, and the specific operation process is as follows:
1) Based on the input time series I, N is utilized H Single head (where N is set up) H Weight parameter matrix of=8), according to the Query, key (Key) and Value (Value) matrix corresponding to each head is obtained:
Q k ＝W k Q I
K k ＝W k K I
V k ＝W k V I
wherein Q is k 、K k 、V k Query, key and Value matrices, W, for the kth head, respectively k Q 、W k K 、W k V The kth head is used for calculating the weight matrix of the Query, key and Value matrix respectively.
2) For each Query vector q in the Query matrix for each head i According to the length L of the self-attention layer input sequence, llnL key vectors k are randomly sampled j Composition of the compositionAnd calculate q i Corresponding sparsity measure->
Wherein d=d model /N H According to a preset sampling factor c=5, clnL q with the maximum sparse measurement is selected i Recombination into sparse matrixSelf-attention calculations were performed:
then, filling the rest positions which are not subjected to attention calculation by using the average Value of the Value matrix to obtain an attention calculation result A corresponding to each head k 。
3) The attention calculation results of each head are spliced together and pass through a weight matrix W O Obtaining a final multi-head probability distribution sparse self-attention calculation result:
MultiHead(Q，K，V)＝Concat(A 1 ，...，A 8 )W O the self-attention distillation three-stage sub-module is used to properly shorten the length of the sequence, reduce the complexity of the model, and include a 1D convolution layer and a 1D max pooling layer (as shown in fig. 4). Wherein, settingThe window length of the 1D convolution is 3, the convolution step length is 1, the number of convolution kernels is 512, and an ELU activation function and a padding strategy are adopted; the window size of the 1D max-pooling layer is 3 and the step size is 2.
(3) Decoder
The submodule improves the dynamic decoding strategy of the traditional converter decoder, and the reasoning efficiency is improved by generating long-time sequence prediction in high dimensionality in one step. The sub-module consists of a mask multi-head probability distribution sparse self-attention layer and a multi-head attention layer (as shown in fig. 4). Before multi-head probability distribution sparse self-attention calculation is carried out on the multi-head probability distribution sparse self-attention layer, masking is carried out on the features on future time steps in input, and autoregressive is avoided; and the multi-head attention layer calculates a Query matrix by using the output of the mask multi-head probability distribution sparse self-attention layer, calculates a Key matrix and a Value matrix by using the output of the encoder, and outputs a high-dimension long sequence prediction result by using the traditional multi-head attention calculation.
(4) Fully connected output layer
The part is a fully connected layer with 1 unit and is used for mapping a high-dimensional prediction sequence generated by a decoder to one dimension and outputting a final 10-minute-level solar radiation length time sequence prediction result.
To sum up, the validity of the method is verified by using solar radiation observation data of a ground radiation station-Beijing station in 2017 and Himaware-8 stationary satellite data. The time resolution of ground observation is 10 minutes, the time sequence is complete, and the ground observation is subjected to strict quality inspection; the spatial resolution of satellite observations was 0.05 ° and the temporal resolution was 10 minutes. In this embodiment, a space window with a size of 32×32 is set with the site position as the center, and satellite data (including 4 channels, namely, band 1-3 observations and solar zenith angles) corresponding to ground observations at each time step are extracted. Setting the length of the pre-sequence and the length of the predicted sequence to be 48 time steps (namely 480 minutes), taking the ground solar radiation observation and satellite image sequence in the pre-period as model input, and taking the ground solar radiation observation sequence in the predicted period as a label (label) value. As shown in fig. 5, the effect of the integrated satellite and site observation of the present embodiment for prediction is better than the case of using only ground and satellite observations; meanwhile, the method also shows the advantages compared with the common deep learning sequence prediction model CLSTM.
In addition, the embodiment further takes the cloud layer movement and solar radiation change conditions of 11:20-19:20 of the local time 2017, 9 months and 13 days as an example, so that the effectiveness of the method is further shown. As shown in fig. 6, the radiation drop due to the occurrence of cloud (e.g., 12:00-12:50 or 14:00 time periods) or the radiation rise due to the departure of cloud (e.g., 13:00 time periods) cannot be predicted using only ground observation. In contrast, integrated satellite and terrestrial observations can more accurately capture the effect of the cloud on future radiation fluctuations based on the law of motion of the cloud layer for the preamble period. At the same time, the predictions obtained by the method are generally closer to the actual radiation conditions than the normal CLSTM model, especially over longer periods of time (e.g., 14:30-19:20 periods of time) in advance. The above results demonstrate significant advantages of the present method in solving long-range dependence problems, obtaining fine time resolution long-sequence predictions of radiation.
Therefore, the present embodiment has the following advantages:
firstly, the embodiment discloses a technical method for finely predicting the future ultra-short solar radiation of an area by using a deep learning method by integrating Himaware-8 stationary satellite remote sensing images and ground site observation. The method can rapidly and accurately provide solar radiation length sequence prediction of 10 minutes in the future, and timely and accurately grasp the change condition of the solar radiation in the future, thereby being beneficial to scientifically controlling and dispatching the reserve of the solar photovoltaic power grid and improving the stability of photovoltaic power supply.
Under the deep learning framework, cloud layer change information in satellite observation and solar radiation dynamic evolution information in ground observation are fully integrated, the long-range dependence problem is solved by utilizing a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation decoder mechanism, the solar radiation fine prediction of 0-8 hours and 10 minutes in the future is rapidly and accurately realized, scientific decisions in the aspects of real-time adjustment, storage management, unit combination and the like of a photovoltaic power generation system are facilitated, and technical support is provided for improving the reliability and stability of photovoltaic power supply.
Secondly, compared with the prior art:
1. compared with the existing method based on ground observation only, the method has the advantages that the influence of large-scale cloud layer dynamics on the ground solar radiation is considered in the prediction process through integrating satellite and ground observation, and the accuracy of radiation prediction under the cloudy condition can be improved remarkably.
2. The method solves the problem of long-range dependence in a high-time-resolution long sequence by using a probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation type decoder mechanism, overcomes the defects of insufficient simulation capability, high complexity, low calculation efficiency and the like when the existing model processes the long time sequence, and can rapidly and accurately provide a 10-minute-level fine radiation prediction result with a long lead period.
3. The method is more beneficial to scientific and intelligent decision making in the aspects of real-time adjustment, storage management, unit combination and the like of the photovoltaic power generation system, and provides technical support for improving the reliability and stability of photovoltaic power supply.
As shown in fig. 7, the present invention provides a solar radiation prediction system comprising:
the data acquisition module is used for acquiring initial data; the initial data includes a long sequence of still satellite images and a long sequence of ground radiation observations.
The radiation prediction module is used for determining a 10-minute-level solar radiation length sequence prediction result according to the initial data and the solar radiation prediction model; the solar radiation prediction model comprises a cloud layer feature extraction module, a radiation change feature extraction module and a long-time sequence radiation prediction module; the cloud layer feature extraction module and the radiation change feature extraction module are connected with the long-time-sequence radiation prediction module; the cloud layer feature extraction module is constructed through a plurality of three-dimensional convolution blocks; the radiation change characteristic extraction module is constructed by a plurality of one-dimensional convolution blocks; the long-time sequence radiation prediction module is constructed according to a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation type decoder mechanism; the cloud layer feature extraction module is used for determining cloud layer space features of the long sequence of the static satellite images; the radiation change feature extraction module is used for determining solar radiation time change features of the ground radiation observation long sequence; the long-time-sequence radiation prediction module is used for determining a 10-minute-level solar radiation long-sequence prediction result according to the cloud layer space characteristics and the solar radiation time change characteristics.
The invention provides an electronic device comprising a memory for storing a computer program and a processor for running the computer program to cause the electronic device to perform a solar radiation prediction method according to the above.
The present invention provides a computer readable storage medium storing a computer program which when executed by a processor implements a solar radiation prediction method as described above.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the core concept of the invention; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (8)
1. A method of solar radiation prediction, comprising:
acquiring initial data; the initial data comprises a static satellite image long sequence and a ground radiation observation long sequence;
determining a 10-minute solar radiation length sequence prediction result according to the initial data and a solar radiation prediction model; the solar radiation prediction model comprises a cloud layer feature extraction module, a radiation change feature extraction module and a long-time sequence radiation prediction module; the cloud layer feature extraction module and the radiation change feature extraction module are connected with the long-time-sequence radiation prediction module;
the cloud layer feature extraction module is constructed through a plurality of three-dimensional convolution blocks; the radiation change characteristic extraction module is constructed by a plurality of one-dimensional convolution blocks; the long-time sequence radiation prediction module is constructed according to a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation type decoder mechanism; the cloud layer feature extraction module is used for determining cloud layer space features of the long sequence of the static satellite images; the radiation change feature extraction module is used for determining solar radiation time change features of the ground radiation observation long sequence; the long-time-sequence radiation prediction module is used for determining a 10-minute-level solar radiation long-sequence prediction result according to the cloud layer space characteristics and the solar radiation time change characteristics.
2. The method of claim 1, wherein each of the three-dimensional convolution blocks and each of the one-dimensional convolution blocks comprises a convolution layer and a maximum pooling layer connected in sequence;
the long-time-sequence radiation prediction module comprises a first embedding sub-module, a second embedding sub-module, an encoder, a decoder and a fully-connected output layer; the first embedded submodule, the encoder and the decoder are connected in sequence; the decoder is also connected with the second embedded submodule;
the encoder comprises two multi-head probability distribution sparse self-attention layers and two self-attention distillation layers which are sequentially and alternately connected; the decoder comprises a mask multi-head probability distribution sparse self-attention layer and a multi-head self-attention layer which are connected in sequence.
3. The solar radiation prediction method according to claim 1, wherein determining a 10-minute-level solar radiation length sequence prediction result according to the initial data and a solar radiation prediction model specifically comprises:
extracting features of the long sequence of the static satellite image by utilizing three-dimensional convolution to obtain cloud layer space features;
extracting features of the ground radiation observation long sequence by utilizing one-dimensional convolution to obtain solar radiation time change features;
and carrying out feature prediction according to the cloud layer space features and the solar radiation time change features to obtain a 10-minute solar radiation length sequence prediction result.
4. A solar radiation prediction method according to claim 3, wherein the characteristic prediction is performed according to the cloud layer spatial characteristic and the solar radiation time variation characteristic to obtain a 10-minute solar radiation length sequence prediction result, and the method specifically comprises the following steps:
carrying out space-time fusion pretreatment on the cloud layer space characteristics and the solar radiation time variation characteristics to obtain a front space-time characteristic sequence and a generated decoder input sequence;
performing marker embedding, position embedding and global time embedding on the pre-space-time feature sequence, and adding according to the time steps and feature dimensions of the corresponding sequence to obtain a first input sequence;
extracting the long-range dependence of the first input sequence by utilizing multi-head probability distribution sparse self-attention operation and self-attention distillation operation to obtain a long-range dependence characteristic diagram;
performing mark embedding, position embedding and global time embedding on the input sequence of the generating decoder, and adding according to the time steps and feature dimensions of the corresponding sequence to obtain a second input sequence;
and generating a 10-minute-level solar radiation long sequence prediction result according to the second input sequence and the long-range dependent feature map by using a feature mask method and multi-head probability distribution sparse self-attention operation.
5. The method according to claim 4, wherein preprocessing the cloud space features and the solar radiation time variation features to obtain a pre-space-time feature sequence and a generated decoder input sequence, specifically comprises:
aligning and splicing the cloud layer space characteristics and the solar radiation time change characteristics according to time steps to obtain a front space-time characteristic sequence;
dividing the preposed space-time characteristic sequence into two sections with equal length, and after the second half section sequence, splicing the 0 value sequence with the length of the predicted time steps along the time dimension to obtain the input sequence of the generating decoder.
6. A solar radiation prediction system, comprising:
the data acquisition module is used for acquiring initial data; the initial data comprises a static satellite image long sequence and a ground radiation observation long sequence;
the radiation prediction module is used for determining a 10-minute-level solar radiation length sequence prediction result according to the initial data and the solar radiation prediction model; the solar radiation prediction model comprises a cloud layer feature extraction module, a radiation change feature extraction module and a long-time sequence radiation prediction module; the cloud layer feature extraction module and the radiation change feature extraction module are connected with the long-time-sequence radiation prediction module;
the cloud layer feature extraction module is constructed through a plurality of three-dimensional convolution blocks; the radiation change characteristic extraction module is constructed by a plurality of one-dimensional convolution blocks; the long-time sequence radiation prediction module is constructed according to a multi-head probability distribution sparse self-attention mechanism, a self-attention distillation mechanism and a generation type decoder mechanism; the cloud layer feature extraction module is used for determining cloud layer space features of the long sequence of the static satellite images; the radiation change feature extraction module is used for determining solar radiation time change features of the ground radiation observation long sequence; the long-time-sequence radiation prediction module is used for determining a 10-minute-level solar radiation long-sequence prediction result according to the cloud layer space characteristics and the solar radiation time change characteristics.
7. An electronic device comprising a memory for storing a computer program and a processor that runs the computer program to cause the electronic device to perform the solar radiation prediction method according to claims 1-5.
8. A computer-readable storage medium, characterized in that it stores a computer program which, when executed by a processor, implements the solar radiation prediction method as claimed in claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310498367.XA CN116612393A (en) | 2023-05-05 | 2023-05-05 | Solar radiation prediction method, system, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310498367.XA CN116612393A (en) | 2023-05-05 | 2023-05-05 | Solar radiation prediction method, system, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116612393A true CN116612393A (en) | 2023-08-18 |
Family
ID=87681016
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310498367.XA Pending CN116612393A (en) | 2023-05-05 | 2023-05-05 | Solar radiation prediction method, system, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116612393A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117272002A (en) * | 2023-11-23 | 2023-12-22 | 中国电建集团西北勘测设计研究院有限公司 | Solar radiation amount estimation method and device, electronic equipment and storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114895380A (en) * | 2022-04-22 | 2022-08-12 | 中国科学院地理科学与资源研究所 | Solar radiation prediction method, device, equipment and medium |
| CN115310674A (en) * | 2022-07-14 | 2022-11-08 | 西北师范大学 | Long-time sequence prediction method based on parallel neural network model LDformer |
| CN115423080A (en) * | 2022-09-16 | 2022-12-02 | 中国电信股份有限公司 | Time series prediction method, time series prediction device, electronic device, and medium |
| CN115457389A (en) * | 2022-09-05 | 2022-12-09 | 河海大学 | Ultra-short-term solar radiation prediction method and system based on sparse space-time feature descriptor |
| CN115511220A (en) * | 2022-11-02 | 2022-12-23 | 河海大学 | Ultra-short-term solar radiation prediction method and system based on cross-mode attention mechanism |
| CN115618714A (en) * | 2022-09-09 | 2023-01-17 | 北京思源知行科技发展有限公司 | Solar radiation probability forecasting method and related equipment |
| CN115982567A (en) * | 2022-12-16 | 2023-04-18 | 上海碳索能源服务股份有限公司 | Refrigerating system load prediction method based on sequence-to-sequence model |
-
2023
- 2023-05-05 CN CN202310498367.XA patent/CN116612393A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114895380A (en) * | 2022-04-22 | 2022-08-12 | 中国科学院地理科学与资源研究所 | Solar radiation prediction method, device, equipment and medium |
| CN115310674A (en) * | 2022-07-14 | 2022-11-08 | 西北师范大学 | Long-time sequence prediction method based on parallel neural network model LDformer |
| CN115457389A (en) * | 2022-09-05 | 2022-12-09 | 河海大学 | Ultra-short-term solar radiation prediction method and system based on sparse space-time feature descriptor |
| CN115618714A (en) * | 2022-09-09 | 2023-01-17 | 北京思源知行科技发展有限公司 | Solar radiation probability forecasting method and related equipment |
| CN115423080A (en) * | 2022-09-16 | 2022-12-02 | 中国电信股份有限公司 | Time series prediction method, time series prediction device, electronic device, and medium |
| CN115511220A (en) * | 2022-11-02 | 2022-12-23 | 河海大学 | Ultra-short-term solar radiation prediction method and system based on cross-mode attention mechanism |
| CN115982567A (en) * | 2022-12-16 | 2023-04-18 | 上海碳索能源服务股份有限公司 | Refrigerating system load prediction method based on sequence-to-sequence model |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117272002A (en) * | 2023-11-23 | 2023-12-22 | 中国电建集团西北勘测设计研究院有限公司 | Solar radiation amount estimation method and device, electronic equipment and storage medium |
| CN117272002B (en) * | 2023-11-23 | 2024-02-20 | 中国电建集团西北勘测设计研究院有限公司 | Solar radiation amount estimation method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110070226B (en) | Photovoltaic power prediction method and system based on convolutional neural network and meta-learning | |
| CN113128113A (en) | Poor information building load prediction method based on deep learning and transfer learning | |
| CN117114192B (en) | Offshore wind power prediction method and device based on multi-scale space-time diagram transformation network | |
| CN116128170B (en) | Photovoltaic power station power ultra-short-term prediction method and device and related equipment | |
| CN116865261B (en) | Power load prediction method and system based on twin network | |
| CN116612393A (en) | Solar radiation prediction method, system, electronic equipment and storage medium | |
| CN116451821A (en) | Deep learning-based large-area new energy generation power prediction method and system | |
| CN112653198A (en) | Wind power output scene generation method and system based on prediction box | |
| CN117439045A (en) | Multi-element load prediction method for comprehensive energy system | |
| CN112925824A (en) | Photovoltaic power prediction method and system for extreme weather type | |
| CN115936177A (en) | Photovoltaic output power prediction method and system based on neural network | |
| CN116029419A (en) | Deep learning-based long-term new energy daily average generation power prediction method and system | |
| CN116502074A (en) | Model fusion-based photovoltaic power generation power prediction method and system | |
| Ma et al. | Multi-features fusion for short-term photovoltaic power prediction | |
| Tuyen et al. | A combination of novel hybrid deep learning model and quantile regression for short‐term deterministic and probabilistic PV maximum power forecasting | |
| Wang et al. | Grid load forecasting based on dual attention BiGRU and DILATE loss function | |
| Yao et al. | Very short-term forecasting of distributed PV power using GSTANN | |
| CN117955088A (en) | Attention mechanism-based distributed photovoltaic power online prediction method and system | |
| CN113984198A (en) | Short wave radiation prediction method and system based on convolutional neural network | |
| CN114004405B (en) | Photovoltaic power prediction method and system based on Elman neural network and satellite cloud image | |
| Bendali et al. | Deep learning for very short term solar irradiation forecasting | |
| Peng et al. | A hybrid model for solar radiation forecasting towards energy efficient buildings | |
| Xia et al. | Research on Solar Radiation Estimation based on Singular Spectrum Analysis-Deep Belief Network | |
| Cai et al. | Short-term solar photovoltaic irradiation predicting using a nonlinear prediction method | |
| Guo et al. | Ultra-Short-Term Photovoltaic Power Forecasting Based on Transformer Model |
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 |