CN114037901B - Real-time satellite near infrared image calculation method based on photovoltaic power generation prediction guiding - Google Patents

Real-time satellite near infrared image calculation method based on photovoltaic power generation prediction guiding Download PDF

Info

Publication number
CN114037901B
CN114037901B CN202111243878.4A CN202111243878A CN114037901B CN 114037901 B CN114037901 B CN 114037901B CN 202111243878 A CN202111243878 A CN 202111243878A CN 114037901 B CN114037901 B CN 114037901B
Authority
CN
China
Prior art keywords
photovoltaic power
power generation
time
layer
convolution
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.)
Active
Application number
CN202111243878.4A
Other languages
Chinese (zh)
Other versions
CN114037901A (en
Inventor
程礼临
臧海祥
刘璟璇
张越
卫志农
孙国强
周亦洲
黄蔓云
陈�胜
韩海腾
朱瑛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hohai University HHU
Original Assignee
Hohai University HHU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hohai University HHU filed Critical Hohai University HHU
Priority to CN202111243878.4A priority Critical patent/CN114037901B/en
Publication of CN114037901A publication Critical patent/CN114037901A/en
Application granted granted Critical
Publication of CN114037901B publication Critical patent/CN114037901B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • 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)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Marketing (AREA)
  • Biophysics (AREA)
  • Development Economics (AREA)
  • Water Supply & Treatment (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Primary Health Care (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Public Health (AREA)
  • Educational Administration (AREA)
  • Computing Systems (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a photovoltaic power generation prediction-oriented real-time satellite near infrared image calculation method, which comprises the following steps: collecting satellite near infrared image data as an estimation object and photovoltaic power station power generation data as guide data; constructing a full convolution neural network model aiming at satellite near-infrared image calculation, taking a plurality of frames of satellite near-infrared images before the time to be predicted as input and a predicted image as output, and determining a model structure and parameters thereof; using a partial differential convolution layer to replace a second convolution layer of the model to incorporate a moment estimation error into a mean square error of the image prediction; estimating photovoltaic power generation capacity based on the predicted image, calculating the mean square error of the photovoltaic power generation capacity, and incorporating the calculated total error of the formed satellite near infrared image; training the model based on the collected data and the total error; and applying the trained model to satellite near infrared image estimation and photovoltaic power generation estimation which are carried out in multiple steps in advance. The method can improve the prediction precision of the photovoltaic power generation amount and can provide data support for future photovoltaic power generation amount estimation.

Description

Real-time satellite near infrared image calculation method based on photovoltaic power generation prediction guiding
Technical Field
The invention relates to a photovoltaic power generation prediction oriented real-time satellite near infrared image calculation method, and belongs to the technical field of big data analysis of a power system.
Background
In recent years, the increase speed of the installed capacity of new energy power generation in China is continuously accelerated, and particularly, the solar photovoltaic power generation is applied to a large scale in a distributed micro-grid, a county rural power grid and the like. However, the randomness of the photovoltaic power generation capacity can increase the dispatching pressure of a new energy power system, so that the unreasonable utilization of photovoltaic resources is caused, and the further development of photovoltaic power generation is limited. The photovoltaic power generation prediction technology can obviously reduce the influence of the uncertainty of the output of the photovoltaic power generation prediction technology, and is beneficial to improving the economic benefits of distributed users and power system operators. The prediction of the real-time photovoltaic power generation mainly aims at a time scale of 30-60 minutes to cooperate with real-time power grid scheduling in a corresponding time period, so that the power balance of a power system is ensured.
The real-time photovoltaic power generation output is mainly influenced by shielding of a short-time cloud cover on a photovoltaic panel, so that the cloud cover change condition of a photovoltaic power station area is mastered and important. With the development of satellite remote sensing technology, satellite near-infrared images can acquire solar radiation level and cloud cover level of a large-range geographic area with higher space-time resolution, effective data support is provided for evaluating the power generation capacity of a photovoltaic power station, satellite near-infrared images of future time are calculated in advance and are applied to estimate future photovoltaic power generation capacity, and accuracy and stability of current real-time photovoltaic power generation prediction are expected to be improved.
Satellite near infrared image estimation needs to solve the problem of multi-frame continuous image processing, so that the satellite near infrared image estimation is a research focus in the field of computer vision, and related estimation methods based on convolutional neural networks and cyclic neural networks have been successfully applied. However, such an estimation method is mostly aimed at the problem of general image processing, and special application scenes are not deeply considered. For the photovoltaic power generation prediction problem of the power system, the overall accuracy of cloud image calculation is not necessarily capable of representing the performance level of photovoltaic power generation evaluation, and the complicated image calculation method and model can obviously aggravate the calculation cost of power system model deployment while the photovoltaic power generation prediction effectiveness is possibly not guaranteed. Therefore, the invention provides a photovoltaic power generation prediction-oriented real-time satellite near infrared image calculation method for serving the scheduling application requirements of the power system.
Disclosure of Invention
Aiming at the defects of the image calculation method in the current computer vision field in the photovoltaic power generation prediction application, the invention provides a real-time satellite near infrared image calculation method for guiding photovoltaic power generation prediction, and the validity of the calculation method for photovoltaic power generation estimation is ensured by incorporating the mean square error of the image prediction result for photovoltaic power generation estimation, so that the data support of future photovoltaic power generation capacity estimation is provided for a new energy power system and a photovoltaic power station under the control of the new energy power system, and the real-time scheduling of the power system is realized in cooperation, so that the economical efficiency and the reliability of power grid operation are improved.
The technical scheme adopted by the invention specifically solves the technical problems as follows:
a photovoltaic power generation prediction-oriented real-time satellite near infrared image calculation method comprises the following steps:
step 1: collecting satellite near infrared image data within set geographic resolution and time resolution as an estimation object, and collecting photovoltaic power station power generation data with the same time resolution as guiding data for satellite near infrared image estimation;
step 2: constructing a full convolution neural network model aiming at satellite near infrared image calculation, wherein the full convolution neural network model is built by only using convolution layers and deconvolution layers, and the number of the convolution layers is more than or equal to 2; based on the current time T and the time T predicted in future, taking a d frame satellite near infrared image before the time t+T to be predicted as input and taking a predicted image of the time t+T as output, and determining a model structure and parameters thereof;
step 3: replacing a second convolution layer of the full-convolution neural network model by using a partial differential convolution layer, performing moment estimation on convolution parameters of the partial differential convolution layer, solving a moment matrix of the convolution parameters, calculating moment estimation errors, and incorporating the moment estimation errors into a mean square error of image prediction;
step 4: estimating the photovoltaic power generation amount of a photovoltaic power station at time t+T based on a predicted image of time t+T, comparing the estimated photovoltaic power generation amount estimated value at time t+T with the photovoltaic power generation amount actual value at time t+T, calculating the mean square error of the photovoltaic power generation amount, and incorporating the mean square error of the photovoltaic power generation amount into the mean square error of the moment estimated error and the image prediction to form a satellite near infrared image estimation total error;
step 5: training a full convolution neural network model by collecting satellite near infrared image data of at least one year time span and photovoltaic power station power generation data and calculating total errors based on the satellite near infrared image;
step 6: the values T=1, 2,3, … and K, wherein K is a natural number above 1, satellite near infrared images with rolling calculation time t+1 to t+K are calculated based on the trained full convolution neural network model, satellite near infrared image calculation in K steps in advance is achieved, and the calculation result and the full convolution neural network model performance are checked based on the mean square error of image prediction and the mean square error of photovoltaic power generation.
Further, as a preferred technical solution of the present invention, the satellite near infrared image data collected in the step 1 needs to cover a geographical range of a position where the photovoltaic power station is located, and reflect a solar radiance measurement value of the photovoltaic power station, where a calculation formula of the solar radiance measurement value is as follows:
Figure BDA0003318964040000021
in the method, in the process of the invention,
Figure BDA0003318964040000031
the near infrared image of the satellite with time t and wavelength wl, w is the image width, h is the image height, < ->
Figure BDA0003318964040000032
Is a real number domain range, and (x, y) is an image coordinate; />
Figure BDA0003318964040000033
The pixel value at the (x, y) coordinate in the satellite near-infrared image at time t and wavelength wl represents the solar radiance measurement value at the (x, y) coordinate +.>
Figure BDA0003318964040000034
The unit is Wm –2 sr –1 (cm –1 ) –1 I.e. the intensity of radiation per centimeter of wavelength at a projected area per square meter.
Further, as a preferred technical solution of the present invention, in the step 2, a full convolution neural network model for satellite near infrared image estimation is constructed, and a calculation formula is expressed as follows:
Figure BDA0003318964040000035
Figure BDA0003318964040000036
in the method, in the process of the invention,
Figure BDA0003318964040000037
predicted image for time t+T to t+1, < >>
Figure BDA0003318964040000038
For satellite near infrared images of time t-d+1+T to t, FCN (·) is a full convolution neural network model function, α (·) is an activation function, { W } is a convolution parameter set, { W } T The deconvolution parameter set, +.>
Figure BDA0003318964040000039
For the output of the mth channel of the first layer,/->
Figure BDA00033189640400000310
For the input of the ith channel of the layer 1, ch is the number of input channels, +.>
Figure BDA00033189640400000311
For convolution operator +.>
Figure BDA00033189640400000312
Convolution parameter for mth channel of layer I, < ->
Figure BDA00033189640400000313
For the deconvolution parameter of the mth channel of the first layer,/for the first layer>
Figure BDA00033189640400000314
Is the bias parameter of the mth channel of the first layer.
Further, as a preferable technical scheme of the present invention, the step 3 specifically includes the following steps:
firstly, defining an operation rule of a partial differential convolution layer, enabling partial differential convolution operation to be equal to partial derivative operation, and adopting the following formula:
Figure BDA00033189640400000315
in the method, in the process of the invention,
Figure BDA00033189640400000316
and->
Figure BDA00033189640400000317
The hidden layer output of the mth channel of the first layer, the input of the ith channel of the first-1 layer and the convolution parameters of the mth channel of the first layer are respectively +.>
Figure BDA00033189640400000318
For the convolution operator ix is the abscissa partial derivative order, iy is the ordinate partial derivative order,/-, and>
Figure BDA00033189640400000319
the input of the ith +iy order bias value for the ith channel of the layer 1,/th>
Figure BDA00033189640400000320
Is the ix-th order partial derivative of the image abscissa,>
Figure BDA00033189640400000321
an iy-th order partial derivative value of an ordinate of the image;
secondly, performing moment estimation on convolution parameters of the partial differential convolution layer, and solving a moment matrix of the convolution parameters, wherein a calculation formula is as follows:
Figure BDA00033189640400000322
in the method, in the process of the invention,
Figure BDA0003318964040000041
the matrix of the convolution parameter of the mth channel of the first layer, ix is the abscissa of the matrix, and the value is equal to the offset of the abscissa; iy is the ordinate of the matrix, and the value is equal to the ordinate partial derivative order; />
Figure BDA0003318964040000042
For the element values at the coordinates (ix, iy) in the moment matrix, k is the convolution parameter size, u and v are the convolution parameter abscissa and ordinate corresponding to the moment matrix coordinates (ix, iy), and +.>
Figure BDA0003318964040000043
Element values at coordinates (u, v) in the convolution parameters for the mth channel of the first layer;
then, matrix of moment based on convolution parameters
Figure BDA0003318964040000044
Calculating a moment estimation error, wherein the calculation formula is as follows:
Figure BDA0003318964040000045
Figure BDA0003318964040000046
where LossM is the moment estimation error, Δ ix,iy For matrix of moment estimation standard, real matrix of size k×k is satisfied
Figure BDA0003318964040000047
Δ ix,iy (i, j) is the coordinate in the moment estimation matrix (i, the value of the element at j) is, I.I F Is a matrix F norm function;
finally, calculating the mean square error of image prediction, and incorporating the moment estimation error into the mean square error of image prediction, wherein the formula is as follows:
Figure BDA0003318964040000048
where LossI is the mean square error of the image prediction,
Figure BDA0003318964040000049
and->
Figure BDA00033189640400000410
Satellite near-infrared images and predicted images of time t+T and wavelength wl respectively 2 Is a vector mean square error function.
Further, as a preferred technical scheme of the present invention, the step 4 estimates the photovoltaic power generation amount of the photovoltaic power station at time t+t through a constructed photovoltaic power generation power estimation model based on the multilayer perceptron, and the calculation formula of the photovoltaic power generation power estimation model is as follows:
Figure BDA00033189640400000411
Figure BDA00033189640400000412
in the method, in the process of the invention,
Figure BDA00033189640400000413
predicted image for time t+T, wavelength wl,>
Figure BDA00033189640400000414
for the photovoltaic power generation estimation of time t+T, MLP (·) is a multilayer perceptron function, { W MLP Sum { b } MLP Weight parameter set and bias parameter set of multi-layer perceptron }, +.>
Figure BDA00033189640400000415
For the hidden layer output of the first layer of the multi-layer sensor,/I>
Figure BDA00033189640400000416
Weight of the first layer of the multi-layer sensor, < ->
Figure BDA00033189640400000417
For the bias of the first layer of the multi-layer sensor, α (·) is the activation function.
Further, as a preferred technical solution of the present invention, the mean square error of the photovoltaic power generation amount calculated in the step 4 is calculated by the following calculation formula:
Figure BDA0003318964040000051
where LossP is the mean square error of the photovoltaic power generation,
Figure BDA0003318964040000052
estimated value of photovoltaic power generation amount, P, for time t+T t+T The actual value of the photovoltaic power generation amount at time T + T, I.I 2 As a vector mean square error function, ns is the number of training samples, < +.>
Figure BDA0003318964040000053
And P t+T (i) The photovoltaic power generation amount estimated value and the actual value at time t+T are the ith sample.
Further, as a preferred technical solution of the present invention, the total error calculated by the satellite near infrared image formed in the step 4 is calculated by the following formula:
Figure BDA0003318964040000054
wherein, loss sigma is the total error calculated by the satellite near infrared image; lossI is the mean square error of image prediction, lossM is the moment estimation error, and LossP is the mean square error of photovoltaic power generation;
Figure BDA0003318964040000055
and->
Figure BDA0003318964040000056
Satellite near-infrared images and predicted images of time t+T and wavelength wl respectively 2 Is a vector mean square error function; />
Figure BDA0003318964040000057
For the element value at coordinates (ix, iy) in the moment matrix, delta ix,iy For the matrix of the moment estimation standard, a real matrix of the size k×k is satisfied>
Figure BDA0003318964040000058
||·|| F Is a matrix F norm function; />
Figure BDA0003318964040000059
Estimated value of photovoltaic power generation amount, P, for time t+T t+T The actual value of photovoltaic power generation at time t+T.
By adopting the technical scheme, the invention can produce the following technical effects:
1) Compared with the existing image calculation technology, the method can better adapt to the photovoltaic power generation amount estimation and prediction requirements of the power system;
2) The method can effectively utilize satellite remote sensing data, and improves photovoltaic power generation capacity prediction accuracy of the traditional daily short-term time scale and real-time scale;
3) The satellite near infrared image calculated by the method can provide data support for future photovoltaic power generation amount estimation for a new energy power system and a photovoltaic power station governed by the new energy power system, and the economy of power grid dispatching operation is ensured by improving the accuracy of the future photovoltaic power generation amount estimation.
Drawings
FIG. 1 is a schematic diagram of the method of the present invention.
Fig. 2 is a view of the result of satellite near infrared image estimation according to the method of the present invention.
FIG. 3 is a graph showing the result of photovoltaic power generation estimation 1.5 hours in advance of the method of the present invention.
Detailed Description
Embodiments of the present invention will be described below with reference to the drawings.
As shown in fig. 1, the invention relates to a photovoltaic power generation prediction guiding real-time satellite near infrared image calculation method, which specifically comprises the following steps:
step 1: satellite near infrared image data within set geographic resolution and time resolution are collected as estimation objects, and photovoltaic power station power generation data with the same time resolution are collected as guiding data of satellite near infrared image estimation.
The invention can collect satellite near infrared image data with geographical resolution of 5km or less and time resolution of 1 hour or less as an estimation object. Moreover, the collected satellite near infrared image data needs to cover the geographical range of the position of the photovoltaic power station, and can reflect the solar radiance measurement value of the photovoltaic power station, and the formula is as follows:
Figure BDA0003318964040000061
in the method, in the process of the invention,
Figure BDA0003318964040000062
the near infrared image of the satellite with time t and wavelength wl, w is the image width, h is the image height, < ->
Figure BDA0003318964040000063
In the real number domain range, (x, y) is the image coordinates. />
Figure BDA0003318964040000064
The pixel value at the (x, y) coordinate in the satellite near-infrared image at time t and wavelength wl represents the solar radiance measurement value at the (x, y) coordinate +.>
Figure BDA0003318964040000065
The unit is Wm –2 sr –1 (cm –1 ) –1 I.e. the intensity of radiation per centimeter of wavelength at a projected area per square meter.
Step 2: constructing a full convolution neural network model (fully convolutional network, FCN) aiming at satellite near infrared image calculation, wherein the full convolution neural network model is built by only using convolution layers and deconvolution layers, and the number of the convolution layers is more than or equal to 2; based on the current time T and the time T predicted to the future, the model structure and parameters thereof are determined by taking the near-infrared image of the satellite of d frames before the time t+T to be predicted as input and the near-infrared predicted image of the satellite of the time t+T as output, specifically as follows:
the full convolution neural network model constructed by the method aiming at satellite near infrared image calculation is built by only using convolution layers and deconvolution layers, wherein the number of the convolution layers is more than or equal to 2, and a model calculation formula is expressed as follows:
Figure BDA0003318964040000066
Figure BDA0003318964040000067
in the method, in the process of the invention,
Figure BDA0003318964040000071
for near infrared predicted images of satellites at times T + T to T +1,
Figure BDA0003318964040000072
for satellite near infrared images of time t-d+1+T to t, FCN (·) is a full convolution neural network model function, α (·) is an activation function, { W } is a convolution parameter set, { W } T The deconvolution parameter set, +.>
Figure BDA0003318964040000073
For the output of the mth channel of the first layer,/->
Figure BDA0003318964040000074
For the input of the ith channel of the layer 1, ch is the number of input channels, +.>
Figure BDA0003318964040000075
For convolution operator +.>
Figure BDA0003318964040000076
Convolution parameter for mth channel of layer I, < ->
Figure BDA0003318964040000077
For the deconvolution parameter of the mth channel of the first layer,/for the first layer>
Figure BDA0003318964040000078
Is the bias parameter of the mth channel of the first layer.
Step 3: replacing a second convolution layer of the full convolution neural network model by using a partial differential convolution layer, performing moment estimation on convolution parameters of the partial differential convolution layer, solving a moment matrix of the convolution parameters, calculating a moment estimation error, and incorporating the moment estimation error into a mean square error of image prediction, wherein the moment estimation error is specifically as follows:
firstly, defining an operation rule of a partial differential convolution layer, enabling partial differential convolution operation to be equal to partial derivative operation, and adopting the following formula:
Figure BDA0003318964040000079
in the method, in the process of the invention,
Figure BDA00033189640400000710
and->
Figure BDA00033189640400000711
The hidden layer output of the mth channel of the first layer, the input of the ith channel of the first-1 layer and the convolution parameters of the mth channel of the first layer are respectively +.>
Figure BDA00033189640400000712
For the convolution operator ix is the abscissa partial derivative order, iy is the ordinate partial derivative order,/-, and>
Figure BDA00033189640400000713
the input of the ith +iy order bias value for the ith channel of the layer 1,/th>
Figure BDA00033189640400000714
Is the ix-th order partial derivative of the image abscissa,>
Figure BDA00033189640400000715
is the iy order partial derivative of the ordinate of the image.
Secondly, performing moment estimation on convolution parameters of the partial differential convolution layer, and solving a moment matrix of the convolution parameters, wherein a calculation formula is as follows:
Figure BDA00033189640400000716
in the method, in the process of the invention,
Figure BDA00033189640400000717
the matrix of the convolution parameter of the mth channel of the first layer, ix is the abscissa of the matrix, and the value is equal to the offset of the abscissa; iy is the ordinate of the matrix of moment and the value is equal to the offset of the ordinate；/>
Figure BDA00033189640400000718
For the element values at the coordinates (ix, iy) in the moment matrix, k is the convolution parameter size, u and v are the convolution parameter abscissa and ordinate corresponding to the moment matrix coordinates (ix, iy), and +.>
Figure BDA00033189640400000719
Is the value of the element at the coordinates (u, v) in the convolution parameters of the mth channel of the first layer.
Then, matrix of moment based on convolution parameters
Figure BDA00033189640400000720
Calculating a moment estimation error, wherein the calculation formula is as follows:
Figure BDA0003318964040000081
Figure BDA0003318964040000082
where LossM is the moment estimation error, Δ ix,iy For matrix of moment estimation standard, real matrix of size k×k is satisfied
Figure BDA0003318964040000083
Δ ix,iy (i, j) is the coordinate in the moment estimation matrix (i, the value of the element at j) is, I.I F As a function of the matrix F norm.
Finally, calculating the mean square error of image prediction, and incorporating the moment estimation error into the mean square error of image prediction, wherein the formula is as follows:
Figure BDA0003318964040000084
where LossI is the mean square error of the image prediction,
Figure BDA0003318964040000085
and->
Figure BDA0003318964040000086
Satellite near-infrared images and predicted images of time t+T and wavelength wl respectively 2 Is a vector mean square error function.
Step 4: estimating photovoltaic power generation amount of a photovoltaic power station at time t+T based on a predicted image of time t+T, comparing an estimated photovoltaic power generation amount estimated value of the time t+T with a photovoltaic power generation amount actual value of the time t+T, calculating a mean square error of the photovoltaic power generation amount, and incorporating the mean square error of the photovoltaic power generation amount into a mean square error of moment estimation error and image prediction to form a satellite near infrared image estimation total error, wherein the method comprises the following steps of:
firstly, constructing a photovoltaic power generation power estimation model based on a multilayer perceptron, taking a predicted image of time t+T as input and a photovoltaic power generation amount estimated value of time t+T as output, wherein the formula of the model is as follows:
Figure BDA0003318964040000087
Figure BDA0003318964040000088
in the method, in the process of the invention,
Figure BDA0003318964040000089
predicted image near infrared for satellite at time t+T, wavelength wl, < >>
Figure BDA00033189640400000810
For the photovoltaic power generation estimation of time t+T, MLP (·) is a multilayer perceptron function, { W MLP Sum { b } MLP Weight parameter set and bias parameter set of multi-layer perceptron }, +.>
Figure BDA00033189640400000811
Is the hidden layer output of the first layer of the multi-layer sensor,/>
Figure BDA00033189640400000812
weight of the first layer of the multi-layer sensor, < ->
Figure BDA00033189640400000813
For the bias of the first layer of the multi-layer sensor, α (·) is the activation function.
Then, calculating the mean square error of the photovoltaic power generation amount according to the photovoltaic power generation amount estimated value and the photovoltaic power generation amount actual value at the time t+T, wherein the calculation formula is as follows:
Figure BDA00033189640400000814
where LossP is the mean square error of the photovoltaic power generation,
Figure BDA0003318964040000091
estimated value of photovoltaic power generation amount, P, for time t+T t+T The actual value of the photovoltaic power generation amount at time T + T, I.I 2 As a vector mean square error function, ns is the number of training samples, < +.>
Figure BDA0003318964040000092
And P t+T (i) The photovoltaic power generation amount estimated value and the actual value at time t+T are the ith sample.
And finally, incorporating the calculated mean square error of the photovoltaic power generation into the mean square error of the moment estimation error and the image prediction to form a satellite near infrared image calculation total error, wherein the formula is as follows:
Figure BDA0003318964040000093
in the formula, loss sigma is the total error calculated by the near infrared image of the satellite.
Step 5: and training the full convolution neural network model based on the satellite near infrared image calculation total error by using the satellite near infrared image data and the photovoltaic power station power generation data which are acquired for at least one year time span.
Step 6: the values T=1, 2,3, … and K, wherein K is a natural number above 1, satellite near infrared images with rolling calculation time t+1 to t+K are calculated based on the trained full convolution neural network model, satellite near infrared image calculation in K steps in advance is achieved, and the calculation result and the full convolution neural network model performance are checked based on the mean square error of image prediction and the mean square error of photovoltaic power generation.
The following describes in detail, with reference to specific embodiments, the implementation of probabilistic wind speed prediction using the method of the present invention. The embodiment of the invention selects 2018 satellite near infrared image data disclosed by European meteorological satellite application organization (EUMEASAT) and is used for estimating 2018 photovoltaic power generation data of Belgium nationally. Based on the acquired data, the method of the invention comprises the following specific implementation steps:
step 1: first, satellite near infrared image data of EUMEASAT in 2018 was acquired with a geographical resolution of 3km, a time resolution of 30min, and a center wavelength of 3.9 μm. The satellite near infrared image data needs to cover the geographic range of the position of the Belgium photovoltaic power station, has longitude of 1.95 DEG E-7.35 DEG E and latitude of 48.45 DEG N-52.85 DEG N, can reflect the solar radiance measurement value of the photovoltaic power station, and has the following formula:
Figure BDA0003318964040000094
in the method, in the process of the invention,
Figure BDA0003318964040000095
the near infrared image of the satellite with time t and wavelength wl, w is the image width, h is the image height, < ->
Figure BDA0003318964040000096
In the real number domain range, (x, y) is the image coordinates. />
Figure BDA0003318964040000097
Pixel values at (x, y) coordinates in the satellite near infrared image for time t, wavelength wlThe pixel value represents the solar radiance measurement value +.>
Figure BDA0003318964040000098
The unit is Wm –2 sr –1 (cm –1 ) –1 I.e. the intensity of radiation per centimeter of wavelength at a projected area per square meter. The image width w is 110 and the height h is 80, determined from longitude 1.95 deg. E-7.35 deg. E, latitude 48.45 deg. N-52.85 deg.. Because the time resolution of the satellite near infrared image data is 30min, the total power generation data of the photovoltaic power station in the Belgium nationwide in 2018 is collected according to the time interval of 30min and is used as the guidance data of satellite near infrared image estimation.
Step 2: constructing a full convolution neural network model aiming at satellite near-infrared image calculation, taking a d frame satellite near-infrared image before the time t+T to be predicted as input, taking a predicted image of the time t+T as output, and determining a model structure and parameters thereof. The model is built by only using a convolution layer and a deconvolution layer, wherein the number of layers of the convolution layer is more than or equal to 2, and a model calculation formula is expressed as follows:
Figure BDA0003318964040000101
Figure BDA0003318964040000102
in the method, in the process of the invention,
Figure BDA0003318964040000103
for near infrared predicted images of satellites at times T + T to T +1,
Figure BDA0003318964040000104
for satellite near infrared images of time t-d+1+T to t, FCN (·) is a full convolution neural network model function, α (·) is an activation function, { W } is a convolution parameter set, { W } T The deconvolution parameter set, +.>
Figure BDA0003318964040000105
For the output of the mth channel of the first layer,/->
Figure BDA0003318964040000106
For the input of the ith channel of the layer 1, ch is the number of input channels, +.>
Figure BDA0003318964040000107
For convolution operator +.>
Figure BDA0003318964040000108
Convolution parameter for mth channel of layer I, < ->
Figure BDA0003318964040000109
For the deconvolution parameter of the mth channel of the first layer,/for the first layer>
Figure BDA00033189640400001010
Is the bias parameter of the mth channel of the first layer.
Step 3: and replacing a second convolution layer of the full convolution neural network model by using the partial differential convolution layer, performing moment estimation on convolution parameters of the partial differential convolution layer, solving a moment matrix of the convolution parameters, calculating moment estimation errors, and incorporating the moment estimation errors into the mean square error of image prediction.
Wherein, defining the operation rule of partial differential convolution layer to make partial differential convolution operation equal to partial derivative operation, the formula is:
Figure BDA00033189640400001011
in the method, in the process of the invention,
Figure BDA00033189640400001012
and->
Figure BDA00033189640400001013
Hidden layer output of mth channel of the first layer, input of ith channel of the first-1 layer and convolution parameter of mth channel of the first layer respectively，/>
Figure BDA00033189640400001014
For the convolution operator ix is the abscissa partial derivative order, iy is the ordinate partial derivative order,/-, and>
Figure BDA00033189640400001015
the input of the ith +iy order bias value for the ith channel of the layer 1,/th>
Figure BDA00033189640400001016
Is the ix-th order partial derivative of the image abscissa,>
Figure BDA00033189640400001017
is the iy order partial derivative of the ordinate of the image.
Then, the moment estimation is carried out on the convolution parameters of the partial differential convolution layer, the moment matrix of the convolution parameters is solved, and the calculation formula is as follows:
Figure BDA0003318964040000111
in the method, in the process of the invention,
Figure BDA0003318964040000112
the matrix of the convolution parameter of the mth channel of the first layer, ix is the abscissa of the matrix, and the value is equal to the offset of the abscissa; iy is the ordinate of the matrix, and the value is equal to the ordinate partial derivative order; />
Figure BDA0003318964040000113
For the element values at the coordinates (ix, iy) in the moment matrix, k is the convolution parameter size, u and v are the convolution parameter abscissa and ordinate corresponding to the moment matrix coordinates (ix, iy), and +.>
Figure BDA0003318964040000114
Is the value of the element at the coordinates (u, v) in the convolution parameters of the mth channel of the first layer.
Second, moment based on convolution parametersMatrix array
Figure BDA0003318964040000115
Calculating a moment estimation error, wherein the calculation formula is as follows:
Figure BDA0003318964040000116
Figure BDA0003318964040000117
where LossM is the moment estimation error, Δ ix,iy For matrix of moment estimation standard, real matrix of size k×k is satisfied
Figure BDA0003318964040000118
Δ ix,iy (i, j) is the coordinate in the moment estimation matrix (i, the value of the element at j) is, I.I F As a function of the matrix F norm.
Finally, calculating the mean square error of image prediction, and incorporating the moment estimation error into the mean square error of image prediction, wherein the formula is as follows:
Figure BDA0003318964040000119
where LossI is the mean square error of the image prediction,
Figure BDA00033189640400001110
and->
Figure BDA00033189640400001111
Satellite near-infrared images and predicted images of time t+T and wavelength wl respectively 2 Is a vector mean square error function.
Step 4: then, the photovoltaic power generation amount of the photovoltaic power station at the time t+T is estimated based on the predicted image of the time t+T, the estimated photovoltaic power generation amount estimated value at the time t+T is compared with the photovoltaic power generation amount actual value at the time t+T, and the calculated mean square error of the photovoltaic power generation amount is included in the mean square error of the moment estimated error and the image prediction, so that the satellite near infrared image estimated total error is formed.
The method comprises the steps of constructing a photovoltaic power generation power estimation model based on a multilayer perceptron, taking a predicted image of time t+T as input, taking a photovoltaic power generation amount estimated value of time t+T as output, and adopting a formula of the model as follows:
Figure BDA0003318964040000121
Figure BDA0003318964040000122
in the method, in the process of the invention,
Figure BDA0003318964040000123
predicted image near infrared for satellite at time t+T, wavelength wl, < >>
Figure BDA0003318964040000124
For the photovoltaic power generation estimation of time t+T, MLP (·) is a multilayer perceptron function, { W MLP Sum { b } MLP Weight parameter set and bias parameter set of multi-layer perceptron }, +.>
Figure BDA0003318964040000125
For the hidden layer output of the first layer of the multi-layer sensor,/I>
Figure BDA0003318964040000126
Weight of the first layer of the multi-layer sensor, < ->
Figure BDA0003318964040000127
For the bias of the first layer of the multi-layer sensor, α (·) is the activation function.
Secondly, calculating the mean square error of the photovoltaic power generation amount based on the photovoltaic power generation amount estimated value and the photovoltaic power generation amount actual value of the model calculation time t+T, wherein the calculation formula is as follows:
Figure BDA0003318964040000128
where LossP is the mean square error of the photovoltaic power generation,
Figure BDA0003318964040000129
estimated value of photovoltaic power generation amount, P, for time t+T t+T The actual value of the photovoltaic power generation amount at time T + T, I.I 2 As a vector mean square error function, ns is the number of training samples, < +.>
Figure BDA00033189640400001210
And P t+T (i) The photovoltaic power generation amount estimated value and the actual value at time t+T are the ith sample.
And finally, the mean square error of the photovoltaic power generation amount is included in the mean square error of the moment estimation error and the image prediction to form a satellite near infrared image calculation total error, wherein the formula is as follows:
Figure BDA00033189640400001211
in the formula, loss sigma is the total error calculated by the near infrared image of the satellite.
Step 5: and training the full convolution neural network model based on the satellite near infrared image calculation total error by using the satellite near infrared image data and the photovoltaic power station power generation data which are acquired for at least one year time span.
Step 6: the values T=1, 2,3, … and K are calculated based on the satellite near infrared images of the rolling calculation time t+1 to t+K of the trained full convolution neural network model, and the satellite near infrared image calculation of K steps in advance is realized. As shown in fig. 2, the image estimation result when k=1 is shown, and as can be seen from the comparison between the estimation result in fig. 2 and the actual image, the method of the present invention can obtain a near-infrared satellite estimation image that approximates the actual image. Table 1 shows the mean square error of image prediction at k=3 (3 steps of 0.5 hours, 1 hour and 1.5 hours), and compared with continuous prediction, the method of the invention effectively reduces the mean square error of image prediction, and obtains better model performance. Further, as shown in fig. 3, the results of the photovoltaic power generation amount estimation performed in advance of 1.5 hours by the method of the present invention are shown, and the error results are shown in table 2. As shown by comparison results, the satellite near infrared image calculated by the method can also reduce the mean square error of photovoltaic power generation, and improve the model performance of real-time photovoltaic power generation prediction.
Table 1 comparison of results of image prediction mean square error
Figure BDA0003318964040000131
Table 2 comparison of results of mean square error of photovoltaic power generation
Figure BDA0003318964040000132
In summary, the photovoltaic power generation prediction-oriented real-time satellite near infrared image calculation method can better adapt to the photovoltaic power generation amount estimation prediction demand of the power system, and effectively utilizes satellite remote sensing data to improve the photovoltaic power generation amount prediction precision of the traditional daily short-term time scale and real-time scale, so as to provide data support for future photovoltaic power generation amount estimation for the new energy power system and the photovoltaic power station under the control of the new energy power system, and ensure the economy of power grid dispatching operation by improving the accuracy of the future photovoltaic power generation amount estimation.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the present invention.

Claims (6)

1. The photovoltaic power generation prediction-oriented real-time satellite near infrared image calculation method is characterized by comprising the following steps of:
step 1: collecting satellite near infrared image data within set geographic resolution and time resolution as an estimation object, and collecting photovoltaic power station power generation data with the same time resolution as guiding data for satellite near infrared image estimation;
step 2: constructing a full convolution neural network model aiming at satellite near infrared image calculation, wherein the full convolution neural network model is built by only using convolution layers and deconvolution layers, and the number of the convolution layers is more than or equal to 2; based on the current time T and the time T predicted in future, taking a d frame satellite near infrared image before the time t+T to be predicted as input and taking a predicted image of the time t+T as output, and determining a model structure and parameters thereof;
step 3: replacing a second convolution layer of the full convolution neural network model by using a partial differential convolution layer, performing moment estimation on convolution parameters of the partial differential convolution layer, solving a moment matrix of the convolution parameters, calculating a moment estimation error, and incorporating the moment estimation error into a mean square error of image prediction, wherein the method specifically comprises the following steps:
firstly, defining an operation rule of a partial differential convolution layer, enabling partial differential convolution operation to be equal to partial derivative operation, and adopting the following formula:
Figure FDA0004191287270000011
in the method, in the process of the invention,
Figure FDA0004191287270000012
and->
Figure FDA0004191287270000013
The hidden layer output of the mth channel of the first layer, the input of the ith channel of the first-1 layer and the convolution parameters of the mth channel of the first layer are respectively +.>
Figure FDA0004191287270000014
For the convolution operator ix is the abscissa partial derivative order, iy is the ordinate partial derivative order,/-, and>
Figure FDA0004191287270000015
layer 1The input of the i channels has a bias value of the ix+iy order,/th>
Figure FDA0004191287270000016
Is the ix-th order partial derivative of the image abscissa,>
Figure FDA0004191287270000017
an iy-th order partial derivative value of an ordinate of the image;
secondly, performing moment estimation on convolution parameters of the partial differential convolution layer, and solving a moment matrix of the convolution parameters, wherein a calculation formula is as follows:
Figure FDA0004191287270000018
in the method, in the process of the invention,
Figure FDA0004191287270000019
the matrix of the convolution parameter of the mth channel of the first layer, ix is the abscissa of the matrix, and the value is equal to the offset of the abscissa; iy is the ordinate of the matrix, and the value is equal to the ordinate partial derivative order; />
Figure FDA00041912872700000110
For the element values at the coordinates (ix, iy) in the moment matrix, k is the convolution parameter size, u and v are the convolution parameter abscissa and ordinate corresponding to the moment matrix coordinates (ix, iy), and +.>
Figure FDA00041912872700000111
Element values at coordinates (u, v) in the convolution parameters for the mth channel of the first layer;
then, matrix of moment based on convolution parameters
Figure FDA00041912872700000112
Calculating a moment estimation error, wherein the calculation formula is as follows:
Figure FDA0004191287270000021
Figure FDA0004191287270000022
where LossM is the moment estimation error, Δ ix,iy For matrix of moment estimation standard, real matrix of size k×k is satisfied
Figure FDA0004191287270000023
Δ ix,iy (i, j) is the coordinate in the moment estimation matrix (i, the value of the element at j) is, I.I F Is a matrix F norm function;
finally, calculating the mean square error of image prediction, and incorporating the moment estimation error into the mean square error of image prediction, wherein the formula is as follows:
Figure FDA0004191287270000024
where LossI is the mean square error of the image prediction,
Figure FDA0004191287270000025
and->
Figure FDA0004191287270000026
Satellite near-infrared images and predicted images of time t+T and wavelength wl respectively 2 Is a vector mean square error function;
step 4: estimating the photovoltaic power generation amount of a photovoltaic power station at time t+T based on a predicted image of time t+T, comparing the estimated photovoltaic power generation amount estimated value at time t+T with the photovoltaic power generation amount actual value at time t+T, calculating the mean square error of the photovoltaic power generation amount, and incorporating the mean square error of the photovoltaic power generation amount into the mean square error of the moment estimated error and the image prediction to form a satellite near infrared image estimation total error;
step 5: training a full convolution neural network model by collecting satellite near infrared image data of at least one year time span and photovoltaic power station power generation data and calculating total errors based on the satellite near infrared image;
step 6: the values T=1, 2,3, … and K, wherein K is a natural number above 1, satellite near infrared images with rolling calculation time t+1 to t+K are calculated based on the trained full convolution neural network model, satellite near infrared image calculation in K steps in advance is achieved, and the calculation result and the full convolution neural network model performance are checked based on the mean square error of image prediction and the mean square error of photovoltaic power generation.
2. The method for calculating the real-time satellite near-infrared image guided by photovoltaic power generation prediction according to claim 1, wherein the satellite near-infrared image data collected in the step 1 needs to cover the geographical range of the position of the photovoltaic power station and reflect the solar radiance measurement value of the photovoltaic power station, and the calculation formula of the solar radiance measurement value is as follows:
Figure FDA0004191287270000027
in the method, in the process of the invention,
Figure FDA0004191287270000028
the near infrared image of the satellite with time t and wavelength wl, w is the image width, h is the image height, < ->
Figure FDA0004191287270000029
Is a real number domain range, and (x, y) is an image coordinate; />
Figure FDA00041912872700000210
The pixel value at the (x, y) coordinate in the satellite near-infrared image at time t and wavelength wl represents the solar radiance measurement value at the (x, y) coordinate +.>
Figure FDA0004191287270000031
The unit is Wm –2 sr –1 (cm –1 ) –1 I.e. the intensity of radiation per centimeter of wavelength at a projected area per square meter.
3. The method for estimating the near infrared image of the satellite in real time based on the prediction and guidance of the photovoltaic power generation according to claim 1, wherein the step 2 is characterized in that a full convolution neural network model for estimating the near infrared image of the satellite is constructed, and the calculation formula is as follows:
Figure FDA0004191287270000032
Figure FDA0004191287270000033
in the method, in the process of the invention,
Figure FDA0004191287270000034
predicted image for time t+T to t+1, < >>
Figure FDA0004191287270000035
For the satellite near infrared image of time t to t-d+1+T, FCN (·) is a full convolution neural network model function, α (·) is an activation function, { W } is a convolution parameter set, { W } T The deconvolution parameter set, +.>
Figure FDA0004191287270000036
For the output of the mth channel of the first layer,/->
Figure FDA0004191287270000037
For the input of the ith channel of the layer 1, ch is the number of input channels, +.>
Figure FDA0004191287270000038
For convolution operator +.>
Figure FDA00041912872700000319
Convolution parameter for mth channel of layer I, < ->
Figure FDA0004191287270000039
For the deconvolution parameter of the mth channel of the first layer,/for the first layer>
Figure FDA00041912872700000310
Is the bias parameter of the mth channel of the first layer.
4. The method for estimating the real-time satellite near-infrared image guided by photovoltaic power generation prediction according to claim 1, wherein the step 4 estimates the photovoltaic power generation amount of the photovoltaic power station at time t+t through a constructed photovoltaic power generation power estimation model based on a multilayer perceptron, and the calculation formula of the photovoltaic power generation power estimation model is as follows:
Figure FDA00041912872700000311
Figure FDA00041912872700000312
in the method, in the process of the invention,
Figure FDA00041912872700000313
predicted image for time t+T, wavelength wl,>
Figure FDA00041912872700000314
for the photovoltaic power generation estimation of time t+T, MLP (·) is a multilayer perceptron function, { W MLP Sum { b } MLP Weight parameter set and bias parameter set of multi-layer perceptron }, +.>
Figure FDA00041912872700000315
For the hidden layer output of the first layer of the multi-layer sensor,/I>
Figure FDA00041912872700000316
Weight of the first layer of the multi-layer sensor, < ->
Figure FDA00041912872700000317
For the bias of the first layer of the multi-layer sensor, α (·) is the activation function.
5. The method for calculating the near infrared image of the real time satellite based on the prediction and guidance of the photovoltaic power generation according to claim 1, wherein the mean square error of the photovoltaic power generation is calculated in the step 4, and the calculation formula is as follows:
Figure FDA00041912872700000318
where LossP is the mean square error of the photovoltaic power generation,
Figure FDA0004191287270000041
estimated value of photovoltaic power generation amount, P, for time t+T t+T The actual value of the photovoltaic power generation amount at time T + T, I.I 2 As a vector mean square error function, ns is the number of training samples, < +.>
Figure FDA0004191287270000042
And P t+T (i) The photovoltaic power generation amount estimated value and the actual value at time t+T are the ith sample.
6. The method for estimating a real-time satellite near-infrared image guided by photovoltaic power generation prediction according to claim 1, wherein the total error estimated by the satellite near-infrared image formed in the step 4 is calculated by the following formula:
Figure FDA0004191287270000043
wherein, loss sigma is the total error calculated by the satellite near infrared image; lossI is the mean square error of image prediction, lossM is the moment estimation error, and LossP is the mean square error of photovoltaic power generation;
Figure FDA0004191287270000044
and->
Figure FDA0004191287270000045
Satellite near-infrared images and predicted images of time t+T and wavelength wl respectively 2 Is a vector mean square error function; />
Figure FDA0004191287270000046
For the element value at coordinates (ix, iy) in the moment matrix, delta ix,iy For the matrix of the moment estimation standard, a real matrix of the size k×k is satisfied>
Figure FDA0004191287270000047
||·|| F Is a matrix F norm function;
Figure FDA0004191287270000048
estimated value of photovoltaic power generation amount, P, for time t+T t+T The actual value of photovoltaic power generation at time t+T.
CN202111243878.4A 2021-10-25 2021-10-25 Real-time satellite near infrared image calculation method based on photovoltaic power generation prediction guiding Active CN114037901B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111243878.4A CN114037901B (en) 2021-10-25 2021-10-25 Real-time satellite near infrared image calculation method based on photovoltaic power generation prediction guiding

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111243878.4A CN114037901B (en) 2021-10-25 2021-10-25 Real-time satellite near infrared image calculation method based on photovoltaic power generation prediction guiding

Publications (2)

Publication Number Publication Date
CN114037901A CN114037901A (en) 2022-02-11
CN114037901B true CN114037901B (en) 2023-06-20

Family

ID=80135353

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111243878.4A Active CN114037901B (en) 2021-10-25 2021-10-25 Real-time satellite near infrared image calculation method based on photovoltaic power generation prediction guiding

Country Status (1)

Country Link
CN (1) CN114037901B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116706903B (en) * 2023-08-07 2024-01-26 深圳航天科创泛在电气有限公司 Photovoltaic power generation amount prediction method, device, equipment and medium

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111210095A (en) * 2020-03-12 2020-05-29 深圳前海微众银行股份有限公司 Power generation amount prediction method, device, equipment and computer readable storage medium

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2017200360A (en) * 2016-04-28 2017-11-02 清水建設株式会社 Photovoltaic power generation amount prediction system, photovoltaic power generation amount prediction method, and program
FR3078802B1 (en) * 2018-03-07 2020-10-30 Electricite De France CONVOLUTIONAL NEURON NETWORK FOR THE ESTIMATION OF A SOLAR ENERGY PRODUCTION INDICATOR
US11693378B2 (en) * 2019-03-01 2023-07-04 Alliance For Sustainable Energy, Llc Image-based solar estimates
CN110070226B (en) * 2019-04-24 2020-06-16 河海大学 Photovoltaic power prediction method and system based on convolutional neural network and meta-learning
CN110503153B (en) * 2019-08-26 2021-06-11 哈尔滨工程大学 Photovoltaic system fault diagnosis method based on differential evolution algorithm and support vector machine
CN110633862A (en) * 2019-09-20 2019-12-31 旻投(上海)数据科技有限公司 Optical power prediction algorithm based on satellite cloud picture
CN110909926A (en) * 2019-11-18 2020-03-24 中国计量大学 TCN-LSTM-based solar photovoltaic power generation prediction method
CN111401452B (en) * 2020-03-17 2022-04-26 北京大学 Image classification method of equal-variation convolution network model based on partial differential operator
CN113298303B (en) * 2021-05-19 2021-12-28 河海大学 Photovoltaic power prediction method for meteorological satellite cloud picture dynamic attention domain

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111210095A (en) * 2020-03-12 2020-05-29 深圳前海微众银行股份有限公司 Power generation amount prediction method, device, equipment and computer readable storage medium

Also Published As

Publication number Publication date
CN114037901A (en) 2022-02-11

Similar Documents

Publication Publication Date Title
CN109934423B (en) Photovoltaic power station power prediction method and system based on grid-connected inverter operation data
CN113298303B (en) Photovoltaic power prediction method for meteorological satellite cloud picture dynamic attention domain
CN114265836A (en) All-weather assimilation method and device of satellite microwave hygrothermograph based on cloud area temperature and humidity profile inversion
Kang et al. Development of an observation processing package for data assimilation in KIAPS
CN105974495B (en) It will be averaged the method for cloud amount using classification fitting process anticipation target area future
CN114707688A (en) Photovoltaic power ultra-short-term prediction method based on satellite cloud chart and space-time neural network
CN116865236A (en) Medium-and-long-term power generation capacity prediction method and system based on new energy power generation
CN114881399B (en) Photovoltaic power generation potential and economical efficiency assessment method based on GF7 remote sensing image
CN112559958B (en) Method for inverting total radiation and direct radiation of earth surface and sun based on wind cloud No. 4 satellite
CN111985678A (en) Photovoltaic power short-term prediction method
CN114037901B (en) Real-time satellite near infrared image calculation method based on photovoltaic power generation prediction guiding
Rincón et al. Bias correction of global irradiance modelled with weather and research forecasting model over Paraguay
Shi et al. First estimation of high-resolution solar photovoltaic resource maps over China with Fengyun-4A satellite and machine learning
CN112801413A (en) Photovoltaic power station generated power prediction method and device
Espinosa-Gavira et al. Cloud motion estimation from small-scale irradiance sensor networks: General analysis and proposal of a new method
CN110059972A (en) Day solar radiation stock assessment method based on functional deepness belief network
Liu et al. Method of identifying the lengths of equivalent clear-sky periods in the time series of DNI measurements based on generalized atmospheric turbidity
Ronzio et al. A survey on different radiative and cloud schemes for the solar radiation modeling
CN112561181A (en) Photovoltaic power generation prediction system based on Unet network and foundation cloud picture
Walstad et al. Data assimilation for coastal observing systems
CN115936387B (en) Photovoltaic power station solar energy resource assessment method based on photometry data
CN116260141B (en) Reconstruction method and system for power of photovoltaic power station and reconstruction terminal
CN116843193B (en) Manual heat flux estimation method and system based on flux observation data
Mengen et al. Solar Irradiance Measuring sites in Bangladesh
CN107608008A (en) A kind of detection method of the clear sky period based on broad sense atmospheric turbidity

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant