CN113159466B - Short-time photovoltaic power generation prediction system and method - Google Patents

Short-time photovoltaic power generation prediction system and method Download PDF

Info

Publication number
CN113159466B
CN113159466B CN202110587166.8A CN202110587166A CN113159466B CN 113159466 B CN113159466 B CN 113159466B CN 202110587166 A CN202110587166 A CN 202110587166A CN 113159466 B CN113159466 B CN 113159466B
Authority
CN
China
Prior art keywords
image
layer
data
power generation
cloud
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
CN202110587166.8A
Other languages
Chinese (zh)
Other versions
CN113159466A (en
Inventor
卢小丁
江思伟
刘莉
袁宏亮
林栋�
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.)
Wotai Energy Co ltd
Original Assignee
Wotai Energy Co ltd
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 Wotai Energy Co ltd filed Critical Wotai Energy Co ltd
Priority to CN202110587166.8A priority Critical patent/CN113159466B/en
Publication of CN113159466A publication Critical patent/CN113159466A/en
Application granted granted Critical
Publication of CN113159466B publication Critical patent/CN113159466B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/04—Context-preserving transformations, e.g. by using an importance map
    • G06T3/047—Fisheye or wide-angle transformations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration using local operators
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/80—Geometric correction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Artificial Intelligence (AREA)
  • Economics (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • General Business, Economics & Management (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Public Health (AREA)
  • Primary Health Care (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Water Supply & Treatment (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

A short-time photovoltaic power generation prediction system and method can more accurately track cloud layer movement by adopting an image processing technology; the photovoltaic power generation efficiency under multi-time and weather conditions is distinguished, and stronger adaptability is achieved; the change of the illumination intensity can be accurately predicted, so that a user can make optimization guidance of energy storage battery charging/grid connection planning on a photovoltaic system in advance, and the impact on a power grid is reduced. The system and the method have the advantages of high prediction precision, wide application range and strong interpretability.

Description

Short-time photovoltaic power generation prediction system and method
Technical Field
The invention relates to the field of power management, in particular to a short-time photovoltaic power generation power prediction system and method.
Background
According to traditional photovoltaic power generation power prediction, a ground anemometer is adopted to measure wind speed, and the moving speed and the wind direction of a high-altitude cloud layer cannot be accurately described. In the method for predicting by adopting the image, the image is mostly used for training and predicting, information in the image is not further extracted, and the influence of factors such as the temperature difference of the photovoltaic panel on the power generation efficiency is ignored. The method based on the number of cloud clusters and the area of the cloud clusters occupying the whole sky cannot distinguish cloud layers with different thicknesses, and cannot accurately predict the impact of the severe change of illumination intensity caused by cloud layer shielding on the whole power grid in the photovoltaic grid connection process.
Disclosure of Invention
Aiming at the problems, the invention provides a system and a method for predicting the short-time photovoltaic power generation power, and the adopted image processing technology can more accurately track the movement of cloud layers; the photovoltaic power generation efficiency under multi-time and weather conditions is distinguished, and stronger adaptability is achieved; the change of illumination intensity can be accurately predicted, and the impact is reduced by taking a response measure in advance; and (4) performing optimization guidance of energy storage battery charging/grid connection planning on the photovoltaic system.
A short-time photovoltaic power generation power prediction system comprises a data acquisition module, a local data processing module, a cloud data processing and storing module and an energy storage battery, wherein the local data processing module comprises a cloud layer moving speed estimation module, a sun position calculation module and a photovoltaic power generation efficiency calculation module;
the data acquisition module comprises a camera, a photovoltaic panel temperature sensor and an environment temperature sensor, the camera shoots a sky image, and meanwhile, the temperature sensors acquire and acquire the temperature of the photovoltaic panel and the environment;
the cloud layer moving speed estimation module and the sun position calculation module process and calculate the acquired sky image to obtain the cloud layer moving speed and the sun position, and the photovoltaic power generation efficiency calculation module inputs the calculated data, the real-time sky image and the temperature into a trained neural network of a multi-input data type to finally obtain the photovoltaic power generation efficiency;
and after the local data processing module processes the data, the data is stored in the cloud, and the energy storage battery is managed according to the calculation result.
A short-time photovoltaic power generation prediction method comprises the following steps:
step 1, acquiring images and temperature information by a camera and a temperature sensor, and preprocessing the acquired data; the temperature sensor collects the temperature of the photovoltaic panel and the environment, the camera shoots the sky image, and the preprocessing is to zoom the image through the characteristic pyramid after the image is subjected to distortion correction.
Step 2, estimating the motion of the cloud layer based on computer vision; carrying out feature point detection on the preprocessed image, and matching the feature point detection with the image at the previous moment; rejecting mismatching feature points, calculating the number of moving pixels of the corresponding feature points projected in the same picture as the moving distance of the cloud layer, and setting the angle formed by the connecting line of the two feature points as the moving direction of the cloud layer; after all the matched feature points are subjected to the operation, an average value is obtained;
step 3, carrying out self-adaptive adjustment on the sampling frequency of the cloud layer image; setting the basic sampling interval time as T, controlling the moving distance of cloud layers of two adjacent images to be l/10 image side length, and setting a target difference value e as l/10-d; the difference e is 0 when a PID control strategy is adopted;
step 4, calculating to obtain the position of the sun through local longitude and latitude and time information, and estimating the influence of cloud movement on the energy from solar radiation to the photovoltaic panel according to the calculated cloud movement speed and direction;
and 5, processing the image and the numerical data by adopting a neural network form of a multi-input data type for the collected data comprising the image data and the numerical data to obtain the final photovoltaic power generation efficiency, wherein the photovoltaic power generation efficiency is instantaneous power generation power/rated installed power.
The invention achieves the following beneficial effects: the invention provides a short-time photovoltaic power generation prediction system and a short-time photovoltaic power generation prediction method, which can more accurately track the movement of a cloud layer through an improved image processing technology, can better predict the shielding condition of the movement of the cloud layer on sunlight, and further obtain a better power generation prediction result; the photovoltaic power generation efficiency under the conditions of multiple times and weather is distinguished, the adaptability is stronger, the system can be directly established and directly put into use for prediction without being adjusted according to the actual position of the system when in use, and the operation is more convenient and easy; by analyzing the motion and the irradiation conditions of the cloud layer and the sun, the change of the illumination intensity can be accurately predicted, and the impact is reduced by taking a response measure in advance; and (4) performing optimization guidance of energy storage battery charging/grid connection planning on the photovoltaic system. The system and the method have the advantages of high prediction precision, wide application range, simple operation and strong interpretability.
Drawings
Fig. 1 is a schematic structural diagram of a photovoltaic power generation power prediction system in an embodiment of the present invention.
Fig. 2 is a schematic diagram of a scale space constructed in cloud layer motion estimation according to an embodiment of the present invention.
Fig. 3 is a schematic diagram illustrating feature point positioning in cloud motion estimation according to an embodiment of the present invention.
Fig. 4 is a schematic diagram illustrating distribution of principal directions of feature points in cloud motion estimation according to an embodiment of the present invention.
Fig. 5 is a schematic diagram illustrating cloud layer feature point matching in cloud layer motion estimation according to an embodiment of the present invention.
Fig. 6 is a diagram illustrating adjustment of sampling frequency of a cloud image by PID control according to an embodiment of the present invention.
FIG. 7 is a schematic diagram of the input and output of a BP neural network neuron in an embodiment of the present invention.
FIG. 8 is a schematic diagram of convolutional neural network pooling in an embodiment of the present invention.
FIG. 9 is an overall architecture diagram of the convolutional neural network layer and the BP neural network layer in an embodiment of the present invention.
Fig. 10 is a schematic diagram illustrating the principle of inserting numerical data into image data according to an embodiment of the present invention.
Fig. 11 is a diagram of a neural network architecture after numerical data is inserted into image data according to an embodiment of the present invention.
Fig. 12 is a flowchart illustrating a method for predicting photovoltaic power generation power according to an embodiment of the present invention.
Fig. 13 is a second flowchart of a method for predicting photovoltaic power generation power according to an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the drawings in the specification.
A short-time photovoltaic power generation power prediction system is shown in figure 1 and comprises a data acquisition and collection module, a local data processing module, a cloud data processing and storing module and an energy storage battery, wherein the local data processing module comprises a cloud layer moving speed estimation module, a solar position calculation module and a photovoltaic power generation efficiency calculation module.
A data acquisition module: including image acquisition and temperature acquisition. The image acquisition comprises panoramic cameras such as fisheye cameras or a plurality of groups of cameras, for example, the fisheye cameras are used for shooting sky pictures by adopting fixed fisheye cameras, and after distortion correction is carried out, the pictures are zoomed by the characteristic pyramid, so that the calculated amount is reduced. The temperature acquisition is realized through a temperature sensor, and the photovoltaic panel and the ambient temperature are acquired.
The cloud layer moving speed estimation module performs feature point detection (not limited to Surf, Sift, ORB and the like) on the processed image, and matches the processed image with the image at the previous moment. And eliminating mismatching feature points, calculating the number of moving pixels of the corresponding feature points projected in the same picture as the moving distance of the cloud layer, and setting the angle formed by the connecting lines of the two feature points as the moving direction of the cloud layer. After the above operation is performed on all the matched feature points, an average value is obtained, and a calculation formula is as follows:
Figure BDA0003088151520000051
wherein (x)i,yi) Is the pixel coordinate at the current time instant,
Figure BDA0003088151520000052
the pixel coordinates at the previous time.
The sun position calculation module calculates the sun position according to the collected sky image.
In the horizon coordinate system, the position of the sun can be solved by the elevation angle α and the azimuth angle θ as follows:
Figure BDA0003088151520000053
Figure BDA0003088151520000054
wherein the content of the first and second substances,delta is solar declination angle;
Figure BDA0003088151520000055
the local latitude is; and omega is a time angle.
After the position of the sun is determined, the solar time angle is obtained from ω 15 × (ST-12) from the two image capturing times. The position of the sun on the projection frame at the next moment is calculated by the Gaussian-gram projection. And estimating the influence of the cloud layer motion on the energy from the solar radiation to the photovoltaic panel according to the calculated cloud layer motion speed and direction. Specifically, the sun projection movement at the next moment is equal to the cloud layer movement in size and opposite in direction, and the illumination intensity of the position is the estimated solar radiation intensity.
In the photovoltaic power generation efficiency calculation module, defining the photovoltaic power generation efficiency as instantaneous power generation power/rated installed power, inputting real-time image data and other acquired data into a trained neural network with multiple input data types, finally obtaining the photovoltaic power generation efficiency, and estimating the photovoltaic power generation amount of the actual installed equipment.
A short-time photovoltaic power generation power prediction method refers to flow charts of two modes of a reference graph 12 and a reference graph 13, only the last step of the two methods has difference, the photovoltaic power generation power prediction function can be achieved, and common parts such as cloud layer motion estimation, sampling, operation frequency calculation, sun position calculation and the like exist, so that the input variables are obtained. The generated power solving part of the two algorithms has different ideas and can be replaced mutually.
Specifically, the prediction method comprises the following steps:
step 1, acquiring images and temperature information by a camera and a temperature sensor, and preprocessing the acquired data; the temperature sensor collects the temperature of the photovoltaic panel and the environment, the camera shoots the sky image, and the preprocessing comprises the steps of carrying out distortion correction on the image and then zooming the image through the characteristic pyramid.
And 2, cloud layer motion estimation based on computer vision.
And (3) carrying out feature point detection (not limited to Surf, Sift, ORB and the like) on the processed image, and matching the processed image with the image at the previous moment. And eliminating mismatching feature points, calculating the number of moving pixels of the corresponding feature points projected in the same picture as the moving distance of the cloud layer, and setting the angle formed by the connecting lines of the two feature points as the moving direction of the cloud layer. And after the above operation is carried out on all the matched feature points, the average value is obtained.
Taking surf algorithm as an example, the brief description of cloud layer motion estimation algorithm comprises a-g processes with 7 steps, wherein feature point detection (process c) is a necessary operation for feature matching.
a. After the obtained cloud layer image is subjected to Gaussian filtering, a sea plug Matrix (Hessian Matrix) is constructed, a determinant of the sea plug Matrix is calculated to judge whether the sea plug Matrix is a local feature point of the image, and a box-shaped fuzzy filter is adopted to calculate a Gaussian approximate value for the determinant.
Figure BDA0003088151520000071
b. Constructing a scale space: the scale space of Surf, as shown in fig. 2, keeps the image size unchanged during its sampling, and a scale pyramid is constructed by changing the size of the box filter.
c. Positioning the characteristic points: as shown in fig. 3, each pixel point processed by the Hessian matrix is compared with 26 points in the neighborhood of the two-dimensional image space and the scale space, a key point is preliminarily located, and a final stable feature point is screened out by filtering out a key point with weak energy and an incorrectly located key point.
d. And (3) distributing the main directions of the characteristic points: haar wavelet characteristics are used. As shown in fig. 4, that is, in the circular neighborhood of the feature point, the sum of the horizontal and vertical haar wavelet features of all points in the 60-degree sector is counted, then after the sector is rotated at intervals of 0.2 radian and the haar wavelet feature values in the region are counted again, the direction of the sector with the largest value is finally taken as the main direction of the feature point.
e. Generating a characteristic point descriptor: a 4 × 4 rectangular region block is taken around the feature point, but the direction of the rectangular region taken is along the main direction of the feature point. Each subregion counts haar wavelet features of 25 pixels in both the horizontal and vertical directions, where both the horizontal and vertical directions are relative to the principal direction. The haar wavelet features are 4 directions of the sum of the horizontal direction value, the vertical direction value, the horizontal direction absolute value and the vertical direction absolute value.
f. Matching the characteristic points: as shown in fig. 5, the matching degree is determined by calculating the euclidean distance between two feature points, and the shorter the euclidean distance, the better the matching degree of the two feature points is represented. And judging the Hessian matrix trace, if the signs of the matrix traces of the two characteristic points are the same, representing that the two characteristics have contrast changes in the same direction, if the signs of the matrix traces are different, indicating that the contrast change directions of the two characteristic points are opposite, and directly excluding the characteristic points even if the Euclidean distance is 0.
The Euclidean distance is calculated as:
the key point descriptor in the image at the previous moment:
Ri＝(ri1,ri2,…,ri128)
the current calculation uses the keypoint descriptors in the image:
Si＝(si1,si2,…,si128)
any two descriptor similarity measures (euclidean distance):
Figure BDA0003088151520000081
g. after eliminating the mismatched points, n groups of corresponding points (x) are obtained11,y11)，(x21,y21)， (x12,y12)，(x22,y22)...(x1n,y1n)(x2n,y2n) And calculating the moving distance d and the direction angle alpha of the cloud layer on the image.
Figure BDA0003088151520000082
Figure BDA0003088151520000083
Reference cloud layer moving speed v: and v is d/T.
In step 3, too high a calculation frequency increases the system load, and further increases the hardware cost and power consumption. In this way, the sampling frequency of the cloud layer image is adaptively adjusted. Assuming that the basic sampling interval time is T, (the moving distance of the cloud layers of two adjacent images is controlled to be l/10 image side length (rough value)), and the target difference e is l/10-d, the difference e is 0 by adopting a control strategy such as PID control and the like referring to fig. 6.
Step 4, in order to prevent cloud layer shielding, the position of the sun on the image cannot be identified by the convolutional network, the position of the sun is obtained through extra calculation according to local longitude and latitude and time information, and in the horizon coordinate system, the position of the sun can be obtained by solving an altitude angle alpha and an azimuth angle theta as follows:
Figure BDA0003088151520000091
Figure BDA0003088151520000092
wherein, delta is the declination angle of the sun;
Figure BDA0003088151520000093
the local latitude is; and omega is a time angle.
And after the position of the sun is determined, calculating the position change of the sun on the projection frame at the next moment according to the shooting time of the two images. And estimating the influence of the cloud layer motion on the energy from the solar radiation to the photovoltaic panel according to the calculated cloud layer motion speed and direction. Here, the estimation influence is only calculated in an initial way that the next moment will not be occluded by the cloud layer. And (3) establishing a flag bit covered _ flag, wherein if the cloud layer can block the sun at the next moment according to the current moment movement speed and direction, the flag bit value is 1, and if not, the flag bit value is 0. And finally, inputting the mark position and the position coordinate of the current sun as sun position information into a neural network for further calculation.
And step 5, the acquired data not only comprises image data, but also comprises numerical data. The embodiment of the invention adopts two modes, and adopts a neural network form of multiple input data types to process the image and numerical data to obtain the photovoltaic power generation efficiency, wherein the photovoltaic power generation efficiency is defined as instantaneous power generation power/rated installed power.
The first method is as follows: the method comprises the steps that a convolution neural network is used for inputting image data, a common BP neural network is used for inputting numerical data, after primary processing is conducted, all connection layers are combined together, and finally the value of the photovoltaic power generation efficiency coefficient is output.
The BP neural network comprises an input layer, a hidden layer and an output layer. The input characteristics include sun position information, photovoltaic panel and ambient temperature, cloud layer movement speed and direction, time of day, and the like. The neuron structure is shown in fig. 7, wherein the inputs are x1, x2 and x3, the weights are w1, w2 and w3, the bias term b is added after the multiplication, and the input is substituted into an activation function to be output. The activation function adopts a sigmoid function, and the expression of the activation function is as follows:
Figure BDA0003088151520000101
the convolutional neural network comprises a convolutional layer, a pooling layer and a full-link layer. And (4) performing convolution on the three RGB channels of the original image, and pooling by adopting a maxporoling mode. Pooling is actually a scaling process. And selecting the pixel with the maximum pixel value in the pixel block range to represent the whole pixel block. Padding refers to a way of filling the head and the tail of an image in the pooling process so as to ensure that the image is an integer. Padding 0 around the original image, so that each pixel point can be scanned by the window. The pooling window moving step length stride is 1.
The keras layers conditioner function ties the two together. After the two models are combined, the two models are processed through a multilayer network, partial node data are randomly screened out through dropout, and finally a final output value is obtained through a sigmoid function. An Adam optimizer is adopted in the training process, and the loss function is MAE.
Referring to fig. 7-9, the fully-connected layer in fig. 9 is a shared layer, the connections after the convolution network pooling and the fully-connected of the BP network share hidden layer neurons, the shared layer in this layer is a literal meaning that the neurons share in such layer, and the fully-connected layer represents a connection mode, that is, each node of the layer is connected with all nodes of the previous layer, so as to integrate the extracted features.
The second method comprises the following steps: the method is characterized in that numerical data are normalized to be between (0, 255) and are forcedly converted into integers.
Figure BDA0003088151520000111
Simultaneously inserting the integer numerical data into three channels of RGB of the image data as a plurality of rows/columns, as shown in FIG. 10, the inserting position is determined according to the size of the pooled kernel; for example, a pooling kernel size of 2 x2, the feature values are filled in the top left corner within the 2 x2 pixel box, as shown in fig. 10. For the case that the data amount is less than a whole row/column, 0 value filling is adopted, and the processed image is input into a convolutional neural network to predict the photovoltaic power generation efficiency, as shown in fig. 11.
In the embodiment of the invention, the BP neural network and the convolutional neural network both belong to the classical neural network, and can be expanded on the basis, such as adopting the cyclic neural network and generating a countermeasure network to replace the current neural network. The adjustment of parameters such as network structure can optimize the overall performance.
In the embodiment of the invention, the collected data can be processed and operated in a cloud computing mode.
In the embodiment of the present invention, the preprocessing in step 1 of the prediction method may further include a noise elimination step. And removing the outliers which are in error matching, and keeping the correct matching points. Common methods are KDTREE, BBF, Randac, GTM, etc.
In the embodiment of the present invention, the calculation cycle may be updated by using fuzzy control, neural network, and the like.
The above description is only a preferred embodiment of the present invention, and the scope of the present invention is not limited to the above embodiment, but equivalent modifications or changes made by those skilled in the art according to the present disclosure should be included in the scope of the present invention as set forth in the appended claims.

Claims (9)

1. A short-time photovoltaic power generation power prediction system is characterized in that:
the system comprises a data acquisition module, a local data processing module, a cloud data processing and storing module and an energy storage battery, wherein the local data processing module comprises a cloud layer moving speed estimation module, a solar position calculation module and a photovoltaic power generation efficiency calculation module;
the data acquisition module comprises a camera, a photovoltaic panel temperature sensor and an environment temperature sensor, the camera shoots a sky image, and meanwhile, the temperature sensors acquire the temperature of the photovoltaic panel and the environment;
the cloud layer moving speed estimation module and the sun position calculation module process and calculate the cloud layer moving speed and the sun position according to the collected sky image, and the photovoltaic power generation efficiency calculation module inputs the calculated data, the real-time sky image and the temperature into a trained neural network of a multi-input data type to finally obtain the photovoltaic power generation efficiency; the method comprises the following steps that image data are input through a convolutional neural network, numerical data are input through a common BP neural network, after primary processing is conducted, the numerical data and the numerical data are combined together through a full-connection layer, and finally a photovoltaic power generation efficiency coefficient value is output; the convolutional neural network comprises a convolutional layer, a pooling layer and a full-connection layer; convolving the RGB three channels of the original image, and pooling by adopting a maxporoling mode; pooling is a scaling process, and a pixel with the largest pixel value in a pixel block range is selected to represent the whole pixel block; padding refers to a filling mode for the head and the tail of an image in a pooling process so as to ensure that the image is an integer; filling 0 around the original image so that each pixel point is scanned by a window; the moving step length stride of the pooling window is 1; the BP neural network comprises an input layer, a hidden layer, an output layer, an inputThe characteristics comprise sun position information, photovoltaic panel and environment temperature, cloud layer moving speed and direction and time; the neuron inputs are x1, x2 and x3, and after the neuron inputs are multiplied by weights w1, w2 and w3, a bias term b is added, and the neuron inputs are substituted into an activation function to be output; the activation function adopts a sigmoid function, and the expression of the activation function is as follows:
Figure FDA0003516319460000011
two neural networks are connected together by a keras. Combining two neural networks, then processing the neural networks through a multilayer network, randomly screening partial node data by using dropout in the process, and finally obtaining a final output value through a sigmoid function; an Adam optimizer is adopted in the training process, and a loss function is MAE;
and after the local data processing module processes the data, the data is stored in the cloud, and the energy storage battery is managed according to the calculation result.
2. The system according to claim 1, wherein the system comprises: the camera is a panoramic camera or a plurality of groups of cameras comprising fisheye cameras.
3. The system according to claim 1, wherein the system comprises: the cloud layer moving speed estimation module detects the characteristic points of the image and matches the image at the previous moment; rejecting mismatching feature points, calculating the number of moving pixels of the corresponding feature points projected in the same picture as the moving distance of the cloud layer, and setting the angle formed by the connecting line of the two feature points as the moving direction of the cloud layer; and after the above operation is carried out on all the matched feature points, the average value is obtained.
4. The system according to claim 1, wherein the system comprises: after the sun position calculation module determines the position of the sun, the position change of the sun on the projection frame at the next moment is calculated according to the shooting time of the two images; and estimating the influence of the cloud layer motion on the energy from the solar radiation to the photovoltaic panel according to the calculated cloud layer motion speed and direction.
5. A short-time photovoltaic power generation power prediction method is characterized by comprising the following steps: the method comprises the following steps:
step 1, acquiring images and temperature information by a camera and a temperature sensor, and preprocessing the acquired data; the temperature sensor collects the temperature of the photovoltaic panel and the environment, the camera shoots a sky image, and the preprocessing comprises the steps of carrying out distortion correction on the image and then zooming the image through a characteristic pyramid;
step 2, estimating the motion of the cloud layer based on computer vision; carrying out feature point detection on the preprocessed image, and matching the feature point detection with the image at the previous moment; rejecting mismatching feature points, calculating the number of moving pixels of the corresponding feature points projected in the same picture as the moving distance of the cloud layer, and setting the angle formed by the connecting line of the two feature points as the moving direction of the cloud layer; after all the matched feature points are subjected to the operation, an average value is obtained;
step 3, carrying out self-adaptive adjustment on the sampling frequency of the cloud layer image; setting the basic sampling interval time as T, controlling the moving distance of the cloud layers of two adjacent images to be 1/10 image side length, and setting a target difference value e to be 1/10-d; the difference e is 0 when a PID control strategy is adopted;
step 4, calculating to obtain the position of the sun through local longitude and latitude and time information, and estimating the influence of cloud movement on the energy from solar radiation to the photovoltaic panel according to the calculated cloud movement speed and direction;
step 5, processing the image and the numerical data by adopting a neural network form of a multi-input data type for the collected data comprising the image data and the numerical data to obtain final photovoltaic power generation efficiency, wherein the photovoltaic power generation efficiency is instantaneous power generation power/rated installed power; in the step 5, the input processing of the image data is performed through a convolution neural network, the input of numerical data is performed through a common BP neural network, after the initial processing, the numerical data and the numerical data are combined together through a full-link layer, and finally the photovoltaic power generation efficiency is outputA coefficient value; the convolutional neural network comprises a convolutional layer, a pooling layer and a full-connection layer; convolving the RGB three channels of the original image, and pooling by adopting a maxporoling mode; pooling is a scaling process, and a pixel with the largest pixel value in a pixel block range is selected to represent the whole pixel block; padding refers to a filling mode for the head and the tail of an image in a pooling process so as to ensure that the image is an integer; filling 0 around the original image so that each pixel point is scanned by a window; the moving step length stride of the pooling window is 1; the BP neural network comprises an input layer, a hidden layer and an output layer, and the input characteristics comprise sun position information, photovoltaic panel and environment temperature, cloud layer moving speed and direction and time; the neuron inputs are x1, x2 and x3, and after the neuron inputs are multiplied by weights w1, w2 and w3, a bias term b is added, and the neuron inputs are substituted into an activation function to be output; the activation function adopts a sigmoid function, and the expression of the activation function is as follows:
Figure FDA0003516319460000021
two neural networks are connected together by a keras. Combining two neural networks, then processing the neural networks through a multilayer network, randomly screening partial node data by using dropout in the process, and finally obtaining a final output value through a sigmoid function; an Adam optimizer is adopted in the training process, and the loss function is MAE.
6. The method according to claim 5, wherein the method comprises: in step 2, cloud layer motion estimation is carried out based on a surf algorithm, and the method comprises the following 6 characteristic matching processes of 2a-2 g:
step 2a, after the obtained cloud layer image is subjected to Gaussian filtering, constructing a Hessian Matrix, calculating a determinant thereof to judge whether the obtained cloud layer image is a local feature point of the image, and calculating a Gaussian approximate value by adopting a box-shaped fuzzy filter:
Figure FDA0003516319460000031
step 2b, constructing a scale space: in the scale space of the Surf, the size of an image is kept unchanged in the sampling process, and a scale pyramid is constructed by changing the size of a box-shaped filter;
step 2c, positioning the characteristic points: comparing each pixel point processed by the Hessian matrix with 26 points in the neighborhood of a two-dimensional image space and a scale space, preliminarily positioning a key point, and screening out a final stable characteristic point by filtering out the key point with weaker energy and the key point with wrong positioning;
step 2d, distribution of the main directions of the feature points: adopting haar wavelet characteristics, namely counting the sum of horizontal haar wavelet characteristics and vertical haar wavelet characteristics of all points in a sector of 60 degrees in a circular neighborhood of characteristic points, then rotating the sector at intervals of 0.2 radian, counting haar wavelet characteristic values in the sector again, and finally taking the direction of the sector with the maximum value as the main direction of the characteristic points;
step 2e, generating a feature point descriptor: taking a 4-by-4 rectangular area block around the feature point, and obtaining the direction of the rectangular area along the main direction of the feature point; counting haar wavelet characteristics of 25 pixels in the horizontal direction and the vertical direction of each subregion, wherein the horizontal direction and the vertical direction are relative to the main direction; the haar wavelet features are 4 directions of the sum of the horizontal direction value, the vertical direction value, the horizontal direction absolute value and the vertical direction absolute value;
step 2f, feature point matching: determining the matching degree by calculating the Euclidean distance between two feature points, wherein the shorter the Euclidean distance is, the better the matching degree of the two feature points is represented; judging the Hessian matrix trace, if the signs of the matrix traces of the two characteristic points are the same, representing that the two characteristics have contrast changes in the same direction, if the signs are different, indicating that the contrast change directions of the two characteristic points are opposite, and directly eliminating the characteristic points even if the Euclidean distance is 0;
in step 2f, the calculation of the euclidean distance specifically includes:
the key point descriptor in the image at the previous moment:
Ri＝(ri1，ri2，...，ri128)
the current calculation uses the keypoint descriptors in the image:
Si＝(si1，si2，...，si128)
any two descriptor similarity measures, euclidean distance:
Figure FDA0003516319460000041
step 2g, n groups of corresponding points (x) are obtained after the mismatching points are removed11，y11)，(x21，y21)，(x12，y12)，(x22，y22)...(x1n，y1n)，(x2n，y2n) Calculating the moving distance d and the direction angle alpha of the cloud layer on the image:
Figure FDA0003516319460000042
Figure FDA0003516319460000043
calculating the moving speed v of the cloud layer:
v＝d/T。
7. the method according to claim 5, wherein the method comprises: in step 4, in the horizon coordinate system, the position of the sun is obtained by solving the altitude angle α and the azimuth angle θ, as follows:
Figure FDA0003516319460000044
Figure FDA0003516319460000045
wherein, delta is the declination angle of the sun;
Figure FDA0003516319460000046
the local latitude is; omega is a time angle;
and after the position of the sun is determined, calculating the position change of the sun on the projection frame at the next moment according to the shooting time of the two images.
8. The method according to claim 5, wherein the method comprises: in the step 4, the influence is estimated in a process of creating a flag bit covered _ flag, if the cloud layer can block the sun at the next moment according to the current moment motion speed and direction, the flag bit value is 1, and if not, the flag bit value is 0; and finally, inputting the mark position and the position coordinate of the current sun as sun position information into the neural network in the step 5 for further calculation.
9. The method according to claim 5, wherein the method comprises: in the step 5, normalizing the numerical data to be between (0, 255), and forcibly converting the numerical data into an integer;
Figure FDA0003516319460000047
simultaneously inserting the image data into RGB three channels of the image data as a plurality of rows/columns, wherein the inserting position is determined according to the size of the pooled kernel; and (3) for the condition that the data volume is less than a whole row/column, filling by adopting a 0 value, and inputting the processed image into a convolutional neural network to predict the photovoltaic power generation efficiency.
CN202110587166.8A 2021-05-27 2021-05-27 Short-time photovoltaic power generation prediction system and method Active CN113159466B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110587166.8A CN113159466B (en) 2021-05-27 2021-05-27 Short-time photovoltaic power generation prediction system and method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110587166.8A CN113159466B (en) 2021-05-27 2021-05-27 Short-time photovoltaic power generation prediction system and method

Publications (2)

Publication Number Publication Date
CN113159466A CN113159466A (en) 2021-07-23
CN113159466B true CN113159466B (en) 2022-04-15

Family

ID=76877931

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110587166.8A Active CN113159466B (en) 2021-05-27 2021-05-27 Short-time photovoltaic power generation prediction system and method

Country Status (1)

Country Link
CN (1) CN113159466B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113870531B (en) * 2021-09-17 2023-05-05 成都理工大学 Monitoring frequency adjusting method for geological disaster monitoring and early warning system
CN114296050B (en) * 2022-03-07 2022-06-07 南京鼐云信息技术有限责任公司 Photovoltaic power station short-term power generation power prediction method based on laser radar cloud picture detection
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
CN115759330B (en) * 2022-09-29 2024-04-09 中国大唐集团科学技术研究总院有限公司 High-time-resolution photovoltaic power generation power prediction method
CN116957223B (en) * 2023-06-14 2024-03-08 华能澜沧江水电股份有限公司 Photovoltaic power prediction method and device based on illumination image
CN117411088B (en) * 2023-12-13 2024-02-27 山东鼎鑫能源工程有限公司 Operation control method of photovoltaic power generation system and photovoltaic power generation system

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103218673A (en) * 2013-03-27 2013-07-24 河海大学 Method for predicating short-period output power of photovoltaic power generation based on BP (Back Propagation) neural network
CN106709585A (en) * 2015-07-20 2017-05-24 周锡卫 System for realizing ultra-short-term photovoltaic power prediction by using real-time shooting
US10642301B2 (en) * 2018-04-13 2020-05-05 Richtek Technology Corporation Power conversion apparatus for tracking maximum power point and control method thereof
CN112507793B (en) * 2020-11-05 2023-08-08 上海电力大学 Ultra-short term photovoltaic power prediction method
CN112801184A (en) * 2021-01-28 2021-05-14 江苏中信博新能源科技股份有限公司 Cloud tracking method, system and device

Also Published As

Publication number Publication date
CN113159466A (en) 2021-07-23

Similar Documents

Publication Publication Date Title
CN113159466B (en) Short-time photovoltaic power generation prediction system and method
CN110675418B (en) Target track optimization method based on DS evidence theory
CN111626128B (en) Pedestrian detection method based on improved YOLOv3 in orchard environment
CN113065558A (en) Lightweight small target detection method combined with attention mechanism
CN106886748B (en) TLD-based variable-scale target tracking method applicable to unmanned aerial vehicle
CN109389044B (en) Multi-scene crowd density estimation method based on convolutional network and multi-task learning
CN111986240A (en) Drowning person detection method and system based on visible light and thermal imaging data fusion
WO2019136591A1 (en) Salient object detection method and system for weak supervision-based spatio-temporal cascade neural network
CN110022422B (en) Video frame sequence generation method based on dense connection network
CN112818969A (en) Knowledge distillation-based face pose estimation method and system
CN110992378B (en) Dynamic updating vision tracking aerial photographing method and system based on rotor flying robot
CN113888461A (en) Method, system and equipment for detecting defects of hardware parts based on deep learning
CN115376024A (en) Semantic segmentation method for power accessory of power transmission line
CN116363694A (en) Multi-target tracking method of unmanned system crossing cameras matched with multiple pieces of information
CN112560865A (en) Semantic segmentation method for point cloud under outdoor large scene
CN116363748A (en) Power grid field operation integrated management and control method based on infrared-visible light image fusion
CN111260687B (en) Aerial video target tracking method based on semantic perception network and related filtering
CN113011308A (en) Pedestrian detection method introducing attention mechanism
CN117237884A (en) Interactive inspection robot based on berth positioning
CN111767826A (en) Timing fixed-point scene abnormity detection method
CN111222459A (en) Visual angle-independent video three-dimensional human body posture identification method
CN113065506B (en) Human body posture recognition method and system
CN106845555A (en) Image matching method and image matching apparatus based on Bayer format
CN113989612A (en) Remote sensing image target detection method based on attention and generation countermeasure network
CN111127355A (en) Method for finely complementing defective light flow graph and application thereof

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