Vision-based multi-energy complementary photovoltaic output estimation method

Abstract

The invention discloses a vision-based multi-energy complementary photovoltaic output estimation method. The method comprises the following steps: S1, constructing a vision estimation system; s2, acquiring an estimation parameter of the to-be-detected photovoltaic equipment; s3, constructing an estimation characteristic curve of the reference panel based on visual parameters; s4, obtaining an estimation characteristic curve of the adjustment panel based on the visual parameters, then obtaining images of all the to-be-tested panels, and performing output estimation based on the estimation characteristic curve of the adjustment panel to form a database; s5, performing normalization preprocessing on the data, obtaining an approximate component and a detail component of an output sequence by using wavelet transform, constructing a prediction model of different components based on LSTM, and obtaining output ultra-short-term prediction through data reconstruction and de-normalization; s6, providing an auxiliary function for the multi-energy complementary system in combination with a vision-based output estimation result and an output ultra-short-term prediction result; according to the invention, the real-time output parameters of the photovoltaic power generation equipment in the operation period can be automatically and quickly obtained with low cost.

Description

Technical Field

[0001] The invention relates to the technical field of photovoltaic output estimation, and in particular to a vision-based multi-energy complementary photovoltaic output estimation method. Background Art

[0002] Patent CN117040444A discloses a method for obtaining the irradiance distribution of a photovoltaic panel, including: using infrared thermal imaging to obtain the initial temperature value of the photovoltaic panel surface, and then constructing a relationship curve between the irradiance value and the temperature change rate under the current environment; under the current irradiance conditions, using infrared thermal imaging to obtain the first temperature of the first photovoltaic panel surface, and recording the heating time; determining the first temperature change rate when the first photovoltaic panel surface rises from the first temperature to a preset temperature difference; comparing the first temperature change rate with the relationship curve to obtain the current irradiance value. This patent establishes the relationship between irradiance and temperature change rate, and uses an infrared thermal imager to obtain the temperature change rate for temperature analysis. However, the temperature change rate of the photovoltaic panel may also be affected by the air temperature, and the influencing factors are not fully considered, which will affect the estimation accuracy of the irradiance. At the same time, the relationship between irradiance and output is not established, and the output cannot be directly estimated.

[0003] Patent CN117130394A discloses a photovoltaic equipment control method based on artificial intelligence, including: obtaining illumination image data around the photovoltaic equipment through an image acquisition device to obtain a historical illumination image data set of the photovoltaic equipment; establishing a target ResNet image recognition neural network model; inputting the historical illumination image data set of the photovoltaic equipment into the target ResNet image recognition neural network model for recognition, judging the illumination intensity, and using the photovoltaic equipment control instruction to control the movement and steering of the photovoltaic equipment to obtain a real-time illumination image data set of the photovoltaic equipment; and performing real-time movement and steering control of the photovoltaic equipment according to the real-time illumination status data of the photovoltaic equipment. This patent establishes an image data set and an image recognition model, and inputs the real-time image into the model to obtain the illumination intensity, but can only obtain the intensity of the illumination, and cannot obtain the corresponding value for output estimation.

[0004] Patent CN117216526B discloses a photovoltaic output prediction method and system based on artificial intelligence, including: collecting relevant data of photovoltaic power generation areas; obtaining the spatial distribution array of photovoltaic panels in distributed power generation areas; constructing the photovoltaic output coefficient of each position of the array, obtaining the spatial feature matrix of photovoltaic panels, and then constructing the photovoltaic output spatial stability matrix at each moment; calculating the transmission weight of each position of the cloud block; calculating the cloud layer correction weighted transmission index at each moment; obtaining the corrected photovoltaic output spatial stability matrix at each moment; obtaining the photovoltaic output spatiotemporal correlation matrix at the current moment; and combining the LSTM time series prediction model to predict photovoltaic output. This method needs to form an output sequence based on historical output data, collect light intensity, temperature, and obtain cloud parameters based on remote sensing images, and construct a prediction model based on various environmental parameters. It is impossible to directly estimate the output based on visual means.

[0005] Photovoltaic power generation is an important power source for multi-energy complementary systems, and the output of photovoltaic power stations is volatile. Generally, it is difficult to understand the real-time output of a photovoltaic station without obtaining control and management authority for the photovoltaic station or without grid-connected data of the photovoltaic station. Therefore, how to quickly analyze and estimate the real-time output of a photovoltaic station only through external visual observation is a technical problem that has not yet been solved.

[0006] The existing technology has visual inspection methods for the operating status of photovoltaic panels, such as defect analysis and radiation intensity analysis, but there is currently no practical method for estimating the output intensity of wind power and photovoltaic stations through machine vision. Summary of the invention

[0007] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a vision-based multi-energy complementary photovoltaic output estimation method to solve the problems raised in the background technology.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is: a multi-energy complementary photovoltaic output estimation method based on vision, which comprises the following steps: S1, configure the fusion visual analyzer, reference panel assistant, photovoltaic output estimator and estimated data output to build a visual estimation system; S2, obtaining estimated parameters of the photovoltaic device to be tested; S3, setting specific working conditions, obtaining visual images of the reference panel under different working conditions based on the visual estimation system, obtaining visual parameters, and constructing an estimation characteristic curve of the reference panel based on the visual parameters; S4, during the formal estimation, first obtain the estimated characteristic curve of the calibration panel based on the visual parameters, then obtain the images of all the panels to be tested, perform output estimation based on the estimated characteristic curve of the calibration panel, and form a database; S5. Perform normalization preprocessing on time, temperature, visual parameters, and output data, use wavelet transform to obtain approximate components and detail components of the output sequence, build prediction models of different components based on LSTM, and obtain ultra-short-term output prediction through data reconstruction and denormalization; S6. Combine the vision-based output estimation results and the ultra-short-term output prediction results to provide auxiliary functions for the multi-energy complementary system.

[0009] Preferably, the step S1 specifically includes the following sub-steps: S1.1, configure a fusion visual analyzer to extract and process visual information; S1.2. Configure a reference panel assistant to provide visual reference data of irradiance intensity for visual-based photovoltaic output estimation; S1.2.1, Determine the size of the reference panel: The size of the reference panel is determined according to the shape and size of the target photovoltaic panel to be tested, and the size of the reference panel is determined by geometric scaling with reference to the shape of the target photovoltaic panel to be tested, as shown in the following formula, and the proportional scaling factor range is set to 0.3 to 1.0;

[0010] Where: L r is the length of the reference panel plane size; L s is the length of the plane size of the photovoltaic panel to be tested; W r is the width of the reference panel plane size; W s is the width of the plane size of the photovoltaic panel to be tested; η is the proportional scaling factor; S1.2.2. Determination of reference panel structural material: The reference panel determines its structural material according to the target photovoltaic panel to be tested; S1.2.3, Determination of the working angle of the reference panel: The working angle of the reference panel is adjusted to be consistent with the working angle of the photovoltaic panel to be tested. For the photovoltaic panel to be tested with a variable working angle, the working angle of the reference panel should be changed accordingly; S1.2.4, Determination of the working angle of view of the reference panel: When using the fusion visual analyzer to extract the visual information of the reference panel, ensure that the camera angle and camera distance of the camera device when shooting the reference panel are consistent with the camera angle and camera distance of the camera device when shooting the photovoltaic panel to be tested; S1.3, configure output estimator: configure a processor and realize real-time output estimation of photovoltaic panels according to parameterized visual data input by fusion visual analyzer; S1.4. Configure a data processor: used to receive the output estimation result information of the output estimator, and process and distribute the results according to the requirements of the application end.

[0011] More preferably, in S1.1, the fusion visual analyzer includes one or more color image cameras and one or more thermal imaging cameras.

[0012] More preferably, in said S1.2, the reference panel auxiliary device comprises a panel and a bracket for supporting the panel and having a panel angle adjustment function.

[0013] Preferably, the step S2 specifically includes the following sub-steps: S2.1, obtaining estimated parameters of the photovoltaic equipment to be tested; S2.1.1. To achieve vision-based photovoltaic power generation output estimation, first obtain the power generation information of the target photovoltaic device to be tested; S2.1.2. Classification of characteristic reference temperature of the photovoltaic equipment to be tested: According to the working requirements and working environment of the photovoltaic equipment to be tested, and according to the possible ambient temperature range under its working state, its characteristic reference temperature is defined as follows:

[0014] Where: T i For the i Characteristics reference temperature; Th max is the highest temperature in the operating temperature range; Th min is the lowest temperature in the operating temperature range; n The total number of reference temperatures for the characteristic to be classified; S2.1.3. Then obtain the visual physical parameters of the target photovoltaic device panel to be tested, including: photovoltaic panel images and thermal imaging images taken by the fusion visual analyzer at various characteristic reference temperatures when the irradiation intensity does not reach the start-up condition during the day; photovoltaic panel images and thermal imaging parameters at various characteristic reference temperatures when it is cloudy; S2.2, obtain the estimated parameters of the reference panel: Obtain visual physical parameters of the reference panel: that is, according to the actual photovoltaic panel output prediction function needs, under various irradiation environments and various characteristic reference temperatures, the photovoltaic panel images and thermal imaging images taken by the fusion visual analyzer.

[0015] Preferably, the step S3 specifically includes the following sub-steps: S3.1. Preparation of the estimation basis based on the reference panel: Before performing the visual estimation of PV output, the fusion visual analyzer is prepared based on the reference panel, which includes the following steps: S3.1.1. Determine the characteristic radiation intensity: According to the output characteristic curve of the photovoltaic equipment to be tested, the characteristic irradiation intensity is defined at each characteristic reference temperature; S3.1.2. Preparation of reference panels: In an outdoor unobstructed area adjacent to the location of the photovoltaic equipment to be tested, a reference panel and a fusion visual analyzer are deployed, and a thermometer and an irradiation detector are configured for the reference panel; S3.1.3, Definition and naming of characteristic working conditions: Define that when the light-facing surface of the reference panel reaches a certain characteristic reference temperature and a certain characteristic irradiation intensity at the same time, it is a characteristic working condition, and the working condition is named after the combination of the characteristic reference temperature and the characteristic irradiation intensity; S3.1.4. Visual data collection of reference panel: When the reference panel reaches any characteristic working condition, the fusion visual analyzer is started, and various camera devices are used to obtain visual images under the working condition; Repeat this step until at least one set of visual image data is obtained under all characteristic working conditions; S3.2. Parameterization of the reference panel visual image: Using the image analysis processor of the fusion visual analyzer, the visual images of the reference panel under various characteristic working conditions are uniformly parameterized; the steps include: S3.2.1. For ordinary color pictures, according to the camera resolution, the color temperature parameters of all pixels in the picture are extracted according to the pixel points, and the overall color temperature value of the picture is obtained by analysis. The processing process is as follows:

[0016] Where: K r is the overall color temperature of the image, kr m For picture m The color temperature value of each pixel, m is the total number of pixels in the image; At the same time, the color parameters of all pixels in the image are extracted and analyzed to obtain the overall color parameters of the image; The processing process is as follows:

[0017] Where: RGB p The overall color RGB encoding of the image, R p ,G p , B p They are the overall color codes of red, green and blue for the overall color RGB code of the picture, rp m , gp m , bp m Respectively for pictures m RGB red, green, and blue monochrome encoding of pixels, m is the total number of pixels in the image; S3.2.2. For thermal imaging images, the thermal imaging images are grayed out, and then the gray value parameters of all pixels in the image are extracted, and the overall gray value of the image is obtained by analysis. The processing process is as follows:

[0018] Where: Gs r is the overall gray value of the image, gs m For picture m The gray value of a pixel, m is the total number of pixels in the image; S3.2.3. Correspond the various parameterized data after visual image processing to the characteristic working conditions, organize them according to the format, and finally establish a database for storage, so as to obtain the visual parameters of the reference panel; S3.3. Establishing an estimated characteristic curve based on the visual parameters of the reference panel: Based on the obtained reference panel visual parameters and in combination with the power generation characteristic curve of the photovoltaic panel to be tested, an estimated characteristic curve based on the reference panel visual parameters is established, the steps of which include: S3.3.1. Extract the irradiance intensity-output relationship curve corresponding to the power generation characteristic curve of the photovoltaic panel to be tested at each characteristic reference temperature; S3.3.2. In the relationship curve, find the corresponding characteristic point according to the characteristic irradiation intensity; S3.3.4. Add reference panel visual parameter information to the corresponding characteristic point of characteristic irradiance intensity to form a characteristic irradiance intensity-output-color temperature parameter-color parameter-grayscale parameter relationship, and bind it to the characteristic point; S3.3.5. Based on the power generation characteristic curve of the photovoltaic panel to be tested, the visual parameters of various reference panels are fitted through the characteristic points, and finally the visual parameter-output relationship curve of the visual parameters of various reference panels is formed, and for any visual parameter within the effective range, the corresponding output is obtained through the relationship curve analysis; S3.4. Complete the preparatory work: input the estimated characteristic curve based on the visual parameters of the reference panel into the photovoltaic output estimator, thus completing the preparatory work for the visual estimation of the photovoltaic power generation equipment.

[0019] Preferably, the step S4 specifically includes the following sub-steps: S4.1. Estimation and calibration of fusion visual analyzer: For the target photovoltaic station to be tested, any target photovoltaic panel to be tested is selected as the calibration photovoltaic panel, and then the estimation and calibration of the visual analyzer is carried out based on the reference panel. The specific steps are as follows: S4.1.1. Equipment layout: Place reference panels next to the calibration photovoltaic panels and equip them with fusion visual analyzers; S4.1.2. Synchronous acquisition of image information: At each characteristic reference temperature, various camera devices of the fusion visual analyzer are used to extract and adjust the working visual images of the photovoltaic panel and the reference panel, and collect working visual images under various lighting conditions; after completion, the reference panel is removed; S4.1.3, respectively parameterize the working visual images of the calibration photovoltaic panel and the reference panel; S4.1.4, comparing the visual image parameters of the calibrated photovoltaic panel and the reference panel, mapping and analyzing the estimated characteristic curve of the reference panel, and obtaining the estimated characteristic curve for the calibrated photovoltaic panel; S4.1.5, inputting the estimated characteristic curve of the calibrated photovoltaic panel obtained by mapping in the previous step into the photovoltaic output estimator for updating; S4.2. Visual estimation of the photovoltaic device under test; S4.2.1. Configure a fusion visual analyzer for all target photovoltaic panels to be tested; S4.2.2. When all target photovoltaic panels to be tested are working, various types of image information are continuously collected through the fusion visual analyzer; S4.2.3, after parameterizing various types of image information of each photovoltaic panel to be tested, substitute them into the estimated characteristic curve of the corresponding type of calibrated photovoltaic panel, and perform real-time output estimation according to step S4.1.4; Then, the predicted values ​​of various image parameters are comprehensively analyzed to obtain the real-time output estimation result of the photovoltaic panel to be tested; as shown in the following formula:

[0020] Where: Eo It is the real-time output estimation result of the photovoltaic panel to be tested; Eo K is the output prediction result of the color temperature parameter; Eo RGB is the output prediction result of color parameters; Eo Gsis the output prediction result of the grayscale parameter; c is the total number of prediction results of each category; S4.3. The visual estimation results of the photovoltaic equipment to be tested can be input into the estimation data processor after being sorted according to the time series, distributed, applied, and stored in a database.

[0021] Preferably, the step S5 specifically includes the following sub-steps: S5.1. Based on the established time, temperature, visual parameter-output relationship database, the time, temperature, visual parameter and output data are obtained, and the normalization method is used to perform normalization processing, and the time, temperature, visual parameter and output data are normalized to between [0,1] respectively; S5.2. For a certain acquisition length of output time series, the photovoltaic output time series is decomposed into low-frequency component CA1 and high-frequency components CD1 and CD2 based on wavelet transform. The Mallat algorithm is used for wavelet decomposition. The approximate signal Hjf of the signal f at a resolution of 2j is expressed as:

[0022] In the formula, H j f is the signal sequence, here it is the photovoltaic output sequence, H j-1 f for H j f The resolution of H is 2 j-1 The low-frequency approximation signal under D j-1 f for H j f The resolution is 2 by high-pass filter G. j-1 and 2 j High-frequency detail signals between At resolution 2 j-1 The expressions of approximate signal and detail signal are as follows: ; In the formula, Φ(x) and Ψ(x) are the wavelet scaling function and wavelet mother function respectively. a j-1 、d j-1 are the approximation coefficient and the detail coefficient respectively; Approximation coefficient a j-1,k and detail factor d j-1,k At resolution 2 j-1The following expression is:

[0023] in, h(k) is the impulse response of the low-pass filter H, g(k) is the impulse response of the high-pass filter G, M is the wavelet decomposition layer number, j is the decomposition layer number, and k is an integer controlling the wavelet translation; S5.3. Establish a photovoltaic output prediction model, as follows: S5.3.1. Select the output data before the prediction time ts minutes to construct the historical output input matrix, together with the temperature, time and visual parameters at time t as the input vector, and the different components at time t as the output values ​​to construct the data set, where s is the length of the historical output time series, s≥60, as shown below:

[0024]

[0025] in, x i is the input vector, E 0t-s for t-s The effort of the moment, t i for i The normalized minute number corresponding to the time, T i for i The temperature of the moment, K ri for i The color temperature value at the moment, RGB i for i The color parameters of the moment, G sri for i Infrared gray value at the moment; y 1i , y 2i , y 3i They are i The output component vector at time , CA1 i , CD1 i , CD2 i for i The low frequency component, high frequency component 1 and high frequency component 2 at the moment; S5.3.2. Construct prediction model: Use LSTM algorithm. The LSTM algorithm model specifically includes forget gate, input gate and output gate. The forget gate formula is as follows:

[0026] in f t for t Always choose the information to keep, W f is the weight matrix of the forget gate, x t express t The input content at the time, h t-1 for t-1 Short-term memory of the moment, b f represents the bias parameter of the forget gate, σ is the sigmoid function; The input gate formula is as follows:

[0027] In the formula, for t Initial long-term memory of the moment, i t for t The weight of information that long-term memory chooses to retain at any given moment, W i is the information weight matrix of the input gate, W c is the weight matrix of the initial long-term memory, x t express t The input content at the time, h t-1 express t-1 Short-term memory of the moment, b i is the bias parameter of the input gate, b c is the bias parameter of the preliminary long-term memory; The output gate formula is as follows:

[0028] in C t express t Long-term memory of the moment, C t-1 for t-1 Long-term memory of the moment, o t for t Output content at all times. h t expresst Short-term memory of the moment; The prediction models of CA1, CD1, and CD2 were constructed respectively, and the data sets were divided into training set, test set, and validation set at a ratio of 70%, 10%, and 20%, respectively, and substituted into the LSTM model for pre-training; S5.4. Use the trained prediction model to predict each component, then reconstruct it to get the normalized output prediction result, and finally denormalize the prediction result to get the final predicted output. The formula is as follows:

[0029] E 0t for t The effort of the moment, CA1 t , CD1 t , CD2 t for t The low-frequency component, high-frequency component 1 and high-frequency component 2 at the moment, E Otmax and E Otmax are the minimum and maximum output respectively.

[0030] Preferably, the step S6 specifically includes the following sub-steps: S6.1. Input the photovoltaic output estimation result into the multi-energy complementary system, and realize the multi-energy complementary function according to the multi-energy complementary method and the system goal; S6.2. Input the ultra-short-term prediction results of photovoltaic output into the multi-energy complementary system, and realize the multi-energy complementary function according to the multi-energy complementary method and system goals; S6.3. Based on the photovoltaic output estimation results and combined with ultra-short-term forecasts, the volatility index of regional photovoltaic output is analyzed to describe the volatility; the volatility index is used to improve the ultra-short-term forecast accuracy, as shown in the following formula:

[0031] Where: v is the photovoltaic output volatility index, Eo It is the real-time output estimation result of the photovoltaic panel to be tested; Eo’ It is the ultra-short-term prediction result of the photovoltaic panel to be tested at the next moment; S6.4. The photovoltaic ultra-short-term prediction result at the previous time point is determined by the photovoltaic output estimation result at the next time point, and combined with the aforementioned volatility index to improve the accuracy of the prediction, as shown in the following formula:

[0032] Where: yest The prediction correction coefficient of photovoltaic output at the moment; v is the PV output volatility index; a, b is the adjustment factor for the forecast correction factor calculation; Eo t The photovoltaic panel to be tested is t Output estimation results at the moment; Eo t+1 is the estimated output result of the photovoltaic panel to be tested at time t+1; Eo’ t The photovoltaic panel to be tested is t The ultra-short-term prediction result of the next moment of the moment prediction; Eo’ t+1 It is the ultra-short-term prediction result of the next moment predicted by the photovoltaic panel to be tested at time t+1; Eo φ t+1 yes t +1 The revised photovoltaic ultra-short-term forecast results at time; a , b}~min( Eo t+2 - Eo φ t+1 ) refers to the goal of minimizing the difference between the photovoltaic output estimation result at time t+2 and the photovoltaic ultra-short-term prediction result corrected at time t+1, and the adjustment coefficient a, b ; That is, by adjusting a, b , improving the accuracy of ultra-short-term forecasts.

[0033] Beneficial effects of the present invention: 1. The present invention can automatically, quickly and at low cost obtain the real-time output parameters of the photovoltaic power generation equipment during operation without the management and control authority of the photovoltaic station or its grid-connected information.

[0034] 2. The present invention can automatically, quickly and at low cost analyze the real-time output of photovoltaic power generation equipment in operation only through appearance observation based on video camera, and realize non-contact estimation; 3. The solution of the present invention proposes auxiliary equipment for reference panels, and integrates ordinary images and thermal imaging analysis results. The ordinary image results can be corrected by assisting with thermal imaging results, which can enhance the accuracy of photovoltaic output estimation.

[0035] 4. The present invention can achieve advance estimation and analysis of light resources in target areas where photovoltaic stations have not yet been established, and provide photovoltaic output prediction parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flowchart of a vision-based multi-energy complementary photovoltaic output estimation method; Figure 2 A schematic diagram of the visual estimation system; Figure 3 Visual parameter diagram for reference panel; Figure 4 Schematic diagram of equipment layout. DETAILED DESCRIPTION

[0037] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0038] Embodiment 1: Photovoltaic power generation is an important component of the new power system, which is mainly achieved through photovoltaic panels and related equipment. The output of photovoltaic power generation is affected by the weather and the change of day and night. It has the characteristics of strong volatility and real-time changes. That is, the output of photovoltaic power generation is directly related to the radiation intensity received by the panel.

[0039] Since the real-time output information of photovoltaic stations is an important basic parameter for photovoltaic power grid connection, it is necessary to quickly predict the real-time output of photovoltaic stations and obtain output information in order to achieve multi-energy complementarity.

[0040] Currently, if one has the management and control authority over a PV station or its grid-connected information, one can directly obtain the real-time output of the PV station. However, if one does not have the management and control authority or the grid parameters, it is difficult to understand the output of the PV station and it is impossible to obtain real-time PV station output data information.

[0041] This embodiment can use machine vision to utilize the proposed reference panel auxiliary system to quickly estimate the real-time output of photovoltaic equipment in operation through multiple types of image information from multiple camera devices. In addition, for photovoltaic power generation, based on ordinary image machine vision and combined with images of thermal imaging technology, fused photovoltaic real-time output estimation is achieved through data analysis, thereby improving the rationality and accuracy of vision-based wind power and photovoltaic estimation.

[0042] In addition, this embodiment can achieve early estimation and analysis of light resources for target areas where photovoltaic stations have not yet been established.

[0043] like Figure 1 As shown, a method for estimating the output of multi-energy complementary photovoltaic power based on vision includes the following steps: S1. Establish a visual estimation system for photovoltaic station output intensity: The system includes: a fusion visual analyzer, a reference panel assistant, a photovoltaic output estimator, and an estimated data processor.

[0044] The relationship between the four main parts of the system (such as Figure 2) is that the fusion visual analyzer extracts visual parameters of the working panel and reference panel auxiliary of the target photovoltaic equipment at the same time, and performs parameterization processing; then the processed visual parameterization information is submitted to the photovoltaic output estimator for photovoltaic power generation output estimation; the final photovoltaic output estimation result will be transmitted to the estimation data processor for practical application.

[0045] S1.1 Configuration of Fusion Visual Analyzer; The fusion visual analyzer is the main component for realizing the vision-based output estimation of this scheme. Its function is to extract and process visual information. Its objects include the target photovoltaic equipment to be tested and the reference panel auxiliary.

[0046] In this solution, the fusion visual analyzer should be configured with, but not limited to: one or more ordinary color image cameras, one or more thermal imaging cameras, and more types of other camera equipment can be configured as needed.

[0047] At the same time, the image camera and the thermal imaging camera should be respectively configured with relevant image analysis processors to realize the parametric analysis of specific image indicators. The specific principle is described in S3.

[0048] S1.2 Refer to the configuration of the panel auxiliaries; The reference panel assistant proposed in this scheme refers to a panel-shaped device that can provide visual reference data of irradiation intensity for vision-based photovoltaic output estimation, mainly including a panel and a bracket that supports the panel and has a panel angle adjustment function. The angle adjustment function of the bracket includes but is not limited to electric drive, manual operation, etc.

[0049] Specifically, the device is a panel device with the same color and similar structural materials as the general photovoltaic panel working surface, but it does not need to have photovoltaic power generation function. Its configuration requirements include: S1.2.1 Determination of the size of the reference panel. The size of the reference panel shall be determined according to the shape and size of the target photovoltaic panel to be tested. Generally, the reference panel shall be the same shape as the target photovoltaic panel to be tested, but smaller in size.

[0050] The reference panel size can be determined by geometric scaling with reference to the shape of the target photovoltaic panel to be tested, as shown in the following formula. This solution sets the proportional scaling factor range to be 0.3 to 1.0.

[0051]

[0052] Where: L r is the length of the reference panel plane size; L s is the length of the plane size of the photovoltaic panel to be tested; W ris the width of the reference panel plane size; W s is the width of the plane size of the photovoltaic panel to be tested; η is the proportional scaling factor.

[0053] S1.2.2 Determination of reference panel structural materials. The reference panel shall determine its structural materials based on the target photovoltaic panel to be tested. That is, the reference panel shall use the same or at least similar structural materials as the photovoltaic panel to be tested.

[0054] For example, if a photovoltaic panel to be tested uses a certain type of silicon material and is protected by a glass panel, forming a two-layer structure, then the reference panel should also use a two-layer structure of silicon material and glass. The silicon material should at least be consistent in color with the silicon material of the photovoltaic panel to be tested; the glass panel should have similar light transmittance and thickness to the photovoltaic panel to be tested.

[0055] S1.2.3 Determination of the working angle of the reference panel. The working angle of the reference panel shall be adjusted to be consistent with the working angle of the PV panel to be tested. For PV panels to be tested with variable working angles, the working angle of the reference panel shall be changed accordingly.

[0056] S1.2.4 Determination of the working angle of reference panel. When using the fusion visual analyzer to extract the visual information of the reference panel, it should be ensured that the camera angle and camera distance of the camera device when shooting the reference panel are consistent with the camera angle and camera distance of the camera device when shooting the photovoltaic panel to be tested.

[0057] S1.3 Configuration of output estimator; It is mainly a configured processor, and can realize real-time output estimation of photovoltaic panels based on the parameterized visual data input by the fusion visual analyzer.

[0058] S1.4 Configuration of data processors; It is used to receive the output estimation result information from the output estimator, and process and distribute the results according to the application requirements.

[0059] S2. Prepare visual estimation parameters of photovoltaic power generation equipment; S2.1 Obtaining estimated parameters of the photovoltaic equipment to be tested; S2.1.1 To achieve vision-based photovoltaic power generation output estimation, the power generation information of the target photovoltaic equipment to be tested should be obtained first, including: starting irradiation intensity; rated output; power generation characteristic curve (temperature-irradiation intensity-output).

[0060] S2.1.2 Classification of characteristic reference temperatures of the photovoltaic equipment to be tested. According to the working requirements and working environment of the photovoltaic equipment to be tested, and according to the possible ambient temperature range under its working state, its characteristic reference temperature is defined (in this scheme, it is set within its working temperature range, and 3 or more characteristic reference temperatures need to be obtained). The specific formula is as follows:

[0061] Where: T i For the i Characteristics reference temperature; Th max is the highest temperature in the operating temperature range; Th min is the lowest temperature in the operating temperature range; n The total number of reference temperatures for the characteristic to be partitioned.

[0062] S2.1.3 then obtains the visual physical parameters of the target photovoltaic device panel to be tested, including: photovoltaic panel images and thermal imaging images taken by a fusion visual analyzer at various characteristic reference temperatures when the starting irradiation intensity condition is not reached during the day.

[0063] For example, on a cloudy day, the photovoltaic panel images and thermal imaging parameters at various characteristic reference temperatures.

[0064] S2.2 Obtaining estimated parameters of reference panel; Obtain visual physical parameters of the reference panel. According to the actual photovoltaic panel output prediction function requirements, the photovoltaic panel images and thermal imaging images taken by the fusion visual analyzer under various irradiation environments and various characteristic reference temperatures are used. See S3 for details.

[0065] S3. Principles and preparations for visual estimation of photovoltaic power generation equipment; S3.1 Preparation of the basis for estimates based on the reference panel; To implement this solution, it is necessary to prepare the fusion visual analyzer based on the reference panel before performing the visual estimation of photovoltaic output to achieve more accurate output estimation, which includes the following steps: S3.1.1 Determine the characteristic irradiance intensity; According to the output characteristic curve of the photovoltaic equipment to be tested, the characteristic irradiation intensity is defined at each characteristic reference temperature. In this scheme, 3 or more characteristic irradiation intensities need to be obtained at each characteristic reference temperature, and one of the characteristic irradiation intensities should be the irradiation intensity corresponding to the rated output of the photovoltaic panel at the characteristic reference temperature.

[0066] S3.1.2 Preparation of reference panels; In an outdoor unobstructed area adjacent to the location of the photovoltaic equipment to be tested, a reference panel and a fusion visual analyzer are arranged as described in S1.2. A thermometer and an irradiation detector are configured for the reference panel.

[0067] S3.1.3 Definition and naming of characteristic conditions; The characteristic working condition is defined as when the light-facing surface of the reference panel reaches a certain characteristic reference temperature and a certain characteristic irradiation intensity at the same time. The working condition is named after the combination of the characteristic reference temperature and the characteristic irradiation intensity.

[0068] For example, if three characteristic reference temperatures and three characteristic irradiation intensities are determined, there are nine characteristic operating conditions, of which the first characteristic operating condition is named (characteristic reference temperature 1, characteristic irradiation intensity 1).

[0069] S3.1.4 Visual data collection of reference panels; When the reference panel reaches any characteristic working condition, the fusion visual analyzer is started, and various types of camera equipment are used to obtain visual images under this working condition.

[0070] For example, a fusion visual analyzer equipped with a common color camera and a thermal imaging device needs to obtain both a common color image and a thermal imaging image of the reference panel under characteristic working conditions.

[0071] Repeat this step until at least one set of visual image data is obtained under all characteristic working conditions.

[0072] S3.2 Parameterization of the visual picture of the reference panel; The image analysis processor of the fusion visual analyzer is used to uniformly parameterize the visual images of the reference panel under various characteristic working conditions. The steps include: S3.2.1 For ordinary color pictures. According to the camera resolution, the color temperature parameters of all pixels in the picture are extracted according to the pixel points, and the overall color temperature value of the picture is obtained by analysis. The processing process is as follows:

[0073] Where: K r is the overall color temperature of the image, kr m For picture m The color temperature value of each pixel, m is the total number of pixels in the image.

[0074] At the same time, the color parameters of all pixels of the image are extracted and analyzed to obtain the overall color parameters of the image. The color parameters should follow various existing regular color encoding rules such as RGB, YUV encoding and other existing encodings.

[0075] Taking RGB as an example, the processing process is as follows:

[0076] Where: R p , G p , B p They are the overall color codes of red, green and blue for the overall color RGB code of the picture, rp m , gp m , bp m Respectively for pictures m RGB red, green, and blue monochrome encoding of pixels, m is the total number of pixels in the image.

[0077] S3.2.2 For thermal imaging images. Perform grayscale processing on the thermal imaging images, then extract the grayscale value parameters of all pixels in the image, and analyze to obtain the overall grayscale value of the image. The processing process is as follows:

[0078] Where: Gs r is the overall gray value of the image, gs m For picture m The gray value of a pixel, m is the total number of pixels in the image.

[0079] S3.2.3 The various parameterized data after visual image processing are matched with characteristic working conditions, and are sorted in the format of (characteristic reference temperature, characteristic irradiation intensity, color temperature parameter, color parameter, grayscale parameter), and finally a database is established for storage. That is, the reference panel visual parameters (such as Figure 3 as shown).

[0080] S3.3 Estimated characteristic curves based on visual parameters of reference panels; Based on the obtained visual parameters of the reference panel and combined with the power generation characteristic curve (temperature-irradiation intensity-output) of the photovoltaic panel to be tested, an estimated characteristic curve based on the visual parameters of the reference panel is established. The steps include: S3.3.1 At each characteristic reference temperature, extract the (irradiation intensity-output) relationship curve corresponding to the power generation characteristic curve of the photovoltaic panel to be tested.

[0081] S3.3.2 In the (irradiation intensity-output) relationship curve, find the corresponding characteristic point based on the characteristic irradiation intensity.

[0082] S3.3.4 Add the reference panel visual parameter information to the corresponding characteristic point of the characteristic irradiance to form a (characteristic irradiance intensity-output-color temperature parameter-color parameter-grayscale parameter) relationship and bind it to the characteristic point.

[0083] S3.3.5 Based on the power generation characteristic curve of the photovoltaic panel to be tested, the visual parameters of various reference panels are fitted through the characteristic points. Finally, the (visual parameter-output) relationship curve of the visual parameters of various reference panels is formed. And for any visual parameter within the effective range, the corresponding output can be obtained through the relationship curve analysis.

[0084] For example: (color temperature parameter-output), (color parameter-output), (grayscale parameter-output) and other relationships.

[0085] Here, we take the (grayscale-output) parameter as an example to show the fitting process, as shown in the following formula:

[0086] Where: μ The power generation characteristic curve of the photovoltaic panel can be expressed at each characteristic reference temperature. Tp (0- j The corresponding (irradiation intensity-output) relationship curve under j ), among which Ra is the irradiance value, Eo is the output value; π The grayscale-output relationship curve is obtained based on the above analysis, which is the characteristic curve μ Corresponding irradiation intensity values ​​through characteristic points Ra and characteristic gray value Gs Transformation fitting obtained.

[0087] function f ( Tp j , Gs t , π ) indicates that at a specific temperature Tp t In this case, any gray value within the valid range Gs t Substitute the (grayscale-output) relationship and obtain the corresponding output parameter value Eo t process.

[0088] S3.4 Complete the preparation work. Input the estimated characteristic curve based on the reference panel visual parameters into the photovoltaic output estimator, and the preparation work for the visual estimation of the photovoltaic power generation equipment is completed.

[0089] S4. Visual estimation process of photovoltaic power generation equipment; S4.1 Estimation and calibration of the fusion visual analyzer; For the target photovoltaic station to be tested, any target photovoltaic panel to be tested is selected as the calibration photovoltaic panel, and then the estimation and calibration of the visual analyzer can be carried out based on the reference panel. The specific steps are as follows: S4.1.1 Equipment layout. Next to the calibration photovoltaic panel, lay out the reference panel according to the above S1.2 steps, and configure the fusion visual analyzer. Figure 4 As shown in the figure: 1. Target photovoltaic panel to be tested (adjustment photovoltaic panel); 2. Reference panel; Camera of the vision system.

[0090] S4.1.2 Image information is collected synchronously. At each characteristic reference temperature, various camera devices of the fusion visual analyzer are used to extract and adjust the working visual images of the photovoltaic panel and the reference panel. Collect working visual images under various lighting conditions (the specific number of working conditions set in this plan is determined according to actual needs and should be at least 5).

[0091] Once completed, remove the reference panel.

[0092] S4.1.3 Parameterize the working visual images of the calibration photovoltaic panel and the reference panel respectively as described in S3.2.

[0093] S4.1.4 Compare the visual image parameters of the calibrated photovoltaic panel and the reference panel, and in combination with S3.3, perform mapping analysis on the estimated characteristic curve of the reference panel to obtain the estimated characteristic curve for the calibrated photovoltaic panel.

[0094] Taking the (grayscale-output) parameter as an example, the fitting process is shown as follows:

[0095] Where: Gs Tpj It is to adjust the photovoltaic panel in the j Characteristics Reference Temperature Tp The visual grayscale parameter value; Gs | Tpj The reference panel is j Characteristics Reference Temperature Tp The visual grayscale parameter value; D is a parameter of the fitting process, representing the distance between the reference panel visual parameters and the calibrated PV panel visual parameters.

[0096] π β is the (grayscale-output) relationship curve obtained by analyzing the visual parameters of the reference panel as described in S3.3; πα This is the (grayscale-output) relationship curve of the calibrated photovoltaic panel obtained through mapping analysis.

[0097] According to the (grayscale-output) relationship curve of the calibrated photovoltaic panel, the real-time photovoltaic output can be estimated according to the grayscale parameters of the calibrated photovoltaic panel measured at any time. As shown in the following formula:

[0098] Where: Function f ( Tp j , Gs t , π α ) indicates that at a specific temperature Tp t In this case, any gray value within the valid range Gs t Substitution π α Adjust the photovoltaic panel (grayscale-output) relationship and obtain the corresponding output parameter value Eo t process.

[0099] S4.1.5 The (grayscale-output) relationship curve of the calibrated photovoltaic panel obtained by mapping in the previous step is input and updated into the photovoltaic output estimator.

[0100] S4.2 Visual estimation of the photovoltaic device under test; S4.2.1 For all target PV panels to be tested, configure the fusion vision analyzer as described in S1.2.4.

[0101] S4.2.2 When all target photovoltaic panels to be tested are operating, various types of image information are continuously collected through the fusion visual analyzer.

[0102] S4.2.3 After parameterizing various types of image information of each photovoltaic panel to be tested according to step S3.2, substitute them into the estimated characteristic curve of the corresponding type of calibrated photovoltaic panel, and perform real-time output estimation according to step S4.1.4.

[0103] Then, the predicted values ​​of various image parameters are comprehensively analyzed to obtain the real-time output estimation result of the photovoltaic panel to be tested. As shown in the following formula:

[0104] Where: Eo It is the real-time output estimation result of the photovoltaic panel to be tested; Eo K is the output prediction result of the color temperature parameter; Eo RGBis the output prediction result of color parameters; Eo Gs is the output prediction result of the grayscale parameter; c is the total number of prediction results of each category.

[0105] S4.3 The visual estimation results of the photovoltaic equipment to be tested can be input into the estimation data processor after being sorted according to the time series, distributed, applied, and stored in the database.

[0106] S5. Ultra-short-term prediction of photovoltaic output for multi-energy complementary systems; S5.1 obtains time, temperature, visual parameters and output data based on the established "time, temperature, visual parameters-output" relationship database, and uses normalization methods for normalization. Normalization methods include but are not limited to linear normalization, Z-Score normalization, mean normalization, etc. Taking linear normalization as an example, normalize the time, temperature, visual parameters, and output data to [0,1] respectively, and the formula is as follows:

[0107] In the formula, is the raw minute number of the day, 0:01 corresponds to For 1, 0:30 corresponds to 30, 24:00 corresponds to is 1440, and so on; , T min 、T max are original temperature, minimum temperature, and maximum temperature respectively; , K rmin 、K rmax are the original color temperature, the minimum color temperature and the maximum color temperature, , RGB min , RGB max are the original, minimum and maximum color parameters, , G srmin 、G srmax are the original, minimum and maximum infrared grayscale values, , E Omin 、E Omax For original, minimum and maximum output.

[0108] S5.2 For a certain acquisition length of output time series, the photovoltaic output time series is decomposed into low-frequency component CA1 and high-frequency components CD1 and CD2 based on wavelet transform. The Mallat algorithm can be used for wavelet decomposition. j The signal below f The approximate signal H j f It can be expressed as:

[0109] In the formula, H j f is the signal sequence, here it is the photovoltaic output sequence, H j-1 f for H j f The resolution of H is 2 j-1 The low-frequency approximation signal under D j-1 f for H j f The resolution is 2 by high-pass filter G. j-1 and 2 j High-frequency detail signals between At resolution 2 j-1 The expressions of approximate signal and detail signal are as follows: ; In the formula, Φ(x) and Ψ(x) are the wavelet scaling function and wavelet mother function respectively. a j-1 、d j-1 are the approximation coefficient and the detail coefficient respectively; Approximation coefficient a j-1,k and detail factor d j-1,k At resolution 2 j-1 The following expression is:

[0110] in, h(k) is the impulse response of the low-pass filter H, g(k) is the impulse response of the high-pass filter G, M is the wavelet decomposition layer number, j is the decomposition layer number, and k is an integer controlling the wavelet translation; S5.3 Establish a photovoltaic output prediction model, as follows: S5.3.1 Selecting the prediction time t-s ( s is the length of the historical output time series, s≥60 ) minutes ago to build the historical output input matrix, together with the temperature, time and visual parameters at time t as the input vector, and the different components at time t as the output values ​​to build the data set, as shown below:

[0111]

[0112] in, x i is the input vector, E 0t-s for t-s The effort of the moment, t i for i The normalized minute number corresponding to the time, T i for i The temperature of the moment, K ri for i The color temperature value at the moment, RGB i for i The color parameters of the moment, G sri for i Infrared gray value at the moment; y 1i , y 2i , y 3i They are i The output component vector at time , CA1 i , CD1 i , CD2 i for i The low frequency component, high frequency component 1 and high frequency component 2 at the moment; S5.3.2 Use machine learning models (such as random forest algorithm, support vector machine algorithm, etc.) to build a prediction model. For example, use LSTM algorithm. The LSTM model specifically includes a forget gate, an input gate, and an output gate. The forget gate formula is as follows:

[0113] in f t for t Always choose the information to keep, W f is the weight matrix of the forget gate,x t express t The input content at the time, h t-1 for t-1 Short-term memory of the moment, b f represents the bias parameter of the forget gate, σ is the sigmoid function; The input gate formula is as follows:

[0114] In the formula, for t Initial long-term memory of the moment, i t for t The weight of information that long-term memory chooses to retain at any given moment, W i is the weight matrix of the input gate, W c is the weight matrix of the initial long-term memory, x t express t The input content at the time, h t-1 express t-1 Short-term memory of the moment, b i is the weight matrix of the input gate, b c is the weight matrix of the preliminary long-term memory; The output gate formula is as follows:

[0115] in C t express t Long-term memory of the moment, C t-1 for t-1 Long-term memory of the moment, o t for t Output content at all times; The prediction models of CA1, CD1, and CD2 were constructed respectively, and the data sets were divided into training set, test set, and validation set at a ratio of 70%, 10%, and 20%, respectively, and substituted into the LSTM model for pre-training; S5.4 Use the trained prediction model to predict each component, then reconstruct to obtain the normalized output prediction result, and finally denormalize the prediction result to obtain the final predicted output. The formula is as follows:

[0116] E 0t for t The effort of the moment, CA1 t , CD1 t , CD2 t for t The low-frequency component, high-frequency component 1 and high-frequency component 2 at the moment, E Otmax and E Otmax are the minimum and maximum output respectively; S6. Visual multi-energy complementary photovoltaic output estimation results and prediction results application; Based on the photovoltaic output estimation and ultra-short-term output prediction that can be achieved by this solution, the results can be applied to multi-energy complementary systems. The steps are as follows: S6.1 Input the photovoltaic output estimation result into the multi-energy complementary system, and realize the multi-energy complementary function according to the existing multi-energy complementary method and the system goal.

[0117] S6.2 Input the ultra-short-term prediction results of photovoltaic output into the multi-energy complementary system, and realize the multi-energy complementary function according to the existing multi-energy complementary method and system goals.

[0118] More accurate ultra-short-term prediction results can better serve the multi-energy complementary system.

[0119] S6.3 Based on the photovoltaic output estimation results and combined with ultra-short-term forecasts, the volatility index of regional photovoltaic output can be analyzed to describe the volatility. This volatility index can be used to improve the accuracy of ultra-short-term forecasts. As shown in the following formula:

[0120] Where: v is the photovoltaic output volatility index, Eo It is the real-time output estimation result of the photovoltaic panel to be tested; Eo’ It is the ultra-short-term prediction result of the photovoltaic panel to be tested at the next moment.

[0121] S6.4 The ultra-short-term photovoltaic forecast result at the previous time point can be judged by the photovoltaic output estimation result at the next time point, and combined with the above volatility index, the accuracy of the forecast can be improved. As shown in the following formula:

[0122] Where: yes t The prediction correction coefficient of photovoltaic output at the moment; v is the PV output volatility index; a, b is the adjustment factor for the forecast correction factor calculation;Eo t The photovoltaic panel to be tested is t Output estimation results at the moment; Eo t+1 is the estimated output result of the photovoltaic panel to be tested at time t+1; Eo’ t The photovoltaic panel to be tested is t The ultra-short-term prediction result of the next moment of the moment prediction; Eo’ t+1 It is the ultra-short-term prediction result of the next moment predicted by the photovoltaic panel to be tested at time t+1; Eo φ t+1 yes t +1 The revised photovoltaic ultra-short-term forecast results at time; a , b}~min( Eo t+2 - Eo φ t+1 ) refers to the goal of minimizing the difference between the photovoltaic output estimation result at time t+2 and the photovoltaic ultra-short-term prediction result corrected at time t+1, and the adjustment coefficient a, b ; That is, by adjusting a, b , improving the accuracy of ultra-short-term prediction of this scheme.

[0123] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for estimating the output of multi-energy complementary photovoltaic power based on vision, characterized in that: It includes the following steps: S1, configure the fusion visual analyzer, reference panel assistant, photovoltaic output estimator and estimated data output to build a visual estimation system; S2, obtaining estimated parameters of the photovoltaic device to be tested; S3, setting specific working conditions, obtaining visual images of the reference panel under different working conditions based on the visual estimation system, obtaining visual parameters, and constructing an estimation characteristic curve of the reference panel based on the visual parameters; S4, during the formal estimation, first obtain the estimated characteristic curve of the calibration panel based on the visual parameters, then obtain the images of all the panels to be tested, perform output estimation based on the estimated characteristic curve of the calibration panel, and form a database; S5. Perform normalization preprocessing on time, temperature, visual parameters, and output data, use wavelet transform to obtain approximate components and detail components of the output sequence, build prediction models of different components based on LSTM, and obtain ultra-short-term output prediction through data reconstruction and denormalization; S6. Combine the vision-based output estimation results and the ultra-short-term output prediction results to provide auxiliary functions for the multi-energy complementary system.

2. The method for estimating the output of multi-energy complementary photovoltaic power based on vision according to claim 1, characterized in that: The step S1 specifically includes the following sub-steps: S1.1, configure a fusion visual analyzer to extract and process visual information; S1.

2. Configure a reference panel assistant to provide visual reference data of irradiance intensity for visual-based photovoltaic output estimation; S1.2.1, Determine the size of the reference panel: The size of the reference panel is determined according to the shape and size of the target photovoltaic panel to be tested, and the size of the reference panel is determined by geometric scaling with reference to the shape of the target photovoltaic panel to be tested, as shown in the following formula, and the proportional scaling factor range is set to 0.3 to 1.0; ； Where: L r is the length of the reference panel plane size; L s is the length of the plane size of the photovoltaic panel to be tested; W r is the width of the reference panel plane size; W s is the width of the plane size of the photovoltaic panel to be tested; η is the proportional scaling factor; S1.2.

2. Determination of reference panel structural material: The reference panel determines its structural material according to the target photovoltaic panel to be tested; S1.2.3, Determination of the working angle of the reference panel: The working angle of the reference panel is adjusted to be consistent with the working angle of the photovoltaic panel to be tested. For the photovoltaic panel to be tested with a variable working angle, the working angle of the reference panel should be changed accordingly; S1.2.4, Determination of the working angle of view of the reference panel: When using the fusion visual analyzer to extract the visual information of the reference panel, ensure that the camera angle and camera distance of the camera device when shooting the reference panel are consistent with the camera angle and camera distance of the camera device when shooting the photovoltaic panel to be tested; S1.3, configure output estimator: configure a processor and realize real-time output estimation of photovoltaic panels according to parameterized visual data input by fusion visual analyzer; S1.

4. Configure a data processor: used to receive the output estimation result information of the output estimator, and process and distribute the results according to the requirements of the application end.

3. The method for estimating the output of multi-energy complementary photovoltaic power based on vision according to claim 2, characterized in that: In S1.1, the fusion visual analyzer includes one or more color image cameras and one or more thermal imaging cameras.

4. The method for estimating the output of multi-energy complementary photovoltaic power based on vision according to claim 2, characterized in that: In the above S1.2, the reference panel auxiliary device includes a panel and a bracket for supporting the panel and having a panel angle adjustment function.

5. The method for estimating the output of multi-energy complementary photovoltaic power based on vision according to claim 1, characterized in that: The step S2 specifically includes the following sub-steps: S2.1, obtaining estimated parameters of the photovoltaic equipment to be tested; S2.1.

1. To achieve vision-based photovoltaic power generation output estimation, first obtain the power generation information of the target photovoltaic device to be tested; S2.1.

2. Classification of characteristic reference temperature of the photovoltaic equipment to be tested: According to the working requirements and working environment of the photovoltaic equipment to be tested, and according to the possible ambient temperature range under its working state, its characteristic reference temperature is defined as follows: ； Where: T i For the i Characteristics reference temperature; Th max is the highest temperature in the operating temperature range; Th min is the lowest temperature in the operating temperature range; n The total number of reference temperatures for the characteristic to be classified; S2.1.

3. Then obtain the visual physical parameters of the target photovoltaic device panel to be tested, including: photovoltaic panel images and thermal imaging images taken by the fusion visual analyzer at various characteristic reference temperatures when the irradiation intensity does not reach the start-up condition during the day; photovoltaic panel images and thermal imaging parameters at various characteristic reference temperatures when it is cloudy; S2.2, obtain the estimated parameters of the reference panel: Obtain visual physical parameters of the reference panel: that is, according to the actual photovoltaic panel output prediction function needs, under various irradiation environments and various characteristic reference temperatures, the photovoltaic panel images and thermal imaging images taken by the fusion visual analyzer.

6. The method for estimating the output of multi-energy complementary photovoltaic power based on vision according to claim 1, characterized in that: The step S3 specifically includes the following sub-steps: S3.

1. Preparation of the estimation basis based on the reference panel: Before performing the visual estimation of PV output, the fusion visual analyzer is prepared based on the reference panel, which includes the following steps: S3.1.

1. Determine the characteristic radiation intensity: According to the output characteristic curve of the photovoltaic equipment to be tested, the characteristic irradiation intensity is defined at each characteristic reference temperature; S3.1.

2. Preparation of reference panels: In an outdoor unobstructed area adjacent to the location of the photovoltaic equipment to be tested, a reference panel and a fusion visual analyzer are deployed, and a thermometer and an irradiation detector are configured for the reference panel; S3.1.3, Definition and naming of characteristic working conditions: Define that when the light-facing surface of the reference panel reaches a certain characteristic reference temperature and a certain characteristic irradiation intensity at the same time, it is a characteristic working condition, and the working condition is named after the combination of the characteristic reference temperature and the characteristic irradiation intensity; S3.1.

4. Visual data collection of reference panel: When the reference panel reaches any characteristic working condition, the fusion visual analyzer is started, and various camera devices are used to obtain visual images under the working condition; Repeat this step until at least one set of visual image data is obtained under all characteristic working conditions; S3.

2. Parameterization of the reference panel visual image: Using the image analysis processor of the fusion visual analyzer, the visual images of the reference panel under various characteristic working conditions are uniformly parameterized; the steps include: S3.2.

1. For ordinary color pictures, according to the camera resolution, the color temperature parameters of all pixels in the picture are extracted according to the pixel points, and the overall color temperature value of the picture is obtained by analysis. The processing process is as follows: ； Where: K r is the overall color temperature of the image, kr m For picture m The color temperature value of each pixel, m is the total number of pixels in the image; At the same time, the color parameters of all pixels in the image are extracted and analyzed to obtain the overall color parameters of the image; The processing process is as follows: ； Where: RGB p The overall color RGB encoding of the image, R p , G p , B p They are the overall color codes of red, green and blue for the overall color RGB code of the picture, rp m , gp m , bp m Respectively for pictures m RGB red, green, and blue monochrome encoding of pixels, m is the total number of pixels in the image; S3.2.

2. For thermal imaging images, the thermal imaging images are grayed out, and then the gray value parameters of all pixels in the image are extracted, and the overall gray value of the image is obtained by analysis. The processing process is as follows: ； Where: Gs r is the overall gray value of the image, gs m For picture m The gray value of a pixel, m is the total number of pixels in the image; S3.2.

3. Correspond the various parameterized data after visual image processing to the characteristic working conditions, organize them according to the format, and finally establish a database for storage, so as to obtain the visual parameters of the reference panel; S3.

3. Establishing an estimated characteristic curve based on the visual parameters of the reference panel: Based on the obtained reference panel visual parameters and in combination with the power generation characteristic curve of the photovoltaic panel to be tested, an estimated characteristic curve based on the reference panel visual parameters is established, the steps of which include: S3.3.

1. Extract the irradiance intensity-output relationship curve corresponding to the power generation characteristic curve of the photovoltaic panel to be tested at each characteristic reference temperature; S3.3.

2. In the relationship curve, find the corresponding characteristic point according to the characteristic irradiation intensity; S3.3.

4. Add reference panel visual parameter information to the corresponding characteristic point of characteristic irradiance intensity to form a characteristic irradiance intensity-output-color temperature parameter-color parameter-grayscale parameter relationship, and bind it to the characteristic point; S3.3.

5. Based on the power generation characteristic curve of the photovoltaic panel to be tested, the visual parameters of various reference panels are fitted through the characteristic points, and finally the visual parameter-output relationship curve of the visual parameters of various reference panels is formed, and for any visual parameter within the effective range, the corresponding output is obtained through the relationship curve analysis; S3.

4. Complete the preparatory work: input the estimated characteristic curve based on the visual parameters of the reference panel into the photovoltaic output estimator, thus completing the preparatory work for the visual estimation of the photovoltaic power generation equipment.

7. The method for estimating the output of multi-energy complementary photovoltaic power based on vision according to claim 1, characterized in that: The step S4 specifically includes the following sub-steps: S4.

1. Estimation and calibration of fusion visual analyzer: For the target photovoltaic station to be tested, any target photovoltaic panel to be tested is selected as the calibration photovoltaic panel, and then the estimation and calibration of the visual analyzer is carried out based on the reference panel. The specific steps are as follows: S4.1.

1. Equipment layout: Place reference panels next to the calibration photovoltaic panels and equip them with fusion visual analyzers; S4.1.

2. Synchronous acquisition of image information: At each characteristic reference temperature, various camera devices of the fusion visual analyzer are used to extract and adjust the working visual images of the photovoltaic panel and the reference panel, and to collect working visual images under various lighting conditions; Once completed, remove the reference panel; S4.1.3, respectively parameterize the working visual images of the calibration photovoltaic panel and the reference panel; S4.1.4, comparing the visual image parameters of the calibrated photovoltaic panel and the reference panel, mapping and analyzing the estimated characteristic curve of the reference panel, and obtaining the estimated characteristic curve for the calibrated photovoltaic panel; S4.1.5, inputting the estimated characteristic curve of the calibrated photovoltaic panel obtained by mapping in the previous step into the photovoltaic output estimator for updating; S4.

2. Visual estimation of the photovoltaic device under test; S4.2.

1. Configure a fusion visual analyzer for all target photovoltaic panels to be tested; S4.2.

2. When all target photovoltaic panels to be tested are working, various types of image information are continuously collected through the fusion visual analyzer; S4.2.3, after parameterizing various types of image information of each photovoltaic panel to be tested, substitute them into the estimated characteristic curve of the corresponding type of calibrated photovoltaic panel, and perform real-time output estimation according to step S4.1.4; Then, the predicted values ​​of various image parameters are comprehensively analyzed to obtain the real-time output estimation result of the photovoltaic panel to be tested; as shown in the following formula: ； Where: Eo It is the real-time output estimation result of the photovoltaic panel to be tested; Eo K is the output prediction result of the color temperature parameter; Eo RGB is the output prediction result of color parameters; Eo Gs is the output prediction result of the grayscale parameter; c is the total number of prediction results of each category; S4.

3. The visual estimation results of the photovoltaic equipment to be tested can be input into the estimation data processor after being sorted according to the time series, distributed, applied, and stored in a database.

8. The method for estimating the output of multi-energy complementary photovoltaic power based on vision according to claim 1, characterized in that: The step S5 specifically includes the following sub-steps: S5.

1. Based on the established time, temperature, visual parameter-output relationship database, the time, temperature, visual parameter and output data are obtained, and the normalization method is used to perform normalization processing, and the time, temperature, visual parameter and output data are normalized to between [0,1] respectively; S5.

2. For a certain acquisition length of output time series, the photovoltaic output time series is decomposed into low-frequency component CA1 and high-frequency components CD1 and CD2 based on wavelet transform. The Mallat algorithm is used for wavelet decomposition. The approximate signal Hjf of the signal f at a resolution of 2j is expressed as: ； In the formula, H j f is the signal sequence, here it is the photovoltaic output sequence, H j-1 f for H j f The resolution of H is 2 j-1 The low-frequency approximation signal under D j-1 f for H j f The resolution is 2 by high-pass filter G. j-1 and 2 j High-frequency detail signals between At resolution 2 j-1 The expressions of approximate signal and detail signal are as follows: ； In the formula, Φ(x) and Ψ(x) are the wavelet scaling function and wavelet mother function respectively. a j-1 、d j-1 are the approximation coefficient and the detail coefficient respectively; Approximation coefficient a j-1,k and detail factor d j-1,k At resolution 2 j-1 The following expression is: ； in, h(k) is the impulse response of the low-pass filter H, g(k) is the impulse response of the high-pass filter G, M is the wavelet decomposition layer number, j is the decomposition layer number, and k is an integer controlling the wavelet translation; S5.

3. Establish a photovoltaic output prediction model, as follows: S5.3.

1. Select the output data before the prediction time ts minutes to construct the historical output input matrix, together with the temperature, time and visual parameters at time t as the input vector, and the different components at time t as the output values ​​to construct the data set, where s is the length of the historical output time series, s≥60, as shown below: ； ； in, x i is the input vector, E 0t-s for ts The effort of the moment, t i for i The normalized minute number corresponding to the time, T i for i The temperature of the moment, K ri for i The color temperature value at the moment, RGB i for i The color parameters of the moment, G sri for i Infrared gray value at the moment; y 1i , y 2i , y 3i They are i The output component vector at time , CA1 i , CD1 i , CD2 i for i The low frequency component, high frequency component 1 and high frequency component 2 at the moment; S5.3.

2. Construct prediction model: Use LSTM algorithm. The LSTM algorithm model specifically includes forget gate, input gate and output gate. The forget gate formula is as follows: ； in f t for t Always choose the information to keep, W f is the weight matrix of the forget gate, x t express t The input content at the time, h t-1 for t-1 Short-term memory of the moment, b f represents the bias parameter of the forget gate, σ is the sigmoid function; The input gate formula is as follows: ； In the formula, for t Initial long-term memory of the moment, i t for t The weight of information that long-term memory chooses to retain at any given moment, W i is the information weight matrix of the input gate, W c is the weight matrix of the initial long-term memory, x t express t The input content at the time, h t-1 express t-1 Short-term memory of the moment, b i is the bias parameter of the input gate, b c is the bias parameter of the preliminary long-term memory; The output gate formula is as follows: ； in C t express t Long-term memory of the moment, C t-1 for t-1 Long-term memory of the moment, o t for t Output content at all times. h t express t Short-term memory of the moment; The prediction models of CA1, CD1, and CD2 were constructed respectively, and the data sets were divided into training set, test set, and validation set in the ratio of 70%, 10%, and 20%, respectively, and substituted into the LSTM model for pre-training; S5.

4. Use the trained prediction model to predict each component, then reconstruct it to get the normalized output prediction result, and finally denormalize the prediction result to get the final predicted output. The formula is as follows: ； E 0t for t The effort of the moment, CA1 t , CD1 t , CD2 t for t The low-frequency component, high-frequency component 1 and high-frequency component 2 at the moment, E Otmax and E Otmax are the minimum and maximum output respectively.

9. The method for estimating the output of multi-energy complementary photovoltaic power based on vision according to claim 1, characterized in that: The step S6 specifically includes the following sub-steps: S6.

1. Input the photovoltaic output estimation result into the multi-energy complementary system, and realize the multi-energy complementary function according to the multi-energy complementary method and the system goal; S6.

2. Input the ultra-short-term prediction results of photovoltaic output into the multi-energy complementary system, and realize the multi-energy complementary function according to the multi-energy complementary method and the system objectives; S6.

3. Based on the photovoltaic output estimation results and combined with ultra-short-term forecasts, the volatility index of regional photovoltaic output is analyzed to describe the volatility; the volatility index is used to improve the ultra-short-term forecast accuracy, as shown in the following formula: ； Where: v is the photovoltaic output volatility index, Eo It is the real-time output estimation result of the photovoltaic panel to be tested; Eo' It is the ultra-short-term prediction result of the photovoltaic panel to be tested at the next moment; S6.

4. The photovoltaic ultra-short-term prediction result at the previous time point is determined by the photovoltaic output estimation result at the next time point, and combined with the aforementioned volatility index to improve the accuracy of the prediction, as shown in the following formula: ； Where: yes t The prediction correction coefficient of photovoltaic output at the moment; v is the PV output volatility index; a, b is the adjustment factor for the forecast correction factor calculation; Eo t The photovoltaic panel to be tested is t Output estimation results at the moment; Eo t+1 is the estimated output result of the photovoltaic panel to be tested at time t+1; Eo' t The photovoltaic panel to be tested is t The ultra-short-term prediction result of the next moment of the moment prediction; Eo' t+1 It is the ultra-short-term prediction result of the next moment predicted by the photovoltaic panel to be tested at time t+1; Eo φ t+1 yes t +1 The revised photovoltaic ultra-short-term forecast results at time; a , b }~min( Eo t+2 -Eo φ t+1 ) refers to the goal of minimizing the difference between the photovoltaic output estimation result at time t+2 and the photovoltaic ultra-short-term prediction result corrected at time t+1, and the adjustment coefficient a, b ; That is, by adjusting a, b , improving the accuracy of ultra-short-term forecasts.

Images