Ultra-short-term prediction system and method based on image analysis

Abstract

The invention relates to the technical field of photovoltaic power prediction, in particular to an ultra-short-term prediction system and method based on image analysis, and the system comprises an all-sky imaging data processing module, a cloud layer dynamic feature calculation module, a solar irradiance change prediction module, a photovoltaic power generation power calculation module and a prediction model optimization training module. According to the method, the cloud layer thickness, transparency, movement direction and coverage rate change are calculated, the cloud layer dynamic characteristic analysis precision is enhanced, and the radiation scattering coefficient is calculated in combination with the direct solar light shielding ratio, the cloud layer characteristics and the solar angle, so that the irradiance calculation is more accurate, the illumination change rate is analyzed, and the short-term irradiance trend is dynamically adjusted; the total power is calculated and the shielding influence is adjusted in combination with component arrangement, so that the power prediction is more practical, the input data is normalized, the time sequence sample is enhanced, the dynamic adjustment of the model parameters is optimized, and the adaptive capacity and the prediction stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power prediction, and in particular to an ultra-short-term prediction system and method based on image analysis. Background Art

[0002] The field of photovoltaic power prediction technology includes the use of mathematical modeling, meteorological data analysis and image processing methods to predict the power generation of photovoltaic power stations. The core content of this technical field is to obtain and analyze environmental parameters at different time scales, combined with physical modeling and data-driven methods, to infer the output power of photovoltaic modules. Overall, photovoltaic power prediction technology mainly covers long-term predictions based on numerical weather forecasts, short-term predictions based on historical power data, and ultra-short-term predictions based on real-time environmental information. Among them, ultra-short-term predictions rely on high-frequency collected meteorological data, such as solar radiation, cloud changes, etc., to dynamically calculate and update photovoltaic power on a time scale of minutes, thereby improving the ability to adapt to intermittency and volatility.

[0003] Among them, the ultra-short-term prediction system based on image analysis refers to the use of a full sky imager to collect sky images, and based on image processing methods to analyze cloud distribution and movement trends, so as to predict the power generation of photovoltaic power stations. This patent subject is aimed at the problem of power fluctuations in photovoltaic power stations within a short time scale, covering technical matters such as image acquisition, cloud segmentation, cloud motion analysis and irradiance estimation. The specific method includes using a full sky imager to obtain sky images, using color segmentation and edge detection methods to extract cloud areas, calculating cloud motion vectors based on the optical flow method, and combining the solar irradiance model to estimate the photovoltaic irradiance changes in a short period of time, so as to predict the photovoltaic power output in the next few minutes to tens of minutes.

[0004] Relying on traditional cloud analysis and irradiance estimation methods, the lack of joint consideration of cloud thickness and transparency leads to reduced prediction accuracy under complex weather conditions. Real-time data is not fully utilized for dynamic adjustment in photovoltaic power forecasting, and power forecasting is unstable under rapidly changing cloud cover and light conditions. In addition, existing models mostly use fixed parameters and lack adaptive adjustments to environmental changes, which affects the reliability and adaptability of long-term forecasts. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose an ultra-short-term prediction system and method based on image analysis.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: The ultra-short-term prediction system based on image analysis includes:

[0007] The all-sky imaging data processing module acquires all-sky imaging data, sets the shooting frequency, adjusts the exposure time, performs denoising and filtering processing, segments the sky area, and obtains denoised and segmented image data;

[0008] The cloud dynamic feature calculation module calculates the cloud thickness, transparency, analyzes the cloud movement direction, calculates the change in coverage rate based on the denoised and segmented image data, obtains cloud movement feature parameters, and establishes an initial prediction model and its parameters;

[0009] The solar irradiance change prediction module calculates the ratio of direct solar light occlusion based on the cloud movement feature parameters, combines the cloud thickness, transparency, and solar angle to calculate the radiation scattering coefficient, adjusts the clear sky irradiance curve, calculates the light attenuation ratio, analyzes the light change rate in adjacent time windows, and obtains the short-term irradiance change trend;

[0010] The photovoltaic power calculation module calculates the power generation per unit area based on the short-term irradiance change trend, combines the module layout to calculate the total power, adjusts the occlusion effect, and obtains the photovoltaic power prediction value;

[0011] The prediction model optimization and training module normalizes the input data, enhances the time series samples, dynamically adjusts the prediction model and its parameters based on the photovoltaic power prediction value, and records the dynamically optimized prediction model parameters.

[0012] The denoised and segmented image data includes denoised images, segmented images, and sky area images. The cloud movement feature parameters include cloud thickness, transparency, movement direction, and change in coverage rate. The short-term irradiance change trend includes the ratio of direct solar light occlusion, radiation scattering coefficient, light attenuation ratio, and light change rate. The photovoltaic power prediction value includes power generation per unit area, total power generation, and occlusion-adjusted power. The dynamically optimized prediction model parameters include normalized input data, time series samples, prediction model structure, and prediction model parameters.

[0013] As a further solution of the present invention, the all-sky imaging data processing module includes:

[0014] The imaging data acquisition sub-module acquires all-sky imaging data, sets the shooting frequency, adjusts the exposure time, ensures that the imaging in multiple time periods has the same lighting reference, and obtains imaging data;

[0015] The denoising and filtering processing sub-module calculates the pixel gray value distribution based on the imaging data, screens out abnormal pixel areas, adjusts the abnormal pixel values according to the mean value of adjacent pixels, and uses the formula:

[0016]

[0017] Denoises all pixel points to obtain denoised image data;

[0018] Among them, D adj represents the pixel adjustment value after denoising, I orig is the original grayscale value, N is the number of neighborhood pixels, I′ i is the pixel value of the i-th neighborhood pixel, W noise is the noise adjustment weight, calculated based on the grayscale uniformity;

[0019] The sky area segmentation sub-module calls the denoised image data, calculates the color channel ratio, screens out the areas with prominent proportions of blue and gray, sets the area boundaries, excludes the areas that do not meet the range, and obtains the denoised segmented image data.

[0020] As a further solution of the present invention, the cloud dynamic feature calculation module includes:

[0021] The cloud thickness calculation sub-module is based on the denoised segmented image data, calls the pixel intensity distribution of the cloud-covered area, calculates the cloud optical thickness of different height layers, and uses the formula:

[0022]

[0023] Performs operations to obtain cloud thickness data;

[0024] Among them, H c represents the cloud thickness, I i represents the pixel intensity of the i-th layer, λ i represents the attenuation coefficient of the i-th layer corresponding wavelength, P i represents the air pressure value of the i-th layer, T i represents the temperature value of the i-th layer, V i represents the wind speed value of the i-th layer, Δd represents the height layer interval, n′ c represents the number of height layers;

[0025] The cloud transparency calculation sub-module calls the cloud thickness data, calculates the transmittance of different bands, obtains the average value of the transmittance, and obtains the cloud transparency data;

[0026] The cloud motion parameter analysis sub-module calls the cloud transparency data, combines the time series images to calculate the cloud motion direction and the change rate of the coverage rate, and obtains the cloud motion characteristic parameters.

[0027] As a further solution of the present invention, the solar irradiance change prediction module includes:

[0028] The cloud occlusion calculation sub-module is based on the cloud motion characteristic parameters, calculates the proportion of the cloud blocking sunlight, combines the cloud thickness and transparency to correct the occlusion ratio, and obtains the cloud occlusion ratio;

[0029] The radiation scattering coefficient calculation sub-module calculates the radiation scattering rate based on the cloud occlusion ratio, in combination with cloud thickness, transparency, and solar angle, using the formula:

[0030]

[0031] Performs operations to obtain the radiation scattering coefficient, and adjusts the light attenuation ratio in combination with the clear sky irradiance curve to obtain the adjusted irradiance;

[0032] Among them, S c represents the radiation scattering coefficient, T′ c represents the cloud transparency, H c represents the cloud thickness, D s represents the cloud occlusion ratio, A c,i represents the scattering influence factor within the differential time window, n′ c represents the number of time windows;

[0033] The short-term irradiance change trend analysis sub-module calculates the irradiance change rate of adjacent time windows based on the adjusted irradiance, and analyzes and obtains the short-term irradiance change trend.

[0034] As a further solution of the present invention, the photovoltaic power calculation module includes:

[0035] The unit area power generation calculation sub-module calls the photovoltaic module conversion efficiency and ambient temperature correction coefficient based on the short-term irradiance change trend, using the formula:

[0036]

[0037] to obtain the unit area power generation value;

[0038] Among them, P u represents the unit area power generation value, η represents the photovoltaic module conversion efficiency, I r represents the irradiance value calculated from the short-term irradiance change trend, k t represents the temperature correction coefficient, T′ c represents the current temperature of the module, T r represents the standard test temperature, and A represents the unit area of the module;

[0039] The total power calculation sub-module calculates the total power of the photovoltaic system based on the unit area power generation value, in combination with the photovoltaic module layout and the total area of the modules, to obtain the total power value of the photovoltaic modules;

[0040] The occlusion influence adjustment sub-module calls the total power value of the photovoltaic modules, combines the module occlusion influence coefficient, corrects the calculation result, and adjusts the photovoltaic power output to obtain the photovoltaic power prediction value.

[0041] As a further solution of the present invention, the prediction model optimization training module includes:

[0042] The data normalization sub-module calculates the mean normalization value based on the photovoltaic power prediction value, adjusts the scale range by using min-max normalization, and obtains the normalized power data;

[0043] The time series enhancement sub-module constructs a time sliding window based on the normalized power data, extracts the change trend, calculates the change rate, cumulative change amount and increment, dynamically weights and adjusts the weights, enhances the short-term change characteristics, and obtains the enhanced time series samples;

[0044] The dynamic parameter optimization sub-module calculates the model prediction error based on the enhanced time series samples, using the formula:

[0045]

[0046] Dynamically adjusts the model parameters, optimizes the update step size, reduces the prediction error, and obtains the dynamically optimized prediction model parameters;

[0047] Wherein, P opt represents the dynamically optimized prediction model parameters, P t represents the power data at the current time step, P t-1 represents the power data at the previous time step, E t represents the current error value, E min represents the previous minimum error, α represents the error influence factor, β represents the error sensitivity factor, n′ c represents the total number of time steps.

[0048] Ultra-short-term prediction method based on image analysis, the ultra-short-term prediction method based on image analysis is executed based on the above-mentioned ultra-short-term prediction system based on image analysis, and includes the following steps:

[0049] S1: Obtain all-sky imaging data, set the shooting frequency, adjust the exposure time, perform denoising and filtering processing in combination with the illuminance distribution, segment the sky area based on the color distribution and edge detection algorithm, and screen the pixel areas containing cloud information to obtain the denoised segmentation image data;

[0050] S2: Based on the denoised segmentation image data, extract the cloud edge contour, calculate the pixel displacement vector between consecutive frames, obtain the cloud movement direction and speed, call the cloud coverage rate to calculate the proportion of the cloud area, calculate the direct solar radiation occlusion ratio in combination with the transparency change, and analyze the influence of clouds at different heights on the occlusion ratio to obtain the cloud movement characteristic parameters;

[0051] S3: Based on the cloud motion characteristic parameters, calculate the ratio of direct sunlight occlusion, calculate the radiation scattering coefficient by combining the cloud thickness, transparency and solar angle, call the clear sky irradiance curve to adjust the direct light intensity, calculate the light attenuation rate, analyze the light change rate within a continuous time window, and obtain the short-term irradiance change trend;

[0052] S4: Based on the short-term irradiance change trend, calculate the power generation power of a unit area of photovoltaic modules, calculate the power generation power of the overall photovoltaic array by combining the module layout, adjust the occlusion effect, and obtain the photovoltaic power prediction value;

[0053] S5: Based on the photovoltaic power prediction value, normalize the input data, enhance the time series samples, adjust the time step, retrain the prediction model for different weather conditions, update the parameter weights, and obtain the dynamically optimized prediction model parameters.

[0054] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0055] In the present invention, by calculating the changes in cloud thickness, transparency, motion direction and coverage rate, the dynamic characteristics of clouds are made more comprehensive, and combined with the initial prediction model, the trend analysis on a short time scale is more accurate. Calculate the ratio of direct sunlight occlusion, and deduce the radiation scattering coefficient by combining cloud characteristics and solar angle, making the irradiance calculation more targeted. Analyze the light change rate of adjacent time windows, realize the dynamic adjustment of the short-term irradiance change trend, and improve the accuracy of light attenuation prediction. Calculate the power generation power per unit area, calculate the total power by combining the module layout, and adjust the occlusion effect, so that the power prediction fully considers the module layout characteristics and environmental factors, and improves the practicability. Normalize the input data and enhance the time series samples, optimize the dynamic adjustment of model parameters, improve the model adaptability, and enhance the prediction stability. Description of the Drawings

[0056] Figure 1 is the system flow chart of the present invention;

[0057] Figure 2 is the flow chart of the all-sky imaging data processing module of the present invention;

[0058] Figure 3 is the flow chart of the cloud dynamic characteristic calculation module of the present invention;

[0059] Figure 4 is the flow chart of the solar irradiance change prediction module of the present invention;

[0060] Figure 5 is the flow chart of the photovoltaic power generation calculation module of the present invention;

[0061] Figure 6 is the flow chart of the prediction model optimization training module of the present invention. Detailed implementation manners

[0062] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0063] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In addition, in the description of the present invention, "a plurality of" means two or more, unless otherwise specifically defined.

[0064] Embodiment 1

[0065] Please refer to Figure 1 , the present invention provides a technical solution: the ultra-short-term prediction system based on image analysis includes:

[0066] The all-sky imaging data processing module acquires all-sky imaging data, sets the shooting frequency, adjusts the exposure time, performs denoising and filtering processing, segments the sky area, and obtains denoised and segmented image data;

[0067] The cloud dynamic feature calculation module calculates the cloud thickness, transparency, analyzes the cloud movement direction, calculates the change in coverage rate based on the denoised and segmented image data, obtains cloud movement feature parameters, and establishes an initial prediction model and its parameters;

[0068] The solar irradiance change prediction module calculates the ratio of direct sunlight occlusion based on the cloud movement feature parameters, combines the cloud thickness, transparency, and solar angle to calculate the radiation scattering coefficient, adjusts the clear sky irradiance curve, calculates the light attenuation ratio, and analyzes the light change rate in adjacent time windows to obtain the short-term irradiance change trend;

[0069] The photovoltaic power calculation module calculates the power generation per unit area based on the short-term irradiance change trend, combines the component layout to calculate the total power, adjusts the occlusion effect, and obtains the photovoltaic power prediction value;

[0070] The prediction model optimization and training module normalizes the input data, enhances the time series samples, dynamically adjusts the prediction model and its parameters based on the photovoltaic power prediction value, and records the dynamically optimized prediction model parameters.

[0071] The denoised and segmented image data includes denoised images, segmented images, and sky region images. The cloud motion feature parameters include cloud thickness, transparency, motion direction, and coverage change. The short-term irradiance change trend includes direct sunlight occlusion ratio, radiation scattering coefficient, light attenuation ratio, and light change rate. The predicted photovoltaic power values include power generation per unit area, total power generation, and occlusion adjustment power. The dynamic optimization prediction model parameters include normalized input data, time series samples, prediction model structure, and prediction model parameters.

[0072] Please refer to Figure 2 , the all-sky imaging data processing module includes:

[0073] The imaging data acquisition sub-module acquires all-sky imaging data, sets the shooting frequency, adjusts the exposure time, and ensures that the imaging in multiple time periods has the same lighting reference to obtain the imaging data;

[0074] First, for all-sky imaging in different time periods, a fixed shooting frequency needs to be set to ensure the continuity of the imaging data in each time period. For example, the shooting interval can be set to 10 seconds, 30 seconds, or 60 seconds, and the shooting frequency is adjusted according to weather conditions to avoid data discontinuity caused by cloud movement. Subsequently, the exposure time is adjusted, which needs to be adjusted in real time according to the ambient light intensity to maintain the consistency of the imaging data. For example, when the light intensity is high during the day, the exposure time can be set to 5ms, while when the light is weak at night or on cloudy days, the exposure time can be set to 50ms. The unification of the lighting reference can be achieved by setting a gray-scale reference value. For example, the average brightness value in the image is selected and the deviation is calculated. During the imaging process in different time periods, the exposure time is adjusted so that the average brightness of each image remains in the same range, such as set to 120 - 130 gray levels (0 - 255 scale). The specific operation can adopt the method of histogram matching, that is, calculate the image histogram and adjust the exposure parameters to make the brightness distribution of all images as consistent as possible, and obtain all-sky imaging data that meets the lighting reference.

[0075] The denoising and filtering processing sub-module calculates the pixel gray value distribution based on the imaging data, screens out abnormal pixel regions, and adjusts the abnormal pixel values according to the mean of adjacent pixels, using the formula:

[0076]

[0077] Denoise all pixel points to obtain denoised image data;

[0078] Among them, D adj represents the adjusted pixel value after denoising, I orig is the original gray value, N is the number of neighboring pixels, I′ i is the i-th neighboring pixel value, and W noise is the noise adjustment weight, calculated according to the gray-scale uniformity;

[0079] First, to calculate the distribution of pixel grayscale values, all pixel points in the image need to be traversed, and the frequencies of each grayscale level are counted to generate a grayscale histogram. For example, the grayscale values of pixels in a certain area may be concentrated between 80 and 150, while the grayscale values of some abnormal pixels may be higher than 220 or lower than 30. Next, to screen the abnormal pixel areas, the mean deviation of each pixel from its neighboring pixels can be calculated. If the deviation exceeds the set abnormal threshold, the pixel is determined to be abnormal. For example, if the grayscale deviation threshold is set to 30, when the grayscale value of a certain pixel is 200 and the mean of its neighboring pixels is 160, then its deviation is 40, which is greater than 30, and it is determined to be an abnormal pixel. Subsequently, the abnormal pixel values are adjusted based on the mean of the neighboring pixels, adopting a weighted mean adjustment strategy. For example, for an abnormal pixel, its adjustment value D adj is calculated as follows:

[0080]

[0081] Assume the original pixel grayscale value I orig is 200, the number of neighboring pixels N is 8, and the grayscale values of the neighboring pixels are respectively

[0082] [160, 162, 159, 161, 163, 165, 160, 162], then the mean of the neighboring pixels is

[0083]

[0084] If the noise adjustment weight W noise is set to 0.8, then the adjusted pixel value is calculated as follows:

[0085] D adj = |200 - 161.5| × 0.8 = 30.8;

[0086] Finally, the pixel value after denoising is adjusted to

[0087] I adj = I orig - D adj = 200 - 30.8 = 169.2;

[0088] After this process, all abnormal pixel points are denoised, and the denoised image data is obtained.

[0089] The results show that after denoising filtering, the abnormal pixels in the original image are effectively adjusted, making the pixel gray values closer to the average of their surrounding pixels, thereby reducing the mutated pixel points caused by noise and making the overall gray distribution of the image smoother. At the same time, by calculating the deviation of the gray value mean between the abnormal pixels and their neighboring pixels and using the noise adjustment weight to correct the abnormal pixels, the data distribution of the denoised image becomes more uniform. From the calculation results of the finally adjusted pixel values, the abnormally high or low pixel values are effectively corrected, making the data of the entire image more stable and providing more accurate input data for the subsequent sky region segmentation.

[0090] The sky region segmentation sub-module calls the denoised image data, calculates the color channel ratio, filters out the regions with prominent blue and gray proportions, sets the region boundaries, excludes the regions that do not meet the range, and obtains the denoised segmentation image data.

[0091] First, extract the RGB channel data of the image and calculate the blue channel ratio of each pixel using the following calculation formula:

[0092]

[0093] For example, for a certain pixel with RGB values of (50, 100, 200) respectively, its blue channel ratio is calculated as follows:

[0094]

[0095] Filter out the regions with prominent blue and gray proportions. Set the blue region determination threshold to 0.5, that is, pixels with a blue channel ratio greater than 0.5 are determined as sky pixels. At the same time, for the gray region, the R, G, and B channel values need to be similar. For example, if the difference between the R, G, and B channels is less than 15, it is determined as a gray region. Set the region boundaries and, based on the connected region analysis method, determine whether adjacent pixels belong to the same region. For example, if the continuous pixels in a certain region meet the blue channel ratio or gray determination conditions, they are classified into the same region. Finally, exclude the regions that do not meet the range. For example, if a region is smaller than the set minimum area threshold (such as 100 pixels), it is determined as a noise region and excluded, and finally the denoised segmentation image data is obtained.

[0096] Table 1 Color Channel Ratio Screening Parameter Table

[0097]

[0098] As shown in Table 1, the blue channel ratio threshold, the gray region determination standard, and the minimum region area threshold can all be used to screen the sky region.

[0099] Please refer to Figure 3 , the cloud dynamic feature calculation module includes:

[0100] The cloud thickness calculation sub-module calculates the cloud optical thickness of different height layers by calling the pixel intensity distribution of the cloud-covered area based on the denoised and segmented image data, using the formula:

[0101]

[0102] Performs operations to obtain cloud thickness data;

[0103] where H c represents the cloud thickness, I i represents the pixel intensity of the i-th layer, λ i represents the attenuation coefficient corresponding to the wavelength of the i-th layer, P i represents the air pressure value of the i-th layer, T i represents the temperature value of the i-th layer, V i represents the wind speed value of the i-th layer, Δd represents the height layer interval, and n′ c represents the number of height layers;

[0104] First, extract the pixel intensity distribution of the cloud-covered area. This process filters out the pixel points with brightness values exceeding the set threshold as the cloud area by traversing all pixel points in the image data. During this process, the brightness threshold is set to 200 (taking the 0 - 255 gray level), and the pixel points greater than 200 are selected as the cloud area. Subsequently, these pixel points are classified and stratified according to the height layer interval Δd. Set Δd = 100m for height layer division. For each height layer, obtain its pixel intensity I i , wavelength attenuation coefficient λ i , air pressure P i , temperature T i and wind speed V i and calculate their contributions to the cloud thickness respectively. During the calculation process, an optical band with a wavelength range of 0.4μm - 1.2μm is used for measurement. The attenuation coefficient λ i of different wavelengths takes values in the range of 0.02 - 0.05 according to the measured data. The air pressure value P i is adjusted according to the standard atmospheric pressure of different height layers. For example, the sea-level air pressure is taken as 1013.25hPa, and the air pressure drops by about 12hPa for every 100m increase in height. Therefore, when the height layer is 1000m, P i = 1013.25 - 12×10 = 893.25hPa. The temperature T i is assigned according to the actual measured value of the height layer. For example, the temperature of the 1000m height layer is set to 10℃ (283.15K), and the wind speed V iTake the measurement range of 5m / s - 15m / s according to the laminar or turbulent flow situation. During the calculation process, accumulate each contribution term layer by layer, and finally multiply by the height layer interval Δd for cumulative calculation to obtain the cloud layer thickness H c , for example, when I 1 = 22, λ 1 = 0.03, P 1 = 1000hPa, T 1 = 285K, V 1 = 6m / s, the calculation result is The specific calculation is as follows:

[0105]

[0106] This result indicates that under the current parameter conditions such as pixel intensity, air pressure, temperature, wind speed, etc., the optical thickness of the cloud layer is relatively large. This means that the cloud layer in this area may be relatively dense, having a greater impact on the transmission of light, and thus may lead to a lower surface light intensity. At the same time, this value can be used for subsequent calculation of cloud layer transparency to further analyze the transmission situation of different bands.

[0107] Table 2 Example table of cloud layer thickness calculation parameters

[0108]

[0109] As shown in Table 2, for the parameter values of different height layers, the formula is used to calculate and accumulate the thickness contributions of each layer, and finally the overall cloud layer thickness is obtained.

[0110] The cloud layer transparency calculation sub-module calls the cloud layer thickness data, calculates the transmittance of different bands, obtains the average value of the transmittance, and gets the cloud layer transparency data;

[0111] First, calculate the transmittance within each band range. The transmittance of each band is obtained by calculating the proportion of the cloud layer thickness affecting light transmission. The specific calculation method is to calculate the transmittance for each band (0.4μm, 0.6μm, 0.8μm, 1.0μm, 1.2μm) one by one. The transmittance calculation formula is where τ i represents the band attenuation factor, and its value range of 0.01 - 0.03 is determined according to experimental measurements. For example, when the band is 0.6μm, take τ i = 0.02, and when the calculated value of the cloud layer thickness is 738672, the transmittance calculation is as follows:

[0112] T 0.6 = e -0.02×738672 ;

[0113] = e -14773.44 ;

[0114] ≈0;

[0115] The calculated transmittance in this wavelength band is close to 0, indicating that the cloud layer almost completely absorbs the light in this band. Subsequently, the cloud layer transparency is obtained by calculating the transmittance of all wavelength bands and then taking the average value. For example, after calculating the transmittance of multiple wavelength bands, assuming that the transmittance of the 0.4μm wavelength band is 0.05, the transmittance of the 0.8μm wavelength band is 0.02, and the transmittance of the 1.2μm wavelength band is 0.08, the cloud layer transparency is calculated as follows:

[0116]

[0117] This result shows that the cloud layer has a weak ability to transmit light of different wavelength bands, most of the light is absorbed or scattered, and only 5% of the light can pass through the cloud layer and reach the ground. This low transparency value indicates that the cloud layer is thick and has a strong attenuation effect on light in the visible and near-infrared wavelength bands. This result can be used for subsequent analysis of cloud movement characteristics and combined with time-series images to analyze the change in cloud coverage.

[0118] The cloud movement parameter analysis sub-module calls the cloud layer transparency data and combines time-series images to calculate the cloud movement direction and the change rate of the coverage rate, and obtains the cloud movement characteristic parameters.

[0119] First, perform inter-frame difference processing on the time-series images. By comparing the cloud layer transparency distributions of two consecutive frames of images, calculate the cloud movement direction. In this process, the pixel point coordinate differences are used to calculate the cloud displacement vector. For example, assume that at time t 1 the coordinate of the center point of a certain cloud layer is (x1, y1), and at time t 2 it moves to (x2, y2), then calculate the cloud movement vector:

[0120] ΔX = x2 - x1, ΔY = y2 - y1;

[0121] Assume that (x1, y1) = (100, 200), (x2, y2) = (120, 210), then:

[0122] ΔX = 120 - 100 = 20, ΔY = 210 - 200 = 10;

[0123] The calculated cloud movement vector is (20, 10). Further, the change trend of this vector is calculated through multiple frames of images to obtain the cloud movement direction. Subsequently, by statistically calculating the change rate of the proportion of the cloud coverage area in consecutive images, calculate the change rate of the cloud coverage rate. For example, at time t 1 the area of the cloud coverage area is 5000 pixel points, and at time t 2 it decreases to 4800 pixel points, then the change rate of the coverage rate:

[0124]

[0125] The results indicate that the cloud layer moves in the direction of (20, 10) in the continuous time series, and at the same time, the cloud cover area decreases by 4%. This result implies that the cloud layer has undergone displacement during this time period, and the cloud layer may be dissipating, resulting in a decrease in the coverage rate. This data is of great significance for fields such as weather forecasting, aviation navigation, and satellite monitoring, and can be further used to analyze the overall movement trend of the cloud layer and the weather change pattern.

[0126] Please refer to Figure 4 , the solar irradiance change prediction module includes:

[0127] Based on the cloud movement characteristic parameters, the cloud occlusion calculation sub-module calculates the proportion of sunlight occluded by the cloud layer, combines the cloud thickness and transparency to correct the occlusion ratio, and obtains the cloud occlusion ratio;

[0128] First, it is necessary to collect basic data such as the cloud movement speed, cloud cover range, and cloud density within the target area. Among them, the cloud movement speed can be measured by an optical observation device for the position change of the cloud layer at different times. For example, the longitude and latitude changes of the cloud layer recorded at two times t1 = 12:00 and t2 = 12:05 can be used to calculate its average movement speed. Assuming that the cloud layer moves 2 kilometers in 5 minutes, the calculated cloud speed is The cloud cover range can be obtained by satellite remote sensing image segmentation processing to obtain the coverage rate. For example, if the cloud cover area over a certain area at the current moment is 800 km2, and the total area of this area is 1000 km2, then the cloud coverage rate C c is 80%. The cloud density can be obtained by an optical thickness measuring instrument. Assuming the measured value is H c = 3.5. After obtaining these parameters, according to the current solar azimuth angle and its projection position in the sky, combined with the solar irradiance intensity I s (such as the solar irradiance I s = 1000 W / m2 under clear sky conditions at noon in a certain area), calculate the proportion of sunlight occluded by the cloud layer, and use the change in the illumination intensity of the occluded area for comparison. For example, compare the surface irradiance changes under cloudless conditions and cloud cover conditions. Assuming that the surface irradiance measured without clouds is 900 W / m2 and it drops to 450 W / m2 after being occluded by the cloud layer, then the preliminary calculated cloud occlusion ratio is Combined with the cloud thickness H c and transparency T' c for correction. For example, the transparency T' c = 0.6 affects the occlusion ratio, making the final cloud occlusion ratio D s ' = D s × T' c = 0.5 × 0.6 = 0.3, and finally the cloud occlusion ratio 0.3 is obtained.

[0129] This ratio indicates that under the influence of the current cloud thickness and transparency, approximately 30% of the solar radiation is blocked by the clouds, affecting the light intensity on the earth's surface. This value can be used as an input parameter for subsequent radiation scattering calculations to further analyze the specific impact of clouds on light attenuation.

[0130] Based on the cloud occlusion ratio, the radiation scattering coefficient calculation sub-module calculates the radiation scattering rate by combining the cloud thickness, transparency, and solar angle, using the formula:

[0131]

[0132] Performs operations to obtain the radiation scattering coefficient, adjusts the light attenuation ratio in combination with the clear sky irradiance curve, and obtains the adjusted irradiance;

[0133] Among them, S c represents the radiation scattering coefficient, T′ c represents the cloud transparency, H c represents the cloud thickness, D s represents the cloud occlusion ratio, A c,i represents the scattering influence factor within the differential time window, n′ c represents the number of time windows;

[0134] Formula:

[0135]

[0136] First, the combined influence of the cloud transparency T′ c and thickness H c on the scattering process needs to be considered. For example, the transparency determines the intensity of the transmitted light, while the thickness affects the multiple scattering effect. Combining the influence of the solar azimuth angle θ s on the scattering angle (such as the current solar elevation angle θ s = 60°), calculate the radiation scattering rate. Let the number of time windows n′ c = 3, and monitor the scattering influence factor A c,i within each time window. For example, the A c,i values for the three time windows are 0.2, 0.3, and 0.25 respectively, then calculate the root mean square value Substitute into the formula where T′ c = 0.6, H c = 3.5, D s ′ = 0.3, then the calculation gives Adjust the light attenuation ratio in combination with the clear sky irradiance curve. For example, the irradiance at a certain moment in clear sky conditions is 800 W / m2, and the calculated scattering coefficient S c = 0.343, then the adjusted irradiance is Iadj = 800×(1 - S c ) = 800×(1 - 0.343) = 800×0.657 = 525.6 W / m²。

[0137] This result indicates that under the combined action of the current cloud transparency, thickness, and occlusion ratio, the actual irradiance on the ground is reduced by 34.3% compared to the clear sky condition, that is, it is significantly affected by scattering and occlusion. The adjusted irradiance can be used as the basic data for subsequent analysis of the short-term irradiance change trend.

[0138] The short-term irradiance change trend analysis sub-module calculates the irradiance change rate of adjacent time windows based on the adjusted irradiance, and analyzes and obtains the short-term irradiance change trend.

[0139] First, it is necessary to record its change rate within multiple consecutive time windows. For example, let the time window interval Δt = 10 min, and the adjusted irradiances measured at three consecutive moments are I adj1 = 550 W / m², I adj2 = 500 W / m², I adj3 = 450 W / m², then calculate the irradiance change rate R adj :

[0140] The first window:

[0141] The second window:

[0142] According to the trend of the change rate for analysis, if R adj is negative in two consecutive time windows and the absolute values are maintained in a similar range, it indicates that the irradiance shows a continuous downward trend. Combining historical data, it can be further judged whether the change amplitude exceeds the set threshold. For example, set the change threshold R thresh = -6 W / m²·min, and all current R adj values do not exceed this threshold, indicating that the downward trend of the irradiance is within the normal range. If the R adj calculated for a certain window is lower than R thresh , then it is necessary to further analyze the influence caused by cloud changes. For example, combine the change of the cloud occlusion ratio D s ' within the same time window to verify whether the light attenuation is related to the rapid movement of the clouds, and finally obtain the short-term irradiance change trend.

[0143] The calculated irradiance change rates R adj1 = -5 W / m²·min, R adj2 = -5 W / m²·min, and neither exceeds the set change threshold R thresh=-6 W / m2·min, indicating that the current short-term irradiance change trend is stable, within the normal decline range, and there is no abnormally drastic light attenuation. If the change rate exceeds the threshold within the subsequent time window, it may mean that the rapid change of clouds has a more significant impact on the light, and it is necessary to further analyze the specific impact of cloud movement characteristics on light change.

[0144] Please refer to Figure 5 , the photovoltaic power calculation module includes:

[0145] The power generation power per unit area calculation sub-module, based on the short-term irradiance change trend, calls the photovoltaic module conversion efficiency and environmental temperature correction coefficient, and uses the formula:

[0146]

[0147] to obtain the power generation power value per unit area;

[0148] where, P u represents the power generation power value per unit area, η represents the photovoltaic module conversion efficiency, I r represents the irradiance value calculated from the short-term irradiance change trend, k t represents the temperature correction coefficient, T′ c represents the current temperature of the module, T r represents the standard test temperature, A represents the unit area of the module;

[0149] In the specific execution process, first, obtain the short-term irradiance change trend data, which is continuously collected by the light intensity monitoring device within a certain period of time. The instantaneous irradiance value is recorded every 5 minutes, and the average irradiance change rate within the past 30 minutes is calculated. For example, at a certain moment, the irradiance values recorded by the light intensity monitoring device at 5-minute intervals are 800 W / ㎡, 820 W / ㎡, 830 W / ㎡, 850 W / ㎡, 860 W / ㎡, 880 W / ㎡ respectively. Then the irradiance change rate within the past 30 minutes is calculated as follows:

[0150]

[0151] Secondly, call the photovoltaic module conversion efficiency η, which is usually determined according to the module type and material. For example, the conversion efficiency of monocrystalline silicon modules is usually in the range of 18%-22%, and that of polycrystalline silicon modules is about 15%-18%. Assume that in this example, monocrystalline silicon modules are used and the conversion efficiency is set to 20%. Then call the temperature correction coefficient k t , which characterizes the impact of temperature change on the output efficiency of the photovoltaic module. The general value range is 0.003 - 0.005 / ℃. In this example, k t is set to 0.004 / ℃. Next, obtain the current temperature T’ of the photovoltaic modulec With the standard test temperature T r , assuming the standard test temperature T r = 25 °C, and the current temperature is measured as 45 °C, then calculate the temperature correction factor:

[0152] 1 - k t ·|T' c - T r | = 1 - 0.004×|45 - 25| = 1 - 0.08 = 0.92;

[0153] Finally, according to the formula:

[0154]

[0155] Calculate the power generation per unit area, where the unit area A of the component is set to 1.6 m², and substitute the known values:

[0156]

[0157] This result shows that under the current irradiance level and temperature conditions, the power generation of each square meter of photovoltaic module is 101.2 W.

[0158] The total power calculation sub-module calculates the total power of the photovoltaic system based on the power generation per unit area value, combined with the arrangement of photovoltaic modules and the total area of the modules, and obtains the total power value of the photovoltaic modules;

[0159] In the specific implementation process, first determine the arrangement of the photovoltaic modules, such as horizontal arrangement, vertical arrangement or matrix arrangement, and count the effective light-receiving area of the entire photovoltaic system. Assume that a certain photovoltaic power station adopts a horizontal arrangement method, with 50 photovoltaic modules arranged in each row, a total of 10 rows, and the area of each module is 1.6 m², then the total light-receiving area is calculated as follows:

[0160] A total = 50×10×1.6 = 800 m²;

[0161] Secondly, calculate the total power of the photovoltaic system based on the power generation per unit area:

[0162] P total = P u ×A total ;

[0163] Substitute the known values:

[0164] P total = 101.2 W / m²×800 m² = 80960 W = 80.96 kW;

[0165] This result shows that under the current irradiance and temperature conditions, the total power generation of the photovoltaic system is 80.96 kW.

[0166] The occlusion influence adjustment sub-module calls the total power value of the photovoltaic module, combines the component occlusion influence coefficient, corrects the calculation result, adjusts the photovoltaic power output, and obtains the photovoltaic power prediction value.

[0167] In the specific execution process, first determine the component occlusion situation, and use the occlusion monitoring system to record the occlusion ratio of the photovoltaic module. This ratio is calculated based on the proportion of the occluded area on the component surface to the entire component area. Suppose it is found through drone inspection and analysis that due to occlusion by surrounding buildings, the average occlusion ratio of some components is 10%, then the occlusion influence coefficient α s is set to 0.90, that is:

[0168] α s = 1 - 0.10 = 0.90;

[0169] Then calculate the corrected photovoltaic power prediction value:

[0170] P adjusted = P total ×α s ;

[0171] Substitute the known values:

[0172] P adjusted = 80.96kW × 0.90 = 72.86kW;

[0173] This result shows that after considering the occlusion factor, the predicted value of the actual power generation of the photovoltaic system is reduced to 72.86kW.

[0174] Table 3: Calculation parameters of the photovoltaic system

[0175]

[0176] As shown in Table 3, the parameters involved in this embodiment and their values are listed.

[0177] Please refer to Figure 6 , the prediction model optimization training module includes:

[0178] The data normalization sub-module calculates the mean normalization value based on the photovoltaic power prediction value, adjusts the scale range using min-max normalization, and obtains the normalized power data;

[0179] Using the min-max normalization method, all data is converted into the interval [0,1]. Suppose the original photovoltaic power prediction value at a certain moment is P = [420, 500, 610, 390, 580]W. First, calculate the mean Standard deviation Calculate the mean normalization value After that, min-max normalization is performed, and the minimum value P is calculated. min = 390W, the maximum value P max = 610W, and the normalization formula is The normalized data P′ = [0.136, 0.5, 1, 0, 0.864] is obtained, and finally the normalized power data is acquired.

[0180] The calculation results show that the original photovoltaic power data has been successfully converted into [0.136, 0.5, 1, 0, 0.864], and the normalized data retains the relative relationship of the original data, ensuring the accuracy of subsequent time series analysis.

[0181] The time series enhancer sub-module constructs a time sliding window based on the normalized power data, extracts the change trend, calculates the change rate, cumulative change amount and increment, dynamically adjusts the weights, enhances the short-term change characteristics, and obtains the enhanced time series samples;

[0182] Set the window size w = 3 steps, and each time take three consecutive time step data to form a new sample. Extract the first window [0.136, 0.5, 1] from the normalized power dataset, calculate the change trend, and define the trend change rate as the difference between adjacent data. For example, ΔP 1 = 0.5 - 0.136 = 0.364, ΔP 2 = 1 - 0.5 = 0.5, and then calculate the change rate Calculate the cumulative change amount The incremental change is calculated as By dynamically adjusting the weights, define the adjustment coefficient γ as the mean of past changes. Assume that the incremental mean of the previous time window is 0.3, then the current adjustment coefficient Adjust the weights of the current time step data to obtain the enhanced time series sample [0.136×0.366, 0.5×0.366, 1×0.366] = [0.0497, 0.183, 0.366].

[0183] The calculation results show that the enhanced time series sample values are [0.0497, 0.183, 0.366], which are adjusted compared with the original data, highlighting the short-term power fluctuation situation, enabling the subsequent prediction model to better adapt to the change trend and improving the short-term prediction accuracy.

[0184] The dynamic parameter optimization sub-module calculates the model prediction error based on the enhanced time series samples, using the formula:

[0185]

[0186] Dynamically adjust the model parameters, optimize the update step size, reduce the prediction error, and obtain the dynamically optimized prediction model parameters;

[0187] Among them, P opt represents the parameters of the dynamic optimization prediction model, P t represents the power data at the current time step, P t-1 represents the power data at the previous time step, E t represents the current error value, E min represents the previous minimum error, α represents the error influence factor, β represents the error sensitivity factor, n′ c represents the total number of time steps.

[0188] Formula:

[0189]

[0190] Assume that the true power data at a certain time step is P = [420, 500, 610, 390, 580] W, and the model prediction data is Calculate the error to get [10, 10, 10, 10, 10] W, and obtain the historical minimum error E min = 5 W, take the error influence factor α = 0.1, the error sensitivity factor β = 0.05, and the total number of time steps n′ c = 5, substitute into the formula:

[0191]

[0192] Calculate the power change at each time step:

[0193] |P 2 - P 1 | = |500 - 420| = 80 W;

[0194] |P 3 - P 2 | = |610 - 500| = 110 W;

[0195] |P 4 - P 3 | = |390 - 610| = 220 W;

[0196] |P 5 - P 4 | = |580 - 390| = 190 W;

[0197] Calculate the error factor:

[0198] e -0.05·(10-5) = e -0.25 ≈ 0.7788;

[0199] 1 + 0.1×0.7788 = 1.07788;

[0200] Calculation of optimization parameters:

[0201]

[0202] The result shows that the parameters of the dynamic optimization prediction model are P opt = 556.4, which can be used to update the step size to adjust the prediction error, thereby obtaining the parameters of the dynamic optimization prediction model.

[0203] The ultra-short-term prediction method based on image analysis is executed based on the above ultra-short-term prediction system based on image analysis, and includes the following steps:

[0204] S1: Obtain all-sky imaging data, set the shooting frequency, adjust the exposure time, perform denoising and filtering processing in combination with the illuminance distribution, segment the sky area based on the color distribution and edge detection algorithm, and screen the pixel areas containing cloud information to obtain denoised and segmented image data;

[0205] S2: Based on the denoised and segmented image data, extract the cloud edge contour, calculate the pixel displacement vector between consecutive frames, obtain the cloud movement direction and speed, call the cloud coverage rate to calculate the proportion of the cloud area, calculate the direct solar radiation occlusion ratio in combination with the transparency change, and analyze the influence of clouds at different heights on the occlusion ratio to obtain cloud movement characteristic parameters;

[0206] S3: Based on the cloud movement characteristic parameters, calculate the direct solar radiation occlusion ratio, calculate the radiation scattering coefficient in combination with the cloud thickness, transparency and solar angle, call the clear sky irradiance curve to adjust the direct light intensity, calculate the light attenuation rate, and analyze the light change rate within a continuous time window to obtain the short-term irradiance change trend;

[0207] S4: Based on the short-term irradiance change trend, calculate the power generation power of a unit area of photovoltaic modules, calculate the overall power generation power of the photovoltaic array in combination with the module arrangement, and adjust the occlusion effect to obtain the photovoltaic power prediction value;

[0208] S5: Based on the photovoltaic power prediction value, normalize the input data, enhance the time series samples, adjust the time step, retrain the prediction model for different weather conditions, and update the parameter weights to obtain the parameters of the dynamic optimization prediction model.

[0209] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still belong to the protection scope of the technical solution of the present invention.

Claims

1. An ultra-short-term prediction system based on image analysis, characterized in that: The system comprises: The full sky imaging data processing module obtains full sky imaging data, sets the shooting frequency, adjusts the exposure time, performs denoising filtering, segments the sky area, and obtains denoised segmented image data; The cloud layer dynamic feature calculation module calculates the cloud layer thickness and transparency, analyzes the cloud layer movement direction, calculates the coverage change, obtains the cloud layer movement feature parameters, and establishes the initial prediction model and its parameters based on the denoised segmented image data; The solar irradiance change prediction module calculates the solar direct light shielding ratio based on the cloud motion characteristic parameters, calculates the radiation scattering coefficient in combination with the cloud thickness, transparency, and sun angle, adjusts the clear sky irradiance curve, calculates the light attenuation ratio, analyzes the light change rate in adjacent time windows, and obtains the short-term irradiance change trend; The photovoltaic power calculation module calculates the power generation per unit area based on the short-term irradiance change trend, calculates the total power in combination with the component arrangement, adjusts the shading effect, and obtains the photovoltaic power prediction value; The prediction model optimization training module normalizes input data, enhances time series samples, dynamically adjusts the prediction model and its parameters based on the photovoltaic power prediction value, and records the dynamically optimized prediction model parameters.

2. The ultra-short-term prediction system based on image analysis according to claim 1, characterized in that: The denoising and segmented image data includes a denoising image, a segmented image, and a sky area image. The cloud motion characteristic parameters include cloud thickness, transparency, motion direction, and coverage change. The short-term irradiance change trend includes the direct sunlight shielding ratio, radiation scattering coefficient, light attenuation ratio, and light change rate. The photovoltaic power prediction value includes the power generation per unit area, the total power generation, and the shielding adjustment power. The dynamic optimization prediction model parameters include normalized input data, time series samples, prediction model structure, and prediction model parameters.

3. The ultra-short-term prediction system based on image analysis according to claim 2, characterized in that: The all-sky imaging data processing module comprises: The imaging data acquisition submodule acquires all-sky imaging data, sets the shooting frequency, adjusts the exposure time, ensures that imaging in multiple time periods has the same illumination benchmark, and obtains imaging data; The denoising filter processing submodule calculates the pixel gray value distribution based on the imaging data, screens the abnormal pixel area, and adjusts the abnormal pixel value according to the mean value of the adjacent pixels, using the formula: De-noise all pixels to obtain de-noised image data; Among them, D adj Represents the pixel adjustment value after denoising, I orig is the original gray value, N is the number of neighborhood pixels, I′ i is the value of the ith neighborhood pixel, W noise Adjust the weight for noise, calculated based on grayscale uniformity; The sky area segmentation submodule calls the denoised image data, calculates the color channel ratio, selects the areas with prominent blue and gray proportions, sets the area boundaries, excludes the areas that do not meet the range, and obtains the denoised segmented image data.

4. The ultra-short-term prediction system based on image analysis according to claim 3 is characterized in that: The cloud layer dynamic feature calculation module includes: The cloud thickness calculation submodule uses the pixel intensity distribution of the cloud coverage area based on the denoised segmented image data to calculate the cloud optical thickness of the differentiated height layer using the formula: Calculate and obtain cloud thickness data; Among them, H c Represents the cloud thickness, I i represents the pixel intensity of the i-th layer, λ i represents the attenuation coefficient of the i-th layer corresponding to the wavelength, P i represents the air pressure value of the i-th layer, T i Represents the temperature value of the i-th layer, V i represents the wind speed value of the i-th layer, Δd represents the altitude layer interval, n′ c Represents the number of altitude layers; The cloud layer transparency calculation submodule calls the cloud layer thickness data, calculates the transmittance of the differentiated bands, obtains the transmittance average value, and obtains the cloud layer transparency data; The cloud motion parameter analysis submodule calls the cloud transparency data, calculates the cloud motion direction and coverage change rate in combination with the time series images, and obtains cloud motion characteristic parameters.

5. The ultra-short-term prediction system based on image analysis according to claim 4, characterized in that: The solar irradiance change prediction module includes: The cloud occlusion calculation submodule calculates the proportion of sunlight blocked by clouds based on the cloud motion characteristic parameters, and corrects the occlusion ratio by combining the cloud thickness and transparency to obtain the cloud occlusion ratio; The radiation scattering coefficient calculation submodule calculates the radiation scattering coefficient based on the cloud occlusion ratio, combined with cloud thickness, transparency and sun angle, using the formula: The radiation scattering coefficient is obtained by calculation, and the light attenuation ratio is adjusted according to the clear sky irradiance curve to obtain the adjusted irradiance; Among them, S c represents the radiation scattering coefficient, T′ c Represents cloud transparency, H c Represents the cloud thickness, D s Represents the cloud cover ratio, A c,i represents the scattering influence factor within the differentiated time window, n′ c Represents the number of time windows; The short-term irradiance change trend analysis submodule calculates the irradiance change rate of adjacent time windows based on the adjusted irradiance, and analyzes and obtains the short-term irradiance change trend.

6. The ultra-short-term prediction system based on image analysis according to claim 5, characterized in that: The photovoltaic power generation power calculation module includes: The unit area power generation calculation submodule uses the photovoltaic module conversion efficiency and ambient temperature correction coefficient based on the short-term irradiance change trend, and adopts the formula: Get the power generation value per unit area; Among them, P u represents the power generation value per unit area, η represents the conversion efficiency of photovoltaic modules, I r Represents the irradiance value calculated from the short-term irradiance change trend, k t Represents the temperature correction factor, T′ c Represents the current temperature of the component, T r represents the standard test temperature, A represents the unit area of ​​the component; The total power calculation submodule calculates the total power of the photovoltaic system based on the power generation value per unit area, combined with the photovoltaic module arrangement mode and the total area of ​​the modules, and obtains the total power value of the photovoltaic modules; The shielding influence adjustment submodule calls the total power value of the photovoltaic component, combines the component shielding influence coefficient, corrects the calculation result, adjusts the photovoltaic power output, and obtains the photovoltaic power prediction value.

7. The ultra-short-term prediction system based on image analysis according to claim 6, characterized in that: The prediction model optimization training module includes: The data normalization submodule calculates a mean normalization value based on the photovoltaic power prediction value, uses minimum-maximum normalization to adjust the scale range, and obtains normalized power data; The time series enhancement submodule constructs a time sliding window based on the normalized power data, extracts the change trend, calculates the change rate, cumulative change amount and increment, dynamically adjusts the weight, enhances the short-term change characteristics, and obtains enhanced time series samples; The dynamic parameter optimization submodule calculates the model prediction error based on the enhanced time series samples using the formula: Dynamically adjust model parameters, optimize update step size, reduce prediction error, and obtain dynamically optimized prediction model parameters; Among them, P opt represents the parameters of the dynamic optimization prediction model, P t Represents the power data of the current time step, P t-1 Represents the power data of the previous time step, E t Represents the current error value, E min represents the previous minimum error, α represents the error influencing factor, β represents the error sensitivity factor, and n′ c Represents the total number of time steps.

8. An ultra-short-term prediction method based on image analysis, characterized in that: The ultra-short-term prediction system based on image analysis according to any one of claims 1 to 7 comprises the following steps: S1: Acquire all-sky imaging data, set the shooting frequency, adjust the exposure time, perform denoising filtering based on the illumination distribution, segment the sky area based on the color distribution and edge detection algorithm, filter the pixel area containing cloud information, and obtain denoised segmented image data; S2: Based on the denoised and segmented image data, extract the edge contour of the cloud layer, calculate the pixel displacement vector between consecutive frames, obtain the cloud layer movement direction and speed, call the cloud layer coverage rate to calculate the cloud layer area ratio, calculate the direct sunlight occlusion ratio in combination with the transparency change, analyze the influence of the clouds with different heights on the occlusion ratio, and obtain the cloud layer movement characteristic parameters; S3: Based on the cloud motion characteristic parameters, calculate the solar direct light shielding ratio, calculate the radiation scattering coefficient in combination with the cloud thickness, transparency and sun angle, call the clear sky irradiance curve to adjust the direct light intensity, calculate the light attenuation rate, analyze the light change rate in the continuous time window, and obtain the short-term irradiance change trend; S4: Based on the short-term irradiance variation trend, the power generation of the photovoltaic components per unit area is calculated, the power generation of the entire photovoltaic array is calculated in combination with the component arrangement, and the shading effect is adjusted to obtain the photovoltaic power prediction value; S5: Based on the photovoltaic power prediction value, the input data is normalized, the time series samples are enhanced, the time step is adjusted, the prediction model is retrained for differentiated weather conditions, and the parameter weights are updated to obtain dynamically optimized prediction model parameters.

Images