A method and device for predicting photovoltaic power generation in heavy fog weather based on a UAV

Abstract

The present application relates to the technical field of new energy power generation, and discloses a new energy power generation field, in particular to a heavy fog weather photovoltaic power generation power prediction method and device based on a UAV.The present application can realize fine identification of local and rapidly changing heavy fog by introducing a UAV to collect power station image data, and then obtain accurate global fog concentration data through spatial interpolation and time series difference prediction, which can accurately represent the real-time distribution and evolution law of the fog layer; based on the fog concentration, the extinction coefficient is calculated and the light intensity is corrected, which can accurately quantify the attenuation effect of heavy fog on solar radiation, solving the defect that the prior art lacks fog weather irradiance optimization; finally, the power prediction is carried out with the corrected light intensity, so that the model input conforms to the real irradiation condition of fog weather, thereby overcoming the poor adaptability of the prior art to local and rapidly changing heavy fog from the root, and improving the prediction accuracy of photovoltaic power generation power under heavy fog weather.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation, specifically to a method and device for predicting photovoltaic power generation in foggy weather based on unmanned aerial vehicles (UAVs). Background Technology

[0002] As a clean energy source reliant on solar radiation, photovoltaic (PV) power generation is susceptible to the effects of high-aerosol weather conditions such as fog and sandstorms. These weather conditions cause complex attenuation of solar irradiance through strong extinction, leading to significant deviations in power generation predictions. Existing PV power prediction technologies mostly rely on clear-sky models or use satellite remote sensing data to address large-scale cloud cover issues. They lack sophisticated identification and correction methods for foggy weather, which is highly localized and changes rapidly in time and space. This makes it difficult to accurately characterize the extinction attenuation patterns of fog layers, resulting in low power prediction accuracy under foggy conditions, which fails to meet the requirements for safe and stable operation of PV power plants and grid dispatch. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a method and apparatus for predicting photovoltaic power generation in foggy weather based on unmanned aerial vehicles (UAVs), in order to solve the problem that the existing technology lacks a refined identification and irradiance optimization method for localized and rapidly changing foggy weather, resulting in insufficient accuracy in predicting photovoltaic power generation in foggy weather.

[0004] In a first aspect, embodiments of the present invention provide a method for predicting photovoltaic power generation in foggy weather based on unmanned aerial vehicles (UAVs), the method comprising: Multi-source data of a photovoltaic power station is acquired and processed to obtain fog concentration characteristic data. The multi-source data includes image data of the photovoltaic power station collected by a drone, meteorological data of the photovoltaic power station, and operating condition data. Spatial interpolation is performed on the fog concentration of each pixel in the fog concentration feature data to obtain a time series sequence. Differential analysis is then performed on the time series sequence to predict the global fog concentration of the photovoltaic power station. The global fog concentration prediction data includes the spatial and temporal distribution of different fog concentration levels in different local areas at future times. The extinction coefficient, which characterizes the ability of fog weather to attenuate sunlight, is calculated using the global fog concentration prediction data and fog layer optical parameter data. The extinction coefficient is fitted based on the spatial and temporal distribution of different fog concentration levels in different local areas at different future times to obtain the irradiance correction coefficient for different future times and different local areas. The irradiance correction coefficient is then used to correct the basic solar radiation parameters to obtain the predicted value of the actual light intensity that can reach the photovoltaic panel during foggy weather. The photovoltaic power generation capacity of the photovoltaic power station at future times is predicted based on the predicted light intensity value.

[0005] Furthermore, the processing of the multi-source data to obtain fog concentration characteristic data includes: Calculate the dark channel value of the image data in the multi-source data, and substitute the dark channel value into the fog imaging model to derive the transmittance calculation formula to obtain the atmospheric transmittance of each pixel in the image data. The fog concentration value of each pixel is calculated based on the atmospheric transmittance of each pixel. The fog concentration values ​​of each pixel are classified according to their numerical range to obtain quantified fog concentration feature data.

[0006] Furthermore, the step of performing differential analysis on the time series to obtain the global fog concentration prediction data for the photovoltaic power station includes: Spatial interpolation calculation is performed on the fog concentration of each frame pixel in the fog concentration feature data to obtain the global fog concentration distribution data of the photovoltaic power station, and statistical averaging calculation is performed on the global fog concentration distribution data to obtain the global average fog concentration data of the photovoltaic power station. Based on the global average fog concentration data and the historical monitoring data of the photovoltaic power station, a fog concentration time series is constructed. The fog concentration time series is input into the ARIM model, and the fog concentration time series is subjected to differential stabilization and time series prediction to obtain the fog concentration differential prediction value. Using the current average fog concentration across the entire region as a benchmark, the cumulative fog concentration difference prediction values ​​are extrapolated over time to obtain the predicted fog concentration data for the photovoltaic power station at future times.

[0007] Furthermore, the extinction coefficient, which characterizes the ability of fog weather to attenuate sunlight, is calculated using the global fog concentration prediction data and fog layer optical parameter data, including: The atmospheric transmittance conversion formula is used to perform reverse conversion on the global fog concentration prediction data to obtain the corresponding atmospheric transmittance data. The effective path length of the fog layer is obtained by calculating the drone's flight altitude and shooting angle. The atmospheric transmittance data is logarithmically processed using the natural logarithm operation function to obtain the logarithmic transmittance value. The extinction coefficient, which characterizes the ability of fog to attenuate sunlight, is calculated by using a negative ratio between the logarithmic value of the transmittance and the effective path length of the fog layer.

[0008] Furthermore, the extinction coefficient is fitted based on the spatial and temporal distribution of different fog concentration levels in different local areas at different future times to obtain irradiance correction coefficients for different future times and different local areas. These irradiance correction coefficients are then used to correct the basic solar radiation parameters to obtain the predicted irradiance intensity that can actually reach the photovoltaic panel during foggy weather, including: The theoretical irradiance on a clear day under fog-free conditions is obtained by performing a sinusoidal product operation on the aforementioned basic solar radiation parameters. Within a predetermined time period in the future, the extinction coefficient will first be fitted using an exponential model to obtain the irradiance correction coefficient. The extinction coefficient and the measured ground irradiance are recorded in real time using a preset frequency, and a scatter plot of the extinction coefficient and the irradiance correction coefficient is plotted. Determine whether the data distribution characteristics in a scatter plot exhibit an exponential trend; If so, the exponential model described above will continue to be used to determine the irradiance correction coefficient; If not, the extinction coefficient is fitted in stages according to the spatial distribution of fog concentration levels using a piecewise quadratic function of fog concentration levels to obtain the irradiance correction coefficient. The initial irradiance prediction data is obtained by multiplying the theoretical irradiance on a sunny day and the irradiance correction coefficient. The initial irradiance prediction data is then spatially averaged to obtain the predicted irradiance intensity of the photovoltaic panel under foggy weather.

[0009] Furthermore, the step of using a piecewise quadratic function based on fog concentration levels to perform graded fitting on the extinction coefficient to obtain the irradiance correction coefficient for future times includes: By comparing the extinction coefficient with a preset threshold, the target level of the current fog concentration is determined; Obtain the piecewise quadratic function and fitting coefficients corresponding to the target level; Substitute the extinction coefficient into the piecewise quadratic function, and solve the quadratic polynomial using the fitting coefficient to obtain the irradiance correction coefficient for a single point. The irradiance correction coefficient for the single point is used for time-series smoothing to obtain the irradiance correction coefficient for future times.

[0010] Furthermore, predicting the photovoltaic power generation capacity of the photovoltaic power station at future times based on the predicted light intensity value includes: Based on the predicted light intensity, the temperature and humidity data of the photovoltaic power station, and the extinction coefficient, dimensional splicing and normalization processing are performed to obtain the power input feature data. Using historical power data as labels and the input feature data as model input, supervised learning training and parameter iterative optimization are performed on the power prediction model to obtain the trained power prediction model. Using the predicted light intensity and the meteorological data to train the power prediction model, forward inference and regression calculations are performed to obtain the initial power data for future times. The initial power data and the installed capacity of the photovoltaic power station are range-calibrated to obtain the photovoltaic power generation power of the photovoltaic power station at future times.

[0011] Secondly, embodiments of the present invention provide a photovoltaic power generation prediction device based on a drone in foggy weather, the device comprising: The acquisition module is used to acquire multi-source data from the photovoltaic power station and process the multi-source data to obtain fog concentration characteristic data. The multi-source data includes image data of the photovoltaic power station collected by the UAV, meteorological data of the photovoltaic power station, and operating condition data. The analysis module is used to perform spatial interpolation on the fog concentration of each pixel in the fog concentration feature data to obtain a time series sequence, and to perform differential analysis and prediction on the time series sequence to obtain the global fog concentration prediction data of the photovoltaic power station. The calculation module is used to calculate the extinction coefficient, which characterizes the ability of fog weather to attenuate sunlight, using the global fog concentration prediction data and fog layer optical parameter data. Based on the spatial and temporal distribution of different fog concentration levels in different local areas at future times, the extinction coefficient is fitted to obtain the irradiance correction coefficient for different future times and different local areas. The irradiance correction coefficient is used to correct the basic solar radiation parameters to obtain the predicted value of the actual light intensity that can shine on the photovoltaic panel during fog weather. The prediction module is used to predict the photovoltaic power generation capacity of the photovoltaic power station at future times based on the predicted light intensity value.

[0012] Thirdly, embodiments of the present invention provide a computer device, including: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method described in the first aspect or any corresponding embodiment thereof.

[0013] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer instructions that cause a computer to perform the method described in the first aspect or any of its corresponding embodiments.

[0014] This invention introduces image data of power plants collected by drones, enabling refined identification of localized and rapidly changing fog. Accurate global fog concentration data is then obtained through spatial interpolation and temporal difference prediction, accurately characterizing the real-time distribution and evolution of the fog layer. Based on the fog concentration, the extinction coefficient is calculated and the irradiance is corrected, precisely quantifying the attenuation effect of fog on solar radiation and addressing the lack of optimization for foggy irradiance in existing technologies. Finally, power prediction is performed using the corrected irradiance, ensuring the model input closely matches real-world irradiance conditions in foggy weather. This fundamentally overcomes the poor adaptability of existing technologies to localized and rapidly changing fog, improving the prediction accuracy of photovoltaic power generation under foggy conditions. Attached Figure Description

[0015] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a method for predicting photovoltaic power generation in foggy weather based on unmanned aerial vehicles (UAVs) according to some embodiments of the present invention. Figure 2 This is a schematic diagram of a framework for predicting photovoltaic power generation in foggy weather based on unmanned aerial vehicles (UAVs) according to some embodiments of the present invention. Figure 3 This is a structural block diagram of a drone-based photovoltaic power generation prediction device for foggy weather according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] According to an embodiment of the present invention, a method and apparatus for predicting photovoltaic power generation in foggy weather based on unmanned aerial vehicles are provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0019] This embodiment provides a method for predicting photovoltaic power generation in foggy weather based on unmanned aerial vehicles (UAVs). Figure 1 This is a flowchart of a method for predicting photovoltaic power generation in foggy weather based on unmanned aerial vehicles (UAVs) according to an embodiment of the present invention, as shown below. Figure 1 As shown, the process includes the following steps: Step S101: Acquire multi-source data of the photovoltaic power station and process the multi-source data to obtain fog concentration characteristic data. The multi-source data includes image data of the photovoltaic power station collected by UAV, meteorological data of the photovoltaic power station, and operating condition data.

[0020] In this embodiment, the multi-source data includes power plant image data collected by UAVs according to fixed cycles and paths, meteorological data reflecting environmental conditions, and operating condition data containing historical power and installed capacity. The dark channel prior refers to the fact that the grayscale value of at least one color channel in a local area of ​​a clear-sky image tends to be close to 0; heavy fog will significantly increase this value, which can be used to characterize fog concentration. The fog imaging model is used to describe the imaging patterns of foggy images and is the theoretical basis for calculating atmospheric transmittance. After achieving spatiotemporal alignment of the data through preprocessing, the dark channel values ​​of the image are calculated and substituted into the fog imaging model to obtain the atmospheric transmittance of each pixel. Then, based on the conversion relationship between transmittance and fog concentration, the values ​​are calculated and graded, ultimately forming fog concentration characteristic data that can be used for subsequent analysis.

[0021] Specifically, the multi-source data is processed to obtain fog concentration feature data, including: calculating the dark channel value of the image data in the multi-source data, and substituting the dark channel value into the fog imaging model to derive the transmittance calculation formula to obtain the atmospheric transmittance of each pixel in the image data; calculating the fog concentration value of each pixel based on the atmospheric transmittance of each pixel; and classifying the fog concentration value of each pixel according to the numerical range to obtain quantified and graded fog concentration feature data.

[0022] Understandably, the image data of photovoltaic power plants collected by drones is preprocessed to remove noise and abnormal pixels. Based on the dark channel prior principle, each local window of the image is traversed to extract the minimum grayscale value among the red, green, and blue color channels, and the dark channel value of the entire image is calculated. This dark channel value is then substituted into the fog imaging model, and a transmittance calculation relationship is constructed by combining it with the fog attenuation coefficient. A transmittance calculation formula suitable for foggy scenes is derived, and the image values ​​are then used to solve for the atmospheric transmittance corresponding to each pixel. This value characterizes the ability of light to penetrate the fog layer.

[0023] Based on the physical correspondence between atmospheric transmittance and fog concentration, the atmospheric transmittance of each pixel is converted using the formula FC=1-t, and pixel-level fog concentration values ​​are calculated point by point. The values ​​range from 0 to 1, with larger values ​​indicating higher fog concentration and stronger radiation attenuation at that location. After completing the pixel calculation for the entire image, the fog concentration values ​​are classified according to preset ranges: 0 to 0.2 is defined as the edge of no fog / light fog, 0.2 to 0.4 as light fog, 0.4 to 0.7 as heavy fog, and 0.7 to 1.0 as dense fog. Through classification processing, continuous values ​​are converted into discrete level labels, ultimately yielding quantitatively graded fog concentration feature data that can be directly used for subsequent analysis.

[0024] The complete calculation process is as follows: Based on the dark channel prior dehazing algorithm, atmospheric transmittance t(x,y) is extracted: According to the dark channel prior principle, in a clear day natural image, at least one color channel gray value is close to 0 (dark channel) in any local area; the dark channel gray value of a foggy image is significantly increased, and the dark channel value can be approximated as the fog concentration.

[0025] The fog imaging model is the theoretical basis for defogging algorithms, and its related formulas can be simplified as follows: .

[0026] For defogging image data The expression for the dark channel is as follows: .

[0027] in, is the dark channel gray value of the pixel with coordinates (x,y) in the image, with a value range of [0,255]. The smaller the value, the higher the corresponding fog density; c represents the three channels of red (r), green (g), and blue (b) in the image data. The dark channel of each channel is calculated separately, and the minimum value is taken as the final dark channel value. This represents a local window centered at (x, y). This represents the grayscale value of the pixel at coordinates (i,j) in channel c under fog-free conditions.

[0028] Due to the dark channel value of fog-free images The transmittance approaches 0. By introducing a buffer factor, the following formula for calculating transmittance is derived: .

[0029] in, t is the atmospheric transmittance at pixel coordinates (x,y), with a value range of [0,1]. It represents the proportion of light remaining after passing through the fog layer. The closer t is to 0, the higher the fog concentration and the more severe the irradiance attenuation. ω is the fog attenuation coefficient. The denser the fog particles, the larger the value of ω. It is 0.9 for regular heavy fog, 1.0 for strong dense fog, and 0.8 for light fog. This represents the actual grayscale value of the pixel at coordinates (i,j) in the c channel of a real-time fog image captured by a drone. This represents the grayscale value corresponding to the fog-free / fog-sparsest region in a foggy image in channel c.

[0030] The fog concentration FC(x,y) is calculated based on transmittance. The relevant formula is as follows: .

[0031] in, This represents the fog concentration at pixel coordinates (x,y), with a value range of [0,1]. It directly characterizes the density of the fog, i.e., the proportion of light attenuated by the fog layer. Generally, 0≤FC<0.2 (no fog / light fog edge), 0.2≤FC<0.4 (light fog), 0.4≤FC<0.7 (dense fog), and 0.7≤FC≤1.0 (strong dense fog).

[0032] As an example, taking a frame of drone imagery captured by a photovoltaic power station in heavy fog, the image is first preprocessed with denoising and normalization. The dark channel value of each local window is calculated and substituted into the fog imaging model to obtain the atmospheric transmittance of each pixel. Subsequently, the fog concentration value of the entire pixel is obtained by converting the transmittance. The values ​​were found to be concentrated between 0.5 and 0.65 in the central area of ​​the power station and between 0.3 and 0.45 in the edge area. After classification according to the preset threshold, the central area is determined to be heavy fog and the edge area is determined to be light fog. Finally, quantitative fog concentration feature data containing level labeling and spatial distribution is generated, which can be directly used for subsequent global interpolation and temporal prediction.

[0033] Step S102: Spatial interpolation is performed on the fog concentration of each pixel in the fog concentration feature data to obtain a time series sequence, and differential analysis is performed on the time series sequence to obtain the global fog concentration prediction data of the photovoltaic power station.

[0034] In this embodiment, spatial interpolation is a calculation method that extends the fog concentration of discrete pixels to cover the entire spatial distribution of the photovoltaic power station, thus obtaining the global average fog concentration. The time series consists of multiple sets of global average fog concentration data arranged chronologically to reflect the changing pattern of fog concentration over time. ARIMA model differential prediction refers to stabilizing the time series data through differential operations, then using the model to fit historical patterns and extrapolate future values. After spatial interpolation of the fog concentration frame by frame, a time series is constructed. The ARIMA model is used for differential analysis and training to obtain the differential fog concentration prediction value. The differential results are accumulated based on the current value to finally obtain the global fog concentration prediction data for the future period of the power station.

[0035] In reality, the overall fog concentration prediction data represents the average fog concentration across the entire power plant area over a future period. However, in actual scenarios, the fog concentration across the entire power plant area is highly likely not evenly distributed. Therefore, when using the overall fog concentration prediction data for subsequent calculations, ground-measured irradiance can be collected at a preset frequency. Based on the overall fog concentration prediction data, the ground-measured irradiance can be used to correct the overall fog concentration prediction data by region, obtaining the spatial and temporal distribution of different fog concentration levels in different local areas at future times. For example, taking the overall fog concentration prediction data as the prediction data for the next 4 hours as an example, the ground-measured irradiance is collected at a preset frequency of 15 minutes / time. Within the first 15 minutes, the measured irradiance of multiple local areas is collected, and the average measured irradiance is calculated. Using the overall fog concentration prediction data and the average measured irradiance as a basis, the trend of the average measured irradiance is plotted / determined according to the trend of the overall fog concentration prediction data over the next preset period (4 hours). Alternatively, using 15-minute intervals, the change in average irradiance is determined within each 15-minute interval based on the variation area of ​​the overall fog concentration prediction data, thus determining the average measured irradiance every 15 minutes for the next 4 hours. Then, every 15 minutes, the difference between the measured irradiance collected in each local area and the corresponding average measured irradiance every 15 minutes is calculated. If the difference is greater than a preset difference, the corresponding local area in the overall fog concentration prediction data is corrected according to the change trend of that local area, obtaining the spatial and temporal distribution of different fog concentration levels in different local areas at future times. Specifically, differential analysis is performed on the time series to obtain the overall fog concentration prediction data for the photovoltaic power station, including: Step A1: Spatial interpolation calculation is performed based on the fog concentration of each pixel in the fog concentration feature data to obtain the global fog concentration distribution data of the photovoltaic power station. Then, statistical averaging is performed based on the global fog concentration distribution data to obtain the global average fog concentration data of the photovoltaic power station.

[0036] Based on the fog concentration values ​​frame by frame and pixel by pixel in the fog concentration feature data, spatial interpolation calculations are performed on the entire area of ​​the photovoltaic power station to expand the discrete, point-like pixel fog concentration into a continuous, area-like spatial distribution result, forming full-area fog concentration distribution data that can completely cover the power station area.

[0037] Based on this, a comprehensive statistical average calculation is performed on the fog concentration distribution data across the entire area to eliminate numerical fluctuations and anomalies in local pixels, resulting in a global average fog concentration that represents the fog condition level of the entire photovoltaic power station. This process achieves a scale transformation of fog concentration from pixel-level to power station-wide level, providing a stable and unified input basis for subsequent time-series modeling and ensuring that subsequent fog concentration prediction results can reflect the overall fog condition changes of the power station.

[0038] Step A2: Construct a time series sequence of fog concentration based on the average fog concentration data of the entire region and the historical monitoring data of the photovoltaic power station.

[0039] Using the obtained global average fog concentration data as the core, and combining it with historical fog concentration monitoring data from photovoltaic power plants during the same period, the data are arranged in an orderly manner according to fixed time intervals and chronological order to construct a fog concentration time series that meets the requirements of time series analysis. During the construction process, the sequence length is ensured to cover at least 1 to 2 hours of previous monitoring data, so that the sequence contains sufficient fog concentration change trends and fluctuation characteristics, meeting the basic conditions for ARIMA model training and prediction.

[0040] Simultaneously, the time series is preprocessed with time alignment and missing value imputation to ensure data continuity and regularity, accurately characterizing the evolution of fog concentration over time, and providing a reliable data carrier for subsequent differential stabilization processing and time series prediction.

[0041] Step A3: Input the fog concentration time series into the ARIM model, perform differential stabilization on the fog concentration time series and perform time series prediction to obtain the fog concentration differential prediction value.

[0042] This step inputs the constructed fog concentration time series into the ARIMA model. First, the series undergoes difference stabilization to eliminate trends and non-stationary features, ensuring the data meets model input requirements. Based on this stabilization, the ARIMA model is used to fit and predict the time series, extracting historical patterns of fog concentration variation and extrapolating them to obtain the difference-based predicted fog concentration values ​​for each future time step. These predicted values ​​reflect the magnitude and direction of fog concentration changes between adjacent times; positive values ​​indicate increasing fog concentration, while negative values ​​indicate decreasing fog concentration, providing accurate data for the final time series extrapolation calculation.

[0043] Step A4: Using the current average fog concentration across the entire region as a benchmark, accumulate the fog concentration difference prediction values ​​and perform time-series extrapolation calculations to obtain the predicted fog concentration data for the photovoltaic power station at future times.

[0044] Using the current measured average fog concentration across the entire area as the calculation benchmark, the predicted fog concentration differences at each future time step are sequentially accumulated to perform time-series extrapolation calculations. By superimposing the benchmark value and the changes, the average fog concentration level across the entire photovoltaic power station at different future times is directly derived, ultimately forming complete future fog concentration prediction data for the entire area. This prediction data can cover an ultra-short-term period of up to 4 hours, accurately reflecting the development, dissipation, and concentration change trends of fog in the power station area, providing crucial fog condition input for subsequent extinction coefficient calculation, irradiance correction, and power prediction.

[0045] Understandably, spatial interpolation is performed on the FC(x,y) extracted frame by frame to obtain the average fog concentration over the entire photovoltaic power station area. Forming a time series (t1~t) n For the initial monitoring time points, n≥1-8 groups, i.e., 1-2 hours of data from the previous period, to ensure the convergence of the subsequent extrapolation model. The ARIMA model is used for time-series extrapolation; the formula for extrapolating fog concentration for the next 4 hours is as follows: .

[0046] in, This represents the predicted fog concentration at time τ, with a value range of [0,1]. The last moment of the previous monitoring (t) n The average fog concentration across the entire power station area; For from t n The sum of the fog concentration difference predictions up to time τ is the total change in fog concentration over a future period of time. The fog concentration difference prediction value at the k-th time step reflects the change in fog concentration at that time step. A positive value indicates an increase in fog concentration, and a negative value indicates a decrease in fog concentration.

[0047] Step S103: Using the global fog concentration prediction data and fog layer optical parameter data, the extinction coefficient, which characterizes the ability of fog weather to attenuate sunlight, is calculated. Based on the spatial and temporal distribution of different fog concentration levels in different local areas at different future times, the extinction coefficient is fitted to obtain the irradiance correction coefficient for different future times and different local areas. The irradiance correction coefficient is used to correct the basic solar radiation parameters to obtain the predicted value of the actual light intensity that can shine on the photovoltaic panel during fog weather.

[0048] In this embodiment, the extinction coefficient is a physical quantity characterizing the ability of fog to absorb and scatter sunlight; a larger value indicates stronger attenuation. Fog layer optical parameters include the effective path length, calculated using trigonometric functions based on the drone's flight altitude and shooting angle. The irradiance correction coefficient is used to convert the theoretical irradiance in fog-free conditions into the actual irradiance in foggy weather. Different models are used to fit different regions, fog concentration levels, and the temporal variation of fog concentration levels; for example, an exponential model or a stepwise piecewise quadratic function can be used. Atmospheric transmittance is calculated based on fog concentration prediction data, and the extinction coefficient is obtained by combining it with the path length. The theoretical irradiance is then calculated using the solar constant and solar altitude angle. The extinction coefficient is matched with the correction coefficient and multiplied, and the result is averaged across the entire region to obtain the predicted irradiance intensity actually illuminating the photovoltaic panel.

[0049] Specifically, using global fog concentration prediction data and fog layer optical parameter data, the extinction coefficient, which characterizes the ability of foggy weather to attenuate sunlight, is calculated, including: Step B1: Use the atmospheric transmittance conversion formula to perform reverse conversion on the global fog concentration prediction data to obtain the corresponding atmospheric transmittance data.

[0050] Based on the physical conversion relationship between atmospheric transmittance and fog concentration, the atmospheric transmittance conversion formula is used to perform reverse conversion processing on the fog concentration prediction data of the entire photovoltaic power station area. The fog concentration value is converted into the corresponding value through the formula t=1-FC, and atmospheric transmittance data that can reflect the ability of light to penetrate fog is obtained. The value of this data is between 0 and 1. The smaller the value, the stronger the blocking effect of fog on light. This provides the core optical parameter basis for subsequent extinction coefficient calculation and ensures that the conversion process conforms to the basic laws of atmospheric optical transmission.

[0051] Step B2: Calculate the drone's flight altitude and shooting angle to obtain the effective path length of the fog layer.

[0052] Based on the actual flight parameters of the UAV during image acquisition, the flight altitude and shooting angle of the UAV are accurately calculated using trigonometric functions. The effective path length of the fog layer is obtained by solving the formula d≈h / cosθ. This parameter represents the effective distance that light actually travels in the fog layer and is a key geometric parameter for calculating the extinction coefficient. The calculation process fully considers the spatial relationship between the UAV's inspection attitude and the fog layer distribution to ensure that the path length value can truly reflect the transmission path of light in the fog field, thereby improving the accuracy of subsequent optical parameter calculations.

[0053] Step B3: Use the natural logarithm function to perform logarithmic calculation on the atmospheric transmittance data to obtain the logarithmic transmittance value.

[0054] The obtained atmospheric transmittance data is used as the computational object, and the natural logarithm operation function is used to perform logarithmic calculation to obtain the corresponding logarithmic transmittance value. This operation is consistent with the Beer-Lambert law in atmospheric extinction theory, which can transform the linear relationship of transmittance into a logarithmic form that can be directly used in extinction calculation. The calculation process strictly follows the requirements of the physical model, eliminates the influence of nonlinear changes in transmittance values, and enables the results to accurately characterize the degree of light attenuation by the fog layer, providing a standardized intermediate calculation for the final extinction coefficient calculation.

[0055] Step B4: Based on the negative ratio of the logarithmic transmittance to the effective path length of the fog layer, the extinction coefficient, which characterizes the ability of fog to attenuate sunlight, is calculated.

[0056] Based on the obtained logarithmic transmittance value and the obtained effective path length of the fog layer, the negative ratio value is calculated according to the extinction coefficient calculation formula. The final extinction coefficient is obtained through the formula σ=-ln(t) / d. This coefficient is the core physical quantity that quantifies the ability of fog to absorb and scatter sunlight. The larger the value, the stronger the extinction effect of the fog and the more serious the irradiance attenuation. The calculation process is completely consistent with the atmospheric optical model, and the results can be directly used for subsequent irradiance correction and photovoltaic power generation prediction.

[0057] It should be noted that, based on classical atmospheric optics theory, the visual characteristics of UAV images during foggy weather are converted into a fog extinction coefficient to characterize the degree of attenuation of solar irradiance by fog, which is used for subsequent correction of predicted irradiance. The extinction coefficient σ(t) represents the ability of fog to attenuate sunlight, mainly including scattering and absorption. The larger the extinction coefficient, the stronger the extinction effect and the stronger the attenuation of irradiance. Using atmospheric transmittance t(x,y) as an intermediate quantity, combined with the effective path length d of the fog layer captured by the UAV, the following derivation is performed. The relevant formulas are as follows: .

[0058] in, The average extinction coefficient of the entire power station at time t is expressed in km². - ¹ indicates the degree of light attenuation as it travels 1 km through the fog layer. The larger the fog, the higher the fog concentration and the more severe the light attenuation; The effective path length of the drone through the fog layer is measured in km and is determined by the drone's flight altitude. The calculation formula is d≈h / cosθ (where h is the drone's flight altitude and θ is the shooting angle).

[0059] In this embodiment of the application, the predicted value of the actual light intensity that can reach the photovoltaic panel during foggy weather is obtained by calculating using the extinction coefficient and basic solar radiation parameters, including: Step C1: Perform a sinusoidal product operation on the basic solar radiation parameters to obtain the theoretical irradiance under fog-free conditions on a clear day.

[0060] Using the solar constant and the future solar altitude angle as the core fundamental parameters of solar radiation, a sinusoidal product calculation is performed based on an atmospheric optical model, using the formula G0=G s The theoretical irradiance under clear weather conditions is calculated using c×sin(h). The solar constant is taken as the standard value of 1367 W / m², and the solar altitude angle changes dynamically with time. This calculation strictly follows the physical laws of solar radiation transmission, accurately reflecting the solar irradiance level reaching the surface of the photovoltaic power station in an ideal clear weather environment. It provides a benchmark reference value for irradiance correction in foggy weather, and the calculation results are stable and reliable.

[0061] Step C2: Within a preset time period, first, the extinction coefficient is fitted using an exponential model to obtain an irradiance correction coefficient; then, the extinction coefficient and the measured ground irradiance are recorded in real time at a preset frequency, and a scatter plot of the extinction coefficient and the irradiance correction coefficient is plotted; determine whether the data distribution characteristics in the scatter plot show an exponential trend; if so, continue to use the exponential model to determine the irradiance correction coefficient; if not, use a piecewise quadratic function based on fog concentration levels to perform a graded fitting of the extinction coefficient according to the spatial distribution of fog concentration levels to obtain the irradiance correction coefficient.

[0062] In this embodiment, the extinction coefficient is fitted to a piecewise quadratic function based on the fog concentration level to obtain the irradiance correction coefficient for future times. This includes: comparing the extinction coefficient with a preset threshold to determine the target level of the current fog concentration; obtaining the piecewise quadratic function and fitting coefficients corresponding to the target level; substituting the extinction coefficient into the piecewise quadratic function and solving the quadratic polynomial with the fitting coefficients to obtain the irradiance correction coefficient for a single point; and performing time-series smoothing based on the irradiance correction coefficient for the single point to obtain the irradiance correction coefficient for future times.

[0063] The process involves comparing the calculated extinction coefficient with preset grading thresholds one by one. Based on the numerical value, the target grading level of the current fog concentration is determined, classifying the extinction coefficient into three levels: light fog, heavy fog, and dense fog. This precise classification of fog conditions provides a basis for the grading and fitting correction coefficients. Next, based on the determined target grading level, the corresponding piecewise quadratic function and polynomial coefficients fitted using the least squares method are retrieved from a pre-stored model library to ensure that the function and coefficients match the irradiance attenuation law of the current fog concentration level. Then, the extinction coefficient is substituted as an input variable into the corresponding piecewise quadratic function, and the fitting coefficients are used to complete the quadratic polynomial solution, calculating the irradiance correction coefficient for the current single-point location. Finally, the irradiance correction coefficient for the single-point location undergoes time-series smoothing to eliminate abrupt changes and fluctuations in the time series, ensuring that the coefficient change trend conforms to the natural evolution characteristics of fog concentration, thus obtaining stable and continuous irradiance correction coefficients for future times.

[0064] Step C3: Multiply the theoretical irradiance on a sunny day with the irradiance correction coefficient to obtain the initial irradiance prediction data, and then perform spatial averaging on the initial irradiance prediction data to obtain the predicted irradiance intensity of the photovoltaic panel under foggy weather.

[0065] The theoretical irradiance for clear weather is multiplied by the irradiance correction coefficient to obtain the initial irradiance prediction data for foggy scenarios. Since the initial data is a single-point calculation and cannot comprehensively reflect the irradiance distribution across the entire photovoltaic power station area, spatial averaging is performed to integrate regional numerical differences and eliminate errors caused by local outliers, making the results more closely reflect the actual sunlight received by the power station. The final predicted irradiance value can accurately represent the actual solar irradiance illuminating the photovoltaic panel surface under foggy weather conditions.

[0066] It should be noted that the ultra-short-term extinction coefficient is calculated based on the fog concentration data extrapolated over time, and an irradiance correction model based on the extinction coefficient is constructed. The larger the extinction coefficient, the more severe the irradiance attenuation. The relevant formula is as follows.

[0067] .

[0068] in, This represents the predicted average irradiance of the entire photovoltaic power station at time τ in the future, expressed in W / m², and is the core input for photovoltaic power prediction. The theoretical solar irradiance at time τ in a clear sky, expressed in W / m², represents the solar irradiance at that time under fog-free conditions. The relevant calculation formula is as follows: ( The solar constant, (The solar altitude angle at a future time τ). Let be the irradiance correction coefficient at a future time τ, with a value range of (0,1], representing the ratio of irradiance on a foggy day to irradiance on a sunny day at the same time, determined by the extinction coefficient. Matched.

[0069] Because the relationship between fog concentration and irradiance attenuation varies across different fog concentration ranges, thicker fog layers result in multiple scattering of photons, potentially leading to higher actual radiation reaching the ground than theoretically expected, causing the curve to deviate from a purely exponential pattern. Generally, light fog exhibits linear attenuation, while dense fog and heavy fog show non-linear attenuation. Furthermore, using quadratic function-based hierarchical modeling significantly improves the prediction efficiency and accuracy of the correction coefficients, providing an empirical approximation of the actual physical process within a specific range. To enhance the model's practicality and adaptability, this invention initially employs a simplified exponential model for irradiance correction, gradually introducing a piecewise fitting strategy after data accumulation. (1) Due to insufficient initial data, an exponential model was used to calculate the irradiance correction factor: .

[0070] in, The extinction coefficient is an empirical value, and L is the effective thickness of the fog. The value of L is first estimated by back-calculating the measured data of typical foggy days; secondly, the average thickness of the fog layer can be determined by local meteorological data as the value of L.

[0071] The derived extinction coefficient and measured ground irradiance for foggy weather were recorded in real time at a frequency of 15 minutes per recording. Data was accumulated, and scatter plots of the extinction coefficient and irradiance correction coefficient under different fog concentrations were plotted. The fitting method was selected based on the data distribution characteristics. If an exponential decay trend was observed, the exponential model was continued, and the L parameter was optimized. If the trend significantly deviated from the exponential trend, a quadratic polynomial was used to fit the parameters. The relevant formulas are as follows: .

[0072] in, The fitting coefficients for different fog levels were obtained by matching the data of "extinction coefficient derived from fog concentration - ratio of measured irradiance to theoretical irradiance" from the previous drone monitoring data using the least squares method. The number of samples under each fog concentration condition was at least 30 groups, and the data covered the entire fog season as much as possible.

[0073] Specifically, data preprocessing is performed to remove outliers and group the data, minimizing the sum of squared residuals for each data segment: .

[0074] Step S104: Predict the photovoltaic power generation capacity of the photovoltaic power station at future times based on the predicted light intensity value.

[0075] In this embodiment, multi-feature fusion refers to splicing and normalizing multi-dimensional features such as light intensity, temperature, relative humidity, and extinction coefficient to form a standardized model input. The XGBoost model is a gradient boosting decision tree model with strong fitting and generalization capabilities, suitable for photovoltaic power generation regression prediction. Range calibration applies upper limit constraints and proportional corrections to the initial power results based on the installed capacity to ensure reasonable and reliable output. The fused features are input into the trained XGBoost model, forward inference is performed to obtain the initial power, and then range calibration is used to eliminate abnormal deviations, ultimately outputting the ultra-short-term power generation prediction results for the photovoltaic power plant at future times.

[0076] Specifically, predicting the photovoltaic power generation capacity of a photovoltaic power plant at future times based on predicted light intensity values ​​includes: Step D1 involves performing dimensional splicing and normalization based on the predicted light intensity, temperature and humidity data of the photovoltaic power station, and extinction coefficient to obtain the power input feature data.

[0077] Using the predicted irradiance of photovoltaic panels under foggy weather as the core, and combining it with real-time temperature, relative humidity data, and extinction coefficient data from the photovoltaic power station, multiple heterogeneous data sets are spliced ​​together in a unified dimension to form a complete feature combination. The spliced ​​data is then normalized to eliminate calculation biases caused by differences in the dimensions and numerical ranges of different parameters, mapping all features to a unified numerical range. Finally, standardized, stable, and directly inputtable power prediction feature data is obtained, providing a high-quality data foundation for subsequent machine learning model training and inference.

[0078] Step D2 involves using historical power data as labels and input feature data as model input to perform supervised learning training and parameter iterative optimization on the power prediction model, resulting in a trained power prediction model.

[0079] Supervised learning was adopted for model training. Historical measured power generation data of photovoltaic power plants were used as training labels, and the obtained input feature data were used as model input. The XGBoost power prediction model was loaded and hyperparameters were initialized. Through multiple rounds of iterative training, the internal weights and decision structure of the model were continuously optimized to minimize the prediction error and loss function value. At the same time, key parameters such as learning rate, tree depth, and regularization coefficient were adaptively tuned to improve the model's fitting ability and generalization ability under complex foggy conditions. After the model converged and the error met the preset requirements, a power prediction model that had been trained and could be used for actual inference was obtained.

[0080] Step D3: Using the predicted light intensity and meteorological data as inputs to the trained power prediction model, forward inference and regression calculations are performed to obtain the initial power data for future times.

[0081] The predicted irradiance for future moments, along with corresponding meteorological data such as temperature, humidity, and extinction coefficient, are input into the trained power prediction model. The model performs forward inference calculations and completes regression predictions based on the learned mapping relationship between foggy irradiance and power. The decision tree calculations and results are accumulated layer by layer without introducing additional external data interference. The initial power data corresponding to the future moment is directly output. This data is the model's original prediction result, which can initially reflect the power generation capacity of the photovoltaic power station under the influence of fog conditions, but there is still a certain range of numerical fluctuations and risks of exceeding the range.

[0082] Step D4 involves calibrating the initial power data against the installed capacity of the photovoltaic power station to obtain the photovoltaic power generation capacity of the photovoltaic power station at future times.

[0083] Based on the initial power data, range calibration is carried out in conjunction with the rated installed capacity parameters of the photovoltaic power station. Values ​​exceeding the upper limit of the installed capacity are constrained and corrected, and values ​​that are too low are calibrated according to a reasonable ratio. This ensures that the predicted power is always within a reasonable range that conforms to actual operation, while smoothing out instantaneous fluctuation errors. This guarantees that the prediction results are continuous, stable, and consistent with the on-site operating patterns, ultimately yielding accurate, reliable, and directly usable photovoltaic power generation prediction data for future photovoltaic power stations that can be used for dispatch reference.

[0084] Understandably, the corrected irradiance prediction value will be... Using temperature, relative humidity, extinction coefficient, and historical power data from the previous two hours as core inputs, an ultra-short-term forecast of photovoltaic power generation is calculated based on the XGBoost machine learning model, yielding the ultra-short-term forecast value of photovoltaic power generation for the next 4 hours. .

[0085] Specifically, the sample data undergoes preprocessing, including time alignment, forward imputation for handling missing and outlier values, and feature importance analysis based on the Pearson correlation coefficient. The XGBoost model hyperparameters are set and trained, and the predicted values ​​are used as input for the next time step for rolling prediction. The core hyperparameter settings are as follows: number of trees (500), learning rate (0.05), maximum tree depth (5), minimum weight of leaf nodes (2), sample sampling ratio (0.8), feature sampling ratio (0.8), L1 regularization coefficient (0.5), L2 regularization coefficient (1.5), and task type (regression task).

[0086] In this embodiment, the method further includes: acquiring the latest extracted global average fog concentration data based on the preset image capture frequency of the UAV; performing a sliding window update on the fog concentration time series sample set by removing the oldest data and supplementing with the latest data based on the global average fog concentration data, resulting in an updated fog concentration time series sample set with a fixed number of frames; iteratively optimizing and training the fog concentration extrapolation model based on the updated fog concentration time series sample set, resulting in an optimized fog concentration extrapolation model adapted to real-time fog conditions; and continuously predicting the fog concentration, irradiance, and photovoltaic power generation at future times based on the optimized fog concentration extrapolation model, removing past time steps and supplementing with new prediction steps, to obtain prediction results that closely match the real-time changes in the fog layer.

[0087] Specifically, a strategy combining conditional triggering and sliding window sample updates is adopted. Based on a drone shooting frequency of 15 minutes per session, after each acquisition of new data, a sliding window mechanism is used to supplement the time-series sample set with the latest extracted global average fog concentration data, while the oldest data set is discarded. At the model update level, a conditional triggering mechanism is used. The ARIMA model is only retrained when the fog concentration changes significantly, i.e., the change exceeds a set threshold of 10% for multiple consecutive time intervals. This involves substituting data containing significant changes into the model for retraining, ensuring that the prediction results always closely reflect the real-time changes in the fog layer, continuously optimizing accuracy. The power irradiance correction model and the power prediction model do not need to adjust their own parameters during rolling iterations; they only need to receive the updated prediction data as input. The time-series sample update formula is as follows: .

[0088] in, To update the fog concentration time series sample set, the earliest data set (t1) was removed, and the most recently captured data set (t) was added. n+1 This avoids slow model updates due to excessive samples. This is the latest global average fog concentration data extracted from drone footage.

[0089] This invention introduces image data of power plants collected by drones, enabling refined identification of localized and rapidly changing fog. Accurate global fog concentration data is then obtained through spatial interpolation and temporal difference prediction, accurately characterizing the real-time distribution and evolution of the fog layer. Based on the fog concentration, the extinction coefficient is calculated and the irradiance is corrected, precisely quantifying the attenuation effect of fog on solar radiation and addressing the lack of optimization for foggy irradiance in existing technologies. Finally, power prediction is performed using the corrected irradiance, ensuring the model input closely matches real-world irradiance conditions in foggy weather. This fundamentally overcomes the poor adaptability of existing technologies to localized and rapidly changing fog, improving the prediction accuracy of photovoltaic power generation under foggy conditions.

[0090] As a complete example, such as Figure 2 As shown, taking the fog inspection scenario of a photovoltaic power station in a plain as an example, the drone conducts low-altitude patrols at a fixed frequency of 15 minutes / time, with preset altitude and flight path, to collect images of the entire power station area. Simultaneously, it acquires numerical weather prediction (NWP) data, power station layout parameters, historical power generation, on-site measured temperature and humidity, and installed capacity, forming a multi-source dataset containing visual, meteorological, and power station operation information, providing raw data support for subsequent fog feature analysis and power prediction.

[0091] The multi-source data collected by UAVs undergoes unified preprocessing. First, spatiotemporal matching is performed to align images, meteorological data, and operational data from different collection frequencies and timestamps to the same time reference. Then, the image data is denoised and contrast enhanced, and the meteorological and operational data are de-identified and filled with missing values. Finally, resolution unification and dimension normalization are completed to eliminate format and numerical differences between different data sources and ensure data consistency and validity.

[0092] Based on the dark channel prior dehazing algorithm, the local windows of the preprocessed image are traversed to extract the minimum gray value of the red, green, and blue channels to obtain the dark channel image. The atmospheric transmittance calculation formula is derived by combining the fog imaging model, and the atmospheric transmittance data is calculated pixel by pixel. Then, the transmittance is converted into fog concentration value by the formula FC=1-t. According to the value range, it is divided into four levels: no fog / light fog edge, light fog, heavy fog, and strong dense fog. The pixel-level fog concentration is quantitatively extracted to characterize the spatial distribution and concentration difference of heavy fog in the power station area.

[0093] Spatial interpolation is performed on pixel-level fog concentration data to expand it into average fog concentration data across the entire photovoltaic power station. A time series is constructed based on at least 1-2 hours of previous monitoring data. The time series is then input into the ARIMA model. First, the non-stationarity of the data is eliminated through difference operations. Then, the historical fog concentration variation pattern is fitted to obtain the differential prediction value of fog concentration at each future time step. The differential prediction value is accumulated based on the current average fog concentration across the entire area to finally obtain the fog concentration prediction result for the entire power station area for the next 4 hours, thus realizing the time series extrapolation of fog characteristics.

[0094] Based on the time-series prediction results of fog concentration, the corresponding atmospheric transmittance data is first obtained by reverse conversion. Then, combined with the flight altitude and shooting angle of the UAV, the effective path length of the fog layer is derived through trigonometric functions. According to the Beer-Lambert law, the natural logarithm of atmospheric transmittance is calculated, and the logarithm value is calculated by making a negative ratio with the effective path length of the fog layer to obtain the extinction coefficient. This coefficient can quantitatively characterize the ability of fog to absorb and scatter sunlight.

[0095] First, by combining the solar constant and the solar altitude angle at future times, the theoretical irradiance under fog-free conditions is calculated. Then, based on the fog concentration level corresponding to the extinction coefficient, the piecewise quadratic functions and fitting coefficients corresponding to light fog, heavy fog, and dense fog are retrieved, and the extinction coefficient is substituted into the function to obtain the irradiance correction coefficient. Finally, the theoretical irradiance under clear weather is multiplied by the correction coefficient to refine the initial irradiance, thereby obtaining the predicted value of the actual light intensity irradiated on the photovoltaic panel surface under foggy weather, eliminating the irradiance deviation caused by the extinction effect of the fog layer.

[0096] The corrected irradiance prediction is used as the core input, combined with characteristic data such as temperature, humidity, and extinction coefficient, and then input into the trained ultra-short-term photovoltaic power generation prediction model (such as the XGBoost model). Through forward inference and regression calculation, the initial prediction value of photovoltaic power generation for the next 4 hours is obtained. Then, the initial power is calibrated according to the installed capacity of the power station to constrain the value within a reasonable operating range, and finally, a stable and reliable photovoltaic power generation prediction result for foggy days is output.

[0097] Based on a drone shooting frequency of 15 minutes per shot, the latest extracted global average fog concentration data is added to the time series sample set, while the earliest set of data is removed to maintain a fixed number of frames in the sample set. When the fog concentration change exceeds the 10% threshold for multiple consecutive time moments, the ARIMA model is retrained, and the model parameters are optimized based on the updated time series sample set. Subsequently, the fog concentration, irradiance, and power prediction results for the next 4 hours are continuously updated, and the time steps that have passed are removed and new prediction steps are added to ensure that the prediction results always fit the real-time changes of the fog layer and continuously improve the prediction accuracy.

[0098] This embodiment also provides a drone-based photovoltaic power generation prediction device for foggy weather. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0099] This embodiment provides a photovoltaic power generation prediction device based on unmanned aerial vehicles (UAVs) for foggy weather, such as... Figure 3 As shown, it includes: The acquisition module 301 is used to acquire multi-source data of the photovoltaic power station and process the multi-source data to obtain fog concentration characteristic data. The multi-source data includes image data of the photovoltaic power station collected by UAV, meteorological data of the photovoltaic power station, and operating condition data. The analysis module 302 is used to perform spatial interpolation on the fog concentration of each pixel in the fog concentration feature data to obtain a time series sequence, and to perform differential analysis and prediction on the time series sequence to obtain the global fog concentration prediction data of the photovoltaic power station. The calculation module 303 is used to calculate the extinction coefficient, which characterizes the ability of fog weather to attenuate sunlight, using global fog concentration prediction data and fog layer optical parameter data. Based on the spatial and temporal distribution of different fog concentration levels in different local areas at future times, the extinction coefficient is fitted to obtain the irradiance correction coefficient for different future times and different local areas. The irradiance correction coefficient is used to correct the basic solar radiation parameters to obtain the predicted value of the actual light intensity that can be irradiated on the photovoltaic panel during fog weather. The prediction module 304 is used to predict the photovoltaic power generation of a photovoltaic power station at a future time based on the predicted value of light intensity.

[0100] In this embodiment, the acquisition module 301 is used to calculate the dark channel value of the image data in the multi-source data, and substitute the dark channel value into the fog imaging model to derive the transmittance calculation formula to obtain the atmospheric transmittance of each pixel in the image data; calculate the fog concentration value of each pixel based on the atmospheric transmittance of each pixel; classify the fog concentration value of each pixel according to the value range to obtain the quantified and classified fog concentration feature data.

[0101] In this embodiment, the analysis module 302 is used to perform spatial interpolation calculation based on the fog concentration of each frame of the fog concentration feature data to obtain the global fog concentration distribution data of the photovoltaic power station, and to perform statistical averaging calculation based on the global fog concentration distribution data to obtain the global average fog concentration data of the photovoltaic power station; to construct a fog concentration time series based on the global average fog concentration data and the historical monitoring data of the photovoltaic power station; to input the fog concentration time series into the ARIM model, to perform differential stabilization processing on the fog concentration time series and to perform time series prediction to obtain the fog concentration differential prediction value; and to perform time series extrapolation calculation by accumulating the fog concentration differential prediction value based on the current global average fog concentration to obtain the global fog concentration prediction data of the photovoltaic power station at future times.

[0102] In this embodiment, the calculation module 303 is used to perform reverse conversion processing on the global fog concentration prediction data using the atmospheric transmittance conversion formula to obtain the corresponding atmospheric transmittance data; calculate the flight altitude and shooting angle of the UAV to obtain the effective path length of the fog layer; perform logarithmic calculation processing on the atmospheric transmittance data using the natural logarithm operation function to obtain the logarithmic transmittance value; and calculate the extinction coefficient, which characterizes the ability of fog to attenuate sunlight, based on the negative ratio of the logarithmic transmittance value and the effective path length of the fog layer.

[0103] In this embodiment, the calculation module 303 is used to perform a sinusoidal product operation on the basic solar radiation parameters to obtain the theoretical irradiance under fog-free conditions. Within a preset future time period, an extinction coefficient is first fitted using an exponential model to obtain an irradiance correction coefficient. The extinction coefficient and the measured ground irradiance are recorded in real time at a preset frequency, and a scatter plot of the extinction coefficient and the irradiance correction coefficient is plotted. It is determined whether the data distribution characteristics in the scatter plot exhibit an exponential trend. If so, the exponential model is used to determine the irradiance correction coefficient. If not, a piecewise quadratic function based on fog concentration levels is used to perform a graded fitting of the extinction coefficient according to the spatial distribution of fog concentration levels to obtain the irradiance correction coefficient. The theoretical irradiance under clear skies and the irradiance correction coefficient are multiplied to obtain initial irradiance prediction data, and the initial irradiance prediction data is spatially averaged to obtain the predicted light intensity of the photovoltaic panel actually irradiated during foggy weather.

[0104] In this embodiment, the calculation module 303 is used to compare the extinction coefficient with a preset threshold to determine the target level of the current fog concentration; obtain the piecewise quadratic function and fitting coefficients corresponding to the target level; substitute the extinction coefficient into the piecewise quadratic function, and use the fitting coefficients to solve the quadratic polynomial to obtain the irradiance correction coefficient of a single point; perform time-series smoothing processing based on the irradiance correction coefficient of the single point to obtain the irradiance correction coefficient at future times.

[0105] In this embodiment, the prediction module 304 is used to perform dimensional splicing and normalization processing based on the predicted light intensity, temperature and humidity data of the photovoltaic power station, and extinction coefficient to obtain power input feature data; using historical power data as labels and the input feature data as model input, supervised learning training and parameter iterative optimization are performed on the power prediction model to obtain the trained power prediction model; using the predicted light intensity and meteorological data as input to the trained power prediction model, forward inference and regression calculations are performed to obtain the initial power data at future times; the initial power data and the installed capacity of the photovoltaic power station are range-calibrated to obtain the photovoltaic power generation power of the photovoltaic power station at future times.

[0106] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 4As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system).

[0107] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0108] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0109] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device as shown by a landing page for an app. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0110] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0111] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0112] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0113] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for predicting photovoltaic power generation in foggy weather based on unmanned aerial vehicles (UAVs), characterized in that, The method includes: Multi-source data of a photovoltaic power station is acquired and processed to obtain fog concentration characteristic data. The multi-source data includes image data of the photovoltaic power station collected by a drone, meteorological data of the photovoltaic power station, and operating condition data. Spatial interpolation is performed on the fog concentration of each pixel in the fog concentration feature data to obtain a time series sequence. Differential analysis is then performed on the time series sequence to predict the overall fog concentration of the photovoltaic power station. The overall fog concentration prediction data includes the spatial and temporal distribution of different fog concentration levels in different local areas at future times. Based on the overall fog concentration prediction data, ground-measured irradiance is used to correct the overall fog concentration prediction data according to region, thus obtaining the spatial and temporal distribution of different fog concentration levels in different local areas at future times. The extinction coefficient, which characterizes the ability of fog weather to attenuate sunlight, is calculated using the global fog concentration prediction data and fog layer optical parameter data. The extinction coefficient is fitted based on the spatial and temporal distribution of different fog concentration levels in different local areas at different future times to obtain the irradiance correction coefficient for different future times and different local areas. The irradiance correction coefficient is then used to correct the basic solar radiation parameters to obtain the predicted value of the actual light intensity that can reach the photovoltaic panel during foggy weather. The photovoltaic power generation capacity of the photovoltaic power station at future times is predicted based on the predicted light intensity value.

2. The method according to claim 1, characterized in that, The process of processing the multi-source data to obtain fog concentration characteristic data includes: Calculate the dark channel value of the image data in the multi-source data, and substitute the dark channel value into the fog imaging model to derive the transmittance calculation formula to obtain the atmospheric transmittance of each pixel in the image data. The fog concentration value of each pixel is calculated based on the atmospheric transmittance of each pixel. The fog concentration values ​​of each pixel are classified according to their numerical range to obtain quantified fog concentration feature data.

3. The method according to claim 1, characterized in that, The step of performing differential analysis on the time series to obtain the global fog concentration prediction data for the photovoltaic power station includes: Spatial interpolation calculation is performed on the fog concentration of each frame pixel in the fog concentration feature data to obtain the global fog concentration distribution data of the photovoltaic power station, and statistical averaging calculation is performed on the global fog concentration distribution data to obtain the global average fog concentration data of the photovoltaic power station. Based on the global average fog concentration data and the historical monitoring data of the photovoltaic power station, a fog concentration time series is constructed. The fog concentration time series is input into the ARIM model, and the fog concentration time series is subjected to differential stabilization and time series prediction to obtain the fog concentration differential prediction value. Using the current average fog concentration across the entire region as a benchmark, the cumulative fog concentration difference prediction values ​​are extrapolated over time to obtain the predicted fog concentration data for the photovoltaic power station at future times.

4. The method according to claim 1, characterized in that, The extinction coefficient, which characterizes the ability of fog weather to attenuate sunlight, is calculated using the global fog concentration prediction data and fog layer optical parameter data, including: The atmospheric transmittance conversion formula is used to perform reverse conversion on the global fog concentration prediction data to obtain the corresponding atmospheric transmittance data. The effective path length of the fog layer is obtained by calculating the drone's flight altitude and shooting angle. The atmospheric transmittance data is logarithmically processed using the natural logarithm operation function to obtain the logarithmic transmittance value. The extinction coefficient, which characterizes the ability of fog to attenuate sunlight, is calculated by using a negative ratio between the logarithmic value of the transmittance and the effective path length of the fog layer.

5. The method according to claim 1, characterized in that, The extinction coefficient is fitted based on the spatial and temporal distribution of fog concentration levels in different local areas at different future times to obtain irradiance correction coefficients for different future times and different local areas. These irradiance correction coefficients are then used to correct the basic solar radiation parameters to obtain the predicted irradiance intensity that can actually reach the photovoltaic panels during foggy weather, including: The theoretical irradiance on a clear day under fog-free conditions is obtained by performing a sinusoidal product operation on the aforementioned basic solar radiation parameters. Within a predetermined time period in the future, the extinction coefficient will first be fitted using an exponential model to obtain the irradiance correction coefficient. The extinction coefficient and the measured ground irradiance are recorded in real time using a preset frequency, and a scatter plot of the extinction coefficient and the irradiance correction coefficient is plotted. Determine whether the data distribution characteristics in a scatter plot exhibit an exponential trend; If so, the exponential model described above will continue to be used to determine the irradiance correction coefficient; If not, the extinction coefficient is fitted in stages according to the spatial distribution of fog concentration levels using a piecewise quadratic function of fog concentration levels to obtain the irradiance correction coefficient. The initial irradiance prediction data is obtained by multiplying the theoretical irradiance on a sunny day and the irradiance correction coefficient. The initial irradiance prediction data is then spatially averaged to obtain the predicted irradiance intensity of the photovoltaic panel under foggy weather.

6. The method according to claim 5, characterized in that, The step of using a piecewise quadratic function based on fog concentration levels to perform a graded fitting of the extinction coefficient to obtain the irradiance correction coefficient for future times includes: By comparing the extinction coefficient with a preset threshold, the target level of the current fog concentration is determined; Obtain the piecewise quadratic function and fitting coefficients corresponding to the target level; Substitute the extinction coefficient into the piecewise quadratic function, and solve the quadratic polynomial using the fitting coefficient to obtain the irradiance correction coefficient for a single point. The irradiance correction coefficient for the single point is used for time-series smoothing to obtain the irradiance correction coefficient for future times.

7. The method according to claim 1, characterized in that, The prediction of the photovoltaic power generation capacity of the photovoltaic power station at future times based on the predicted light intensity includes: Based on the predicted light intensity, the temperature and humidity data of the photovoltaic power station, and the extinction coefficient, dimensional splicing and normalization processing are performed to obtain the power input feature data. Using historical power data as labels and the input feature data as model input, supervised learning training and parameter iterative optimization are performed on the power prediction model to obtain the trained power prediction model. Using the predicted light intensity and the meteorological data to train the power prediction model, forward inference and regression calculations are performed to obtain the initial power data for future times. The initial power data and the installed capacity of the photovoltaic power station are range-calibrated to obtain the photovoltaic power generation power of the photovoltaic power station at future times.

8. A photovoltaic power generation prediction device based on unmanned aerial vehicles (UAVs) in foggy weather, characterized in that, The device includes: The acquisition module is used to acquire multi-source data from the photovoltaic power station and process the multi-source data to obtain fog concentration characteristic data. The multi-source data includes image data of the photovoltaic power station collected by the UAV, meteorological data of the photovoltaic power station, and operating condition data. The analysis module is used to perform spatial interpolation on the fog concentration of each pixel in the fog concentration feature data to obtain a time series sequence, and to perform differential analysis and prediction on the time series sequence to obtain the global fog concentration prediction data of the photovoltaic power station. The global fog concentration prediction data includes the spatial and temporal distribution of different fog concentration levels in different local areas at future times. Based on the global fog concentration prediction data, the global fog concentration prediction data is corrected according to the region using the ground measured irradiance to obtain the spatial and temporal distribution of different fog concentration levels in different local areas at future times. The calculation module is used to calculate the extinction coefficient, which characterizes the ability of fog weather to attenuate sunlight, using the global fog concentration prediction data and fog layer optical parameter data. Based on the spatial and temporal distribution of different fog concentration levels in different local areas at future times, the extinction coefficient is fitted to obtain the irradiance correction coefficient for different future times and different local areas. The irradiance correction coefficient is used to correct the basic solar radiation parameters to obtain the predicted value of the actual light intensity that can shine on the photovoltaic panel during fog weather. The prediction module is used to predict the photovoltaic power generation capacity of the photovoltaic power station at future times based on the predicted light intensity value.

9. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the method of any one of claims 1 to 7.

Images