A new energy power station snow accumulation dynamic identification and power prediction method and system

Abstract

The application discloses a new energy power station accumulated snow dynamic identification and power prediction method and system in the field of new energy power generation, and the method comprises the following steps: S1, acquiring a satellite cloud image, and preprocessing the satellite cloud image containing accumulated snow of photovoltaic components; S2, determining whether the image contains accumulated snow in combination with a visible light cloud image and a short-wave infrared band to distinguish accumulated snow from cloud layers; S3, extracting features from the accumulated snow image, including shape features and size features; training an ARIMA model based on the features, and obtaining future accumulated snow area changes through the ARIMA model prediction; and S4, training an LSTM prediction model according to the future accumulated snow area changes and numerical weather forecast information, and obtaining future power changes of the photovoltaic power station through the LSTM prediction model. The application not only improves the adaptability of the photovoltaic power station under complex meteorological conditions, but also provides strong technical support for intelligent management and control of the photovoltaic power station.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation, specifically to a method and system for dynamic identification of snow accumulation and power prediction in new energy power plants. Background Technology

[0002] Currently, satellite cloud imagery technology has made significant progress in extreme weather forecasting for wind and solar power plants. By analyzing satellite cloud imagery data, the movement, development, and changes of clouds can be monitored in real time, thereby predicting extreme weather events that may affect power plants, such as storms, hail, and thunderstorms. This data is typically provided by meteorological satellites, which can cover a wide geographical area and have high resolution and high-frequency update capabilities.

[0003] Existing forecasting systems typically combine multiple meteorological models and algorithms, such as numerical weather prediction (NWP) models, machine learning, and deep learning techniques, to improve the accuracy and reliability of forecasts. These systems can provide forecasts ranging from hours to several days, helping operators of renewable energy power plants respond promptly to reduce the potential damage to equipment and operations caused by extreme weather.

[0004] However, existing technologies still have certain limitations in terms of accuracy, forecast timeliness, and adaptability to complex weather patterns. For example, forecasting models may not be accurate enough when dealing with small-scale weather phenomena, or may contain errors in predicting the intensity and duration of extreme weather events. In addition, existing systems also face challenges in real-time data integration and processing, especially when dealing with large amounts of data and rapid update speeds.

[0005] Chinese patent CN121983947A discloses a photovoltaic prediction method that integrates high-precision satellite cloud images with ground monitoring, but it is not accurate enough when dealing with small-scale weather phenomena. Summary of the Invention

[0006] This invention addresses the limitations of existing technologies in terms of accuracy, forecast timeliness, and adaptability to complex weather patterns. It proposes a method and system for dynamic snow accumulation identification and power prediction in new energy power plants, which not only improves the adaptability of photovoltaic power plants under complex weather conditions but also provides strong technical support for the intelligent management and control of photovoltaic power plants.

[0007] To achieve the above objectives, the present invention adopts the following technical solution.

[0008] A method for dynamic snow accumulation identification and power prediction in new energy power plants includes the following steps: S1, acquire satellite cloud images and preprocess the satellite cloud images containing snow accumulation on photovoltaic modules; S2, combine visible light cloud image and shortwave infrared band to determine whether there is snow in the image in order to distinguish snow from clouds; S3 extracts features from snow images, including shape and size features; trains an ARIMA model based on these features, and uses the ARIMA model to predict future changes in snow area. S4. Based on future changes in snow cover area and numerical weather forecast information, train an LSTM prediction model and obtain the future power changes of the photovoltaic power station through the LSTM prediction model.

[0009] In this technical solution, the satellite cloud image containing snow cover on photovoltaic modules is first preprocessed, and then snow cover is identified based on visible light cloud image and shortwave infrared band. Relevant features are extracted from the image, and an ARIMA model is trained and constructed to predict the change in snow cover area on photovoltaic modules. Finally, the power of the photovoltaic power station is predicted based on the LSTM prediction model.

[0010] The present invention is further configured such that step S1 includes: S11, homomorphic filtering is applied to the satellite cloud image to obtain an enhanced image; S12, Adaptive filtering is used to denoise the image, and the effect of adaptive filtering is quantitatively analyzed; S13, perform threshold segmentation on the image to generate a binary image containing both snow-covered and non-snow-covered regions.

[0011] In this technical solution, the preprocessing mainly includes image enhancement, noise removal, and threshold segmentation in preparation for subsequent snow accumulation identification.

[0012] The present invention is further configured such that step S2 includes: The Normalized Difference Snow Index (NDSI) is calculated. NDSI is the difference between the reflectance of the visible light band and the reflectance of the shortwave infrared band, divided by the sum of the reflectance of the visible light band and the shortwave infrared band. At the same time, the brightness temperature of the infrared band is compared with the low and high critical values. Combining NDSI and brightness temperature, snow cover and cloud cover are distinguished.

[0013] In this technical solution, by combining visible light and 1.6μm shortwave infrared band for analysis, snow cover and clouds can be effectively distinguished.

[0014] The present invention is further configured such that step S3 includes: S31, determine whether the time series data has stabilized. If yes, proceed to S31; otherwise, proceed to S32. S32, perform the difference operation; S33, Use the ADF test to check the stationarity of the time series; S34, select the ARIMA model using either AIC or BIC; S35, each feature parameter is fitted to the ARIMA model; S36 uses the ARIMA model to predict future changes in snow cover area.

[0015] In this technical solution, the extracted features are input in the form of time series to construct an ARIMA model to predict future changes in snow cover area.

[0016] The present invention is further configured such that step S4 includes: The future snow cover change and numerical weather prediction information are used as the input data training set for the LSTM prediction model, while the future output power of the photovoltaic power station is used as the output data training set. After completing the construction of the LSTM prediction model, the output power of the photovoltaic power plant in the future is predicted using the LSTM prediction model.

[0017] The present invention is further configured such that the satellite cloud image is specifically acquired through the Fengyun-4 satellite.

[0018] The present invention is further configured such that: the satellite cloud image includes a visible light cloud image, a water vapor cloud image and an infrared cloud image, wherein the visible light cloud image is made using the principle of cloud tops reflecting sunlight, and the infrared cloud image is made by satellite measuring the infrared radiation emitted from the Earth's surface and cloud surfaces.

[0019] The present invention is further configured such that: the quantitative analysis of the adaptive filtering effect adopts several evaluation indicators, including signal-to-noise ratio and mean square error; by calculating the signal-to-noise ratio and mean square error of the images before and after filtering, the effect of the adaptive filter is quantitatively analyzed.

[0020] The present invention is further configured such that: the shape feature can identify the boundary of the snow-covered area and describe its outline; the size feature is the area of ​​the snow-covered area, which can be quantified and determined by pixel count and image resolution.

[0021] A dynamic snow accumulation identification and power prediction system for a new energy power plant, applicable to the aforementioned dynamic snow accumulation identification and power prediction method for a new energy power plant, comprising: The image preprocessing module preprocesses the acquired satellite cloud images; The snow cover identification module distinguishes between snow cover and cloud cover using visible light cloud images and shortwave infrared bands. The snow cover area prediction module predicts future changes in snow cover area by constructing an ARIMA model. The photovoltaic power plant power prediction module uses an LSTM prediction model to obtain the future power changes of the photovoltaic power plant.

[0022] In this technical solution, the image preprocessing module, snow accumulation identification module, snow accumulation area prediction module, and photovoltaic power station power prediction module are connected in sequence and work together to complete the dynamic identification of snow accumulation and power prediction of new energy power stations.

[0023] The present invention can bring about the following beneficial effects.

[0024] This invention can predict snow cover changes in the future, thereby accurately predicting power variations in photovoltaic (PV) power plants. These predictions are crucial for adjusting and managing power plant output, as snow cover directly impacts the power generation efficiency of PV panels. Real-time monitoring and prediction can effectively improve the operational efficiency and reliability of PV power plants, reduce power generation losses due to weather changes, and ensure the stability and economic viability of power supply. This invention not only enhances the adaptability of PV power plants to complex weather conditions but also provides strong technical support for the intelligent management and control of PV power plants. Attached Figure Description

[0025] Figure 1 This is a flowchart of step S3 of the method for dynamic identification of snow accumulation and power prediction of a new energy power station according to the present invention.

[0026] Figure 2 This is a flowchart of step S4 of the method for dynamic identification of snow accumulation and power prediction of a new energy power plant according to the present invention. Detailed Implementation

[0027] Example 1:

[0028] To address the limitations of existing technologies in terms of accuracy, forecast timeliness, and adaptability to complex weather patterns, this embodiment proposes a method for dynamic snow cover identification and power prediction for new energy power plants. This method can accurately identify the land surface, mid-to-high clouds, low clouds, and snow cover in the area where the photovoltaic power plant is located via satellite. Image recognition and analysis of snow cover changes at the photovoltaic power plant are performed, and based on this data, snow cover changes in future time periods are predicted, thereby accurately predicting power changes at the photovoltaic power plant. These predictions are crucial for adjusting and managing the power plant's output, as snow cover directly affects the power generation efficiency of photovoltaic panels. Real-time monitoring and prediction can effectively improve the operational efficiency and power generation reliability of photovoltaic power plants, reduce power generation losses due to weather changes, and ensure the stability and economy of power supply.

[0029] The present invention provides a method for dynamic identification of snow accumulation and power prediction in a new energy power plant, which mainly includes the following steps.

[0030] Step S1: Obtain satellite cloud images and preprocess the satellite cloud images containing snow accumulation on photovoltaic modules.

[0031] In this embodiment, the satellite cloud imagery is specifically acquired through the Fengyun-4 satellite. The Fengyun-4 satellite provides accurate meteorological data, offering strong support for weather forecasting, climate research, and natural disaster early warning. Particularly in monitoring snow cover and cloud cover changes at photovoltaic power plants, the multi-band remote sensing capabilities of Fengyun-4 enable precise identification of the land surface, mid-to-high clouds, low clouds, and snow cover, thereby improving the accuracy of power prediction.

[0032] Step S1 includes the following sub-steps.

[0033] Step S11: Perform homomorphic filtering on the satellite cloud image to obtain an enhanced image.

[0034] Homomorphic filtering can significantly improve the local contrast of an image, making the edges and textures of snow clearer.

[0035] Homomorphic filtering is an image enhancement method performed in the frequency domain that improves image contrast and sharpness by adjusting the image's frequency response. In snow-covered images, homomorphic filtering can effectively enhance details in snow-covered areas while reducing the effects of shadows and uneven lighting caused by terrain or other factors.

[0036] Homomorphic filtering first transforms the image from the spatial domain to the frequency domain, then adjusts the frequency components, and finally transforms the image back to the spatial domain.

[0037] In more detail, homomorphic filtering of satellite cloud images includes the following process.

[0038] Amplitude and Phase Representation: First, the image is represented in complex form, where amplitude (A) represents the brightness of the image, and phase (θ) represents the texture and edge information of the image. For each pixel (x,y), it can be represented as: I(x,y) equals A(x,y) multiplied by e. jθ(x,y) ; where j is the imaginary unit.

[0039] Logarithmic transformation of an image: Taking the natural logarithm of the amplitude portion of the image compresses bright areas and expands dark areas, enhancing the image contrast: L(x,y) equals log(A(x,y) plus C), where C is a small constant to avoid instability when the logarithm is negative or when taking the logarithm.

[0040] Fourier Transform: Performing a Fourier transform on an image that has undergone a logarithmic transform converts it from the spatial domain to the frequency domain: F(u,v) equals .

[0041] High-pass filtering: In the frequency domain, a high-pass filter can be applied to further enhance the high-frequency components of an image, namely edges and details. equal Multiply by F(u,v), where, It is the transfer function of the high-pass filter.

[0042] Inverse Fourier Transform: Converts a high-pass filtered frequency domain image back to the spatial domain. equal .

[0043] Exponential transform: Performing an exponential transform on an image in the spatial domain recovers the image's amplitude information. equal Multiply .

[0044] Combination of amplitude and phase: The phase information of the original image is combined with the amplitude information after homomorphic filtering to obtain the final enhanced image. equal .

[0045] Step S12: Adaptive filtering is used to denoise the image, and the effect of adaptive filtering is quantitatively analyzed.

[0046] Several evaluation metrics are used to quantitatively analyze the effect of adaptive filtering, including signal-to-noise ratio (SNR) and mean square error (MSE). The effect of the adaptive filter is quantitatively analyzed by calculating the SNR and MSE of the images before and after filtering.

[0047] Noise is a common problem in remote sensing images, and it can originate from electronic noise in the sensor, atmospheric interference, or other random factors. In the analysis of snow-covered images, the presence of noise can obscure the true characteristics of the snow, leading to inaccurate identification and quantification. Therefore, noise removal is an indispensable step in image preprocessing.

[0048] Adaptive filtering is an intelligent noise removal technique that dynamically adjusts filter parameters based on the local characteristics of an image. This means that the strength and type of the filter are adjusted according to the noise level and features of each region in the image. For example, if a region has a high noise level, the adaptive filter may apply a stronger smoothing operation; while at the edges or in textured areas of the image, the filter may reduce smoothing to avoid overly blurring image details.

[0049] The process of denoising an image using adaptive filtering includes the following steps.

[0050] Determining filter parameters: The parameters of an adaptive filter need to be determined based on the local characteristics of the image. For example, the local variance or gradient of the image can be used to adjust the filter strength. In snow-covered images, these parameters can help identify flat areas (such as snow-covered surfaces) and areas with distinct edges (such as snow boundaries).

[0051] Applying adaptive filters: Adaptive filters are applied to the image based on defined parameters. This may involve dynamically adjusting the kernel size, shape, or weights of the filter. In a snow-covered image, the filter might use a larger kernel in flat areas to smooth noise, while using a smaller kernel in edge areas to preserve detail.

[0052] Quantitative evaluation of filtering effect: Quantitative analysis of the filtered snow image to determine whether snow features are preserved or improved. This includes whether the area and shape of the snow-covered region remain consistent; whether snow boundaries are sharp and whether false identification due to noise is reduced; and whether the texture of the snow surface is preserved and whether noise-induced artifacts are reduced. Evaluation metrics used include: Signal-to-noise ratio (SNR): A measure of the ratio of signal strength to noise strength in an image. A filtered image has a higher SNR value, indicating a reduced noise level. The formula for calculating SNR is: SNR equals 10 multiplied by... Among them, the signal power in the snow-covered area equal ,in, This is the average value of all pixel values ​​within the signal area. Total number of pixels in the signal area; noise power equal Multiply , The variance of pixel values ​​in the noise region. This represents the total number of pixels in the signal area.

[0053] Mean Squared Error (MSE): Measures the difference between the original image and the filtered image. After filtering, the MSE should decrease. Its calculation formula is: MSE = ... Multiply ,in, These are the pixel values ​​in the original image. N represents the pixel value in the filtered image, and N is the total number of pixels.

[0054] Quantitative analysis of the effectiveness of adaptive filters can be performed by calculating the SNR and MSE of the images before and after filtering. The parameters of the adaptive filter, such as kernel size, shape, or weights, can be dynamically adjusted based on the local characteristics of the snow-covered image. For example, in snow-covered areas, the kernel size can be increased to smooth noise; in snow-covered boundaries or non-snow-covered areas, the kernel size can be decreased to preserve edges and details.

[0055] Step S13: Threshold segmentation is performed on the image to generate a binary image containing both snow-covered and non-snow-covered regions.

[0056] In image segmentation of snow cover areas in photovoltaic power plants, thresholding is a simple and fast method that separates snow from the background by setting a brightness threshold. First, the global threshold T of the image is calculated, which can be automatically determined by analyzing the image's histogram and applying Otsu's method. The goal is to maximize the inter-class variance T. Next, this threshold is used to classify each pixel as either snow-covered or non-snow-covered. If the pixel intensity... If the pixel's size is greater than a threshold T, it is considered part of the snow area; otherwise, it is considered part of the background. In pixel intensity When the value is greater than the threshold T, the value equals 1. In pixel intensity When the value is not greater than the threshold T, the value is equal to 0, thus generating a binary image in which the snow-covered area is marked as white (or 1), and the remaining area is black (or 0).

[0057] Step S2: Combine visible light cloud image and shortwave infrared band to determine whether there is snow in the image in order to distinguish snow from clouds.

[0058] On visible light cloud images, snow and clouds often appear white, making them difficult to distinguish. However, in the shortwave infrared band, the reflectivity of snow decreases significantly with increasing wavelength, while the reflectivity of clouds changes relatively little. Particularly at a wavelength of 1.6 μm, the reflectivity of snow and clouds shows a clear difference. Therefore, by combining analysis using both visible light and the 1.6 μm shortwave infrared band, it is possible to effectively distinguish between snow and clouds.

[0059] Step S2 includes the following processes: The Normalized Difference Snow Index (NDSI) is calculated. NDSI is the difference between the reflectance of the visible light band and the reflectance of the shortwave infrared band, divided by the sum of the reflectance of the visible light band and the shortwave infrared band. At the same time, the brightness temperature of the infrared band is compared with the low and high critical values. Combining NDSI and brightness temperature, snow cover and cloud cover are distinguished.

[0060] Specifically, the Normalized Snow Index (NDSI) is one of the most effective methods for snow cover identification and extraction. Satellites provide the 0.55-0.75 μm visible light band and the 1.6 μm shortwave infrared band suitable for NDSI calculation. Therefore, satellite snow cover identification can be achieved through NDSI.

[0061] It also includes the following criteria: R is greater than R th NDSI is greater than NDSI th T is less than T thH And greater than T thLWhere R represents the reflectance reading in the visible light band, and its corresponding identification threshold is R0. th Snow and clouds, due to their high reflectivity in the visible light band, can be clearly distinguished from other land cover types, such as water and bare ground. The Normalized Differential Snow Index (NDSI) was calculated for the 1.6 μm band and the visible light band, and a threshold NDSI was set for its identification. th Typically, the NDSI value of snow cover is greater than 0.3, further confirming its presence. Simultaneously, the brightness temperature (T) in the infrared band and a preset low threshold value (T) are used... thL ) and high critical value (T) thH A comparison of these snow indexes and far-infrared brightness temperature data can reveal the reflectance characteristics of snow cover and clouds in the shortwave infrared band. Snow cover has low reflectance in the 1.6 μm band, while clouds maintain high reflectance. Combining these snow cover indices with far-infrared brightness temperature data allows for a more accurate distinction between high clouds, mid-to-low clouds, and snow cover.

[0062] Step S3: Extract features from the snow image, including shape and size features; train an ARIMA model based on the features, and predict future changes in snow area using the ARIMA model.

[0063] Shape features can identify the boundaries of snow-covered areas and describe their contours; size features are the area of ​​snow-covered areas, which can be quantified by pixel count and image resolution.

[0064] For shape features, quantization can be achieved through boundary tracking, with the main purpose of identifying the boundaries of snow-covered areas and describing their contours.

[0065] Specifically, identifying the edges of snow-covered areas on photovoltaic modules helps to more clearly define the boundaries of snow in the image. The Laplacian operator is used to detect edges by identifying local extrema of image brightness. It calculates the zero-crossing points of the second derivative of the image brightness. The Laplacian operator can be expressed as: L equals I xx Add I yy , among which, I xx and I yy It is the second derivative of the image in the horizontal and vertical directions, which can be obtained through convolution operations. The edge detection algorithm outputs a binary image, where edges are marked as white (or non-zero values) and the remaining areas are black (or zero values).

[0066] Size, in this context, typically refers to the area of ​​the snow-covered region, serving as a direct indicator of snow cover. Size can be quantified using pixel counting and area measurement methods.

[0067] Pixel count: In a binarized or segmented image of a snow-covered region, the number of white (or non-zero) pixels is counted, which is the pixel count of the snow-covered region.

[0068] Area measurement: Convert pixel counts to actual area and set the image resolution. The actual area A of the snow-covered region can be calculated using the following formula: A equals Pixel_Count multiplied by Resolution. 2 .

[0069] In this embodiment, in order to accurately predict the change in the snow area on the surface of the photovoltaic module, the future change in the snow area of ​​the photovoltaic module is predicted by using an autoregressive integral moving average (ARIMA) model.

[0070] First, it is necessary to ensure the stationarity of the feature time series data, because most time series are non-stationary in their original form.

[0071] refer to Figure 1 Step S3 includes the following process.

[0072] Step S31: Determine whether the time series data is stationary. If yes, proceed to S31; otherwise, proceed to S32. A stationary time series means that its statistical properties (such as mean and variance) remain unchanged throughout the time.

[0073] Step S32: Perform the differencing operation. Most time series data are not stationary in their initial state, hence the differencing operation; differencing is a technique to transform a non-stationary series into a stationary one. By calculating the difference between consecutive observations, the trend and seasonal components of the series can be reduced or eliminated, resulting in a more stationary series.

[0074] Step S33: Use the ADF test to check the stationarity of the time series. Use the Augmented Dickey-Fuller Test (ADF) to further check the stationarity of the time series. The ADF test is a statistical test that can determine whether a time series has a unit root, i.e., whether it is non-stationary. If the test result indicates that the series has a unit root, further differencing is needed until the series becomes stationary.

[0075] Step S34: Select the ARIMA model using either AIC or BIC. To select the optimal ARIMA model, the Minimum Information Criterion (AIC) or the Bayesian Information Criterion (BIC) is used. Both AIC and BIC are metrics for measuring model goodness of fit, while penalizing model complexity. When the sample size is large, BIC is more suitable because it penalizes model complexity more severely, helping to avoid overfitting.

[0076] Step S35: Fit each feature parameter to an ARIMA model. After determining the appropriate differencing order and model parameters, we fit each feature parameter to an ARIMA model. The ARIMA model combines three components: autoregression (AR), differencing (I), and moving average (MA), which can capture the dynamic characteristics and periodicity of time series.

[0077] Step S36: Predict future snow cover changes using the ARIMA model. The fitted ARIMA model is used to predict the snow cover change for the next hour. The prediction results will be updated every 15 minutes to reflect the latest changes in snow cover. This high-frequency prediction update is crucial for the operation and maintenance management of photovoltaic power plants, helping to take timely measures to ensure the efficiency and safety of photovoltaic modules.

[0078] Through this series of steps, the ARIMA model can effectively predict changes in the snow cover area on the surface of photovoltaic modules, providing a scientific basis for snow management and optimization of photovoltaic power plants.

[0079] Step S4: Train an LSTM prediction model based on future changes in snow cover area and numerical weather forecast information, and obtain the future power changes of the photovoltaic power station through the LSTM prediction model.

[0080] refer to Figure 2 Step S4 includes: The future snow cover change and numerical weather prediction information are used as the input data training set for the LSTM prediction model, while the future output power of the photovoltaic power station is used as the output data training set. After completing the construction of the LSTM prediction model, the output power of the photovoltaic power plant in the future is predicted using the LSTM prediction model.

[0081] The snow cover area characteristics of photovoltaic power plant modules within the next hour and numerical weather prediction (NWP) information are used as the input data training set for a Long Short-Term Memory (LSTM) artificial neural network model, while the output power of the photovoltaic power plant within the next hour is used as the output data training set. By constructing and training this LSTM-based model, the power changes of the photovoltaic power plant can be accurately predicted.

[0082] In this embodiment, the Fengyun-4 satellite cloud image data has a resolution that updates every 15 minutes. Therefore, whenever the satellite cloud image data is updated, the LSTM prediction model will also make a forecast every 15 minutes, predicting the output power of the photovoltaic power station within the next hour. Such high-frequency prediction can reflect changes in snow cover and meteorological conditions in real time, thereby more accurately adjusting the operation strategy of the photovoltaic power station and improving power generation efficiency and reliability.

[0083] Satellite cloud images include visible light cloud images, water vapor cloud images, and infrared cloud images. Visible light cloud images are created using the principle of sunlight reflecting off cloud tops and are mainly used for observing daytime clouds, dust storms, and aerosols. Infrared cloud images are created by satellites measuring the infrared radiation emitted from the Earth's surface and cloud surfaces, and are also displayed in black and white tones. Water vapor cloud images are images of radiation emitted by water vapor in the atmosphere, received by the satellite and displayed in black and white tones. Their brightness is inversely proportional to the amount of radiation received by the satellite radiometer. The higher the water vapor content, the less radiation it emits, and the whiter the color on the water vapor image. The radiation received by the satellite is only related to temperature; the lower the temperature, the less radiation, and the whiter the color on the cloud image, and vice versa.

[0084] This embodiment also proposes a dynamic snow accumulation identification and power prediction system for new energy power plants, including an image preprocessing module, a snow accumulation identification module, a snow accumulation area prediction module, and a photovoltaic power plant power prediction module. The image preprocessing module, snow accumulation identification module, snow accumulation area prediction module, and photovoltaic power plant power prediction module are connected in sequence and work together to complete the dynamic snow accumulation identification and power prediction of new energy power plants.

[0085] The image preprocessing module is designed to preprocess the acquired satellite cloud images.

[0086] The snow cover identification module functions to distinguish between snow cover and clouds using visible light cloud images and shortwave infrared bands.

[0087] The snow cover area prediction module is designed to predict future changes in snow cover area by constructing an ARIMA model.

[0088] The photovoltaic power plant power prediction module functions as follows: it uses an LSTM prediction model to obtain the future power changes of the photovoltaic power plant.

Claims

1. A method for dynamic snow accumulation identification and power prediction in new energy power plants, characterized in that, Includes the following steps: S1, acquire satellite cloud images and preprocess the satellite cloud images containing snow accumulation on photovoltaic modules; S2, combine visible light cloud image and shortwave infrared band to determine whether there is snow in the image in order to distinguish snow from clouds; S3 extracts features from the snow image, including shape and size features; ARIMA model is trained based on features, and future changes in snow cover area are predicted using the ARIMA model. S4. Based on future changes in snow cover area and numerical weather forecast information, train an LSTM prediction model and obtain the future power changes of the photovoltaic power station through the LSTM prediction model.

2. The method for dynamic snow accumulation identification and power prediction of a new energy power plant according to claim 1, characterized in that, Step S1 includes: S11, homomorphic filtering is applied to the satellite cloud image to obtain an enhanced image; S12, Adaptive filtering is used to denoise the image, and the effect of adaptive filtering is quantitatively analyzed; S13, perform threshold segmentation on the image to generate a binary image containing both snow-covered and non-snow-covered regions.

3. A method for dynamic snow accumulation identification and power prediction in a new energy power plant according to claim 1 or 2, characterized in that, Step S2 includes: The Normalized Difference Snow Index (NDSI) is calculated. NDSI is the difference between the reflectance of the visible light band and the reflectance of the shortwave infrared band, divided by the sum of the reflectance of the visible light band and the shortwave infrared band. At the same time, the brightness temperature of the infrared band is compared with the low and high critical values. Combining NDSI and brightness temperature, snow cover and cloud cover are distinguished.

4. The method for dynamic snow accumulation identification and power prediction of a new energy power station according to claim 3, characterized in that, Step S3 includes: S31, determine whether the time series data has stabilized. If yes, proceed to S31; otherwise, proceed to S32. S32, perform the difference operation; S33, Use the ADF test to check the stationarity of the time series; S34, select the ARIMA model using either AIC or BIC; S35, each feature parameter is fitted to the ARIMA model; S36 uses the ARIMA model to predict future changes in snow cover area.

5. The method for dynamic snow accumulation identification and power prediction of a new energy power station according to claim 4, characterized in that, Step S4 includes: The future snow cover change and numerical weather prediction information are used as the input data training set for the LSTM prediction model, while the future output power of the photovoltaic power station is used as the output data training set. After completing the construction of the LSTM prediction model, the output power of the photovoltaic power plant in the future is predicted using the LSTM prediction model.

6. The method for dynamic snow accumulation identification and power prediction of a new energy power station according to claim 1, characterized in that, The satellite cloud image was specifically acquired through the Fengyun-4 satellite.

7. A method for dynamic snow accumulation identification and power prediction in a new energy power plant according to claim 1 or 2, characterized in that, The satellite cloud images include visible light cloud images, water vapor cloud images, and infrared cloud images. The visible light cloud images are made using the principle of cloud tops reflecting sunlight, and the infrared cloud images are made by satellites measuring the infrared radiation emitted from the Earth's surface and cloud surfaces.

8. The method for dynamic snow accumulation identification and power prediction of a new energy power station according to claim 2, characterized in that, The quantitative analysis of the adaptive filtering effect uses several evaluation indicators, including signal-to-noise ratio (SNR) and mean square error (MSE). The effect of the adaptive filter is quantitatively analyzed by calculating the SNR and MSE of the images before and after filtering.

9. A method for dynamic snow accumulation identification and power prediction in a new energy power plant according to claim 1 or 2, characterized in that, The shape feature can identify the boundary of the snow-covered area and describe its outline; the size feature is the area of ​​the snow-covered area, which can be quantified by pixel count and image resolution.

10. A dynamic snow accumulation identification and power prediction system for a new energy power plant, applicable to the dynamic snow accumulation identification and power prediction method for a new energy power plant as described in any one of claims 1-9, characterized in that, include: The image preprocessing module preprocesses the acquired satellite cloud images; The snow cover identification module distinguishes between snow cover and cloud cover using visible light cloud images and shortwave infrared bands. The snow cover area prediction module predicts future changes in snow cover area by constructing an ARIMA model. The photovoltaic power plant power prediction module uses an LSTM prediction model to obtain the future power changes of the photovoltaic power plant.

Images