Method for predicting power loss of photovoltaic power station covered with snow

Abstract

The present disclosure relates to a method and device for predicting snow-covered power loss of a photovoltaic power station, and a medium, wherein the method comprises: extracting snow cover rate features of visible and infrared dual-mode images through an improved convolutional neural network; screening historical similar daily and hourly meteorological features based on grey correlation analysis and weighted Euclidean distance using a multi-resolution selector; constructing a double-branch network to process daily and hourly features respectively and output fused features; establishing an ensemble learning framework to fuse the snow cover rate features, double-branch output and temperature change rate features, and output a power loss rate through weighted decision. According to the technical solution of the present disclosure, high-precision, high-timeliness and physically interpretable power loss prediction can be achieved.

Description

Technical Field

[0001] This disclosure relates to the field of photovoltaic power generation monitoring technology, and in particular to a method, device, equipment and medium for predicting power loss due to snow cover in photovoltaic power plants. Background Technology

[0002] Power prediction for photovoltaic power plants in snow-affected areas faces severe challenges. Traditional methods rely heavily on visible light sensors to detect snow cover, but visible light cannot penetrate the internal structure of the snow layer, making it difficult to effectively identify abnormal hot spots in the modules, resulting in significant biases in power generation loss assessment. More critically, meteorological monitoring data, such as temperature change rate and snow depth fluctuations, and photovoltaic panel properties, such as tilt angle and surface reflectivity, have inherent differences in time scale. The former changes frequently, while the latter exhibits long-term stability. This temporal inconsistency leads to feature conflicts during the fusion of multi-source data, directly manifesting as a lag in the prediction model's response to sudden snow disasters and a decrease in accuracy.

[0003] While current mainstream solutions combine infrared and visible light data, they employ a static fusion mechanism with fixed weights, failing to adaptively adjust the contribution of dual-modal features. Furthermore, they lack dynamic modeling of snow cover physical properties and have not yet established a collaborative analysis framework at the feature engineering level. These systemic deficiencies prevent power plant operation and maintenance from accurately predicting power generation efficiency under extreme snow and ice conditions. Therefore, a method for predicting power loss from snow cover in photovoltaic power plants is needed to address the limitations of static fusion mechanisms in feature adaptability, the lack of cross-period correlation of meteorological features, and the blind spots in dynamic modeling of snow layer properties. Summary of the Invention

[0004] To address the aforementioned technical issues, this disclosure provides a method, apparatus, equipment, and medium for predicting power loss due to snow cover in photovoltaic power plants.

[0005] In a first aspect, embodiments of this disclosure provide a method for predicting power loss due to snow cover in photovoltaic power plants, including: Acquire multimodal data to be predicted; the multimodal data includes visible light images, infrared thermal imaging, and meteorological monitoring data; A four-channel fused image is generated based on the visible light image and the infrared thermal image, and snow cover features are extracted based on the fused image; Based on grey relational analysis, daily meteorological data similar to the meteorological monitoring data are determined from historical meteorological data; based on weighted Euclidean distance, time-level meteorological data similar to the meteorological monitoring data are determined from the historical meteorological data. The daily and hourly meteorological data are processed by a dual-branch network to generate daily trend features and hourly fluctuation features. The dual-branch network includes a daily branch and an hourly branch, both of which use a one-dimensional CNN-LSTM. The daily branch is used to process daily-scale sequence data, and the hourly branch is used to process hourly-scale sequence data. A fused feature vector is generated based on the snow cover characteristics, the daily trend characteristics and the time-level fluctuation characteristics, and the radiation compensation value; The fused feature vector is input into the integrated prediction model for prediction to obtain the power loss rate of the photovoltaic power plant.

[0006] Secondly, embodiments of this disclosure provide a photovoltaic power plant snow-covered power loss prediction device, comprising: The acquisition module is used to acquire the multimodal data to be predicted; the multimodal data includes visible light images, infrared thermal imaging, and meteorological monitoring data. An extraction module is used to generate a four-channel fused image based on the visible light image and the infrared thermal image, and to extract snow cover features based on the fused image; The determination module is used to determine daily meteorological data similar to the meteorological monitoring data from historical meteorological data based on grey relational analysis, and to determine time-level meteorological data similar to the meteorological monitoring data from historical meteorological data based on weighted Euclidean distance; The generation module is used to process the daily meteorological data and the hourly meteorological data through a dual-branch network to generate daily trend features and hourly fluctuation features; wherein, the dual-branch network includes a daily branch and an hourly branch, both of which adopt a one-dimensional CNN-LSTM, the daily branch is used to process daily-scale sequence data, and the hourly branch is used to process hourly-scale sequence data; The fusion module is used to generate a fused feature vector based on the snow cover rate characteristics, the daily trend characteristics and the time-level fluctuation characteristics, and the radiation compensation value. The prediction module is used to input the fused feature vector into the integrated prediction model for prediction, so as to obtain the power loss rate of the photovoltaic power plant.

[0007] Thirdly, embodiments of this disclosure provide an electronic device, including: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the photovoltaic power plant snow cover power loss prediction method described in the first aspect above.

[0008] Fourthly, embodiments of this disclosure provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the photovoltaic power plant snow-covered power loss prediction method described in the first aspect.

[0009] Compared with the prior art, the technical solution provided in this disclosure has the following advantages: a four-channel fused image is generated by visible light image and infrared thermal imaging, and snow cover features are extracted from the fused image. Then, based on grey relational analysis, daily meteorological data similar to meteorological monitoring data are determined from historical meteorological data, and hourly meteorological data similar to meteorological monitoring data are determined from historical meteorological data based on weighted Euclidean distance. The daily and hourly meteorological data are processed by a dual-branch network to generate daily trend features and hourly fluctuation features. Furthermore, a fused feature vector is generated based on the snow cover features, daily trend features, hourly fluctuation features, and radiation compensation value. The fused feature vector is input into the integrated prediction model for prediction to obtain the power loss rate of the photovoltaic power station. Thus, high-precision, high-timeliness, and physically interpretable power loss prediction can be achieved. Attached Figure Description

[0010] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0011] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic flowchart illustrating a method for predicting power loss due to snow cover in a photovoltaic power plant, provided in an embodiment of this disclosure; Figure 2 A flowchart illustrating a multimodal data preprocessing method provided in this embodiment of the disclosure; Figure 3 A flowchart for calculating snow cover rate is provided as an embodiment of this disclosure; Figure 4 A flowchart of feature integration and prediction provided in an embodiment of this disclosure; Figure 5 This is a flowchart of a snow-covered power loss prediction algorithm for photovoltaic power plants based on multimodal deep learning, provided in an embodiment of this disclosure. Detailed Implementation

[0013] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0014] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0015] Figure 1 This is a flowchart illustrating a method for predicting power loss due to snow cover in a photovoltaic power plant, as provided in an embodiment of this disclosure. The method provided in this embodiment can be executed by a device for predicting power loss due to snow cover in a photovoltaic power plant. This device can be implemented using software and / or hardware and can be integrated into any electronic device with computing capabilities.

[0016] like Figure 1 As shown in the embodiments of this disclosure, the method for predicting power loss due to snow cover in photovoltaic power plants may include: Step 101: Obtain the multimodal data to be predicted.

[0017] The multimodal data includes visible light images, infrared thermal imaging, and meteorological monitoring data, including temperature, wind speed, solar irradiance, snow depth, and temperature change rate.

[0018] In one embodiment of this disclosure, after acquiring the multimodal data to be predicted, the multimodal data is preprocessed. The preprocessing steps include image denoising, data normalization, and missing value imputation. The resolution of the visible light image is not less than 1920×1080, and the wavelength of the infrared thermal image is 8–14 μm. Optionally, Gaussian filtering is used to denoise the visible light image, and non-uniformity correction is performed on the infrared thermal image; the visible light image and the infrared thermal image are uniformly resampled to 256×256 pixels using a bilinear interpolation algorithm; missing data in the meteorological monitoring data is imputed using time-series linear interpolation; and Z-score normalization is performed on the meteorological monitoring data.

[0019] As an example, image processing involves: denoising the visible light image using Gaussian filtering, and correcting the non-uniformity of the infrared thermal imaging data. Then, both are uniformly resampled to 256×256 pixels using a bilinear interpolation algorithm. The specific calculation logic is as follows: Coordinate mapping: Establishing the coordinates of the target image Coordinates of the source image The geometric mapping relationship between them;

[0020] in, and These represent the width and height of the original image, respectively.

[0021] Pixel interpolation: due to the mapped source coordinates Typically a floating-point number, the target pixel value. Distance in the source image The four nearest integer coordinates The weighted sum of the pixel values ​​yields:

[0022] Among them, weight It is inversely proportional to the Euclidean distance from the sampling point to the four neighboring pixels, and satisfies .

[0023] Missing values ​​in meteorological monitoring data were filled using linear interpolation:

[0024] in, The feature values ​​to be filled. The timestamp of the missing point. and These are the nearest valid observation data before and after the missing point, along with their timestamps.

[0025] The standardization of meteorological monitoring data adopts the Z-score method:

[0026] in, This represents the arithmetic mean of the meteorological features in the historical training dataset. The standard deviation of the meteorological features in the historical training dataset. These are the raw observation values ​​of meteorological characteristics. These are the standardized input feature values.

[0027] Data alignment: Based on a unified NTP time server, all data is timestamped to the second, and a spatial mapping relationship between multi-source data at the same time is established through database join operations, forming well-organized data packets for subsequent steps. The multimodal data preprocessing workflow is as follows: Figure 2 As shown.

[0028] Step 102: Generate a four-channel fused image based on the visible light image and infrared thermal imaging, and extract snow cover features based on the fused image.

[0029] In this embodiment, to address the shortcomings of static fusion mechanisms, this scheme constructs a cross-spectral feature fusion layer, breaking through the simple mode of traditional image channel cascading. Specifically, generating a four-channel fused image based on visible light images and infrared thermal imaging includes: mapping single-channel thermal imaging data into a pseudo-RGB three-channel image using a trainable 1×1 convolutional kernel, achieving semantic alignment of thermal physical information in visual space. , where the weight matrix bias , For single-channel thermal imaging data, such as initialization And optimize during training; weighted fusion of visible light image and pseudo-RGB three-channel image to obtain four-channel fused image: ,in, To integrate weights, , This is three-channel data for a visible light image.

[0030] Among them, the fusion weight Based on ambient lighting conditions The piecewise adaptive mapping function (unit: lux) is designed to dynamically adjust the weights of thermal imaging features based on the imaging quality of the visible light sensor.

[0031] Therefore, by introducing a dynamic adjustment coefficient related to the ambient light intensity, the adaptive weighted fusion of infrared and visible light features is achieved, effectively utilizing thermal imaging information to penetrate the snow cover layer and enhancing the model's feature expressiveness and robustness under complex lighting conditions such as reflection and shadow.

[0032] Furthermore, The input is an improved ResNet-34 network, which includes convolutional layers, batch normalization layers, ReLU activation functions, and residual blocks. Its output feature map is mapped to snow cover using global average pooling layers and fully connected layers. Optionally, the network is pre-trained on ImageNet, and the final fully connected layer is replaced with a neuron of output dimension 1, using a sigmoid activation function to map the output value to the [0, 1] interval, representing the snow cover as a percentage. The loss function used is binary cross-entropy. The process for calculating snow cover is as follows: Figure 3 As shown.

[0033] Step 103: Based on grey relational analysis, determine daily meteorological data similar to meteorological monitoring data from historical meteorological data, and based on weighted Euclidean distance, determine hourly meteorological data similar to meteorological monitoring data from historical meteorological data.

[0034] Step 104: Process the daily and hourly meteorological data using a dual-branch network to generate daily trend features and hourly fluctuation features.

[0035] The dual-branch network includes a daily branch and a time-scale branch. Both the daily and time-scale branches use a one-dimensional CNN-LSTM (Convolutional Neural Network) (Long Short-Term Memory) network. The daily branch is used to process daily-scale sequence data, and the time-scale branch is used to process time-scale sequence data.

[0036] In this embodiment, the formula for grey relational analysis is as follows:

[0037] in, The resolution coefficient, , To be the minimum difference, For the maximum difference, denoted as the absolute difference between the k-th meteorological data point on the day to be predicted and the historical data point.

[0038] Specifically, .

[0039] As an example, grey relational analysis was used to select the 15 historical days most similar to the current day from 365 days of historical meteorological data.

[0040] In this embodiment, the formula for the weighted Euclidean distance is as follows:

[0041] in, Let be the weight of the i-th meteorological data; if Indicates the snow depth weight, then ;like If the weight represents the rate of temperature change, then... For example, snow depth is weighted at 3.2 and temperature change rate at 2.2 to quantify the contribution of different meteorological parameters to power loss.

[0042] In this embodiment, daily trend features are generated through the following steps: local features of daily meteorological data are extracted through a 3-layer one-dimensional convolutional neural network, wherein the kernel size of each layer is 3 and the stride is 1. The input feature dimension of the daily meteorological data is [5,15,288], representing 5 features, 15 days, and 288 time points per day; the time dependency is processed through a bidirectional LSTM to output daily trend features, wherein the number of hidden units of the bidirectional LSTM is 64.

[0043] As an example, the input to the daily branch is a selection of meteorological data from 15 historical days. Each daily sequence is 288 bytes long, with a 5-minute interval and 5 feature dimensions, including temperature, wind speed, irradiance, snow depth, and temperature change rate. Features are first extracted using a 3-layer one-dimensional convolution, with a kernel size of 3, a stride of 1, and the number of filters [32, 64, 128]. The activation function is ReLU, as shown in the following formula: ,in This represents a convolution operation. A BiLSTM with 64 hidden units is then used to capture temporal dependencies, outputting daily trend features. .

[0044] In this embodiment, time-level fluctuation features are generated through the following steps: local features of time-level meteorological data are extracted through a 3-layer one-dimensional convolutional neural network, wherein the kernel size of each layer is 3 and the stride is 1. The input feature dimension of the time-level meteorological data is [24,20,24], representing 24 features, 20 hours, and 24 time points per hour; the time dependency is processed through bidirectional LSTM to output time-level fluctuation features.

[0045] As an example, the time-level branch can be processed similarly to the daily-level branch, focusing on intraday high-frequency meteorological fluctuations. The input of the daily-level branch is the feature [24, 20, 24], and the output is the time-level fluctuation feature. .

[0046] Therefore, in response to the lack of cross-period correlation of meteorological characteristics, a synergistic analysis of the long-term evolution of meteorological conditions and short-term sudden fluctuations has been achieved.

[0047] Step 105: Generate a fused feature vector based on snow cover characteristics, daily trend characteristics, time-level fluctuation characteristics, and radiation compensation values.

[0048] In one embodiment of this disclosure, the multimodal data also includes snow cover duration data. Before generating the fused feature vector, snow quality is classified based on the snow cover duration data to determine the snow surface reflectivity. Then, the snow surface reflectivity compensation feature is calculated based on the difference between the snow surface reflectivity and the photovoltaic panel reflectivity to generate a radiation compensation value.

[0049] The snow surface reflectance compensation feature is calculated using the following compensation feature formula:

[0050] in, This is the radiation compensation value, which quantifies the additional radiation gain or loss caused by the change in surface reflectivity of the module due to snow cover (unit: ), This is the angle compensation coefficient. , Install the tilt angle for the photovoltaic panels; Solar irradiance, The reflectivity of the snow surface The reflectivity of the photovoltaic panel.

[0051] In this embodiment, snow quality is first classified and reflectivity is dynamically assigned, based on the duration of snow cover. (Unit: hour) Classify snow quality and assign snow surface reflectance values. :

[0052] As an example, if the snow quality classification result is new snow, the snow surface reflectance is 0.85; if the snow quality classification result is old snow, the snow surface reflectance is 0.55; the photovoltaic panel reflectance is 0.2.

[0053] Optionally, the compensation feature formula also includes a correction for the solar altitude angle:

[0054] in, This is the solar altitude angle.

[0055] Step 106: Input the fused feature vector into the integrated prediction model for prediction to obtain the power loss rate of the photovoltaic power plant.

[0056] In this embodiment, the features generated in the aforementioned steps are concatenated with key instantaneous meteorological data to form the final feature vector. :

[0057] in, For temperature, For wind speed, This represents the rate of temperature change.

[0058] The LightGBM framework is used as the ensemble predictor to achieve the final mapping from the high-dimensional feature space to the power loss rate. The model is configured with a maximum tree depth of ≤7 and a leaf node count of ≤64.

[0059] As an example, the framework configuration for the ensemble prediction model includes: a maximum tree depth of 7, 64 leaf nodes, a learning rate of 0.05, 100 training epochs, a feature sampling ratio of 0.8, and a loss function of mean absolute error. In this example, the model input... Output the predicted power loss rate every 15 minutes for the next 4 hours, with a value range of 0% to 100%. The feature integration and prediction process is as follows: Figure 4 As shown, the complete prediction process is as follows: Figure 5 As shown.

[0060] According to the technical solution of this disclosure, the cross-spectral adaptive interactive fusion mechanism effectively utilizes the penetration capability of thermal imaging information into the snow cover layer, overcomes the limitations of a single visible light image under severe weather conditions, significantly enhances the accuracy and environmental robustness of the model in identifying snow cover status, and improves the reliability of visual recognition. By using a dual-scale spatiotemporal feature modeling network to collaboratively analyze the long-term evolution law and short-term sudden fluctuations of meteorological conditions, the complex spatiotemporal dependence affecting power loss is effectively captured, reducing the systematic bias of prediction results and optimizing the prediction time-series response mechanism. By integrating the reflectivity compensation strategy of snow quality time-series classification, the clear prior knowledge of the physical optical properties of the snow layer is embedded into the data-driven model, which not only makes up for the lack of physical mechanisms in the pure black box model, but also significantly improves the prediction robustness and the transparency and interpretability of the model decision-making process in complex dynamic scenarios such as freeze-thaw cycles and snow aging, thereby improving environmental adaptability and physical interpretability.

[0061] This disclosure also proposes a photovoltaic power plant snow cover power loss prediction device, which includes: The acquisition module is used to acquire the multimodal data to be predicted; the multimodal data includes visible light images, infrared thermal imaging, and meteorological monitoring data. An extraction module is used to generate a four-channel fused image based on the visible light image and the infrared thermal image, and to extract snow cover features based on the fused image; The determination module is used to determine daily meteorological data similar to the meteorological monitoring data from historical meteorological data based on grey relational analysis, and to determine time-level meteorological data similar to the meteorological monitoring data from historical meteorological data based on weighted Euclidean distance; The generation module is used to process the daily meteorological data and the hourly meteorological data through a dual-branch network to generate daily trend features and hourly fluctuation features; wherein, the dual-branch network includes a daily branch and an hourly branch, both of which adopt a one-dimensional CNN-LSTM, the daily branch is used to process daily-scale sequence data, and the hourly branch is used to process hourly-scale sequence data; The fusion module is used to generate a fused feature vector based on the snow cover rate characteristics, the daily trend characteristics and the time-level fluctuation characteristics, and the radiation compensation value. The prediction module is used to input the fused feature vector into the integrated prediction model for prediction, so as to obtain the power loss rate of the photovoltaic power plant.

[0062] The photovoltaic power plant snow-covered power loss prediction device provided in this disclosure can execute any photovoltaic power plant snow-covered power loss prediction method provided in this disclosure, and has the corresponding functional modules and beneficial effects for executing the method. Content not described in detail in the device embodiments of this disclosure can be referred to the description in any method embodiment of this disclosure.

[0063] This disclosure also provides an electronic device including one or more processors and a memory. The processor may be a central processing unit (CPU) or other processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may execute the program instructions to implement the methods of the embodiments of this disclosure above and / or other desired functions. Various contents such as input signals, signal components, and noise components may also be stored in the computer-readable storage medium.

[0064] In one example, the electronic device may also include input and output devices, which are interconnected via a bus system and / or other forms of connection. Furthermore, the input device may include, for example, a keyboard, a mouse, etc. The output device can output various information to the outside, including determined distance information, direction information, etc. The output device may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc. In addition, depending on the specific application, the electronic device may include any other suitable components such as a bus, input / output interfaces, etc.

[0065] In addition to the methods and apparatus described above, embodiments of this disclosure may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform any of the methods provided in the embodiments of this disclosure.

[0066] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of embodiments of this disclosure. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0067] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer program instructions that, when executed by a processor, cause the processor to perform any of the methods provided in the embodiments of this disclosure.

[0068] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0069] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0070] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting power loss due to snow cover in photovoltaic power plants, characterized in that, The method includes: Acquire multimodal data to be predicted; the multimodal data includes visible light images, infrared thermal imaging, and meteorological monitoring data; A four-channel fused image is generated based on the visible light image and the infrared thermal image, and snow cover features are extracted based on the fused image; Based on grey relational analysis, daily meteorological data similar to the meteorological monitoring data are determined from historical meteorological data; based on weighted Euclidean distance, time-level meteorological data similar to the meteorological monitoring data are determined from the historical meteorological data. The daily and hourly meteorological data are processed by a dual-branch network to generate daily trend features and hourly fluctuation features. The dual-branch network includes a daily branch and an hourly branch, both of which use a one-dimensional CNN-LSTM. The daily branch is used to process daily-scale sequence data, and the hourly branch is used to process hourly-scale sequence data. A fused feature vector is generated based on the snow cover characteristics, the daily trend characteristics and the time-level fluctuation characteristics, and the radiation compensation value; The fused feature vector is input into the integrated prediction model for prediction to obtain the power loss rate of the photovoltaic power plant.

2. The method as described in claim 1, characterized in that, The method further includes: The visible light image is denoised using Gaussian filtering, and the infrared thermal image is corrected for non-uniformity. The visible light image and the infrared thermal image are uniformly resampled to 256×256 pixels using a bilinear interpolation algorithm; The missing data in the meteorological monitoring data were filled using time series linear interpolation. The meteorological monitoring data were standardized using Z-score.

3. The method as described in claim 1, characterized in that, The step of generating a four-channel fused image based on the visible light image and the infrared thermal image includes: Single-channel thermal imaging data is mapped to pseudo-RGB three-channel images using trainable 1×1 convolutional kernels: , where the weight matrix bias , This is single-channel thermal imaging data; The visible light image and the pseudo-RGB three-channel image are weighted and fused to obtain the four-channel fused image: ,in, To integrate weights, , This is three-channel data for a visible light image.

4. The method as described in claim 1, characterized in that, The formula for the grey relational analysis is as follows: in, The resolution coefficient, , To be the minimum difference, For the maximum difference, This represents the absolute difference between the k-th meteorological data point on the day to be predicted and the historical data point. The formula for the weighted Euclidean distance is as follows: in, Let be the weight of the i-th meteorological data; if Indicates the snow depth weight, then ;like If the weight represents the rate of temperature change, then... .

5. The method as described in claim 1, characterized in that, The daily trend features are generated through the following steps: Local features of daily meteorological data are extracted using a three-layer one-dimensional convolutional neural network. Each layer has a kernel size of 3 and a stride of 1. The input feature dimension of the daily meteorological data is [5, 15, 288], representing 5 features, 15 days, and 288 time points per day. The time dependency is processed by a bidirectional LSTM to output the daily trend features; the number of hidden units in the bidirectional LSTM is 64. The time-level fluctuation characteristics are generated through the following steps: Local features of time-level meteorological data are extracted using a three-layer one-dimensional convolutional neural network. Each layer has a kernel size of 3 and a stride of 1. The input feature dimension of the time-level meteorological data is [24, 20, 24], representing 24 features, 20 hours, and 24 time points per hour. The time dependence is processed by bidirectional LSTM to output the time-level fluctuation characteristics.

6. The method as described in claim 1, characterized in that, The multimodal data also includes snow cover duration data, and the method further includes: Snow quality is classified and snow surface reflectivity is determined based on the snow cover duration data. The snow surface reflectivity compensation characteristics are calculated based on the difference between the snow surface reflectivity and the photovoltaic panel reflectivity, and a radiation compensation value is generated. The snow surface reflectance compensation feature is calculated using the following compensation feature formula: in, This is the radiation compensation value. This is the angle compensation coefficient. , Install the tilt angle for the photovoltaic panels; Solar irradiance, The reflectivity of the snow surface The reflectivity of the photovoltaic panel.

7. The method as described in claim 6, characterized in that, If the snow quality classification result is new snow, the snow surface reflectance is 0.85; if the snow quality classification result is old snow, the snow surface reflectance is 0.55; the photovoltaic panel reflectance is 0.

2. The compensation feature formula also includes solar altitude angle correction: in, This is the solar altitude angle.

8. The method as described in claim 1, characterized in that, The meteorological monitoring data includes temperature, wind speed, solar irradiance, snow depth, and temperature change rate.

9. The method as described in claim 1, characterized in that, The framework configuration of the integrated prediction model includes: a maximum tree depth of 7, 64 leaf nodes, a learning rate of 0.05, 100 training rounds, a feature sampling ratio of 0.8, and a loss function of mean absolute error.

10. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the photovoltaic power plant snow-covered power loss prediction method according to any one of claims 1-9.

Images