A method and system for identifying shadow blocking of a photovoltaic power station based on multi-source data

Abstract

This invention provides a method and system for identifying shading in photovoltaic (PV) power plants based on multi-source data, relating to the field of PV power plant technology. The method includes acquiring multi-source data from the PV power plant, including bispectral images, IV curve time-series data, meteorological data, and solar radiation parameters; partitioning the PV power plant using the multi-source data; generating a guiding mask for the PV power plant by combining prior information about shading objects and the posterior probability of regional shading; extracting heuristic cross-modal features from the multi-source data using the guiding mask; performing probabilistic cross-attention fusion on the cross-modal features to obtain a probabilistic fusion feature vector; and performing Bayesian multi-classification based on the probabilistic fusion feature vector to generate the shading identification result for the PV power plant. This invention solves the problem that complex lighting and meteorological conditions are not considered when using multi-source data for shading identification.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power plant technology, and more specifically, to a method and system for identifying shading in photovoltaic power plants based on multi-source data. Background Technology

[0002] In the technological evolution of shading identification in photovoltaic power plants, single-modal recognition is no longer sufficient to meet the requirements for accurate detection. Pure electrical parameter methods can only achieve a rough judgment at the string level, unable to locate the specific shading position, and are also susceptible to interference from environmental factors. Pure image recognition, while able to capture shadow contours or temperature anomalies, struggles to distinguish between shading and false anomalies such as component aging or dust reflection, and the false detection rate increases significantly in complex scenarios such as rainy days and strong sunlight. Therefore, the industry is gradually shifting towards multimodal fusion of electrical parameters and images. However, existing solutions often employ simple feature vector concatenation, failing to establish a deep interaction mechanism between the temporal dimension features of electrical parameters and the spatial dimension features of images. This results in the underutilization of their complementarity and an inability to effectively distinguish between temporary and long-term shading.

[0003] Meanwhile, existing technologies generally do not incorporate weather information into the identification system. Meteorological factors such as light intensity, cloud cover, and wind speed directly affect the electrical output characteristics of photovoltaic modules and the visual effect of images. For example, sudden changes in light intensity on cloudy days can cause fluctuations in electrical parameters, and rainy days can reduce image contrast. This further exacerbates the difficulty of misjudging shading and distinguishing types, making it impossible to provide targeted processing basis for operation and maintenance, and making it difficult to meet the actual needs of power plants to reduce operation and maintenance costs and improve power generation efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for identifying shading in photovoltaic power plants based on multi-source data, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0005] Firstly, this application provides a method for identifying shading in photovoltaic power plants based on multi-source data, including:

[0006] Acquire multi-source data from photovoltaic power plants, including bispectral images, IV curve time-series data, meteorological data, and solar radiation parameters. The bispectral images are image pairs composed of visible light images and infrared images.

[0007] Photovoltaic power plants are partitioned using multi-source data, and a boot mask for photovoltaic power plants is generated by combining prior information on obstructions and posterior probability of regional obstruction.

[0008] Heuristic cross-modal feature extraction is performed on multi-source data using a guided mask to obtain cross-modal features, which include image feature vectors, electrical parameter time feature vectors, illumination stability coefficients, and meteorological parameter reliability probabilities.

[0009] Probabilistic cross-attention fusion is performed on cross-modal features to obtain a probabilistic fused feature vector;

[0010] Bayesian multi-classification is performed based on probabilistic fusion feature vectors to generate shadow occlusion recognition results for photovoltaic power plants. The shadow occlusion recognition results include shadow occlusion location and shadow occlusion type.

[0011] Secondly, this application also provides a photovoltaic power station shading identification system based on multi-source data, including:

[0012] The acquisition module is used to acquire multi-source data from the photovoltaic power station. The multi-source data includes bispectral images, IV curve time-series data, meteorological data, and solar radiation parameters. The bispectral images are image pairs composed of visible light images and infrared images.

[0013] The generation module is used to partition the photovoltaic power station through multi-source data and generate the guiding mask of the photovoltaic power station by combining the prior information of the shading object and the posterior probability of the regional shading.

[0014] The extraction module is used to perform heuristic cross-modal feature extraction on multi-source data through a guided mask to obtain cross-modal features, which include image feature vectors, electrical parameter time feature vectors, illumination stability coefficients, and meteorological parameter reliability probabilities.

[0015] The fusion module is used to perform probabilistic cross-attention fusion on cross-modal features to obtain a probabilistic fused feature vector;

[0016] The identification module is used to perform Bayesian multi-classification based on probabilistic fusion feature vectors to generate shadow occlusion identification results for photovoltaic power plants. The shadow occlusion identification results include shadow occlusion location and shadow occlusion type.

[0017] The beneficial effects of this invention are as follows:

[0018] (1) This invention utilizes multi-source data fusion and a partitioned guided masking strategy to fully leverage the complementary information from visible light and infrared images, electrical parameter time-series data, solar radiation parameters, and meteorological data to perform targeted feature extraction and weighted processing on different areas of a photovoltaic power station. The central concave area uses a small receptive field convolutional kernel to extract local detail features, while the peripheral area uses a large receptive field convolutional kernel to capture global shadow trends. Simultaneously, it combines electrical parameter anomaly features and meteorological reliability probabilities to generate cross-modal features, and then uses probabilistic cross-attention fusion to form a robust probabilistic fusion feature vector. This effectively improves the sensitivity to shadow features at different spatial levels, while enhancing the reliability and expressive power of cross-modal features, making shadow recognition more accurate and robust in complex environments.

[0019] (2) This invention also uses Bayesian multi-classification based on probabilistic fusion feature vectors, enabling the shading type and location of each partition unit of the photovoltaic power station to be scientifically estimated based on historical shading events, scene priors, and conditional probabilities. By combining prior probability, likelihood probability, and posterior probability calculations, temporary and long-term shading can be distinguished simultaneously, accurately marking the shading location and reducing misjudgments or omissions. Furthermore, the method of this invention has strong adaptability, adapting to different photovoltaic power station layouts and meteorological conditions, achieving high-precision, real-time shading identification, and providing a reliable basis for photovoltaic power station operation and maintenance, energy efficiency optimization, and operation and maintenance decisions.

[0020] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the process for identifying shading in a photovoltaic power station based on multi-source data, as described in an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the structure of the photovoltaic power station shadow occlusion recognition system based on multi-source data as described in an embodiment of the present invention. Detailed Implementation

[0024] 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] Example 1:

[0027] This embodiment provides a method for identifying shading in photovoltaic power plants based on multi-source data.

[0028] See Figure 1 The figure shows that the method includes steps S1, S2, S3, S4 and S5.

[0029] Step S1: Obtain multi-source data from the photovoltaic power station. The multi-source data includes bispectral images, IV curve time-series data, meteorological data, and solar radiation parameters. The bispectral images are image pairs composed of visible light images and infrared images.

[0030] In this step, the synchronously acquired visible light and infrared images are registered to ensure that the corresponding pixels of the same component of the photovoltaic power station are consistent in both the visible light and infrared images. Real-time IV curve time-series data, including current, voltage, and power components, are acquired from the photovoltaic array inverter or combiner box. Meteorological data, including ambient temperature, relative humidity, wind speed, and cloud cover, is obtained from weather stations or meteorological APIs. Solar radiation parameters include real-time irradiance, solar altitude, and azimuth. Simultaneously, the IV curve time-series data, meteorological data, and solar radiation parameters are aligned with the bispectral images using timestamps to form multi-source data that is time-synchronized and spatially registered.

[0031] Step S2: Divide the photovoltaic power station into zones using multi-source data, and generate a boot mask for the photovoltaic power station by combining prior information on obstructions and posterior probability of regional obstruction.

[0032] In this step, a guiding mask is generated by partitioning and fusing multi-source information, which clearly marks the potential areas occluded by shadows, providing both spatial structural constraints and enhancing adaptability to dynamic environmental changes.

[0033] In step S2, the step of generating a pilot mask for the photovoltaic power station by combining prior information about obstructions and posterior probability of regional obstruction includes:

[0034] Step S21: Based on the arrangement pattern of photovoltaic modules in the photovoltaic power station, the bispectral image is divided into regular grids to obtain partitioned units;

[0035] Step S22: Calculate the current deviation coefficient, fill factor stability, and local power anomaly index of each partition unit within the sampling period using IV curve time series data, and generate an electrical parameter anomaly map;

[0036] In this step, the current deviation coefficient, fill factor stability, and local power anomaly index of each partition cell are calculated using IV curve time-series data:

[0037]

[0038]

[0039]

[0040]

[0041] In the formula, Indicates the current deviation coefficient. Represents current. This represents the average current over the period. This indicates taking the absolute value. Indicates the stability of the fill factor. Represents variance. Indicates time The fill factor, Indicates time The maximum output power of the component at that time Indicates short-circuit current. Indicates open-circuit voltage. This represents the mean of the fill factor within the sampling period. Indicates the local power anomaly index. Indicates power, This represents the average power of adjacent normal components within the cycle.

[0042] Understandably, the current deviation coefficient is used to measure the degree of fluctuation in component current over time. Shading causes unstable current output and exacerbates fluctuations. A higher fill factor indicates a greater likelihood of shading. Fill factor stability reflects the stability of the module's power conversion characteristics over time; shading can distort the shape of the IV curve, thus... The volatility has increased. And... It combines three-dimensional characteristics of current, voltage, and power, making it more sensitive to nonlinear characteristics caused by shading and able to distinguish between local mismatch and global attenuation. This allows for the detection of relative anomalies in shading from a spatial perspective, reducing false alarms caused by fluctuations in overall irradiance.

[0043] The degree of electrical parameter anomaly is calculated using the current deviation coefficient, fill factor stability, and local power anomaly index.

[0044]

[0045] In the formula, Indicates the degree of abnormality of electrical parameters. , and This represents the abnormal weighting factor.

[0046] The electrical parameter anomaly degree is plotted as a grayscale image by partition unit to obtain the electrical parameter anomaly map, and the current deviation coefficient, fill factor stability and local power anomaly index are associated with each partition unit.

[0047] Step S23: Construct a time-by-time shadow projection analysis model based on solar radiation parameters and prior information on shading objects, and generate potential shadow projection areas based on the time-by-time shadow projection analysis model;

[0048] In this step, areas that may be obscured are obtained in advance as spatial prior constraints. Specifically, the time is obtained from the solar radiation parameters. solar altitude angle at time With azimuth Obtain prior information about obstructions (such as the height and location of towers, tall trees, and buildings).

[0049] For each occluder, calculate the length and direction of the shadow projection:

[0050]

[0051] In the formula, Indicates obstruction The length of the shadow projection, Indicates obstruction height, Indicates time solar altitude angle at that time Indicates obstruction The direction of the shadow projection, Indicates time The azimuth angle at that time.

[0052] Based on the geometric projection relationship, a time-by-time shadow projection analytical model is constructed to calculate the shadow coverage coordinate range:

[0053]

[0054] In the formula, Indicates the coordinates of the shadow coverage.

[0055] Map the shadow coverage coordinate range to partition cells to generate the potential shadow projection area. ,and Essentially, it's a binary mask; it's 1 if it lies within the projection region, and 0 otherwise. and This represents the x and y coordinates of the partition unit.

[0056] Step S24: Perform probabilistic modeling on the meteorological data, calculate the probability of illumination disturbance in each partition unit, and superimpose it with the brightness temperature difference of the bispectral image to generate a regional occlusion posterior probability map.

[0057] In this step, by using meteorological and infrared brightness temperature difference joint modeling, we can distinguish between illumination disturbances and actual occlusion, and achieve dynamic probability correction driven by the environment.

[0058] Cloud cover change rate in meteorological data and wind speed Establish a probability model for illumination perturbation:

[0059]

[0060] In the formula, Indicates time The probability of illumination disturbance at that time. Represents an exponential function. and Both represent the illumination disturbance factor. Indicates time Cloud cover change rate at time Indicates time Wind speed at that time.

[0061] Calculate the brightness temperature difference of a bispectral image:

[0062]

[0063] In the formula, Indicates position The difference in brightness temperature at the location, Indicates the position of the visible light image. The brightness value at that location, Indicates the position of the infrared image The radiance value at that location.

[0064] It should be noted that visible light intensity and infrared radiant intensity are originally different physical quantities, namely reflected light intensity and thermal radiation energy. However, in this step, both types of images have undergone normalization and local mapping, so differential calculation can be performed directly. and Both values ​​are normalized and mapped to the same [0,1] interval, eliminating differences in units and dimensions. At this point, both are in the same scale space. This represents the difference between the change in light reflection intensity and the thermal radiation response at that location.

[0065] Understandably, shadowed areas affect both visible light and infrared responses, but their response speeds and directions differ. Under normal illumination, the visible light image is bright (high light intensity), while the infrared image is moderately warm (thermal equilibrium). The brightness of the visible light image is relatively small. When temporarily obscured (such as by cloud shadows), the brightness of the infrared image decreases significantly, while the infrared image shows a slight decrease in temperature (thermal change lag). When the light intensity increases and there is prolonged occlusion (such as the shadow of a building or tree), the brightness of the visible light image remains low for a long period, while the infrared image shows a temperature that is too high or too low (local mismatch). Significantly large. Therefore It can be regarded as an indicator of photo-thermal response inconsistency, reflecting the difference in response to the same physical event (light shading, power anomaly) under different modes.

[0066] Then, the two are combined to calculate the posterior probability of region occlusion:

[0067]

[0068] In the formula, Indicates about location The posterior probability of regional occlusion. This represents the Sigmoid function. and All of these represent occlusion weight factors.

[0069] Step S25: Generate the guiding mask for the photovoltaic power station using the electrical parameter anomaly map, the potential shadow projection area, and the posterior probability map of regional shading.

[0070] Step S25 includes:

[0071] Step S251: After normalizing the electrical parameter anomaly map, the potential shadow projection region, and the region occlusion posterior probability map, calculate the shadow confidence of each partition unit using a Bayesian weighted strategy.

[0072] In this step, the potential shadow projection area is... Convert to fuzzy projection intensity:

[0073]

[0074] In the formula, Indicates about location At any moment The intensity of the blurred projection at that time Represents fuzzy weights, Indicates position At any moment The normalized value of the distance to the projection boundary.

[0075] Joint modeling is performed to obtain the shadow confidence score for each partition unit:

[0076]

[0077] In the formula, Indicates position At any moment Confidence of shadow at time Indicates the confidence level of the shadow. All represent confidence weighting factors. Indicates the balance threshold. Indicates about location At any moment Normalized electrical parameter anomaly degree at time, Indicates about location At any moment Normalized blur projection intensity at time, Indicates about location At any moment Normalized posterior probability of region occlusion at time.

[0078] Step S252: Set the shadow threshold;

[0079] Step S253: Divide the region based on the shadow confidence and shadow threshold to obtain the division result of the photovoltaic power station. The division result includes a central concave area and an outer peripheral area.

[0080] Step S254: Perform binary mask encoding on the partitioning result to obtain the guiding mask of the photovoltaic power station.

[0081] In this step, the partitioning result is converted into a binary mask. If it is a central concave region, the guiding mask is 1; if it is a peripheral region, the guiding mask is 0. The guiding mask enables spatial saliency control, providing structural constraints for subsequent attention.

[0082] Step S3: Heuristic cross-modal feature extraction is performed on multi-source data using a guided mask to obtain cross-modal features, which include image feature vectors, electrical parameter time feature vectors, illumination stability coefficients, and meteorological parameter reliability probabilities.

[0083] Compared to traditional parallel extraction of multi-source features, this step utilizes guided masks to enhance the feature response of key areas and suppress interference from noisy areas, thereby maintaining the reliability and recognizability of features even when the quality of the original observation data is uneven, the lighting is unstable, or the meteorological conditions are complex.

[0084] Step S3 includes:

[0085] Step S31: Obtain the central recessed area and outer perimeter area of ​​the photovoltaic power station through the guide mask;

[0086] Step S32: Based on the central concave region and the peripheral region, feature extraction is performed on the bispectral image using different resolutions to obtain the image feature vector;

[0087] In this step, based on the spatial feature differences of the photovoltaic power station module array, the central concave area and the outer peripheral area are extracted respectively based on the guiding mask, and the dual-resolution convolution feature extraction strategy is adopted to extract the image feature vector from the dual-spectral image composed of visible light image and infrared image.

[0088] Specifically, for the central concave region, a small receptive field convolutional kernel (3×3) is used for feature extraction to enhance the ability to extract local defects and detailed features. For the peripheral region, a large receptive field convolutional kernel (7×7) is used for feature extraction to capture the overall shadow pattern and thermal distribution trend. Feature extraction is performed simultaneously on both the visible light and infrared images.

[0089] To integrate the complementary features of visible and infrared light, a feature channel attention mechanism is employed for fusion, yielding cross-spectral fusion features in the central concave region and the peripheral region. These cross-spectral fusion features are then subjected to global average pooling to obtain image feature vectors of a uniform scale.

[0090] By partitioning the central concave region and the peripheral region, the model avoids ignoring central details or the global background by using a single convolutional scale. Simultaneously, dual-resolution convolution enhances the model's responsiveness at different spatial levels, enabling it to detect both small-area shadows and large-area occlusions. Furthermore, since visible light images are sensitive to geometric shadows and infrared images are sensitive to thermal anomalies, the fused image feature vector provides a more robust shadow representation.

[0091] Step S33: Extract abnormal electrical parameters from the abnormal electrical parameter map and assign weights to the abnormal electrical parameters using a guiding mask;

[0092] In this step, electrical parameter anomaly features are constructed by arranging electrical parameter anomalies in chronological order. The electrical parameter anomalies of each partition unit are weighted according to the guiding mask of the central concave region and the outer peripheral region to obtain the weighted electrical parameter anomaly features. The weight of the central concave region is set to 1.0 and the weight of the outer peripheral region is set to 0.8 to balance the feature contributions of different spatial regions.

[0093] Step S34: Obtain the electrical parameter time feature vector by passing the weighted electrical parameter anomaly features through a bidirectional LSTM network;

[0094] In this step, bidirectional LSTM is used to capture past and future information simultaneously, so that the time feature vector of electrical parameters can reflect the dynamic changes of electrical parameter anomalies over time.

[0095] Step S35: Calculate the irradiance change rate using solar radiation parameters, and calculate the light fluctuation index using the luminance variance of the bispectral image;

[0096] In this step, the irradiance change rate is calculated by using the real-time irradiance at different times, and the brightness value is obtained by using the visible light image. The brightness variance is then calculated using the brightness value and used as the light fluctuation index.

[0097] Step S36: Calculate the illumination stability coefficient using the irradiance change rate and the illumination fluctuation index;

[0098] In this step, the overall irradiance change rate With light fluctuation index Define the illumination stability coefficient :

[0099]

[0100] In the formula, and All of these represent the weighting coefficients for illumination stability.

[0101] Light stability coefficient The closer to 1, the more stable the illumination; the smaller the value, the greater the fluctuation in illumination.

[0102] Step S37: Establish a meteorological reliability prior model by using the cloud cover change rate and historical statistical errors;

[0103] In this step, the meteorological reliability prior model can reflect whether the current environment is stable, thereby reducing erroneous shadow identification caused by environmental disturbances. For example, when the illumination is stable and the cloud cover is low, the image reliability is high; when the illumination disturbance is large, the image weight is reduced to avoid misjudging shadows. Therefore, by considering cloud cover and historical errors, the credibility of the current meteorological data can be quantified, reducing the misleading effect of abnormal weather conditions on shadow identification.

[0104] Specifically, obtaining the cloud cover change rate Compared with historical statistical error (e.g., the deviation between long-term observation data and theoretical irradiance). Establish a meteorological reliability prior model:

[0105]

[0106] In the formula, This indicates the preliminary probability of the reliability of meteorological data. These are the weighting coefficients.

[0107] Step S38: Combine the weight allocation of the guiding mask to generate the meteorological parameter reliability probability of the meteorological data through the meteorological reliability prior model.

[0108] In this step, a weighted distribution is applied to different regions (central concave area and outer perimeter area) using a guiding mask to generate the reliability probability of meteorological parameters. .

[0109] Step S4: Perform probabilistic cross-attention fusion on the cross-modal features to obtain a probabilistic fused feature vector;

[0110] In this step, a probabilistic cross-attention fusion mechanism for cross-modal features is introduced, which enables features from different modalities to not only achieve effective information complementarity based on attention weights during the fusion process, but also to dynamically modulate the attention weights using the reliability probability of meteorological parameters and the illumination stability coefficient. This actively weakens the interference of low-reliability modalities and enhances the contribution of high-reliability modalities during feature fusion.

[0111] Step S4 includes:

[0112] Step S41: Calculate the reliability probability of the image feature vector using the reliability probability of meteorological parameters;

[0113] In this step, image feature vectors are more reliable when meteorological conditions are reliable. Calculate the reliability probability of the image feature vector:

[0114]

[0115] In the formula, This represents the Sigmoid function, which maps the input to... , This represents the first trainable scaling factor, used to adjust the strength of the influence of meteorological parameters on the reliability of image features. This represents the first trainable bias term. This represents the reliability probability of an image feature vector.

[0116] Step S42: Calculate the reliability probability of the electrical parameter time feature vector using the illumination stability coefficient;

[0117] In this step, regions with small illumination fluctuations exhibit more stable and reliable electrical parameter time series, resulting in a higher probability of reliability. Therefore, the illumination stability coefficient is utilized. Probabilistically transform the time characteristics of electrical parameters:

[0118]

[0119] In the formula, Represents the reliability probability of the time eigenvector of electrical parameters. This represents a trainable second scaling factor used to adjust the effect of illumination stability on the reliability of electrical parameter features. This represents the trainable second bias term.

[0120] Step S43: Perform linear mapping on the image feature vector and the electrical parameter time feature vector respectively to obtain the corresponding query vector, key vector and value vector;

[0121] Step S44: Construct an image-electrical parameter cross-attention matrix based on the query vector and key vector;

[0122] In this step, the image focuses on the electrical parameters. for:

[0123]

[0124] Attention to images by electrical parameters for:

[0125]

[0126] In the formula, Represents the normalization function. and These represent query vectors for image feature vectors and electrical parameter time feature vectors, respectively. and These represent the transposes of the key vectors, which respectively represent the time feature vectors of electrical parameters and the image feature vectors. This represents the dimension of the key vector.

[0127] pass and Constructing an image-electrical parameter cross-attention matrix can capture the complementary relationship between image features and electrical parameter features, allowing each modality to focus on the most relevant information of the other.

[0128] Step S45: Calculate the probabilistic weights using the image-electrical parameter cross-attention matrix, the reliability probability of the image feature vector, and the reliability probability of the electrical parameter time feature vector;

[0129] In this step, the reliability probability calculation probabilistic weights take into account both the correlation between modes and the reliability of the modes themselves, making the fusion results more robust.

[0130] Specifically, the probabilistic weights for reliability probability calculation are:

[0131]

[0132]

[0133] In the formula, The probabilistic weights representing the image feature vectors. The probabilistic weights represent the time eigenvectors of electrical parameters. This indicates element-wise multiplication.

[0134] Step S46: The image feature vector and the electrical parameter time feature vector are weighted and fused using value vector and probabilistic weights to obtain the probabilistic fused feature vector.

[0135] In this step, the feature vectors are probabilistically fused. for:

[0136]

[0137] In the formula, and These represent the value vectors of the image feature vector and the electrical parameter time feature vector, respectively.

[0138] The fused feature vector integrates image and electrical parameter information, and adjusts their respective contributions according to probabilistic weights. This ensures that the probabilistically fused feature vector retains cross-modal information while reducing noise interference. A reliability-driven adaptive fusion method is implemented to address the time-varying meteorological conditions, lighting environment, and equipment status in photovoltaic power plants, effectively mitigating cross-modal conflict issues caused by cloud cover changes, local overexposure, module aging, or instantaneous current fluctuations.

[0139] Step S5: Perform Bayesian multi-classification based on probabilistic fusion feature vectors to generate shadow occlusion recognition results for photovoltaic power plants. The shadow occlusion recognition results include shadow occlusion location and shadow occlusion type.

[0140] In this step, prior knowledge and feature probability distributions are explicitly introduced into the classification process, making shadow occlusion recognition more stable and interpretable. A Bayesian framework is used, employing scene information from the photovoltaic power station and historical occlusion patterns as priors. This effectively distinguishes between different types of temporary occlusion (such as moving cloud shadows) and long-term occlusion (such as occlusion by fixed structures), improving the reliability of shadow classification. Simultaneously, kernel density estimation is performed on the fusion features of historical samples of each type to obtain a more realistic conditional probability distribution, maintaining robustness even when facing complex light disturbances, current fluctuations, or local noise.

[0141] In step S5, the step of performing Bayesian multi-classification based on probabilistic fusion feature vectors to generate the shading recognition result of the photovoltaic power station includes:

[0142] Step S51: Define the initial prior probability for each shading type based on the photovoltaic power station scene parameters and historical shading events;

[0143] In this step, the shading type is defined by combining the scene parameters of the photovoltaic power station (such as component layout, geographical location, and height of obstructions) and historical shading events. initial prior probability Among them, shadow occlusion types include temporary occlusion and permanent occlusion.

[0144] Step S52: Based on historical data, perform kernel density estimation on the probabilistic fusion feature vectors of different shadow occlusion types to obtain the conditional probability distribution;

[0145] In this step, conditional probabilities are estimated using nonparametric methods to avoid assuming the distribution of features and improve the ability of multimodal features to distinguish shadow types.

[0146] The conditional probability distribution is:

[0147]

[0148] In the formula, Indicates the type of shadow occlusion Under the conditions, The probability distribution of occurrence Indicates the type of shadow occlusion Historical sample size Indicates kernel bandwidth. Indicates the type of shadow occlusion The kernel function (using a Gaussian kernel). This represents a probabilistically fused feature vector. Indicates the first Probabilistic fusion feature vectors of historical samples.

[0149] Step S53: Based on the conditional probability distribution and the probabilistic fusion feature vector, query the likelihood probability that each partition unit of the photovoltaic power station belongs to different shading types;

[0150] Step S54: Calculate the posterior probability of each partition unit of the photovoltaic power station belonging to different shading types using the initial prior probability and the likelihood probability;

[0151] In this step, the likelihood probability comes from real-time observations, reflecting the current data behavior, while the initial prior probability comes from historical statistics, reflecting the long-term distribution of the scene. Combining these two factors ensures stability even under noise and large lighting fluctuations, and also addresses the issue of missing data, allowing for classification even when images are unstable.

[0152] Combined with initial prior probabilities and likelihood probability Calculate the posterior probability:

[0153]

[0154] In the formula, Represents partition unit This belongs to the shadow occlusion type The posterior probability, Represents partition unit This belongs to the shadow occlusion type The likelihood probability, Represents partition unit The corresponding probabilistic fusion feature vector.

[0155] Step S55: Select the shadow occlusion type with the highest posterior probability as the shadow occlusion type of the partition unit;

[0156] Step S56: Mark the shading location of the photovoltaic power station by the shading type of the partition unit, and generate shading identification results.

[0157] In this step, the shadow location is marked on the two-dimensional plane of the photovoltaic power station by the shadow occlusion type of the partition unit, generating the shadow occlusion recognition result of the photovoltaic power station.

[0158] Example 2:

[0159] like Figure 2 As shown, this embodiment provides a photovoltaic power station shading recognition system based on multi-source data. The system includes:

[0160] The acquisition module is used to acquire multi-source data from the photovoltaic power station. The multi-source data includes bispectral images, IV curve time-series data, meteorological data, and solar radiation parameters. The bispectral images are image pairs composed of visible light images and infrared images.

[0161] The generation module is used to partition the photovoltaic power station through multi-source data and generate the guiding mask of the photovoltaic power station by combining the prior information of the shading object and the posterior probability of the regional shading.

[0162] The extraction module is used to perform heuristic cross-modal feature extraction on multi-source data through a guided mask to obtain cross-modal features, which include image feature vectors, electrical parameter time feature vectors, illumination stability coefficients, and meteorological parameter reliability probabilities.

[0163] The fusion module is used to perform probabilistic cross-attention fusion on cross-modal features to obtain a probabilistic fused feature vector;

[0164] The identification module is used to perform Bayesian multi-classification based on probabilistic fusion feature vectors to generate shadow occlusion identification results for photovoltaic power plants. The shadow occlusion identification results include shadow occlusion location and shadow occlusion type.

[0165] The generation module includes:

[0166] The partitioning unit is used to perform regular grid division on the bispectral image based on the arrangement pattern of photovoltaic modules in a photovoltaic power station, resulting in partitioning units;

[0167] The first calculation unit is used to calculate the current deviation coefficient, fill factor stability and local power anomaly index of each partition unit within the sampling period using IV curve time series data, and generate an electrical parameter anomaly map.

[0168] The first building unit is used to construct a time-by-time shadow projection analysis model based on solar radiation parameters and prior information of shading objects, and to generate potential shadow projection areas based on the time-by-time shadow projection analysis model.

[0169] The second computing unit is used to perform probabilistic modeling of meteorological data, calculate the probability of illumination disturbance in each partition unit, and superimpose it with the brightness temperature difference of the bispectral image to generate a regional occlusion posterior probability map.

[0170] The first generation unit is used to generate a guiding mask for a photovoltaic power station from an electrical parameter anomaly map, a potential shadow projection area, and a region occlusion posterior probability map.

[0171] The extraction module includes:

[0172] The acquisition unit is used to acquire the central recessed area and the outer peripheral area of ​​the photovoltaic power station through the guide mask;

[0173] The extraction unit is used to extract features from the bispectral image based on the central concave region and the peripheral region at different resolutions to obtain the image feature vector;

[0174] The allocation unit is used to extract abnormal electrical parameters from the abnormal electrical parameter map and assign weights to the abnormal electrical parameter features using a guiding mask.

[0175] The third computational unit is used to obtain the electrical parameter time feature vector by passing the weighted electrical parameter anomaly features through a bidirectional LSTM network.

[0176] The fourth calculation unit is used to calculate the irradiance change rate through solar radiation parameters and to calculate the light fluctuation index through the luminance variance of the bispectral image.

[0177] The fifth calculation unit is used to calculate the illumination stability coefficient using the irradiance change rate and the illumination fluctuation index;

[0178] The second building unit is used to establish a meteorological reliability prior model by using the cloud cover change rate and historical statistical errors;

[0179] The second generation unit is used to combine the weight allocation of the guiding mask to generate the meteorological parameter reliability probability of the meteorological data through the meteorological reliability prior model.

[0180] The fusion module includes:

[0181] The sixth calculation unit is used to calculate the reliability probability of image feature vectors based on the reliability probability of meteorological parameters.

[0182] The seventh calculation unit is used to calculate the reliability probability of the time characteristic vector of electrical parameters using the illumination stability coefficient;

[0183] The mapping unit is used to perform linear mapping on the image feature vector and the electrical parameter time feature vector respectively to obtain the corresponding query vector, key vector and value vector;

[0184] The third building unit is used to construct an image-electrical parameter cross-attention matrix based on the query vector and the key vector;

[0185] The eighth computational unit is used to calculate probabilistic weights using the image-electrical parameter cross-attention matrix, the reliability probability of the image feature vector, and the reliability probability of the electrical parameter time feature vector.

[0186] The fusion unit is used to perform weighted fusion of image feature vector and electrical parameter time feature vector through value vector and probabilistic weights to obtain probabilistic fused feature vector.

[0187] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0188] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0189] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying shading in photovoltaic power plants based on multi-source data, characterized in that, include: Acquire multi-source data from photovoltaic power plants, including bispectral images, IV curve time-series data, meteorological data, and solar radiation parameters. The bispectral images are image pairs composed of visible light images and infrared images. Photovoltaic power plants are partitioned using multi-source data, and a boot mask for photovoltaic power plants is generated by combining prior information on obstructions and posterior probability of regional obstruction. Heuristic cross-modal feature extraction is performed on multi-source data using a guided mask to obtain cross-modal features, which include image feature vectors, electrical parameter time feature vectors, illumination stability coefficients, and meteorological parameter reliability probabilities. Probabilistic cross-attention fusion is performed on cross-modal features to obtain a probabilistic fused feature vector; Bayesian multi-classification is performed based on probabilistic fusion feature vectors to generate shadow occlusion recognition results for photovoltaic power plants. The shadow occlusion recognition results include shadow occlusion location and shadow occlusion type.

2. The photovoltaic power station shading identification method based on multi-source data according to claim 1, characterized in that, The process of generating a pilot mask for a photovoltaic power station by combining prior information about obstructions and posterior probability of regional obstruction includes: Based on the arrangement pattern of photovoltaic modules in a photovoltaic power plant, the bispectral image is divided into regular grids to obtain partitioned units; The current deviation coefficient, fill factor stability and local power anomaly index of each partition unit are calculated using IV curve time series data to generate an electrical parameter anomaly map. A time-by-time shadow projection analysis model is constructed based on solar radiation parameters and prior information on shading objects, and potential shadow projection regions are generated based on the time-by-time shadow projection analysis model. Probabilistic modeling of meteorological data is performed to calculate the probability of illumination disturbance in each partition unit, and the result is superimposed with the brightness temperature difference of the bispectral image to generate a posterior probability map of regional occlusion. The guiding mask for a photovoltaic power plant is generated by using anomaly maps of electrical parameters, potential shadow projection areas, and posterior probability maps of regional shading.

3. The photovoltaic power station shading identification method based on multi-source data according to claim 2, characterized in that, A guiding mask for a photovoltaic power plant is generated using electrical parameter anomaly maps, potential shadow projection regions, and posterior probability maps of regional shading. This includes: After normalizing the electrical parameter anomaly map, the potential shadow projection region, and the region occlusion posterior probability map, the shadow confidence of each partition unit is calculated using a Bayesian weighted strategy. Set the shadow threshold; The region is divided based on shadow confidence and shadow threshold to obtain the division result of photovoltaic power station, which includes a central concave area and an outer peripheral area; The partitioning results are encoded using a binary mask to obtain the guiding mask for the photovoltaic power station.

4. The photovoltaic power station shading identification method based on multi-source data according to claim 1, characterized in that, The method of heuristically extracting cross-modal features from multi-source data using a guided mask to obtain cross-modal features includes: The central recessed area and outer perimeter area of ​​the photovoltaic power station are obtained by using a guide mask; Based on the central concave region and the outer peripheral region, feature extraction is performed on the bispectral image using different resolutions to obtain the image feature vector; The abnormal features of electrical parameters are extracted from the abnormal electrical parameter map, and weights are assigned to the abnormal electrical parameter features using a guiding mask. The weighted electrical parameter anomalies are passed through a bidirectional LSTM network to obtain the electrical parameter time feature vector. The rate of change of irradiance is calculated using solar radiation parameters, and the light fluctuation index is calculated using the luminance variance of the bispectral image. The light stability coefficient is calculated using the irradiance variation rate and the light fluctuation index. A meteorological reliability prior model was established by using cloud cover change rate and historical statistical error. By combining the weight allocation of the guiding mask, the reliability probability of meteorological parameters in meteorological data is generated through a meteorological reliability prior model.

5. The photovoltaic power station shading identification method based on multi-source data according to claim 1, characterized in that, The probabilistic cross-attention fusion of cross-modal features to obtain a probabilistic fused feature vector includes: The reliability probability of image feature vectors is calculated using the reliability probability of meteorological parameters. The reliability probability of the time eigenvector of electrical parameters is calculated using the illumination stability coefficient. Linear mapping is performed on the image feature vector and the electrical parameter time feature vector to obtain the corresponding query vector, key vector and value vector; Construct an image-electrical parameter cross-attention matrix based on query vectors and key vectors; Probabilistic weights are calculated using the image-electrical parameter cross-attention matrix, the reliability probability of the image feature vector, and the reliability probability of the electrical parameter time feature vector. The image feature vector and the electrical parameter time feature vector are weighted and fused by value vector and probabilistic weights to obtain the probabilistic fused feature vector.

6. The photovoltaic power station shading identification method based on multi-source data according to claim 1, characterized in that, The method of generating shadow occlusion recognition results for photovoltaic power plants based on probabilistic fusion feature vectors for Bayesian multi-class classification includes: The initial prior probability of each shadow occlusion type is defined based on the photovoltaic power plant scenario parameters and historical occlusion events. Based on historical data, kernel density estimation is performed on the probabilistic fusion feature vectors of different shadow occlusion types to obtain conditional probability distributions; Based on the conditional probability distribution and probabilistic fusion feature vector, query the likelihood probability that each partition unit of the photovoltaic power station belongs to different shading types; The posterior probability of each partition unit of the photovoltaic power station belonging to different shading types is calculated by using the initial prior probability and the likelihood probability. The shadow occlusion type with the highest posterior probability is selected as the shadow occlusion type of the partition unit; The location of shadow shading in a photovoltaic power station is marked by the shadow shading type of the partition unit, and shadow shading identification results are generated.

7. A photovoltaic power station shading recognition system based on multi-source data, characterized in that, include: The acquisition module is used to acquire multi-source data from the photovoltaic power station. The multi-source data includes bispectral images, IV curve time-series data, meteorological data, and solar radiation parameters. The bispectral images are image pairs composed of visible light images and infrared images. The generation module is used to partition the photovoltaic power station through multi-source data and generate the guiding mask of the photovoltaic power station by combining the prior information of the shading object and the posterior probability of the regional shading. The extraction module is used to perform heuristic cross-modal feature extraction on multi-source data through a guided mask to obtain cross-modal features, which include image feature vectors, electrical parameter time feature vectors, illumination stability coefficients, and meteorological parameter reliability probabilities. The fusion module is used to perform probabilistic cross-attention fusion on cross-modal features to obtain a probabilistic fused feature vector; The identification module is used to perform Bayesian multi-classification based on probabilistic fusion feature vectors to generate shadow occlusion identification results for photovoltaic power plants. The shadow occlusion identification results include shadow occlusion location and shadow occlusion type.

8. The photovoltaic power station shading identification system based on multi-source data according to claim 7, characterized in that, The generation module includes: The partitioning unit is used to perform regular grid division on the bispectral image based on the arrangement pattern of photovoltaic modules in a photovoltaic power station, resulting in partitioning units; The first calculation unit is used to calculate the current deviation coefficient, fill factor stability and local power anomaly index of each partition unit within the sampling period using IV curve time series data, and generate an electrical parameter anomaly map. The first building unit is used to construct a time-by-time shadow projection analysis model based on solar radiation parameters and prior information of shading objects, and to generate potential shadow projection areas based on the time-by-time shadow projection analysis model. The second computing unit is used to perform probabilistic modeling of meteorological data, calculate the probability of illumination disturbance in each partition unit, and superimpose it with the brightness temperature difference of the bispectral image to generate a regional occlusion posterior probability map. The first generation unit is used to generate a guiding mask for a photovoltaic power station from an electrical parameter anomaly map, a potential shadow projection area, and a region occlusion posterior probability map.

9. The photovoltaic power station shading identification system based on multi-source data according to claim 7, characterized in that, The extraction module includes: The acquisition unit is used to acquire the central recessed area and the outer peripheral area of ​​the photovoltaic power station through the guide mask; The extraction unit is used to extract features from the bispectral image based on the central concave region and the peripheral region at different resolutions to obtain the image feature vector; The allocation unit is used to extract abnormal electrical parameters from the abnormal electrical parameter map and assign weights to the abnormal electrical parameter features using a guiding mask. The third computational unit is used to obtain the electrical parameter time feature vector by passing the weighted electrical parameter anomaly features through a bidirectional LSTM network. The fourth calculation unit is used to calculate the irradiance change rate through solar radiation parameters and to calculate the light fluctuation index through the luminance variance of the bispectral image. The fifth calculation unit is used to calculate the illumination stability coefficient using the irradiance change rate and the illumination fluctuation index; The second building unit is used to establish a meteorological reliability prior model by using the cloud cover change rate and historical statistical errors; The second generation unit is used to combine the weight allocation of the guiding mask to generate the meteorological parameter reliability probability of the meteorological data through the meteorological reliability prior model.

10. The photovoltaic power station shading recognition system based on multi-source data according to claim 7, characterized in that, The fusion module includes: The sixth calculation unit is used to calculate the reliability probability of image feature vectors based on the reliability probability of meteorological parameters. The seventh calculation unit is used to calculate the reliability probability of the time characteristic vector of electrical parameters using the illumination stability coefficient; The mapping unit is used to perform linear mapping on the image feature vector and the electrical parameter time feature vector respectively to obtain the corresponding query vector, key vector and value vector; The third building unit is used to construct an image-electrical parameter cross-attention matrix based on the query vector and the key vector; The eighth computational unit is used to calculate probabilistic weights using the image-electrical parameter cross-attention matrix, the reliability probability of the image feature vector, and the reliability probability of the electrical parameter time feature vector. The fusion unit is used to perform weighted fusion of image feature vector and electrical parameter time feature vector through value vector and probabilistic weights to obtain probabilistic fused feature vector.

Images