Photovoltaic abnormal data processing method and related equipment

Abstract

The invention discloses a photovoltaic abnormal data processing method and related equipment, and the method comprises the steps: firstly collecting the historical total power and solar irradiance data of a target photovoltaic field station, generating a first data set through negative value zeroing and obvious abnormality elimination, and eliminating error data caused by sensor faults and the like; and calculating the neighborhood density deviation degree of data points by using a local abnormal factor algorithm, identifying and eliminating discrete abnormal points generated by instant faults of the sensor and the like to obtain a second data set, and eliminating the interference of discrete data on image identification. Then, the second data set is preprocessed to generate a scatter diagram, accumulation type abnormal areas caused by sudden change of system parameters and the like are recognized by means of image recognition technologies such as edge detection and the like, normal data are marked, and a final data set is output through reverse normalization. According to the method, two algorithms are fused, discrete abnormal points are recognized in a self-adaptive mode, accumulation type anomalies are analyzed based on morphological characteristics, normal data mistaken deletion is reduced, and the anomaly detection accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and more specifically, to a photovoltaic abnormal data processing method and related equipment. Background Art

[0002] As the energy mix shifts toward cleaner energy, distributed photovoltaic power generation, with its clean and flexible characteristics, is becoming increasingly important in the energy sector. Accurate power generation forecasts not only facilitate participation in electricity market transactions and improve economic efficiency, but also enhance the grid's ability to absorb distributed energy and ensure stable power system operation. However, the current development of distributed photovoltaic power generation faces numerous data processing challenges.

[0003] From a construction perspective, distributed photovoltaic panels are installed on residential rooftops, industrial and commercial buildings, and other locations. Each household has a small installation scale and varying specifications, making the deployment of data collection equipment and the construction of transmission links complex, making efficient and unified data collection difficult. Furthermore, the spatial distance between meteorological stations and photovoltaic sites prevents precise matching of meteorological data and photovoltaic power data in both time and space. Sensor measurement errors also affect data accuracy, interfering with subsequent modeling.

[0004] During data collection, factors such as sensor failures, photovoltaic system maintenance, extreme weather, and power system scheduling frequently generate a large amount of abnormal data. Once this abnormal data enters the training and inference stages of prediction models, it distorts the true data distribution, causing the models to learn incorrect patterns, reducing prediction accuracy, and even leading to operational risks and economic losses for the power grid.

[0005] Currently used methods for identifying abnormal photovoltaic power data have significant shortcomings. Traditional probabilistic statistical methods rely on specific probability distribution assumptions and are ineffective at identifying data that does not conform to these assumptions, particularly accumulation-type anomalies. While artificial intelligence methods can identify outliers, they require manual setting of key parameters, making it easy to mistakenly delete normal data. Even when image recognition technology is applied, discrete data in the original scatter plot creates interference noise, which reduces the accuracy of identifying accumulation-type anomalies.

[0006] Therefore, a photovoltaic abnormal data processing method is urgently needed to meet the needs of practical applications. Summary of the Invention

[0007] The present application provides a photovoltaic abnormal data processing method and related equipment, which collects historical total power and solar irradiance data and performs negative value zeroing and obvious anomaly elimination, then uses a local anomaly factor algorithm to identify and eliminate discrete anomalies, and then performs image preprocessing on the processed data to generate a scatter plot and uses an image recognition algorithm to identify accumulated abnormal areas, and finally marks the normal data for denormalization output, thereby combining the advantages of the two algorithms and improving the accuracy and reliability of abnormal data detection.

[0008] A photovoltaic abnormal data processing method, comprising:

[0009] Collect historical total power data and matching historical solar irradiance data of target photovoltaic stations;

[0010] Performing negative value zeroing and obvious anomaly elimination processing on the historical total power data and the historical solar irradiance data to generate a first data set including solar irradiance and photovoltaic power;

[0011] identifying discrete photovoltaic power anomaly points in the first data set by using a local anomaly factor algorithm, and obtaining a second data set by marking and removing the discrete photovoltaic power anomaly points;

[0012] Performing image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power;

[0013] The accumulation-type photovoltaic power anomaly area in the solar irradiance and photovoltaic power scatter plot is identified by an image recognition algorithm, normal data outside the accumulation-type photovoltaic power anomaly area is marked by coordinate transformation, and the final data set is output after denormalization.

[0014] Optionally, identifying discrete photovoltaic power anomaly points in the first data set by using a local anomaly factor algorithm includes:

[0015] performing normalization processing on the first data set;

[0016] Calculating the Euclidean distance between all data points in the first data set, and determining the nearest neighbor distance of each data point based on a preset number of nearest neighbors;

[0017] Calculate the reachable distance of each data point according to the neighborhood distance of the nearest neighbor number;

[0018] Calculate the local reachability density of each data point based on the reachability distance;

[0019] A local anomaly factor is calculated according to the local reachable density, and a data point where the local anomaly factor is greater than a preset threshold is marked as a discrete photovoltaic power anomaly point.

[0020] Optionally, the calculation formula for the reachable distance is:

[0021]

[0022] The calculation formula of the local reachable density is:

[0023]

[0024] The calculation formula of the local anomaly factor is:

[0025]

[0026] in, For data points ( , ) to data point ( , )’s reachable distance, is the solar irradiance at time i, is the photovoltaic power at time i, is the neighborhood distance between the data at time j and the kth neighbor data, For data points ( , ) to data point ( , ), For data points ( , ), is the neighbor dataset, For data points ( , )’s local anomaly factor.

[0027] Optionally, performing image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power includes:

[0028] Normalizing the solar irradiance and photovoltaic power values ​​in the second data set independently so that their numerical ranges are mapped to the interval [0, 1];

[0029] A scatter plot of solar irradiance and photovoltaic power is generated by plotting the normalized solar irradiance as the horizontal axis and the photovoltaic power as the vertical axis.

[0030] Optionally, identifying an accumulation-type photovoltaic power anomaly region in the solar irradiance and photovoltaic power scatter plot by an image recognition algorithm includes:

[0031] Converting the solar irradiance and photovoltaic power scatter plot into a grayscale image, and performing binarization processing on the grayscale image;

[0032] Perform edge detection on the binary image and extract the scattered point distribution outline;

[0033] Performing ellipse fitting on the scatter point distribution profile to generate an ellipse equation representing a normal data distribution boundary;

[0034] The area outside the ellipse is determined as a pile-up type photovoltaic power abnormality area according to the ellipse equation.

[0035] Optionally, performing edge detection on the binarized image to extract the scattered point distribution outline includes:

[0036] Using a Gaussian filter to smooth the binarized image and output a noise-reduced image;

[0037] Applying a Sobel operator to perform a convolution operation on the denoised image to generate a gradient magnitude and gradient direction for each pixel;

[0038] Performing non-maximum suppression based on the gradient magnitude and the gradient direction, and outputting a refined edge response map;

[0039] Performing a double-threshold connection process on the refined edge response map to screen strong edge pixels and connect them to form a candidate edge set;

[0040] A continuous contour is generated based on the candidate edge set by a contour fitting algorithm as the scattered point distribution contour.

[0041] Optionally, the calculation formula for the gradient magnitude and gradient direction of each pixel is:

[0042]

[0043]

[0044] Among them, G is the gradient amplitude of the pixel point, 、 They are the horizontal gradient component obtained by horizontal convolution of the image by the Sobel operator, and the vertical gradient component obtained by vertical convolution of the image, is the gradient direction of the pixel.

[0045] A photovoltaic abnormal data processing device, comprising:

[0046] Photovoltaic data acquisition unit, used to collect historical total power data of the target photovoltaic station and the matching historical solar irradiance data;

[0047] an obvious anomaly processing unit, configured to perform negative value zeroing and obvious anomaly elimination processing on the historical total power data and the historical solar irradiance data, to generate a first data set including solar irradiance and photovoltaic power;

[0048] a discrete anomaly processing unit, configured to identify discrete photovoltaic power anomaly points in the first data set by using a local anomaly factor algorithm, and obtain a second data set by marking and removing the discrete photovoltaic power anomaly points;

[0049] an image processing construction unit, configured to perform image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power;

[0050] The accumulation anomaly processing unit is used to identify the accumulation-type photovoltaic power anomaly area in the solar irradiance and photovoltaic power scatter diagram through an image recognition algorithm, mark the normal data outside the accumulation-type photovoltaic power anomaly area through coordinate transformation, and output the final data set after denormalization.

[0051] A photovoltaic abnormal data processing device, comprising a memory and a processor;

[0052] The memory is used to store programs;

[0053] The processor is configured to execute the program to implement each step of the photovoltaic abnormal data processing method as described above.

[0054] A readable storage medium stores a computer program thereon, wherein when the computer program is executed by a processor, the computer program implements each step of the photovoltaic abnormal data processing method as described in any one of the above items.

[0055] It can be seen from the above technical solutions that the photovoltaic abnormal data processing method and related equipment provided by the embodiment of the present application first collect the historical total power data and historical solar irradiance data of the target photovoltaic station, perform negative value zeroing and obvious abnormality elimination processing on them, and generate a first data set, which effectively eliminates the negative values ​​or obvious erroneous data generated by factors such as sensor failure and transmission abnormality during actual collection. The first data set is processed using a local anomaly factor algorithm. The algorithm quantifies the degree of abnormality of each data point by calculating the density deviation between the data point and its neighborhood points. Discrete anomalies are often isolated data generated by instantaneous sensor failure, communication interruption, etc. The local anomaly factor algorithm can accurately identify discrete photovoltaic power anomalies. By marking and eliminating them, a second data set is obtained, which effectively eliminates the interference of discrete data on the overall data distribution, solves the problem of discrete data affecting image recognition morphological judgment in the prior art, and enables subsequent image recognition to focus more on the true distribution characteristics of the data.

[0056] Image preprocessing is performed on the second dataset to generate a scatter plot of solar irradiance and photovoltaic power. The scatter plot is then analyzed using an image recognition algorithm. This algorithm utilizes edge detection and morphological analysis techniques to identify areas of accumulated photovoltaic power anomalies. Accumulated anomalous data is often caused by sudden changes in system parameters or interference from multiple factors, and manifests itself in the scatter plot as a localized cluster of abnormal data. Normal data outside the abnormal area is marked through coordinate transformation, and the final dataset is output through denormalization. This accurately separates the normal data, overcoming the shortcomings of traditional methods in identifying accumulated anomalous data.

[0057] This application integrates a local anomaly factor algorithm with an image recognition algorithm, avoiding the limitations of a single algorithm. The local anomaly factor algorithm adaptively identifies discrete outliers without requiring complex pre-set parameters; the image recognition algorithm is based on intuitive data morphological feature analysis. The two work together to reduce the problem of inadvertent deletion of normal data due to improper parameter settings, significantly improving the accuracy and reliability of anomaly data detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0059] Figure 1 This is a flow chart of a photovoltaic abnormal data processing method disclosed in an embodiment of the present application;

[0060] Figure 2 A scatter plot of unprocessed solar irradiance and photovoltaic power disclosed in an embodiment of the present application;

[0061] Figure 3 The embodiment disclosed in this application Figure 2 Scatter plot after processing photovoltaic abnormal data;

[0062] Figure 4 A schematic diagram of a photovoltaic abnormality data processing device disclosed in an embodiment of the present application;

[0063] Figure 5 This is a hardware structure block diagram of a photovoltaic abnormal data processing device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0065] The present application can be used in a variety of general or special computing device environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multi-processor devices, and distributed computing environments including any of the above devices or devices.

[0066] Next, we will introduce the application scheme. This application proposes the following technical scheme, please see below for details.

[0067] Figure 1 This is a flowchart of a photovoltaic abnormal data processing method disclosed in an embodiment of the present application.

[0068] like Figure 1 As shown, the method may include:

[0069] Step S1: Collect historical total power data and matching historical solar irradiance data of the target photovoltaic station.

[0070] Specifically, to obtain basic data on the operating status of a photovoltaic station, it is necessary to synchronously collect the historical total power data and corresponding solar irradiance data of the target station within a specific time period. The historical total power data of the target photovoltaic station refers to the total output power value of the photovoltaic system collected by the power monitoring equipment during the operation of the station, reflecting the power generation capacity of the photovoltaic power station at different times; the matching historical solar irradiance data is the solar radiation intensity data at the location of the photovoltaic panels collected synchronously by meteorological monitoring equipment, forming a time series data set. Data collection should maintain consistency in the time dimension, for example, using a 15-minute time interval for sampling to ensure that the two types of data strictly correspond in terms of timestamps. In practical applications, the operating data of a photovoltaic station for several consecutive months can be selected as samples to provide an original data set with time series characteristics for subsequent anomaly identification.

[0071] Step S2: performing negative value zeroing and obvious anomaly elimination processing on the historical total power data and the historical solar irradiance data to generate a first data set including solar irradiance and photovoltaic power.

[0072] Specifically, due to sensor failure or transmission errors, the raw data may contain negative values ​​or obvious abnormal values ​​that are physically unreasonable. The processing process includes:

[0073] ① Negative value zeroing: All data points with PV power less than 0 are uniformly corrected to 0 to eliminate erroneous values ​​caused by sensor abnormalities.

[0074] ② Eliminate obviously abnormal data: Identify and eliminate data that does not conform to physical laws.

[0075] Obviously abnormal data include:

[0076] Periodic anomalies: When solar irradiance is close to 0 in the early morning or evening, the PV power may be high. This is usually caused by sensor mis-triggering or data acquisition module failure.

[0077] Unreasonably low values: During normal daytime hours, solar irradiance or photovoltaic power is significantly lower than historical levels for the same period. This may be due to sensor aging, dust accumulation on the photovoltaic panels, or local circuit failure.

[0078] Contradictory data: When the solar irradiance is 0, the photovoltaic power is greater than 0, which violates the physical law that photovoltaic power generation depends on light. This is often caused by sensor clock asynchrony or data storage errors.

[0079] Missing data: The solar irradiance or photovoltaic power data is missing due to power failure of the collection equipment, communication interruption, etc.

[0080] Through the above preprocessing, a preliminarily cleaned first data set is generated.

[0081] Step S3: identifying discrete photovoltaic power anomaly points in the first data set by using a local anomaly factor algorithm, and obtaining a second data set by marking and removing the discrete photovoltaic power anomaly points.

[0082] Specifically, discrete outliers are often isolated, transient error values, such as peak sensor output caused by electromagnetic interference, distortion caused by fluctuations in data transmission networks, or data generated by temporary sensor disruptions due to extreme weather (severe lightning, sudden strong winds). The local anomaly factor algorithm analyzes the density differences between each data point and its neighbors to identify discrete points that significantly deviate from the normal distribution. For example, if a point in a dense data cluster is significantly distant from its neighbors, it is identified as an anomaly and removed, resulting in a cleaner second dataset.

[0083] To accurately identify discrete PV power anomalies in the first dataset, a local outlier factor (LOF) algorithm was systematically applied. First, data standardization was performed on the first dataset. This normalized the solar irradiance and PV power data to a unified metric space by eliminating dimensional differences. This ensured that different characteristic parameters were comparable during anomaly identification, avoiding biased anomaly determinations due to differences in numerical magnitude.

[0084] Neighborhood analysis is then performed to quantitatively assess the degree of deviation of data points within a local neighborhood by calculating the spatial distance and density distribution relationship between each data point and its neighbors. Normally, data points are densely distributed within a neighborhood, while discrete outliers, due to their isolated existence, have a local neighborhood density significantly lower than normal. This characteristic serves as a key basis for anomaly identification.

[0085] During the anomaly detection phase, the algorithm compares the local density of data points with the average density of their neighborhood to accurately identify discrete outliers that significantly deviate from the normal distribution. Typical anomaly cases include power spikes caused by transient electromagnetic interference on sensors, low values ​​of distortion caused by fluctuations in data transmission links, and isolated data points caused by brief sensor failures in extreme weather conditions. These outliers clearly show up as outliers within the data cluster.

[0086] Finally, data purification is achieved through marking and elimination operations, generating a second data set with significantly improved consistency and reliability. This algorithm does not rely on a preset probability distribution model, but achieves adaptive anomaly recognition through local neighborhood density comparison. It can adapt to scenarios where data distribution is dynamically adjusted due to changes in installed capacity of photovoltaic stations and aging of components. Isolated outliers caused by transient interference such as sensor failures and extreme weather can be accurately eliminated without affecting normal data clusters, effectively avoiding the problem of accidental deletion of normal data caused by global threshold settings in traditional statistical methods. In addition, by removing discrete outliers, the interference of clutter in the scatter plot is significantly reduced, providing a purer data visualization foundation for subsequent image recognition algorithms and ensuring the accuracy of morphological feature analysis of accumulated anomaly areas.

[0087] Step S4: performing image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power.

[0088] Specifically, to transform numerical data into visual analysis objects for subsequent detection of accumulation anomalies using image recognition techniques, a systematic image preprocessing operation is performed on the second dataset. First, the solar irradiance and photovoltaic power parameters in the dataset are normalized. A linear transformation is used to map the numerical ranges of the two data types to the interval [0, 1]. This ensures that the horizontal and vertical coordinates are presented on the same scale and avoids visualization errors caused by differences in numerical spans. Normalization eliminates dimensionality effects, resulting in comparable distributions of irradiance and power in a two-dimensional coordinate system.

[0089] Then, appropriate drawing parameters are selected to generate a scatter plot: a rectangular coordinate system is established with solar irradiance as the horizontal axis and photovoltaic power as the vertical axis. A reasonable coordinate range is set according to the data distribution characteristics to ensure that the distribution form of normal data clusters (such as the trend of nonlinear growth of power with increasing irradiance) is fully presented. During the drawing process, the proportional consistency of point density and the amount of original data must be maintained to avoid distortion of distribution characteristics caused by point overlap. In typical application scenarios, data in clear weather should show a dense clustered distribution extending from the coordinate origin to the upper right, while data in rainy weather or in the morning and evening hours should show a corresponding distribution of low irradiance and low power.

[0090] After the scatter plot is drawn, the image file is saved at its original scale to ensure that the spatial distribution of the data points is not disrupted by the scaling operation. This scatter plot serves as the basis for subsequent image recognition, and its quality directly affects the accuracy of detecting accumulation-type anomaly areas. By removing the discrete anomalies in the second dataset, the scatter plot eliminates the interference of outliers, making the distribution of normal data clearer and laying a visual foundation for the recognition of accumulation-type anomalies based on image morphology.

[0091] Performing image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power specifically includes:

[0092] ① Normalize the solar irradiance and photovoltaic power values ​​in the second dataset independently so that their numerical ranges are mapped to the interval [0, 1];

[0093] ② A scatter plot of solar irradiance and photovoltaic power is generated by plotting the normalized solar irradiance as the horizontal axis and the photovoltaic power as the vertical axis.

[0094] The solar irradiance and photovoltaic power values ​​in the second dataset are normalized independently, and the original numerical ranges of the two types of data are uniformly mapped to the [0,1] interval using a linear mapping method. The specific implementation method is:

[0095] For the second data set Normalize the horizontal and vertical coordinates to [0,1] and get the data set :

[0096] ,

[0097] in, 、 For the second data set Solar irradiance and photovoltaic power in.

[0098] A two-dimensional rectangular coordinate system is constructed with the normalized solar irradiance value as the horizontal axis and the photovoltaic power value as the vertical axis. Each data point is plotted using a fixed point size (e.g., 0.5 pixel diameter) to ensure that the data distribution density remains consistent with the original dataset. The resulting scatter plot intuitively presents the correlation and distribution characteristics of solar irradiance and photovoltaic power. The following parameter setting principles should be followed during the plotting process:

[0099] Coordinate range: The horizontal axis (irradiance) and the vertical axis (power) are both set to [0, 1], strictly matching the range of the normalized data;

[0100] Point density control: Adjust the image resolution according to the size of the dataset to ensure that the data points neither overlap and obscure the distribution characteristics, nor are they too sparse to cause morphological discontinuity;

[0101] Image scale: Keep the physical unit lengths of the horizontal and vertical axes consistent (e.g., 1 unit corresponds to 100 pixels) to avoid visual distortion caused by inconsistent axis scaling.

[0102] Step S5: Identify the accumulation-type photovoltaic power abnormal area in the solar irradiance and photovoltaic power scatter plot using an image recognition algorithm, mark the normal data outside the accumulation-type photovoltaic power abnormal area through coordinate transformation, and output the final data set after denormalization.

[0103] Specifically, accumulation-type photovoltaic power anomaly refers to the local failure of photovoltaic modules (such as series branch failure, inverter abnormality), spatial mismatch between meteorological stations and site stations (systematic deviation between irradiance measurement values ​​and actual received values), or interference from multiple factors (such as dust accumulation, shadow obstruction), which results in the appearance of local dense data areas that deviate from the normal data distribution cluster in the scatter plot. Its characteristic is that the aggregation pattern of data points is inconsistent with the normal power generation law (such as continuously low or high power under the same irradiance).

[0104] To accurately identify these anomalies, the following process is implemented: First, the scatter plot is converted to a grayscale image using a weighted formula to account for the human eye's varying color sensitivity. Next, the image is binarized using the Otsu thresholding method or manually set thresholds to separate the data points from the background. To eliminate discrete noise interference, the binarized image is smoothed using a Gaussian filter. The Sobel operator is then used to calculate the intensity and direction of pixel gradients. Finally, non-maximum suppression and double thresholding are used to extract the edge contours of data clusters.

[0105] Ellipse fitting is then performed on the main contours, using direct least squares to obtain the ellipse parameters in the image processing coordinate system, including the midpoint coordinates, semi-major axis, semi-minor axis, and rotation angle. This ellipse represents the distribution range of normal data. A coordinate transformation is used to rotate the ellipse to the standard coordinate system to eliminate the effects of angular deviation. For each point in the normalized dataset, the ellipse equation is used to determine whether it belongs to the normal area. Points that meet the criteria are marked as normal data.

[0106] Finally, the original scale of the data is restored through denormalization. For the normalized values ​​r' and P', the original solar irradiance and photovoltaic power values ​​are calculated respectively, and finally a pure data set is output that eliminates discrete and accumulation anomalies.

[0107] The processing effect of this application is shown in Figure 2 and Figure 3 , Figure 2 is a scatter plot of unprocessed solar irradiance and photovoltaic power, Figure 3 This application is used to Figure 2 Scatter plot obtained after photovoltaic anomaly data processing.

[0108] It can be seen from the above technical solutions that the photovoltaic abnormal data processing method and related equipment provided by the embodiment of the present application first collect the historical total power data and historical solar irradiance data of the target photovoltaic station, perform negative value zeroing and obvious abnormality elimination processing on them, and generate a first data set, which effectively eliminates the negative values ​​or obvious erroneous data generated by factors such as sensor failure and transmission abnormality during actual collection. The first data set is processed using a local anomaly factor algorithm. The algorithm quantifies the degree of abnormality of each data point by calculating the density deviation between the data point and its neighborhood points. Discrete anomalies are often isolated data generated by instantaneous sensor failure, communication interruption, etc. The local anomaly factor algorithm can accurately identify discrete photovoltaic power anomalies. By marking and eliminating them, a second data set is obtained, which effectively eliminates the interference of discrete data on the overall data distribution, solves the problem of discrete data affecting image recognition morphological judgment in the prior art, and enables subsequent image recognition to focus more on the true distribution characteristics of the data.

[0109] Image preprocessing is performed on the second dataset to generate a scatter plot of solar irradiance and photovoltaic power. The scatter plot is then analyzed using an image recognition algorithm. This algorithm utilizes edge detection and morphological analysis techniques to identify areas of accumulated photovoltaic power anomalies. Accumulated anomalous data is often caused by sudden changes in system parameters or interference from multiple factors, and manifests itself in the scatter plot as a localized cluster of abnormal data. Normal data outside the abnormal area is marked through coordinate transformation, and the final dataset is output through denormalization. This accurately separates the normal data, overcoming the shortcomings of traditional methods in identifying accumulated anomalous data.

[0110] This application integrates a local anomaly factor algorithm with an image recognition algorithm, avoiding the limitations of a single algorithm. The local anomaly factor algorithm adaptively identifies discrete outliers without requiring complex pre-set parameters; the image recognition algorithm is based on intuitive data morphological feature analysis. The two work together to reduce the problem of inadvertent deletion of normal data due to improper parameter settings, significantly improving the accuracy and reliability of anomaly data detection.

[0111] In some embodiments of the present application, the process of identifying discrete photovoltaic power anomaly points in the first data set by using the local anomaly factor algorithm in step S3 is introduced, which may specifically include:

[0112] Step S31: normalizing the first data set;

[0113] Step S32: Calculate the Euclidean distance between all data points in the first data set, and determine the nearest neighbor distance of each data point based on a preset nearest neighbor number;

[0114] Step S33, calculating the reachable distance of each data point according to the neighborhood distance of the nearest neighbor number;

[0115] Step S34: Calculate the local reachability density of each data point based on the reachable distance;

[0116] Step S35: Calculate a local anomaly factor according to the local reachable density, and mark a data point where the local anomaly factor is greater than a preset threshold as a discrete photovoltaic power anomaly point.

[0117] Specifically, the first dataset is normalized to eliminate the dimensional differences between solar irradiance and photovoltaic power due to different physical units. By mapping each data type to a numerical range of 0 to 1, the different features are equally weighted in subsequent calculations, preventing the impact of numerical scale differences on the accuracy of outlier identification.

[0118] Next, the Euclidean distance between all data points in the dataset is calculated—that is, the straight-line distance between each data point and every other point in two-dimensional space. Based on a preset number of neighbors (e.g., 5 or 10), a set of nearest neighbors is determined for each data point. The distance to the farthest point in this set is used as the neighborhood distance for that data point, thus measuring the spatial distribution density around the data point.

[0119] The reachable distance of each data point is then calculated based on the neighborhood distance, which is defined as the larger of the actual distance between the data point and its nearest neighbor and the neighborhood distance. This calculation method avoids the problem of unstable distance calculation caused by data points being too close together, ensuring the reliability of subsequent density calculations.

[0120] Based on the reachable distance, we further calculate the local reachable density of each data point. This is the inverse of the average reachable distance from the data point to all its neighbors. The local reachable density reflects how tightly clustered the data points are within their neighborhood. Normal data points typically have a high local reachable density, while discrete outliers, far from the main data cluster, have a low local reachable density.

[0121] Finally, the local outlier factor is calculated by comparing the local reachable density of each data point with the local reachable density of its nearest neighbors. This factor indicates the degree of anomalousness of a data point relative to its neighborhood. When the local outlier factor exceeds a preset threshold (e.g., 1.5 or 2.0), the data point's local density is significantly lower than that of its nearest neighbors, making it a discrete outlier. By marking and removing these outliers, a higher-quality second dataset is obtained, providing a reliable data foundation for subsequent image preprocessing and analysis.

[0122] Assume that the data set D of solar irradiance and photovoltaic power is:

[0123] Choose the number of nearest neighbors k and the distance metric. Calculate the Euclidean distance d of all data:

[0124]

[0125] For each data point, find its corresponding neighbor , and calculate the Euclidean distance of all data :

[0126]

[0127]

[0128] The calculation formula of the reachable distance is:

[0129]

[0130] The calculation formula of the local reachable density is:

[0131]

[0132] The calculation formula of the local anomaly factor is:

[0133]

[0134] in, For data points ( , ) to data point ( , )’s reachable distance, is the solar irradiance at time i, is the photovoltaic power at time i, is the neighborhood distance between the data at time j and the kth neighbor data, For data points ( , ) to data point ( , ), For data points ( , ), is the neighbor dataset, For data points ( , )’s local anomaly factor.

[0135] In some embodiments of the present application, the process of identifying the accumulation-type photovoltaic power abnormal area in the solar irradiance and photovoltaic power scatter diagram by using the image recognition algorithm in step S5 is introduced, which may specifically include:

[0136] Step S51: converting the solar irradiance and photovoltaic power scatter plot into a grayscale image, and performing binarization processing on the grayscale image.

[0137] Specifically, the color scatter plot is converted into a grayscale image. A single-channel grayscale image is obtained by taking a weighted average of the pixel values ​​of the red, green, and blue channels (for example, the green channel has a higher weight, reflecting the human eye's greater sensitivity to green). This grayscale image is then binarized. By setting an appropriate threshold (for example, using the Otsu algorithm to automatically determine the globally optimal threshold), the pixels in the image are divided into two categories: points above the threshold represent data points (set to white), and points below the threshold represent background (set to black). This process converts the continuous scatter plot into a distinct black-and-white binary image, highlighting the distribution of the data points and facilitating subsequent edge detection.

[0138] The grayscale image conversion formula is as follows:

[0139]

[0140] R, G, and B are the pixel values ​​for the red, green, and blue channels, respectively. OpenCV uses BGR sorting when reading images. The weights in this formula are derived from the differences in the human eye's sensitivity to different colors, with green contributing the most to brightness and blue the least.

[0141] Step S52: perform edge detection on the binary image to extract the scattered point distribution outline.

[0142] Specifically, an edge detection algorithm is applied to the binarized image. First, a Gaussian filter is used to smooth the image to reduce noise. Then, the gradient strength and direction of each pixel are calculated. Non-maximum suppression is used to refine the edges, retaining only the points with local gradient maxima. Finally, a double-thresholding process is used to connect broken edges. Points with gradient strengths above the high threshold are considered strong edges, while points between the high and low thresholds and connected to strong edges are considered weak edges. This extracts a complete outline of the scattered point distribution, forming a closed boundary curve.

[0143] Perform edge detection on the binary image and extract the scattered point distribution outline, including:

[0144] ① Using a Gaussian filter to smooth the binary image and output a noise-reduced image;

[0145] ② Applying the Sobel operator to perform convolution operation on the denoised image to generate the gradient magnitude and gradient direction of each pixel;

[0146] ③ performing non-maximum suppression based on the gradient magnitude and the gradient direction, and outputting a refined edge response map;

[0147] ④ Performing double threshold connection processing on the refined edge response map, screening strong edge pixels and connecting them to form a candidate edge set;

[0148] ⑤ Based on the candidate edge set, a continuous contour is generated by a contour fitting algorithm as the scattered point distribution contour.

[0149] Specifically, to reduce the interference of salt and pepper noise or isolated pixels that may exist in the binary image on edge detection, a Gaussian filter is used for smoothing. The Gaussian filter is centered on the pixel and assigns weights based on a two-dimensional Gaussian function, performing a weighted average of neighboring pixels. By adjusting the standard deviation of the Gaussian kernel (e.g., σ = 1.0), noise is effectively suppressed while retaining edge details, generating a denoised image and making subsequent edge detection results more consistent and accurate. The mathematical expression of the Gaussian filter is as follows:

[0150]

[0151] in, is the standard deviation, are the coordinates relative to the filter center.

[0152] The Sobel operator is applied to the horizontal and vertical directions, and the gradient of each pixel is calculated through convolution. The horizontal operator (Gx) detects vertical edges, and the vertical operator (Gy) detects horizontal edges. For each pixel, the gradient magnitude and gradient direction are calculated as follows:

[0153]

[0154]

[0155] Among them, G is the gradient amplitude of the pixel point, 、 They are the horizontal gradient component obtained by horizontal convolution of the image by the Sobel operator, and the vertical gradient component obtained by vertical convolution of the image, is the gradient direction of the pixel. The gradient magnitude indicates the strength of the edge, and the gradient direction indicates the direction of the edge. The generated gradient magnitude map and direction map provide the basis for subsequent edge refinement.

[0156] To eliminate edge blurring, only the points with the maximum gradient magnitude along the local gradient direction are retained. For each pixel, the two adjacent pixels are checked along its gradient direction. If the gradient magnitude of the current pixel is not greater than that of the adjacent pixel, its gradient magnitude is set to 0; otherwise, its value is retained. This process refines wide edges to single-pixel widths, generating a refined edge response map that highlights the true edge locations.

[0157] Two thresholds are set: pixels with gradient magnitudes greater than the upper threshold are marked as strong edges, pixels between the upper and lower thresholds are marked as weak edges, and pixels with gradient magnitudes less than the lower threshold are suppressed as non-edges. A tracking algorithm connects strong and weak edges. If a weak edge is connected to a strong edge, it is retained as a true edge; otherwise, it is suppressed. This process effectively retains true edges while reducing false edges, forming a set of candidate edges.

[0158] A contour tracing algorithm is used to generate a continuous closed contour for the selected candidate edge pixel set. Starting from the edge starting point, the algorithm traces adjacent edge pixels in a clockwise or counterclockwise direction until a closed curve is formed. The generated contour is smoothed, removing redundant points while retaining key feature points. Ultimately, a continuous contour representing the scattered point distribution is generated, providing accurate boundary data for subsequent ellipse fitting.

[0159] Step S53: performing ellipse fitting on the scatter point distribution profile to generate an ellipse equation representing the boundary of normal data distribution.

[0160] Specifically, the least squares method is used to fit an ellipse to the extracted scattered point distribution profile. The geometric parameters of the ellipse (center coordinates, major axis length, minor axis length, and rotation angle) are determined by calculating the sum of the squared distances from all points on the profile to the ellipse and minimizing the error function. The ellipse equation characterizes the distribution boundary of the normal data points, and its shape reflects the inherent correlation between solar irradiance and photovoltaic power (such as the nonlinear growth trend of power with increasing irradiance).

[0161] Step S54: determining the area outside the ellipse as the accumulation-type photovoltaic power abnormality area according to the ellipse equation.

[0162] Specifically, based on the ellipse equation obtained by fitting, a coordinate transformation relationship is established to map the original scatter plot to the ellipse coordinate system. For each data point in the scatter plot, its position in the ellipse coordinate system is calculated. If the point is outside the ellipse, it is determined to be an abnormal point. These abnormal points outside the ellipse are usually clustered to form a specific area, which may be caused by local failure of the photovoltaic module (such as failure of some panels), measurement equipment deviation or environmental interference (such as local shadows). By marking the area outside the ellipse as a pile-up abnormal area and extracting the normal data points within the ellipse, the original physical dimension of the data is finally restored through the denormalization operation, and a pure final data set is output.

[0163] A photovoltaic abnormal data processing device provided in an embodiment of the present application is described below. The photovoltaic abnormal data processing device described below and the photovoltaic abnormal data processing method described above can refer to each other.

[0164] See also Figure 4 , Figure 4This is a schematic diagram of a photovoltaic abnormal data processing device disclosed in an embodiment of the present application.

[0165] like Figure 4 As shown, the photovoltaic abnormal data processing device may include:

[0166] Photovoltaic data collection unit 110, used to collect historical total power data and matching historical solar irradiance data of the target photovoltaic station;

[0167] An obvious anomaly processing unit 120 is used to perform negative value zeroing and obvious anomaly elimination processing on the historical total power data and the historical solar irradiance data to generate a first data set including solar irradiance and photovoltaic power;

[0168] a discrete anomaly processing unit 130 for identifying discrete photovoltaic power anomaly points in the first data set by using a local anomaly factor algorithm, and obtaining a second data set by marking and removing the discrete photovoltaic power anomaly points;

[0169] An image processing and construction unit 140 is configured to perform image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power;

[0170] The accumulation anomaly processing unit 150 is used to identify the accumulation-type photovoltaic power anomaly area in the solar irradiance and photovoltaic power scatter diagram through an image recognition algorithm, mark the normal data outside the accumulation-type photovoltaic power anomaly area through coordinate transformation, and output the final data set after denormalization.

[0171] It can be seen from the above technical solutions that the photovoltaic abnormal data processing method and related equipment provided by the embodiment of the present application first collect the historical total power data and historical solar irradiance data of the target photovoltaic station, perform negative value zeroing and obvious abnormality elimination processing on them, and generate a first data set, which effectively eliminates the negative values ​​or obvious erroneous data generated by factors such as sensor failure and transmission abnormality during actual collection. The first data set is processed using a local anomaly factor algorithm. The algorithm quantifies the degree of abnormality of each data point by calculating the density deviation between the data point and its neighborhood points. Discrete anomalies are often isolated data generated by instantaneous sensor failure, communication interruption, etc. The local anomaly factor algorithm can accurately identify discrete photovoltaic power anomalies. By marking and eliminating them, a second data set is obtained, which effectively eliminates the interference of discrete data on the overall data distribution, solves the problem of discrete data affecting image recognition morphological judgment in the prior art, and enables subsequent image recognition to focus more on the true distribution characteristics of the data.

[0172] Image preprocessing is performed on the second dataset to generate a scatter plot of solar irradiance and photovoltaic power. The scatter plot is then analyzed using an image recognition algorithm. This algorithm utilizes edge detection and morphological analysis techniques to identify areas of accumulated photovoltaic power anomalies. Accumulated anomalous data is often caused by sudden changes in system parameters or interference from multiple factors, and manifests itself in the scatter plot as a localized cluster of abnormal data. Normal data outside the abnormal area is marked through coordinate transformation, and the final dataset is output through denormalization. This accurately separates the normal data, overcoming the shortcomings of traditional methods in identifying accumulated anomalous data.

[0173] This application integrates a local anomaly factor algorithm with an image recognition algorithm, avoiding the limitations of a single algorithm. The local anomaly factor algorithm adaptively identifies discrete outliers without requiring complex pre-set parameters; the image recognition algorithm is based on intuitive data morphological feature analysis. The two work together to reduce the problem of inadvertent deletion of normal data due to improper parameter settings, significantly improving the accuracy and reliability of anomaly data detection.

[0174] Optionally, identifying discrete photovoltaic power anomaly points in the first data set by using a local anomaly factor algorithm includes:

[0175] performing normalization processing on the first data set;

[0176] Calculating the Euclidean distance between all data points in the first data set, and determining the nearest neighbor distance of each data point based on a preset number of nearest neighbors;

[0177] Calculate the reachable distance of each data point according to the neighborhood distance of the nearest neighbor number;

[0178] Calculate the local reachability density of each data point based on the reachability distance;

[0179] A local anomaly factor is calculated according to the local reachable density, and a data point where the local anomaly factor is greater than a preset threshold is marked as a discrete photovoltaic power anomaly point.

[0180] Optionally, the calculation formula for the reachable distance is:

[0181]

[0182] The calculation formula of the local reachable density is:

[0183]

[0184] The calculation formula of the local anomaly factor is:

[0185]

[0186] in, For data points ( , ) to data point ( , )’s reachable distance, is the solar irradiance at time i, is the photovoltaic power at time i, is the neighborhood distance between the data at time j and the kth neighbor data, For data points ( , ) to data point ( , ), For data points ( , ), is the neighbor dataset, For data points ( , )’s local anomaly factor.

[0187] Optionally, performing image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power includes:

[0188] Normalizing the solar irradiance and photovoltaic power values ​​in the second data set independently so that their numerical ranges are mapped to the interval [0, 1];

[0189] A scatter plot of solar irradiance and photovoltaic power is generated by plotting the normalized solar irradiance as the horizontal axis and the photovoltaic power as the vertical axis.

[0190] Optionally, identifying an accumulation-type photovoltaic power anomaly region in the solar irradiance and photovoltaic power scatter plot by an image recognition algorithm includes:

[0191] Converting the solar irradiance and photovoltaic power scatter plot into a grayscale image, and performing binarization processing on the grayscale image;

[0192] Perform edge detection on the binary image and extract the scattered point distribution outline;

[0193] Performing ellipse fitting on the scatter point distribution profile to generate an ellipse equation representing a normal data distribution boundary;

[0194] The area outside the ellipse is determined as a pile-up type photovoltaic power abnormality area according to the ellipse equation.

[0195] Optionally, performing edge detection on the binarized image to extract the scattered point distribution outline includes:

[0196] Using a Gaussian filter to smooth the binarized image and output a noise-reduced image;

[0197] Applying a Sobel operator to perform a convolution operation on the denoised image to generate a gradient magnitude and gradient direction for each pixel;

[0198] Performing non-maximum suppression based on the gradient magnitude and the gradient direction, and outputting a refined edge response map;

[0199] Performing a double-threshold connection process on the refined edge response map to screen strong edge pixels and connect them to form a candidate edge set;

[0200] A continuous contour is generated based on the candidate edge set by a contour fitting algorithm as the scattered point distribution contour.

[0201] Optionally, the calculation formula for the gradient magnitude and gradient direction of each pixel is:

[0202]

[0203]

[0204] Among them, G is the gradient amplitude of the pixel point, 、 They are the horizontal gradient component obtained by horizontal convolution of the image by the Sobel operator, and the vertical gradient component obtained by vertical convolution of the image, is the gradient direction of the pixel.

[0205] The photovoltaic abnormal data processing device provided in the embodiment of the present application can be applied to photovoltaic abnormal data processing equipment. Figure 5 The hardware structure diagram of the photovoltaic abnormal data processing equipment is shown. Figure 5 ,The hardware structure of the photovoltaic abnormal data processing device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

[0206] In the embodiment of the present application, the number of the processor 1, the communication interface 2, the memory 3, and the communication bus 4 is at least one, and the processor 1, the communication interface 2, and the memory 3 communicate with each other through the communication bus 4;

[0207] The processor 1 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention;

[0208] The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory;

[0209] The memory stores a program, and the processor can call the program stored in the memory, wherein the program is used to:

[0210] Collect historical total power data and matching historical solar irradiance data of target photovoltaic stations;

[0211] Performing negative value zeroing and obvious anomaly elimination processing on the historical total power data and the historical solar irradiance data to generate a first data set including solar irradiance and photovoltaic power;

[0212] identifying discrete photovoltaic power anomaly points in the first data set by using a local anomaly factor algorithm, and obtaining a second data set by marking and removing the discrete photovoltaic power anomaly points;

[0213] Performing image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power;

[0214] The accumulation-type photovoltaic power anomaly area in the solar irradiance and photovoltaic power scatter plot is identified by an image recognition algorithm, normal data outside the accumulation-type photovoltaic power anomaly area is marked by coordinate transformation, and the final data set is output after denormalization.

[0215] Optionally, the refined functions and extended functions of the program may refer to the above description.

[0216] The present application also provides a readable storage medium, which may store a program suitable for execution by a processor, wherein the program is used to:

[0217] Collect historical total power data and matching historical solar irradiance data of target photovoltaic stations;

[0218] Performing negative value zeroing and obvious anomaly elimination processing on the historical total power data and the historical solar irradiance data to generate a first data set including solar irradiance and photovoltaic power;

[0219] identifying discrete photovoltaic power anomaly points in the first data set by using a local anomaly factor algorithm, and obtaining a second data set by marking and removing the discrete photovoltaic power anomaly points;

[0220] Performing image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power;

[0221] The accumulation-type photovoltaic power anomaly area in the solar irradiance and photovoltaic power scatter plot is identified by an image recognition algorithm, normal data outside the accumulation-type photovoltaic power anomaly area is marked by coordinate transformation, and the final data set is output after denormalization.

[0222] Optionally, the refined functions and extended functions of the program may refer to the above description.

[0223] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0224] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0225] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A photovoltaic abnormal data processing method, characterized in that: include: Collect historical total power data and matching historical solar irradiance data of target photovoltaic stations; Performing negative value zeroing and obvious anomaly elimination processing on the historical total power data and the historical solar irradiance data to generate a first data set including solar irradiance and photovoltaic power; identifying discrete photovoltaic power anomaly points in the first data set by using a local anomaly factor algorithm, and obtaining a second data set by marking and removing the discrete photovoltaic power anomaly points; Performing image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power; The accumulation-type photovoltaic power anomaly area in the solar irradiance and photovoltaic power scatter plot is identified by an image recognition algorithm, normal data outside the accumulation-type photovoltaic power anomaly area is marked by coordinate transformation, and the final data set is output after denormalization.

2. The method according to claim 1, characterized in that Identifying discrete photovoltaic power anomaly points in the first data set using a local anomaly factor algorithm includes: performing normalization processing on the first data set; Calculating the Euclidean distance between all data points in the first data set, and determining the nearest neighbor distance of each data point based on a preset number of nearest neighbors; Calculate the reachable distance of each data point according to the neighborhood distance of the nearest neighbor number; Calculate the local reachability density of each data point based on the reachability distance; A local anomaly factor is calculated according to the local reachable density, and a data point where the local anomaly factor is greater than a preset threshold is marked as a discrete photovoltaic power anomaly point.

3. The method according to claim 2, characterized in that The calculation formula of the reachable distance is: The calculation formula of the local reachable density is: The calculation formula of the local anomaly factor is: in, For data points ( , ) to data point ( , )’s reachable distance, is the solar irradiance at time i, is the photovoltaic power at time i, is the neighborhood distance between the data at time j and the kth neighbor data, For data points ( , ) to data point ( , ), For data points ( , ), is the neighbor dataset, For data points ( , )’s local anomaly factor.

4. The method according to claim 1, wherein Performing image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power includes: Normalizing the solar irradiance and photovoltaic power values ​​in the second data set independently so that their numerical ranges are mapped to the interval [0, 1]; A scatter plot of solar irradiance and photovoltaic power is generated by plotting the normalized solar irradiance as the horizontal axis and the photovoltaic power as the vertical axis.

5. The method according to claim 1, wherein Identifying an accumulation-type photovoltaic power anomaly region in the solar irradiance and photovoltaic power scatter plot using an image recognition algorithm includes: Converting the solar irradiance and photovoltaic power scatter plot into a grayscale image, and performing binarization processing on the grayscale image; Perform edge detection on the binary image and extract the scattered point distribution outline; Performing ellipse fitting on the scatter point distribution profile to generate an ellipse equation representing a normal data distribution boundary; The area outside the ellipse is determined as a pile-up type photovoltaic power abnormality area according to the ellipse equation.

6. The method according to claim 5, characterized in that The performing edge detection on the binary image and extracting the scattered point distribution outline includes: Using a Gaussian filter to smooth the binarized image and output a noise-reduced image; Applying a Sobel operator to perform a convolution operation on the denoised image to generate a gradient magnitude and gradient direction for each pixel; Performing non-maximum suppression based on the gradient magnitude and the gradient direction, and outputting a refined edge response map; Performing a double-threshold connection process on the refined edge response map to screen strong edge pixels and connect them to form a candidate edge set; A continuous contour is generated based on the candidate edge set by a contour fitting algorithm as the scattered point distribution contour.

7. The method according to claim 6, characterized in that The calculation formula for the gradient amplitude and gradient direction of each pixel is: Among them, G is the gradient amplitude of the pixel point, 、 They are the horizontal gradient component obtained by horizontal convolution of the image by the Sobel operator, and the vertical gradient component obtained by vertical convolution of the image, is the gradient direction of the pixel.

8. A photovoltaic abnormal data processing device, characterized in that: include: Photovoltaic data acquisition unit, used to collect historical total power data of the target photovoltaic station and the matching historical solar irradiance data; an obvious anomaly processing unit, configured to perform negative value zeroing and obvious anomaly elimination processing on the historical total power data and the historical solar irradiance data, to generate a first data set including solar irradiance and photovoltaic power; a discrete anomaly processing unit, configured to identify discrete photovoltaic power anomaly points in the first data set by using a local anomaly factor algorithm, and obtain a second data set by marking and removing the discrete photovoltaic power anomaly points; an image processing construction unit, configured to perform image preprocessing based on the second data set to generate a scatter plot of solar irradiance and photovoltaic power; The accumulation anomaly processing unit is used to identify the accumulation-type photovoltaic power anomaly area in the solar irradiance and photovoltaic power scatter diagram through an image recognition algorithm, mark the normal data outside the accumulation-type photovoltaic power anomaly area through coordinate transformation, and output the final data set after denormalization.

9. A photovoltaic abnormal data processing device, characterized in that: including memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement each step of the photovoltaic abnormal data processing method according to any one of claims 1 to 7.

10. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the photovoltaic abnormal data processing method according to any one of claims 1 to 7 is implemented.

Images