A Dynamic Monitoring Method for Vegetation Coverage in Photovoltaic Areas Based on Artificial Intelligence

Abstract

This invention discloses an artificial intelligence-based method for dynamic monitoring of vegetation cover in photovoltaic areas, relating to the field of dynamic monitoring technology. The method includes the following steps: integrating geometric information of photovoltaic modules, on-site geographic data, and support height parameters to construct a three-dimensional model; obtaining the solar angle using a solar path algorithm, calculating the shadow boundary using the three-dimensional model, and aggregating this into a daily dynamic shading map; preprocessing remote sensing images of the power station, annotating the shading map, and performing spectral correction to obtain a corrected remote sensing image sequence; using environmental data and the corrected images, outputting the vegetation cover using a vegetation cover model; and building a prediction model based on the vegetation cover, inputting future environmental forecasts, and predicting trends to provide early warning of vegetation anomalies. This invention, through daily dynamic shading maps, solves the monitoring error problem caused by the shadow interference of photovoltaic modules when traditional remote sensing image analysis methods are applied to photovoltaic areas.

Description

Technical Field

[0001] This invention relates to the field of dynamic monitoring engineering, specifically to a method for dynamic monitoring of vegetation coverage in photovoltaic areas based on artificial intelligence. Background Technology

[0002] With the rapid development of the global renewable energy industry, photovoltaic (PV) power generation, as a clean and efficient energy source, faces challenges in ecological and environmental protection during the large-scale construction of PV power plants. Especially in PV power plant areas, vegetation cover not only affects ecosystem stability but also directly impacts the heat dissipation performance and power generation efficiency of the PV system. Therefore, accurately monitoring and assessing vegetation cover in PV areas has become a key technological requirement for achieving ecological PV integration and sustainable development.

[0003] Traditional methods for monitoring vegetation cover primarily rely on remote sensing image analysis. This involves selecting appropriate remote sensing satellites, periodically acquiring multispectral image data of the Earth's surface, and then analyzing this data to determine vegetation cover. However, when using remote sensing image analysis in photovoltaic (PV) areas, the structural characteristics and shading effects of the PV modules significantly interfere with the vegetation spectral signals in the images. This results in substantial errors when applying remote sensing image analysis in PV areas, severely limiting the reliability and practical value of the monitoring data. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based method for dynamic monitoring of vegetation coverage in photovoltaic areas, which solves the problem that the structural characteristics and shadowing effects of photovoltaic modules themselves cause significant interference to vegetation spectral signals in remote sensing images.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic monitoring of vegetation cover in photovoltaic areas based on artificial intelligence, comprising the following steps:

[0006] Step S1: By analyzing the photovoltaic power station design drawings, obtain the geometric information of the photovoltaic modules, collect the geographical information of the photovoltaic power station and the height parameters of the photovoltaic panel support, and combine the geometric information of the photovoltaic modules and the height parameters of the photovoltaic panel support to construct a three-dimensional spatial structure model of the photovoltaic modules;

[0007] Step S2: Calculate the geographical information of the photovoltaic power station on site using the solar path algorithm to obtain the azimuth and elevation angle of the sun. Based on the azimuth and elevation angle of the sun and the three-dimensional spatial structure model of the photovoltaic module, calculate the shadow boundary cast by the photovoltaic module on the ground. By aggregating the shadow boundaries of each photovoltaic module, obtain a daily dynamic shading time map.

[0008] Step S3: Acquire and preprocess images of the photovoltaic power station area to obtain a processed remote sensing image sequence. Match and label the processed remote sensing image sequence with the daily dynamic shading time map to generate a shadow-labeled image sequence. Based on the surface reflectance benchmark library, compensate and correct the shadow-labeled image sequence with a spectral correction algorithm to obtain a corrected remote sensing image sequence.

[0009] Step S4: Collect environmental data of the photovoltaic power station and construct a vegetation coverage learning model. Input the environmental data of the photovoltaic power station and the corrected remote sensing image sequence into the vegetation coverage learning model, and output the vegetation coverage of the photovoltaic power station.

[0010] Step S5: Based on the vegetation coverage of the photovoltaic power station, construct a vegetation coverage change trend prediction model, collect environmental forecast data for the next few days, input the data into the vegetation coverage change trend prediction model, and output the predicted change trend of vegetation coverage. Through the predicted change trend of vegetation coverage, provide early warning of abnormal vegetation growth in the photovoltaic area.

[0011] Preferably, the geometric information of the photovoltaic modules is obtained by analyzing the design drawings of the photovoltaic power station, and the geographical information of the photovoltaic power station and the height parameters of the photovoltaic panel support are collected, including:

[0012] Obtain geometric information on the arrangement, shape, and size of photovoltaic modules from photovoltaic power plant design drawings;

[0013] Use GPS measuring equipment to obtain the longitude, latitude, and altitude geographic coordinates of the power station's location;

[0014] The height H of the photovoltaic panel support and the tilt angle of the photovoltaic panels were determined through on-site measurements. .

[0015] Preferably, constructing a three-dimensional spatial structure model of the photovoltaic module by combining the geometric information of the photovoltaic module and the height parameters of the photovoltaic panel support includes:

[0016] A three-dimensional Cartesian coordinate system is established with the location of the photovoltaic power station as the origin, the x-axis pointing east, the y-axis pointing north, and the z-axis pointing vertically upward;

[0017] Based on the arrangement in the design drawings, calculate the position of each photovoltaic panel in three-dimensional space. For the photovoltaic panel in the i-th row and j-th column, the coordinate of its center in the x-direction can be expressed as:

[0018]

[0019] in, Indicates the first Line 1 Photovoltaic panel center The coordinates of the direction, where i is the row number of the photovoltaic panel. It is the column number where the photovoltaic panel is located. It is the width of the photovoltaic panel;

[0020] The coordinates in the y-direction are:

[0021]

[0022] in, is the coordinate of the center of the photovoltaic panel in the i-th row and j-th column in the y-direction, and L is the length of the photovoltaic panel;

[0023] The coordinate in the z-direction is the support height H;

[0024] Considering that the photovoltaic panels need to receive sunlight, according to the laws of physics, let's assume that the photovoltaic panels are tilted in the north-south direction:

[0025] normal vector of photovoltaic panel Direction in three-dimensional space can be determined by the angle of inclination. Sure:

[0026]

[0027] in, It is the first Line 1 Normal vector of photovoltaic panel ;

[0028] Based on a three-dimensional Cartesian coordinate system, the center of each photovoltaic panel and the normal vector of the photovoltaic panel are used to construct a three-dimensional spatial structure model of the photovoltaic module using 3ds Max three-dimensional modeling software.

[0029] Preferably, the solar path algorithm is used to calculate the geographical information of the photovoltaic power station on-site to obtain the azimuth and elevation angles of the sun, including:

[0030] Calculate the solar azimuth angle The formula is:

[0031]

[0032] The solar altitude angle is calculated based on the local latitude, solar declination angle, and solar hour angle:

[0033]

[0034] in, It is the solar altitude angle. It is the latitude of the photovoltaic power station.

[0035] Preferably, by aggregating the shadow boundaries of each photovoltaic module, a daily dynamic shading time map is obtained, including:

[0036] Iterate through all photovoltaic modules, repeating the above shadow boundary calculation steps for all photovoltaic modules within the photovoltaic power station; assuming there are a total of Each photovoltaic panel, in turn, is processed. Processing is performed to obtain the result at each time step. The shadow boundary polygon below ;

[0037] Ground area gridding: Dividing the ground area where the photovoltaic power station is located into a regular grid, for example, setting the side length of the grid to be... Each grid can be determined by the coordinates of its bottom-left vertex. This is used to identify the entire ground area, which is then discretized into a series of grid cells.

[0038] For each time step Traverse all grid cells Determine whether each grid cell is covered by the shadow of a photovoltaic panel. If a grid cell is covered by the shadow of u photovoltaic panels, then... Simultaneous coverage means that the grid cell is in time The shadow coverage intensity is u;

[0039] In grid cell coordinates As a row index, time step Use it as a column index to construct a two-dimensional matrix. Matrix elements Indicates the location and time The shadow coverage intensity u at time is calculated. Through the above shadow coverage intensity calculation process, the shadow coverage intensity value of each grid cell at different time steps is filled into the matrix, and finally the daily dynamic shading time map T is obtained.

[0040] Preferably, images of the photovoltaic power station area are acquired and preprocessed to obtain a processed remote sensing image sequence. The processed remote sensing image sequence is then matched and labeled using the daily dynamic shading time map to generate a shaded image sequence, including:

[0041] Multi-temporal high-resolution remote sensing images of the photovoltaic power station area were collected, with the number of images collected being [number missing]. The shooting time corresponding to image I is Latitude and longitude coordinates are ;

[0042] These images cover the visible and near-infrared bands, and the spatial resolution of the images is [missing information]. ;

[0043] The acquired images are arranged in chronological order to form a time-series data sequence. ,in Indicates the first A remote sensing image;

[0044] The time series data sequence is preprocessed as follows:

[0045] First, time-series image registration is performed, selecting an image taken close to noon as the reference image, because at this time the sun's altitude angle is large, the illumination is uniform, and there are few shadows;

[0046] For other images Feature point extraction and matching are performed using the Scale Invariant Feature Transform (SIFT) algorithm, assuming the image... The coordinates of the feature points on are In the reference image The coordinates of the feature points matched above are ;

[0047] Through affine transformation matrix , making After transformation Then get as close as possible affine transformation matrix The usual form is The transformation formula is:

[0048]

[0049] in, , , , , , These are elements in the affine transformation matrix. These are the transformed coordinates; time-series image registration is performed on each time-series data sequence to obtain the processed remote sensing image sequence.

[0050] After obtaining the processed remote sensing image sequence, the images are matched and labeled according to the daily dynamic shading time map:

[0051] Shaded areas are marked for each pre-processed remote sensing image. According to its shooting time In the dynamic shading time map Find the shadow distribution data for the corresponding time moment;

[0052] Image The pixel coordinates are ,like Then the image The median coordinate is The pixels are labeled as shadow pixels, and a shadow-labeled image sequence is generated. .

[0053] Preferably, the shaded image sequence is compensated and corrected using a spectral correction algorithm to obtain a corrected remote sensing image sequence, including:

[0054] Based on the surface reflectance benchmark database The database contains standard spectral curves of common vegetation types in photovoltaic areas. Let the vegetation type be... The standard reflectance in band c is ;

[0055] The spectral correction algorithm compensates and corrects the shaded image sequence to obtain the corrected remote sensing image sequence, as follows:

[0056] For shaded image sequences Shadow pixels in Set shadow pixels In the band The original reflectance is Perform band-specific design;

[0057] Three key nonlinear interference factors are introduced to quantify the effects of shadow, light, and vegetation type as follows:

[0058] Shadow intensity factor Quantization grid cells Shadow coverage intensity, single shadow Double shadow Multiple shadows ;

[0059] Illumination angle factor That is, the solar altitude angle at that moment. Standardization ;

[0060] Vegetation type factors Values ​​were assigned to herbaceous plants based on a surface reflectance benchmark library. Shrub plant value assignment ;

[0061] In the visible light band, attenuation is significantly affected by shadow intensity and vegetation type. and The interactive items are defined by the following formula:

[0062]

[0063] in, It is the basic correction factor; It is the shadow intensity correction factor; It is the vegetation type correction coefficient; It is the illumination angle compensation coefficient;

[0064] In the near-infrared band, attenuation is more significantly affected by shadow intensity and illumination angle. quadratic terms and sum The exponent term is given by the following formula:

[0065]

[0066] in, It is the near-infrared baseline correction factor; It is the coefficient of the quadratic term of shadow intensity; It is the illumination angle index compensation coefficient; It is the vegetation type compensation coefficient;

[0067] The above corrections are applied to all shadow pixels to obtain the corrected remote sensing image sequence. Let the corrected remote sensing image sequence be... ,in Indicates time, For time The acquired corrected remote sensing images provide accurate data for subsequent identification of vegetation growth dynamics.

[0068] Preferably, step S4 includes:

[0069] The environmental data collected from the photovoltaic power plant are as follows:

[0070] Meteorological data: A weather station is installed inside the photovoltaic power station, and the temperature is recorded as follows: Rain gauges are installed within photovoltaic power stations to measure rainfall. The rain gauges determine the rainfall amount by collecting and measuring the amount of rainwater falling into them over a specific period. The rainfall amount is recorded as follows: ;

[0071] Soil data: Soil moisture sensors are buried in the soil of the photovoltaic power station, and the soil moisture is recorded as follows: ;

[0072] To ensure consistency of dimensions, the above environmental data were normalized.

[0073] Construction of a vegetation cover learning model:

[0074] Convolutional Neural Network Module: Used to extract spatial features of vegetation in images, for each image Features are extracted through convolutional layers and pooling layers;

[0075] Long Short-Term Memory (LSTM) network module: used to capture the growth and change characteristics of vegetation over time, storing spatial features at different times τ. Input the data sequentially into the LSTM to compute the temporal features;

[0076] Fusion layer: Combines environmental factors to output vegetation cover and temporal characteristics. Environmental data , , By splicing, the fusion features are obtained. ;

[0077] Calculate vegetation cover using fully connected (Dense) layers:

[0078]

[0079] in, It is time vegetation coverage; These are the weights of the fully connected layer. It is a bias; It's the sigmoid function, and its output value range is... .

[0080] Preferably, the predicted trends in vegetation cover include:

[0081] By substituting future environmental data into the vegetation cover change trend prediction model, the predicted value of vegetation cover for the next few days can be calculated. Prediction for a certain day in the future:

[0082]

[0083] in, Indicates time The vegetation cover at that time, here Represents a time increment, that is, from time... Push forward Vegetation cover after a period of time; Indicates time The temperature at that time; Indicates time Rainfall at that time; Indicates time Soil moisture at that time; , , , It is a regression coefficient, used to measure the degree of influence of various environmental factors on vegetation cover; This is the error term, representing the deviation between the model's predicted value and the actual value.

[0084] Preferably, the early warning of abnormal vegetation growth in the photovoltaic area based on the predicted trend of vegetation cover includes:

[0085] Based on statistical analysis of historical vegetation cover data, the normal fluctuation range of vegetation cover is determined. and the threshold of change If the vegetation cover changes by more than [a certain amount] in a short period of time Or, if it deviates from the normal range for several consecutive days. If so, it is determined to be abnormal vegetation growth;

[0086] When abnormal vegetation growth is detected, the possible causes of the abnormality are analyzed. Combined with environmental data from the same period, it is determined whether it is caused by drought, flood, extreme temperature factors, or possible biological factors such as pests and diseases.

[0087] Based on the type and degree of abnormal vegetation growth, corresponding ecological regulation suggestions are proposed, and the ecological regulation decisions are as follows:

[0088] Short-term control measures are formulated to address sudden abnormal vegetation growth, and immediate short-term control plans are developed, such as emergency watering and temporary shading measures, to alleviate the pressure on vegetation growth.

[0089] Long-term regulation strategies should be developed from a long-term perspective, including vegetation management and ecological environment improvement.

[0090] This invention provides a method for dynamic monitoring of vegetation cover in photovoltaic areas based on artificial intelligence, involving machine learning and deep learning technologies, which has the following beneficial effects:

[0091] (1) The dynamic monitoring method for vegetation coverage in photovoltaic areas based on artificial intelligence accurately calculates the shadow boundary of the photovoltaic module on the ground by constructing a three-dimensional spatial structure model of the photovoltaic module and generating a daily dynamic shading time map, which effectively solves the shadow interference problem, significantly improves the accuracy and reliability of vegetation coverage monitoring, and provides high-quality data support for subsequent ecological assessment and management.

[0092] (2) This AI-based dynamic monitoring method for vegetation coverage in photovoltaic areas uses a deep learning model combining convolutional neural networks and long short-term memory networks to fuse and analyze these multi-dimensional data. It considers not only the spatial characteristics of vegetation but also captures the growth and change characteristics of vegetation over time. This multi-dimensional data fusion approach makes the assessment of vegetation coverage more comprehensive and accurate, better reflecting the actual growth status of vegetation in photovoltaic areas and providing a more reliable basis for ecological management and decision-making.

[0093] (3) A dynamic monitoring method for vegetation coverage in photovoltaic areas based on artificial intelligence can predict the trend of vegetation coverage changes by constructing a prediction model based on historical data and combining it with future environmental forecast data. By setting normal fluctuation ranges and thresholds, abnormal vegetation growth can be identified in a timely manner, and targeted short-term control measures and long-term control strategies can be proposed, providing a basis for decision-making for the maintenance of ecological balance and sustainable development in photovoltaic areas. Attached Figure Description

[0094] Figure 1 This is a flowchart of a dynamic monitoring method for vegetation coverage in photovoltaic areas based on artificial intelligence, as proposed in this invention.

[0095] Figure 2 The hierarchical map of the corrected remote sensing image sequence is obtained by the dynamic monitoring method of vegetation coverage in photovoltaic areas based on artificial intelligence proposed in this invention.

[0096] Figure 3 This is a hierarchical diagram showing the predicted trend of vegetation coverage changes in a photovoltaic area dynamic monitoring method based on artificial intelligence proposed in this invention. Detailed Implementation

[0097] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0098] Please see Figure 1 This invention provides a technical solution: a method for dynamic monitoring of vegetation cover in photovoltaic areas based on artificial intelligence. Specifically, the method is described below. Please refer to [link to relevant documentation]. Figure 1 The method includes the following steps:

[0099] Step S1: By analyzing the photovoltaic power station design drawings, obtain the geometric information of the photovoltaic modules, collect the geographical information of the photovoltaic power station and the height parameters of the photovoltaic panel support, and combine the geometric information of the photovoltaic modules and the height parameters of the photovoltaic panel support to construct a three-dimensional spatial structure model of the photovoltaic modules.

[0100] Obtain geometric information such as the arrangement (e.g., matrix arrangement, determining the number of rows and columns), shape (usually rectangular, obtaining length and width dimensions), and dimensions of photovoltaic modules from photovoltaic power plant design drawings.

[0101] On-site geographic information: GPS surveying equipment was used to obtain the geographic coordinates of the power plant's location, including longitude, latitude, and altitude. This data is crucial for determining the position of the sun relative to the photovoltaic panels.

[0102] The height H of the photovoltaic panel support and the tilt angle of the photovoltaic panels were determined through on-site measurements. .

[0103] Select a reference point on the ground. Measure the elevation angle α from the reference point to the top of the photovoltaic panel support, and the horizontal distance L from the reference point to the projection point of the bottom of the photovoltaic panel support, using a theodolite or total station. Based on trigonometric relationships, the support height is: H. tanα.

[0104] Using an electronic inclinometer, place the inclinometer directly on the surface of the photovoltaic panel, ensuring it is stable. The inclinometer will directly display the tilt angle of the photovoltaic panel relative to the horizontal plane. .

[0105] The height of the support structure determines the vertical distance between the photovoltaic panel and the ground, while the tilt angle affects the direction of sunlight reception and shadow projection of the photovoltaic panel.

[0106] The model is constructed as follows:

[0107] Coordinate system establishment: A three-dimensional Cartesian coordinate system is established with the location of the photovoltaic power station as the origin. The x-axis points east, the y-axis points north, and the z-axis points vertically upward.

[0108] Based on the arrangement in the design drawings, calculate the position of each photovoltaic panel in three-dimensional space. For the photovoltaic panel in the i-th row and j-th column, the coordinate of its center in the x-direction is represented as:

[0109]

[0110] in, Indicates the first Line 1 Photovoltaic panel center The coordinates of the direction, where i is the row number of the photovoltaic panel. It is the column number where the photovoltaic panel is located. That is the width of the photovoltaic panel.

[0111] The coordinates in the y-direction are:

[0112]

[0113] in, is the coordinate of the center of the photovoltaic panel in the i-th row and j-th column in the y-direction, and L is the length of the photovoltaic panel.

[0114] The coordinate in the z-direction is the support height H.

[0115] Considering that the photovoltaic panels need to receive sunlight, according to the laws of physics, let's assume that the photovoltaic panels are tilted in the north-south direction:

[0116] normal vector of photovoltaic panel Direction in three-dimensional space can be determined by the angle of inclination. Sure:

[0117]

[0118] in, It is the first Line 1 Normal vector of photovoltaic panel ;

[0119] Based on a three-dimensional Cartesian coordinate system, the center of each photovoltaic panel and the normal vector of the photovoltaic panel are used to construct a three-dimensional spatial structure model of the photovoltaic module using 3ds Max three-dimensional modeling software.

[0120] This step involves analyzing the design drawings of the photovoltaic power station to accurately obtain the geometric information of the photovoltaic modules, while simultaneously collecting geographical information and support structure height parameters of the photovoltaic power station on-site. Based on this, a three-dimensional spatial structure model of the photovoltaic modules was successfully constructed. This model provides the basic framework for subsequent steps. Its significance lies in the fact that subsequent calculations of the photovoltaic module's shadow boundary and the generation of daily dynamic shading time maps all rely on the accurate geometric structure data provided by this three-dimensional model, making it an indispensable spatial architectural foundation for the entire research system.

[0121] Step S2: Calculate the geographical information of the photovoltaic power station using the solar path algorithm to obtain the azimuth and elevation angles of the sun. Based on the azimuth and elevation angles of the sun and the three-dimensional spatial structure model of the photovoltaic modules, calculate the shadow boundary cast by the photovoltaic modules on the ground. By aggregating the shadow boundaries of each photovoltaic module, obtain a daily dynamic shading time map.

[0122] Using a measuring instrument with GPS functionality, its built-in GPS module determines the location by receiving satellite signals, can directly read the latitude and longitude data of the power station (λ represents longitude, φ represents latitude), and records the current date and time through the clock built into the GPS module;

[0123] The solar path algorithm is based on astronomical principles, taking into account the Earth's revolution around the Sun and its rotation. It first calculates the Sun's declination angle δ, which reflects the latitude of the subsolar point on Earth. The calculation formula is as follows:

[0124]

[0125] Where JD stands for Julian Day (a method of calculating dates), and 23.45 is an approximation of the obliquity of the ecliptic, which is the angle between the Earth's orbital plane and the equatorial plane, measured in degrees.

[0126] The calculation of the sun's hour angle, which is time-dependent, uses the following formula:

[0127]

[0128] The hour angle ω is an astronomical angle used to describe the position of the sun relative to the local meridian. When the time t is greater than 12 o'clock, the sun is west of the meridian and the hour angle is positive; when the time t is less than 12 o'clock, the sun is east of the meridian and the hour angle is negative. The hour angle changes by 15° per hour because the Earth rotates 360° every 24 hours, resulting in a rotation of 15° per hour.

[0129] The solar altitude angle can be calculated using the following formula based on the local latitude, solar declination angle, and solar hour angle:

[0130]

[0131] in, This is the atmospheric refraction correction value. ,Right now The value ranges from 0.5 to 1 degree;

[0132] Convert it to a question Format:

[0133]

[0134] in, It is the solar altitude angle. It is the latitude of the photovoltaic power station. Between 0.5° and 1°.

[0135] According to the solar altitude angle and azimuth The unit vector of sunlight (From the sun towards the ground) is:

[0136]

[0137] in, The vector of sunlight on the x-axis is... , The sun's ray vector on the y-axis is... , The sun's ray vector on the z-axis is... x points east, y points north, and z points upward; the negative sign indicates that light travels downward.

[0138] Determine the vertex coordinates of the photovoltaic panels: Based on the constructed 3D spatial structure model of the photovoltaic modules, determine the position and orientation of each photovoltaic panel in 3D space. Let the center coordinates of a certain photovoltaic panel be... Its length is Width is The tilt angle is azimuth angle is (with due north as the reference) (Clockwise rotation is positive).

[0139] The coordinates of the four vertices of the photovoltaic panel in three-dimensional space can be calculated using geometric relationships. , , , .

[0140] For example, if the photovoltaic panel is placed horizontally along the east-west direction (azimuth angle) or ), and the center coordinates are , length direction along Positive axis, width direction along If the axis is positive, then the coordinates of the four vertices can be represented as:

[0141] ;

[0142] ;

[0143] ;

[0144] ;

[0145] in, , , , The coordinates of the four vertices of the photovoltaic panel in three-dimensional space are calculated through geometric relationships. L is the horizontal distance L from the reference point to the projection point at the bottom of the photovoltaic panel support, and W is the width of the photovoltaic panel.

[0146] When the photovoltaic panel is tilted At that time, it is necessary to consider bypassing , , Rotation of the shaft:

[0147] Matrix of rotation about the x-axis for:

[0148] ;

[0149] Matrix of rotation about the y-axis for:

[0150] ;

[0151] Matrix of rotation about the z-axis for:

[0152] ;

[0153] Total 3D rotation matrix yes , , Multiply in sequence (determined by the rotation order, such as first rotating around the x-axis, then around the y-axis, and then around the z-axis).

[0154] When the photovoltaic panel is tilted At this time, a three-dimensional rotation matrix is ​​required. For the above coordinates , , , The actual vertex coordinates are obtained by transforming the original coordinate vector. Left multiplication by a 3D rotation matrix This yields the transformed actual vertex coordinates.

[0155] Projection point calculation, for each photovoltaic panel vertex Calculate its position on the ground ( Projection point of the plane .

[0156] Let from the vertex to projection point The vector of the direction of the sun's rays Parallel, and the projection point On the ground On a plane, then:

[0157]

[0158] Among them, parameters Indicates from vertex to projection point The distance multiple along the direction of sunlight.

[0159] Depend on achievable And then calculate

[0160] ;

[0161] ;

[0162] in, Represents the origin Coordinates in three-dimensional space Point exist Projection points on the plane The coordinates.

[0163] The above calculations are performed sequentially on the four vertices of the photovoltaic panel to obtain four projection points. , , , By connecting these four projection points, the shadow boundary polygon of the photovoltaic panel on the ground at a certain moment can be obtained.

[0164] Repeat the above shadow boundary calculation steps for all photovoltaic modules within the photovoltaic power station. Assume there are a total of [number missing] photovoltaic modules within the power station. Each photovoltaic panel, in turn, is processed. ( This process is performed to obtain the results at each time step. ( , The shaded boundary polygon is the total number of time steps from sunrise to sunset. .

[0165] Ground area gridding: Dividing the ground area where the photovoltaic power station is located into a regular grid, for example, setting the side length of the grid to be... (like Each grid cell can be represented by the coordinates of its bottom-left vertex. This is used to identify the entire ground area, thus discretizing it into a series of grid cells.

[0166] For each time step Traverse all grid cells Determine whether each grid cell is covered by the shadow of a photovoltaic panel. If a grid cell is covered by the shadow of u photovoltaic panels, then... Simultaneously cover ( , (This indicates that the mesh cell is not covered by a shadow), then the mesh cell in time... The shadow coverage intensity is u.

[0167] In grid cell coordinates As a row index, time step Use it as a column index to construct a two-dimensional matrix. Matrix elements Indicates the location and time The shadow coverage intensity u at time is calculated. Through the above shadow coverage intensity calculation process, the shadow coverage intensity value of each grid cell at different time steps is filled into the matrix, and finally the daily dynamic shading time map T (spatiotemporal matrix) is obtained.

[0168] This step utilizes a solar path algorithm to accurately calculate the azimuth and altitude angles of the sun based on the geographical information of the photovoltaic power station. Then, combined with the three-dimensional spatial structure model of the photovoltaic modules constructed in step S1, the shadow boundaries cast by the photovoltaic modules on the ground are successfully calculated. By aggregating the shadow boundaries of each module, a daily dynamic shading time map is obtained. This map is crucial for subsequent image processing and analysis. It provides temporal and spatial shadow references for matching and labeling the processed remote sensing image sequence in step S3, visually presenting the relationship between sunlight and the shading status of the photovoltaic modules in a map format, thus bridging the gap between sunlight information and image processing.

[0169] Step S3: Acquire and preprocess images of the photovoltaic power station area to obtain a processed remote sensing image sequence. Match and label the processed remote sensing image sequence with the daily dynamic shading time map to generate a shadow-labeled image sequence. Based on the surface reflectance benchmark library, compensate and correct the shadow-labeled image sequence with a spectral correction algorithm to obtain a corrected remote sensing image sequence.

[0170] Multi-temporal high-resolution remote sensing images of the photovoltaic power station area were collected, with the number of images collected being [number missing]. Image I ( The corresponding shooting time is Latitude and longitude coordinates are .

[0171] These images cover the visible light and near-infrared bands, and the spatial resolution of the images is [missing information]. (Unit: meters / pixel).

[0172] Constructing a time-series data sequence: Arranging the acquired images in chronological order to form a time-series data sequence. ,in Indicates the first A remote sensing image.

[0173] The temporal image preprocessing is as follows:

[0174] Time-series image registration: Select an image taken close to noon as the reference image (set as...). This is because the sun's altitude angle is high at this time, resulting in uniform illumination and fewer shadows.

[0175] For other images Feature point extraction and matching are performed using the Scale Invariant Feature Transform (SIFT) algorithm, assuming the image... The coordinates of the feature points on are In the reference image The coordinates of the feature points matched above are .

[0176] Through affine transformation matrix , making After transformation Then get as close as possible affine transformation matrix The usual form is The transformation formula is:

[0177]

[0178] in, , , , , , These are elements in the affine transformation matrix. These are the transformed coordinates. Temporal image registration is performed on each element of the time-series data sequence to obtain the processed remote sensing image sequence.

[0179] After obtaining the processed remote sensing image sequence, the images are matched and labeled according to the daily dynamic shading time map:

[0180] Shaded areas are marked: for each pre-processed remote sensing image According to its shooting time In the dynamic shading time map Find the shadow distribution data for the corresponding time moment.

[0181] Image The pixel coordinates are ,like Then the image The median coordinate is The pixels are labeled as shadow pixels, and a shadow-labeled image sequence is generated. .

[0182] Spectral correction and vegetation true spectral reconstruction are based on a land surface reflectance reference library:

[0183] Based on the surface reflectance benchmark database The database contains standard spectral curves of common vegetation types in photovoltaic areas. Let the vegetation type be... ( , (Number of vegetation types) in band c ( , The standard reflectance (for the total number of bands) is .

[0184] Spectral correction algorithm:

[0185] For shaded image sequences Shadow pixels in Set shadow pixels In the band The original reflectance is Perform band-specific design:

[0186] Three key nonlinear interference factors are introduced to quantify the effects of shadow, light, and vegetation type:

[0187] Shadow intensity factor Quantized grid cells Shadow coverage intensity, single shadow Double shadow Multiple shadows (Taken from the daily dynamic shading time spectrum of shadow coverage intensity) );

[0188] Illumination angle factor That is, the solar altitude angle at that moment. (Taken from the solar path algorithm calculation results in step S2), standardized to (Value range 0-1, noon) (Approaching 1, and eventually approaching 0).

[0189] Vegetation type factors Assigning values ​​to herbaceous plants based on a surface reflectance benchmark library. Shrub plant value assignment (The type is determined by matching the vegetation texture features in the image to a reference library).

[0190] In the visible light band (c=1, 2, such as red and green light), the attenuation of visible light is greatly affected by shadow intensity and vegetation type. and The interactive items are defined by the following formula:

[0191]

[0192] in, These are the basic correction coefficients (corresponding to the correction benchmark under no-shading conditions, taken from vegetation types in the benchmark library). In the band Standard reflectance ); It is the shadow intensity correction factor (quantization) Nonlinear effects on visible light attenuation, such as in herbaceous plants ,shrub (The negative sign indicates that the stronger the shading, the greater the correction). It is a vegetation type correction factor (distinguishing between attenuation differences for herbaceous and shrubby vegetation, such as herbaceous vegetation). ,shrub ); This is the illumination angle compensation coefficient (compensation is increased when the solar altitude angle is low, such as...). , The smaller the amount, the greater the compensation.

[0193] In the near-infrared band (c=3, such as the 850nm band), the attenuation of the near-infrared band is more significantly affected by the shadow intensity and the illumination angle. The quadratic term (reflecting non-linear growth rate) and The exponential term (reflecting the exponential effect of the solar altitude angle) is calculated using the following formula:

[0194] in, It is the near-infrared baseline correction coefficient (taken from vegetation types in the benchmark library). Standard reflectance in the near-infrared band ); It is the coefficient of the quadratic term of shadow intensity (reflecting the nonlinear decay under multiple shadows, such as...). , (This indicates that the rate of decay increases with the square of the shadow intensity). It is the illumination angle index compensation coefficient ( middle, ,when At (low solar altitude angle), the compensation range is (5 times that of noon, adapted to the strong attenuation of oblique light); It is the vegetation type compensation coefficient (herbaceous) ,shrub (To adapt to the differences in near-infrared attenuation between the two).

[0195] The above corrections are applied to all shadow pixels to obtain the corrected remote sensing image sequence. Let the corrected remote sensing image sequence be... ,in Indicates time, For time The acquired corrected remote sensing images provide accurate data for subsequent identification of vegetation growth dynamics.

[0196] This step begins with image acquisition and preprocessing of the photovoltaic power station area to obtain a preliminary usable remote sensing image sequence. Then, using the daily dynamic shading time map generated in step S2, the image sequence is matched and labeled to generate a shaded image sequence. Finally, based on a surface reflectance benchmark library, a spectral correction algorithm is used to compensate and correct the shaded image sequence, resulting in a corrected remote sensing image sequence. These corrected image sequences provide a clean and reliable data source for the accurate calculation of vegetation cover in step S4, removing the influence of shading and other interference factors on the image spectrum and improving the accuracy of subsequent vegetation cover analysis.

[0197] Step S4: Collect environmental data of the photovoltaic power station and construct a vegetation coverage learning model. Input the environmental data of the photovoltaic power station and the corrected remote sensing image sequence into the vegetation coverage learning model, and output the vegetation coverage of the photovoltaic power station.

[0198] The environmental data collected from the photovoltaic power plant are as follows:

[0199] Meteorological data: A weather station is installed within the photovoltaic power station. This station is equipped with high-precision temperature sensors that can measure the ambient temperature in real time. The temperature is recorded as follows: Rain gauges are installed within photovoltaic power stations to measure rainfall. Rain gauges determine rainfall by collecting and measuring the amount of rainwater falling into them over a specific period; this rainfall is recorded as... .

[0200] Soil data: Soil moisture sensors are buried in the soil of the photovoltaic power station. These sensors determine the soil moisture by measuring the dielectric constant of the soil. The soil moisture is denoted as... .

[0201] To ensure dimensional consistency, the above environmental data were normalized.

[0202] The core input data is a corrected remote sensing image sequence. , , and .

[0203] Construction of a vegetation cover learning model:

[0204] Convolutional Neural Network (CNN) module: Used to extract spatial features (such as texture and morphology) of vegetation in images, for each image. Through convolutional layers ( ) and pooling layer ( Extract features.

[0205] The convolution kernels are initialized uniformly using Xavier (Glorot) as follows:

[0206] Let the number of input channels of a certain convolutional layer in a CNN be... (Corresponding to the number of remote sensing image bands, i.e., red, green, and near-infrared), the number of output channels is... The kernel size is Then the convolution kernel weights The initialization formula is:

[0207]

[0208] in, It is the number of input channels of the convolutional layer, corresponding to the number of bands in the remote sensing image (such as red, green, and near-infrared, usually taken as 3). It is the number of output channels of the convolutional layer (i.e., the number of convolutional kernels, which determines the number of feature types extracted). These are the convolution kernel weights (parameters that need to be initialized, controlling the "weights" of feature extraction). : Indicates a uniform distribution, i.e., weights From the interval Random sampling is performed within a uniform distribution, meaning the initialization of the convolution kernel weights W follows a uniform distribution;

[0209] For images Middle position , No. Convolution operation formula:

[0210]

[0211] in, It is the first Layer convolution at position eigenvalues; It is the first Layer convolution kernel at position Weights (replacing the original) ); It is the first Layer bias; It is an activation function (such as ReLU).

[0212] After multi-layer processing, spatial features are obtained. .

[0213] Long Short-Term Memory (LSTM) module: Used to capture the growth characteristics of vegetation over time, storing spatial features at different times τ. Input the data sequentially into the LSTM to compute the temporal features.

[0214] The core formula for LSTM is as follows:

[0215] Input Gate: ;

[0216] Forgot the door: ;

[0217] Cell state: ;

[0218] Output gate: ;

[0219] Hidden state: ;

[0220] in, It represents the time step and is used to index data at different points in time; It is in time Spatial features, which are features extracted from the convolutional neural network (CNN) part; It is the value of the input gate, indicating how much new information can enter the cell state at the current moment; It is the sigmoid function, which compresses the input to between 0 and 1, and is used to determine the passage of information. It is the weight matrix of the input gate, used in conjunction with spatial features. Multiply. It is the hidden state weight matrix of the input gate, used to compare with the hidden state at the previous time step. Multiply; It is the bias vector of the input gate, used to adjust the activation value of the input gate. It is the hidden state of the previous time step, containing information about the previous time step. After processing, the temporal features are obtained. .

[0221] Fusion Layer: Combining environmental factors (temperature, precipitation, etc.), it outputs vegetation cover and incorporates temporal characteristics. Environmental data , , By splicing, the fusion features are obtained. .

[0222] Calculate vegetation cover using fully connected (Dense) layers:

[0223]

[0224] in, It is time vegetation coverage; These are the weights of the fully connected layer. It is a bias; It's the sigmoid function, and its output value range is... (0 indicates no vegetation, 1 indicates complete cover).

[0225] The loss function consists of two parts: the basic MSE loss and the spatial regularization term, with the total loss being... The calculation formula is as follows:

[0226]

[0227] in, It is the basic mean squared error loss. It is the space regularization term, and λ is the regularization coefficient = 0.01;

[0228] The basic MSE loss calculation formula is as follows:

[0229]

[0230] in, It is the basic mean square error loss; It refers to the number of samples, i.e., the total number of photovoltaic area images / time series samples participating in the training; It is the vertical (e.g., north-south) row number of the ground grid, representing the number of units in the vertical direction of the ground after gridding; This refers to the number of horizontal (e.g., east-west) columns of the ground grid, indicating the number of units on the ground in the horizontal direction after gridding.

[0231] Calculate vegetation cover using fully connected (Dense) layers:

[0232]

[0233] in, It is time vegetation coverage; These are the weights of the fully connected layer. It is a bias; It's the sigmoid function, and its output value range is... (0 indicates no vegetation, 1 indicates complete cover).

[0234] This step focuses on collecting environmental data from the photovoltaic power station and constructing a vegetation cover learning model. The corrected remote sensing image sequence obtained in step S3, along with the collected environmental data, is input into the model, ultimately outputting the vegetation cover of the photovoltaic power station. The obtained vegetation cover data is the foundational data for predicting vegetation cover change trends in step S5. Its accuracy directly affects the reliability of subsequent predictions and analyses, providing a key quantitative indicator for in-depth research on vegetation growth in the photovoltaic power station area.

[0235] Step S5: Based on the vegetation coverage of the photovoltaic power station, construct a vegetation coverage change trend prediction model, collect environmental forecast data for the next few days, input the data into the vegetation coverage change trend prediction model, and output the predicted change trend of vegetation coverage. Through the predicted change trend of vegetation coverage, provide early warning of abnormal vegetation growth in the photovoltaic area.

[0236] The data is prepared as follows:

[0237] Obtaining future environmental data: Collecting weather forecast data and temperature data for the next few days. Precipitation ) and soil moisture forecast data .

[0238] Use the vegetation cover data obtained in step S4 ,in The data, which indicates the time period, are vegetation cover calculated based on historical imagery and environmental data.

[0239] Environmental data collection includes historical environmental data corresponding to vegetation cover data, including temperature. Precipitation and soil moisture .

[0240] Construction of a model for predicting vegetation cover change trends (based on linear regression):

[0241] Assuming vegetation cover There is a linear relationship between it and environmental factors, and the model takes the following form:

[0242]

[0243] in, These are the coefficients to be estimated. It is temperature, It's the amount of precipitation. It's soil moisture. This is the error term.

[0244] Data processing: Historical vegetation cover data and corresponding environmental data are organized into a data matrix. Let the vegetation cover data vector be... The environmental data matrix is ​​as follows:

[0245]

[0246] The coefficients of a linear regression model are estimated using the least squares method; the coefficient vector is... The estimation formula is:

[0247]

[0248] The coefficients are obtained through calculation. The estimated value.

[0249] The predicted trend of vegetation cover change is as follows:

[0250] By substituting future environmental data into the vegetation cover change trend prediction model, the predicted value of vegetation cover for the next few days can be calculated. Prediction for a certain day in the future:

[0251]

[0252] in, Indicates time The vegetation cover at that time, here Represents a time increment, that is, from time... Push forward Vegetation cover after a period of time; Indicates time The temperature at that time; Indicates time Rainfall at that time; Indicates time Soil moisture at that time; , , , It is a regression coefficient, used to measure the degree of influence of various environmental factors on vegetation cover; This is the error term, representing the deviation between the model's predicted value and the actual value.

[0253] Based on statistical analysis of historical vegetation cover data, the normal fluctuation range of vegetation cover is determined. and the threshold of change If vegetation cover decreases in a short period of time (e.g.) =1 day) The change range exceeds or for several consecutive days (such as =3 days) continuously deviating from the normal range If so, it is determined to be an abnormal vegetation growth.

[0254] When abnormal vegetation growth is detected, the possible causes of the abnormality are analyzed. By combining environmental data from the same period (such as temperature, rainfall, soil moisture, etc.), it is determined whether it is caused by factors such as drought, flooding, extreme temperatures, or biological factors such as pests and diseases.

[0255] Based on the type and extent of abnormal vegetation growth, corresponding ecological regulation suggestions are proposed. For example, if drought is causing a decrease in vegetation cover, irrigation can be recommended; if it is a problem of pests and diseases, pest and disease control measures can be recommended.

[0256] Ecological regulation decisions are as follows:

[0257] Short-term control measures are formulated to address sudden abnormal vegetation growth by developing immediate and implementable short-term control plans. These measures include emergency watering and temporary shading (if caused by excessively high temperatures) to alleviate stress on vegetation growth.

[0258] Long-term control strategies, taking a long-term perspective, involve formulating long-term strategies for vegetation management and ecological environment improvement. Examples include optimizing irrigation systems in photovoltaic areas, adjusting vegetation planting structures, and implementing soil improvement programs to maintain healthy vegetation growth and ensure ecological balance.

[0259] This invention proposes an artificial intelligence-based dynamic monitoring method for vegetation cover in photovoltaic (PV) areas, aiming to solve the monitoring error problem caused by the shadow interference of PV modules when traditional remote sensing image analysis methods are applied to PV areas. The method is implemented through the following steps: First, by analyzing PV power station design drawings and on-site geographic information, the geometric information and support height parameters of the PV modules are obtained, and a three-dimensional spatial structure model is constructed. Second, the azimuth and elevation angles of the sun are calculated using a solar path algorithm, and the shadow boundaries of the PV modules on the ground are calculated using the three-dimensional model to generate a daily dynamic shading time map. Third, multi-temporal high-resolution remote sensing images of the PV power station area are collected, preprocessed, and shadow annotations are performed based on the dynamic shading time map. The images are then compensated and corrected using a spectral correction algorithm. Fourth, a vegetation cover learning model is constructed: environmental data (meteorological, soil, etc.) of the PV power station are collected, and a vegetation cover learning model based on a convolutional neural network (CNN) and a long short-term memory network (LSTM) is constructed to output the vegetation cover. Fifth, a linear regression prediction model is constructed based on historical vegetation cover data and environmental data to predict future vegetation cover change trends, provide early warnings of abnormal vegetation growth, and assist in ecological regulation decisions.

[0260] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0261] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

Claims

1. An artificial intelligence-based photovoltaic area vegetation coverage dynamic monitoring method, characterized in that, The method comprises the following steps: Step S1: Obtain the geometric information of the photovoltaic module by analyzing the photovoltaic power station design drawing, collect the geographic information and photovoltaic panel support height parameters of the on-site photovoltaic power station, combine the geometric information of the photovoltaic module and the photovoltaic panel support height parameters, and construct a photovoltaic module three-dimensional space structure model; Step S2: Calculate the geographic information of the on-site photovoltaic power station by a solar path algorithm to obtain the azimuth and elevation angles of the sun, and calculate the shadow boundary of the photovoltaic module on the ground according to the azimuth, elevation angles of the sun and the photovoltaic module three-dimensional space structure model, and obtain a daily dynamic shading time atlas by aggregating the shadow boundary of each photovoltaic module; Step S3: Image collection and preprocessing are performed on the photovoltaic power station area to obtain a processed remote sensing image sequence, the processed remote sensing image sequence is matched and labeled by the daily dynamic shading time atlas to generate a shadow labeled image sequence, and the shadow labeled image sequence is compensated and corrected by a spectral correction algorithm based on a ground reflectivity benchmark library to obtain a corrected remote sensing image sequence; Step S4: Collect the environmental data of the photovoltaic power station, and construct a vegetation coverage learning model, input the environmental data of the photovoltaic power station and the corrected remote sensing image sequence into the vegetation coverage learning model, and output the vegetation coverage of the photovoltaic power station; Step S5: Based on the vegetation coverage of the photovoltaic power station, a vegetation coverage change trend prediction model is constructed, environmental forecast data for several days in the future is collected and input into the vegetation coverage change trend prediction model, and a predicted change trend of the vegetation coverage is output, and early warning is performed on the abnormal growth of vegetation in the photovoltaic area according to the predicted change trend of the vegetation coverage; The photovoltaic module three-dimensional space structure model is constructed as follows: A three-dimensional Cartesian coordinate system is established with the location of the photovoltaic power station as the origin, the x-axis points east, the y-axis points north, and the z-axis points vertically upward; According to the arrangement mode in the design drawing, the position of each photovoltaic panel in the three-dimensional space is calculated, and for the photovoltaic panel in the ith row and jth column, the center coordinate in the x direction is represented as: ； in, Indicates the first Line number Photovoltaic panel center The coordinates of the direction, where i is the row number of the photovoltaic panel. It is the column number where the photovoltaic panel is located. It is the width of the photovoltaic panel; The center coordinate in the y direction is: ； wherein, is the coordinate of the center of the jth photovoltaic panel in the y direction of the ith row, and L is the length of the photovoltaic panel. The center coordinate in the z direction is the support height H; Considering that the photovoltaic panel needs to receive sunlight, according to the physical law, the photovoltaic panel is inclined in the north-south direction: Normal vector of a photovoltaic panel Direction in three-dimensional space by inclination angle Determination: ； wherein is the first row of the normal vector of the photovoltaic panel ; Based on the three-dimensional Cartesian coordinate system, the center of each photovoltaic panel and the normal vector of the photovoltaic panel, a 3dsMax three-dimensional modeling software is selected to construct a photovoltaic module three-dimensional space structure model; The vegetation coverage learning model is constructed as follows: Convolutional neural network module: used to extract spatial features of vegetation in the image, extract features from each image through the convolutional layer and the pooling layer; Long short-term memory network module: for capturing the growth change characteristics of vegetation over time, and the spatial characteristics of different times τ Sequential input LSTM, calculate the timing characteristics; Fusion layer: combine environmental factors, output vegetation coverage, and time sequence features with environmental data , , splicing to obtain fusion features ; The vegetation coverage is calculated through a fully connected layer (Dense): ； wherein, is the time of day; is the full connection layer weight, is the bias; is the sigmoid function, the output value range ; The vegetation coverage change trend prediction model is as follows: ； wherein, denotes the vegetation cover at time , where denotes a time increment, i.e. the vegetation cover after time; denotes the temperature at time ; denotes the precipitation at time ; denotes the soil moisture at time ; , , , are regression coefficients, measuring the degree of influence of the respective environmental factor on the vegetation cover; is an error term, representing the deviation between the model predicted value and the actual value.​ 2. The method according to claim 1, wherein, By analyzing the photovoltaic power station design drawing, the geometric information of the photovoltaic module is obtained, and the geographic information and photovoltaic panel support height parameters of the on-site photovoltaic power station are collected, including: The geometric information of the arrangement mode, shape and size of the photovoltaic module is obtained from the photovoltaic power station design drawing; The longitude, latitude and altitude geographic coordinate information of the location of the power station is obtained by using a GPS measurement device; Determination of the height H of the photovoltaic panel support and of the inclination angle of the photovoltaic panel by field measurement .

3. The method according to claim 2, wherein, The geographical information of the on-site photovoltaic power station is calculated by a solar path algorithm to obtain the azimuth and elevation angle of the sun, including: Computing the solar azimuth angle The formula is: ； wherein, is the solar azimuth angle; is the solar declination angle; is the latitude of the photovoltaic power plant; is the solar hour angle; is the solar elevation angle; The solar elevation angle is calculated according to the local latitude, solar declination angle and solar hour angle: ； wherein, is an atmospheric refraction correction value, i.e. has a value in the range of 0.5 to 1 degree; Convert this to the form of a request : ； wherein is the solar elevation angle, is the latitude of the photovoltaic plant, is between 0.5° and 1°.

4. The method according to claim 3, wherein, The daily dynamic shading time atlas is obtained by aggregating the shadow boundary of each photovoltaic module, including: Traverse all photovoltaic components, repeat the above shadow boundary calculation steps for all photovoltaic components in the photovoltaic power station; suppose that there are block photovoltaic panels in the power station, process each block photovoltaic panel in turn , and obtain the shadow boundary polygon at each time step . ​ The ground area is gridded, and the ground area where the photovoltaic power station is located is divided into regular grids. The side length of the grid is set as Each grid is identified by the coordinates of the lower left corner vertex of the grid, so that the entire ground area is discretized into a series of grid units, wherein represents the coordinate value of the lower left corner vertex of the grid in the horizontal direction; represents the coordinate value of the lower left corner vertex of the grid in the vertical direction. For each time step , traverse all grid cells , determine whether each grid cell is covered by the shadow of the photovoltaic panel, if a grid cell is covered by the shadow polygon of u photovoltaic panels at the same time, then the shadow coverage intensity of the grid cell at time is u; with grid cell coordinates as row index, time steps as column index, a two-dimensional matrix is constructed , the matrix elements represent the shadow coverage intensity u at position and time , the shadow coverage intensity values of each grid cell at different time steps are filled into the matrix through the above shadow coverage intensity calculation process, and finally the daily dynamic shading time atlas T is obtained.

5. The method according to claim 4, wherein, Image collection and preprocessing are performed on the photovoltaic power station area to obtain a processed remote sensing image sequence, and the processed remote sensing image sequence is matched and labeled by the daily dynamic shading time atlas to generate a shadow labeled image sequence, including: The multi-temporal high-resolution remote sensing images of the photovoltaic power station area are collected, and the number of the collected images is , the shooting time corresponding to the image I is , and the longitude and latitude coordinates are ; These images cover visible light, near-infrared band, and the spatial resolution of the images is ; The collected images are arranged in time sequence to form a time sequence data sequence wherein represents the remote sensing image The time series data sequence is preprocessed as follows: First, the time series image is registered, and the image taken at noon is selected as the reference image because the solar elevation angle is large at this time, the light is uniform, and the shadow is less; For other images , the feature point extraction and matching are performed using the scale invariant feature transform algorithm SIFT, and the feature point coordinates on the image are set as , and the feature point coordinates matched on the reference image are set as ; By an affine transformation matrix such that After transformation as close as possible , affine transformation matrix The general form is The transformation formula is: ； wherein, , , , , , is an element in the affine transformation matrix, is the transformed coordinate; performing time-series image registration processing on each of the time-series data sequences to obtain a processed remote sensing image sequence; After obtaining the processed remote sensing image sequence, the image is matched and labeled according to the daily dynamic shading time atlas: Shadow region annotation: for each pre-processed remote sensing image , according to its shooting time , find the shadow distribution data of the corresponding time in the dynamic shading time atlas ; Set the pixel coordinates of the image to , if , mark the pixel with coordinates in the image as a shadow pixel, and generate a shadow-labeled image sequence .

6. The method according to claim 5, wherein, The shadow labeled image sequence is compensated and corrected by a spectral correction algorithm to obtain a corrected remote sensing image sequence, including: Based on ground reflectance benchmark library , the library contains standard spectral curves of common vegetation types in photovoltaic area, and the vegetation types The standard reflectivity in wave band c is ; The process of compensating and correcting the shadow labeled image sequence by the spectral correction algorithm to obtain the corrected remote sensing image sequence is as follows: For shaded image sequences Shadow pixels in Set shadow pixels In the band The original reflectance is Perform band-specific design; Three key nonlinear interference factors are introduced to quantify the influence of shadow, light and vegetation type as follows: Shadow intensity factor , quantization grid cell shadow coverage intensity, single shadow , double shadow , multiple shadow ; Illumination angle factor i.e. the solar elevation angle at that moment is normalized to ; Vegetation type factor , assigning values to herbaceous plants based on a library of ground reflectance benchmarks , assigning values to shrubs ; The visible light band, visible light band attenuation is affected by the intensity of the shadow and vegetation type, the introduction With the interaction term, the formula is as follows: ； wherein, is a base correction coefficient; is a shadow intensity correction coefficient; is a vegetation type correction coefficient; is an illumination angle compensation coefficient; In the near-infrared band, the near-infrared band attenuation is more significantly affected by the shadow intensity and the illumination angle. The quadratic term and the exponential term are introduced as follows: The formula is as follows:​ ； wherein, is a near-infrared base correction coefficient; is a shadow intensity quadratic term coefficient; is an illumination angle index compensation coefficient; is a vegetation type compensation coefficient; After the above correction is performed on all shadow pixels, a corrected remote sensing image sequence is finally obtained, and the corrected remote sensing image sequence is denoted as wherein represents time, is the time The corrected remote sensing image obtained provides accurate data for subsequent dynamic identification of vegetation growth.

7. The method according to claim 6, wherein, The step S4 includes: The environmental data of the photovoltaic power station is collected as follows: Meteorological data, installing a meteorological station in the photovoltaic power station, temperature is recorded as ; installing a rain gauge in the photovoltaic power station to measure precipitation, the rain gauge determines the precipitation by collecting and measuring the amount of rain falling into it within a certain time, the precipitation is recorded as ; Soil data, soil humidity sensor is buried in the soil of the photovoltaic power station, and the soil humidity is recorded as ; In order to ensure dimensional consistency, the above environmental data is normalized.

8. The method according to claim 7, wherein the method is based on artificial intelligence. The predicted change trend of vegetation coverage is obtained, including: The future environment data is substituted into the vegetation coverage change trend prediction model to calculate the predicted value of the vegetation coverage in future days .

9. The method according to claim 8, wherein, Through the predicted change trend of vegetation coverage, early warning of vegetation growth abnormalities in the photovoltaic area is performed, including: According to statistical analysis of historical vegetation coverage data, the normal fluctuation range of vegetation coverage is determined and a change amplitude threshold If the vegetation coverage changes by more than in a short period of time, or continuously deviates from the normal range for many consecutive days , it is determined that the vegetation growth is abnormal; When detecting vegetation growth abnormalities, analyze the possible causes of the abnormalities, and combine the environmental data of the same period to determine whether it is caused by drought, flood, extreme temperature factors or possible biological factors; According to the type and degree of vegetation growth abnormalities, corresponding ecological regulation suggestions are proposed, and the ecological regulation decisions are as follows: Short-term regulation measures: For sudden vegetation growth abnormalities, develop short-term regulation schemes that can be implemented immediately, including emergency water supply and temporary shading measures to relieve vegetation growth pressure. Long-term regulation strategy: From a long-term perspective, develop long-term strategies for vegetation management and ecological environment improvement.

Images