Efficient photovoltaic power generation sun tracking method based on EVAE and EGAN network

Through the combination of EVAE and EGAN networks, the inefficiency problem of traditional photovoltaic systems in complex environments and light changes is solved, the adaptive adjustment of photovoltaic panels is realized, and the photoelectric capture efficiency and system performance are improved.

CN118963415APending Publication Date: 2024-11-15JINLING INST OF TECH
View PDF 0 Cites 8 Cited by

Patent Information

Application Number
CN202410964315.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

Traditional photovoltaic systems cannot effectively adaptively adjust the angle of the photovoltaic panel when complex environments and light conditions change, resulting in low light energy utilization efficiency.

Method used

Using an efficient photovoltaic power generation tracking method based on EVAE and EGAN networks, the photovoltaic panels are obtained by obtaining geographical information, photovoltaic panel cleanliness and photovoltaic system inclination data, combined with enhanced variational autoencoder and generation adversarial network, the photovoltaic panels are realized at different angles and locations. , adaptive adjustment under time and terrain.

Benefits of technology

It improves the adaptive adjustment capability and light energy capture efficiency of the photovoltaic system, ensures that the photovoltaic panels are always at the optimal light angle, improves the overall performance and stability of the system, and adapts to various complex environmental conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118963415A_ABST
    Figure CN118963415A_ABST
Patent Text Reader

Abstract

The invention discloses an efficient photovoltaic power generation sun tracking method based on an EVAE and an EGAN network, and the method comprises the following steps: S1, obtaining geographic information data, and guaranteeing the accurate geographic position, topographic relief, vegetation coverage and building shielding conditions; s2, the cleanliness of the photovoltaic panel is obtained, efficient operation of a photovoltaic system is ensured, the influence of dust and dirt on the performance of the photovoltaic panel is avoided, and the energy capture efficiency is improved; s3, obtaining the inclination angle of the photovoltaic system and the sunlight obtaining area of the photovoltaic panel at different angles, and optimizing the solar energy capturing efficiency and the system performance; and S4, realizing photovoltaic efficiency maximization at different angles, different places, different times and different terrains by using an adversarial training mechanism of an enhanced variational auto-encoder (EVAE) and GANs. Compared with a traditional photovoltaic tracking system, the method realizes self-adaptive adjustment and efficient solar energy capture of the photovoltaic system in a complex environment by combining high-quality hidden variable representation generated by EVAE and authenticity and diversity of EGAN optimization data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of photovoltaic power generation, and in particular to a high-efficiency photovoltaic power generation sun tracking method based on EVAE and EGAN networks. Background Art

[0002] The present invention relates to the field of photovoltaic power generation, and in particular to a high-efficiency solar tracking photovoltaic system based on EVAE and EGAN networks. In the field of photovoltaic power generation, optimizing the efficiency of solar energy capture is of great significance. Traditional photovoltaic systems have many limitations and defects when dealing with complex environments and changes in lighting conditions. For example, traditional systems cannot adaptively adjust the angle of photovoltaic panels according to environmental changes, resulting in low efficiency in light energy utilization.

[0003] In recent years, the development of artificial intelligence and machine learning technologies has provided new ideas for photovoltaic system optimization. As an advanced machine learning method, enhanced variational autoencoder (EVAE) and generative adversarial network (EGAN) make up for the shortcomings of traditional photovoltaic systems by generating high-quality latent variable representations and optimizing the authenticity and diversity of data. However, there are still many problems with existing photovoltaic systems based on machine learning. For example, the model training process is complicated, the computing resources are consumed a lot, and the generalization ability of the model is insufficient in practical applications.

[0004] Traditional photovoltaic systems have certain limitations when dealing with complex environments and changes in lighting conditions. For example, photovoltaic panels with fixed angles cannot adapt to changes in lighting throughout the day, resulting in low efficiency in light energy utilization. To this end, researchers have proposed enhanced variational autoencoders (EVAE) and generative adversarial networks (EGAN) algorithms, which use advanced machine learning techniques to improve the light energy capture efficiency of photovoltaic systems. However, in practical applications, how to effectively combine these improved algorithms to form a systematic photovoltaic optimization method still requires further research and exploration. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an efficient solar tracking photovoltaic system based on EVAE and EGAN network. By combining EVAE and EGAN technologies, the present invention improves the adaptive adjustment capability and light energy capture efficiency of the photovoltaic system, and solves the problem that the traditional method cannot effectively cope with complex environments and dynamic light changes.

[0006] To achieve the above object, the present invention provides a high-efficiency photovoltaic power generation sun tracking method based on EVAE and EGAN network, comprising the following method steps:

[0007] S1. Obtain geographic information data to ensure accurate geographic location, terrain undulations, vegetation coverage and building obstruction;

[0008] S2. Obtain the cleanliness of photovoltaic panels to ensure the efficient operation of the photovoltaic system, avoid the impact of dust and dirt on the performance of photovoltaic panels, and improve energy capture efficiency;

[0009] S3, obtaining the inclination angle of the photovoltaic system and the sunlight acquisition area of ​​the photovoltaic panel at different angles to optimize the solar energy capture efficiency and system performance;

[0010] S4. Use the adversarial training mechanism of enhanced variational autoencoder (EVAE) and GANs to maximize photovoltaic efficiency at different angles, locations, times, and terrains.

[0011] Optionally, the S1-related parameters include obtaining an accurate geographic location:

[0012] First, using the Differential GPS (DGPS) system, the base station and the rover work together to obtain high-precision geographic location data.

[0013] Δφ=φbase-φrover

[0014] Δλ=λbase-λrover

[0015] φcorrected=φrover+Δφ·log(Δφ+1)

[0016] λcorrected=λrover+Δλ·log(Δλ+1)

[0017] Among them, φbawe represents the latitude of the base station, φrover represents the latitude of the mobile station, λbase represents the longitude of the base station, and λrover represents the longitude of the mobile station. These corrected coordinates can significantly improve the accuracy of the geographic location and provide a reliable data basis for subsequent solar track prediction.

[0018] Then, satellite remote sensing data is acquired, and the area where the photovoltaic system is located is accurately located through high-resolution satellite remote sensing images and image processing technology. Then, in the geographic information system (GIS) platform, differential GPS data and satellite remote sensing data are fused to calculate the final precise location.

[0019]

[0020]

[0021] Among them, φ 2 corrected represents the latitude after differential correction, φ 2 satellite represents the latitude of satellite remote sensing images, λ 2 corrected represents the longitude after differential correction, λ2 Satellite represents the longitude obtained from satellite remote sensing images. This data fusion method combines the advantages of DGPS and satellite remote sensing to provide more accurate geographic location and lay a solid foundation for the accurate prediction of the sun's trajectory.

[0022] Optionally, the S1-related parameters include terrain relief data:

[0023] First, generate a contour map using terrain elevation data (such as digital elevation model DEM data):

[0024]

[0025] Among them, t represents the interpolation parameter, h represents the height of the contour line, and Z1 and Z2 represent the elevation of adjacent points respectively. Elevation data can be obtained through digital elevation models (DEM) or laser radar (LiDAR), which provide three-dimensional information of the terrain. DEM data is usually stored in the form of a raster, and each pixel represents the height value of a certain point on the ground. The goal of this step is to determine the location of the contour lines, that is, to mark all points on the topographic map whose elevation is equal to a specific height value (such as 100 meters). Through these points, contour lines representing the undulations of the terrain can be drawn.

[0026] Then perform the interpolation calculation:

[0027] x=(1-t)*x1+t*x2

[0028] y=(1-t)*y1+t*y2

[0029] Among them, x and y represent the coordinates of the interpolation points, x1 and x2 represent the horizontal coordinates of the adjacent points, and y1 and y2 represent the vertical coordinates of the adjacent points. The purpose of interpolation calculation is to find the exact position of the contour line between each pair of adjacent points. Through these interpolation points, smooth contour lines can be drawn. This step ensures the accuracy and continuity of the contour lines, so that the topographic map can accurately reflect the actual undulations of the terrain.

[0030] Next, the generated contour data is processed and analyzed:

[0031]

[0032] Among them, slope represents the slope, Δz represents the elevation difference, and Δd represents the horizontal distance. Slope calculation is used to evaluate the steepness of the terrain, which is very important for the layout and installation angle selection of the photovoltaic system. A steeper slope may make it impossible to install the photovoltaic panels stably, while a flatter terrain is conducive to the stability and sunlight capture of the photovoltaic panels. By analyzing the slope data, the best installation location for the photovoltaic panels can be determined to avoid shadows and insufficient light caused by the undulating terrain.

[0033] Optionally, the S1-related parameters include the permission to obtain vegetation coverage and building occlusion conditions:

[0034] First, images of the study area were obtained through Google Maps to identify the distribution of vegetation and buildings.

[0035] Then, the image is preprocessed using a simple image processing algorithm:

[0036]

[0037] Among them, Igray represents a grayscale image, and R, G, and B represent the pixel values ​​of red, green, and blue, respectively.

[0038] Next, perform color threshold segmentation:

[0039]

[0040] Among them, Thr represents the binarized image, I represents the pixel value, μ represents the mean of the image, σ represents the standard deviation of the image, and T represents the threshold. This formula uses standardization to make the threshold segmentation more stable.

[0041] Then, calculate the density of vegetation cover and building occlusion:

[0042]

[0043] Where Den represents the density of the target area, represents the sum of the standardized pixel values ​​in the target area, and n represents the number of pixels in the target area.

[0044] Finally, a density detection result map is generated to visualize the density of vegetation coverage and building occlusion, such as Figure 4 As shown in the figure, this figure shows the distribution of vegetation coverage and building shading in the study area. The left figure shows the density of vegetation coverage, and the right figure shows the density of building shading. These density maps intuitively reflect the different types of shading in the study area. By analyzing these maps, the layout and installation of photovoltaic panels can be optimized to ensure maximum sunlight capture and improve the overall efficiency of the system.

[0045] Optionally, the S2-related parameters include the cleanliness of the photovoltaic panel:

[0046] First, collect image data to obtain clean images of photovoltaic panels through drone photography, fixed camera monitoring or manual photography:

[0047] Iclean(x,y)=255

[0048] Among them, Iclean(x,y) represents the value of the pixel (x,y) in the clean photovoltaic panel image, and 255 represents a pure white pixel. Compared with the traditional simple background image formula, this formula uses a fixed high value to represent the clean surface, which improves the standardization and consistency of image generation and innovatively solves the consistency problem of background photovoltaic panel image processing.

[0049] Then, collect images of photovoltaic panels with dust, and use image processing technology to annotate and identify dust-covered areas:

[0050]

[0051] Among them, Idusty(x,y) represents the value of the pixel (x,y) in the dusty photovoltaic panel image, 255 represents the initial white pixel, N and M represent the width and height of the image respectively, x0 and y0 are the dust center positions, and σ controls the range of dust distribution. Compared with the traditional simple binary division formula, this formula uses a combination of logarithmic and Gaussian distribution to improve the accurate simulation of dust distribution and innovatively solve the simulation problem of dust density changes at different positions.

[0052] Next, collect images of photovoltaic panels with dirt and use image processing technology to annotate and identify the dirt-covered areas:

[0053]

[0054] Among them, Idirty(x,y) represents the value of the pixel (x,y) in the dirty photovoltaic panel image, 255 represents the initial white pixel, and N and M represent the width and height of the image respectively. Compared with the traditional simple binary division formula, this formula uses a combination of square root function and sine cosine function to improve the multi-dimensional characteristic simulation of dirt distribution and innovatively solves the complexity of dirt being affected by environmental factors.

[0055] Finally, a cleanliness inspection result map is generated to visualize the classification results of photovoltaic panel images with different cleanliness states.

[0056] Optionally, the S3 includes:

[0057] S31. Obtain the inclination angle of the photovoltaic system to ensure the optimal inclination angle to maximize solar energy capture efficiency and system performance;

[0058] First, select and calibrate the laser rangefinder to ensure that it can accurately measure height difference and horizontal distance. Then, select two measurement points at the bottom and top of the photovoltaic panel to measure the height difference and horizontal distance respectively.

[0059]

[0060] Where Δh represents the height difference and d represents the horizontal distance. Compared with the traditional simple trigonometric function formula, this formula introduces a logarithmic function to solve the measurement error caused by the distance and height difference, thereby improving the measurement accuracy. If necessary, convert the inclination angle in radians to angle units:

[0061] θ degrees = θ × (180 / π)

[0062] Finally, the calculated inclination angle is recorded in a data table, including the measurement date, time, location and other information, for subsequent analysis, as shown in Table 1.

[0063] S32, obtaining the sunlight acquisition area of ​​the photovoltaic panel at different angles;

[0064] First, calculate the sun's altitude (hs). Use the geographic location (latitude, longitude), date, and time to calculate the sun's altitude and azimuth:

[0065] hs=arcsin(sin(φ)sin(δ)+cos(φ)cos(δ)cos(H))

[0066] Among them, φ is the geographic latitude, δ is the solar declination angle, and H is the hour angle.

[0067] Next, when the photovoltaic panel inclination angle (θ degrees) and the solar altitude angle (hs) are known, the effective sunlight acquisition area of ​​the photovoltaic panel at different inclination angles is calculated:

[0068] Aeffective=Acos(θdegrees-hs)

[0069] Among them, Aeffective is the effective sunlight acquisition area, A is the actual area of ​​the photovoltaic panel, θdegrees is the inclination angle of the photovoltaic panel, and hs is the solar altitude angle.

[0070] Then, a further correction is made to account for the difference between the azimuth angle of the PV panel (γp) and the azimuth angle of the sun (γs):

[0071] Aeffective=Acos(θ-hs)cos(γp-γs)

[0072] Among them, γp is the azimuth angle of the photovoltaic panel, and γs is the azimuth angle of the sun. Through these steps, the effective sunlight acquisition area of ​​the photovoltaic panel at different inclination angles can be gradually solved, and its capture efficiency can be approximately calculated under specific conditions.

[0073] Optionally, the S4 includes:

[0074] S41, using enhanced variational autoencoder (EVAE) processing: by introducing advanced regularization technology and dynamic adjustment mechanism, the model's ability to process complex geographic information data and photovoltaic panel cleanliness data is further improved to ensure high-precision prediction;

[0075] First, the geographic information data (including terrain information, vegetation coverage, building shading) and photovoltaic panel cleanliness data are standardized and normalized.

[0076] Then, we construct an enhanced variational autoencoder (EVAE)

[0077] The generation process of the latent variable z uses the approximate posterior distribution qφ(z|x), where φ(z|x) is the parameter of the encoder:

[0078] z~qφ(z∣x)

[0079] For example, the encoder integrates features such as the slope and elevation of the terrain, the density and type of vegetation cover, the height and distribution of buildings, and the cleanliness of photovoltaic panels to generate a latent variable z that condenses the key features of all the input data.

[0080] Next, advanced regularization techniques, including L2 regularization and Dropout, are introduced into the model to prevent overfitting and improve the generalization ability of the model.

[0081] The decoder then remaps the latent variables z back to the data space to generate new data samples x".

[0082] x”~pθ(x|z)

[0083] Where pθ(x|z) represents the generated distribution, and its parameter is θ. Through the decoder, the latent variable z can be restored to data similar to the original input, including features such as terrain information, vegetation coverage, building occlusion, and photovoltaic panel cleanliness. Specifically, the latent variable z can decode the slope and elevation of the terrain, the density and type of vegetation coverage, the height and distribution of buildings, and the cleanliness data of photovoltaic panels, so that the generated data samples are similar to the original input data and contain all important features.

[0084] Finally, the total loss function for model training is constructed

[0085] L(θ,φ;x)=Eqφ(z∣x)[logpθ(x∣z)]-DKL(qφ(z∣x)∥p(z))+Lreg(θ)

[0086] Among them, L(θ,φ;x) is the total loss function, Eqφ(z|x)[logpθ(x|z)] is the reconstruction loss, DKL(qφ(z|x)∥p(z)) is the KL divergence, and Lreg(θ) is the regularization loss.

[0087] S42. Optimization using Generative Adversarial Networks (GAN)

[0088] First, the hidden variable z and data sample x" generated by S41 are used as the input of EGAN. EGAN consists of a generator, a discriminator, and a regulator. The generator G generates a new data sample The discriminator D evaluates the authenticity of the data sample, and the regulator R dynamically adjusts the output of the generator to improve the diversity and authenticity of the generated data. The generator G accepts the hidden variable z and generates new data samples:

[0089]

[0090] Next, the discriminator D evaluates the generated data and the authenticity of the real data x", the goal is to distinguish the real data x" from the generated data Then, the regulator R is introduced to dynamically adjust the output of the generator G so that the generated data is not only realistic but also diverse. The regulator achieves this goal by introducing a diversity constraint term in the loss function of the generator:

[0091] Ldiversity=Ez1,z2~pz(z)[∥G(z1)-G(z2)∥2]

[0092] Among them, z1 and z2 are different latent variables, and Ldiversity measures the diversity between data samples generated by different latent variables.

[0093] Then, the loss function of EGAN is constructed, including the losses of the generator, discriminator and regulator. The loss function LG of the generator, the loss function LD of the discriminator and the loss function LR of the regulator are:

[0094] LG=-Ez~pz(z)[logD(G(z))]+λdiversity*Ldiversity

[0095] LD=-Ex”~pdata(x”)[logD(x”)]-Ez~pz(z)[log(1-D(G(z)))]

[0096] LR=α(∣D(G(z))-0.5*∣+∣Ldiversity-β∣)

[0097] Among them, λdiversity is the weight of the diversity constraint, α is the weight of the regulator loss, and β is the target value of the diversity constraint. The goal of the generator is to generate realistic data samples and ensure the diversity of samples under the diversity constraint. The goal of the discriminator is to maximize its ability to discriminate real data while minimizing its ability to discriminate generated data. The goal of the regulator is to dynamically adjust the output of the generator so that the output of the discriminator is close to 0.5, indicating that the generated data and the real data are indistinguishable, and to ensure that the diversity of the generated data reaches the target value.

[0098] Finally, through repeated adversarial training and dynamic adjustment, the generator, discriminator and regulator are continuously optimized.

[0099] S43. Based on the results of S42, a hybrid model is constructed by combining the latent variables generated by EVAE and the data samples optimized by EGAN to predict the illuminated area of ​​PV panels at different inclination angles and improve the efficiency of PV systems under various environmental conditions. As input, a multi-layer neural network model is constructed by combining terrain information (such as slope and elevation), vegetation cover (such as density and type), building shading (such as height and distribution), and photovoltaic panel cleanliness. The input layer of this model receives the optimized data samples The hidden layer extracts features through nonlinear activation functions (such as ReLU), and the output layer predicts the effective illumination area of ​​the photovoltaic panel at different inclination angles.

[0100] The goal of the model is to maximize the illumination area of ​​the photovoltaic panel, thereby improving the efficiency of the photovoltaic system. To this end, we introduce a loss function that combines the actual illumination data And the illumination area predicted by the model The mean square error (MSE) between them is optimized. The loss function is defined as:

[0101]

[0102] Where n is the number of samples, is the actual illumination area of ​​the ith sample, is the predicted illumination area of ​​the i-th sample.

[0103] Through repeated training and optimization, the model continuously adjusts parameters to make the predicted illumination area as close to the actual value as possible. During the training process, a gradient descent algorithm (such as Adam) is used to optimize the loss function and improve the prediction accuracy of the model. After multiple rounds of training, the model can accurately predict the illumination area of ​​photovoltaic panels at different inclinations, locations, times, and terrains.

[0104] S44, Adaptive Adjustment Strategy

[0105] According to the results of S43, the adaptive adjustment of the tilt angle of photovoltaic panels is achieved through reinforcement learning algorithm to maximize the illumination area and energy capture efficiency.

[0106] First, the hybrid model constructed by S43 is used as input for the prediction of the PV panel angle. The reinforcement learning model includes an intelligent agent that selects the best adjustment strategy to maximize the illuminated area based on the current PV panel inclination angle, environmental information (such as terrain, vegetation cover, building shading, and PV panel cleanliness), and historical data.

[0107] The intelligent agent uses the Q-learning algorithm in reinforcement learning to learn the optimal strategy through interaction with the environment. Specifically, the state s represents the current tilt angle of the photovoltaic panel and environmental information, the action a represents the action of adjusting the tilt angle of the photovoltaic panel, and the reward r represents the change in the illuminated area. The update formula of Q-learning is:

[0108] Q(s,a)←Q(s,a)+α(r+γa'maxQ(s',a')-Q(s,a))

[0109] Among them, α is the learning rate, γ is the discount factor, s′ is the new state after executing action a, and a′ is the action selected in the new state.

[0110] Through multiple training sessions, the intelligent agent gradually learns the optimal adjustment strategy, enabling the photovoltaic panels to always maintain the optimal tilt angle under different environmental conditions to maximize the illuminated area and energy capture efficiency.

[0111] Compared with the prior art, the present invention has the following advantages:

[0112] Adaptive adjustment capability: Through the combination of enhanced variational autoencoder (EVAE) and generative adversarial network (EGAN), the present invention realizes the adaptive adjustment of photovoltaic systems at different angles, locations, times and terrains, thereby improving the efficiency of solar energy capture.

[0113] Efficient data processing: Using EVAE to generate high-quality latent variable representations and EGAN to optimize the authenticity and diversity of data enables the system to more accurately predict and adjust the optimal angle of photovoltaic panels, improving the overall performance and stability of the system.

[0114] Intelligent control: Using advanced machine learning algorithms, the system can analyze and process environmental data in real time, including geographic location, terrain undulations, vegetation coverage, and building obstructions, and automatically adjust the angle of the photovoltaic panels to ensure that the photovoltaic panels are always at the optimal lighting angle.

[0115] Ability to resist environmental changes: The system can make timely adjustments and maintenance based on the real-time cleanliness data of the photovoltaic panels to avoid the impact of dust and dirt on the performance of the photovoltaic panels and ensure the efficient operation of the photovoltaic system.

[0116] Wide applicability: The present invention can adapt to various complex environmental conditions, including different geographical locations, terrains and climatic conditions, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0117] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0118] Figure 1 It is a flow chart of a high-efficiency solar tracking photovoltaic system based on EVAE and EGAN network;

[0119] Figure 2 It is a high-efficiency solar tracking photovoltaic system based on EVAE and EGAN network. The geographical location latitude and longitude map of the photovoltaic system;

[0120] Figure 3 It is a terrain undulation data map of a high-efficiency solar tracking photovoltaic system based on EVAE and EGAN network;

[0121] Figure 4 It is a coverage density map of a high-efficiency solar tracking photovoltaic system based on EVAE and EGAN network;

[0122] Figure 5 It is a high-efficiency solar tracking photovoltaic system based on EVAE and EGAN network.

[0123] Figure 6 It is a sunlight collection area of ​​a high-efficiency solar tracking photovoltaic system based on EVAE and EGAN network;

[0124] Figure 7 This is a network structure diagram of a high-efficiency solar tracking photovoltaic system based on EVAE and EGAN networks. DETAILED DESCRIPTION

[0125] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0126] refer to Figure 1 , a flow chart of a high-efficiency photovoltaic power generation sun tracking method based on EVAE and EGAN network.

[0127] The method comprises the following steps:

[0128] S1. Obtain geographic information data to ensure accurate geographic location, terrain undulations, vegetation coverage and building obstruction;

[0129] S2. Obtain the cleanliness of photovoltaic panels to ensure the efficient operation of the photovoltaic system, avoid the impact of dust and dirt on the performance of photovoltaic panels, and improve energy capture efficiency;

[0130] S3, obtaining the inclination angle of the photovoltaic system and the sunlight acquisition area of ​​the photovoltaic panel at different angles to optimize the solar energy capture efficiency and system performance;

[0131] S4. Use the adversarial training mechanism of enhanced variational autoencoder (EVAE) and GANs to maximize photovoltaic efficiency at different angles, locations, times, and terrains.

[0132] In this implementation, S1 includes the following sub-steps:

[0133] S11. Obtain accurate geographic location;

[0134] First, using the differential GPS (DGPS) system, the base station and the rover work together to obtain high-precision geographic location data. DGPS technology eliminates system errors and environmental noise through differential calculations between the base station and the rover, improving positioning accuracy. The location of the base station is known, and the rover is installed at the photovoltaic system, using the following formula to calculate the precise coordinates after differential calculation:

[0135] Δφ=φbase-φrover

[0136] Δλ=λbase-λrover

[0137] φcorrected=φrover+Δφ*log(Δφ+1)

[0138] λcorrected =λrover+Δλ·log(Δλ+1)

[0139] Among them, φbase represents the latitude of the base station, φrover represents the latitude of the mobile station, λbase represents the longitude of the base station, and λrover represents the longitude of the mobile station. These corrected coordinates can significantly improve the accuracy of the geographic location and provide a reliable data basis for subsequent solar track prediction.

[0140] Then, satellite remote sensing data is acquired, and the area where the photovoltaic system is located is accurately located through high-resolution satellite remote sensing images and image processing technology. Then, in the geographic information system (GIS) platform, the differential GPS data and satellite remote sensing data are integrated to calculate the final precise location. The GIS platform provides powerful data processing and analysis functions, which can effectively integrate data from different sources and improve positioning accuracy:

[0141]

[0142]

[0143] Among them, φ 2 corrected represents the latitude after differential correction, φ 2 satellite represents the latitude of satellite remote sensing images, λ 2 corrected represents the longitude after differential correction, λ 2 Satellite represents the longitude obtained from satellite remote sensing images. This data fusion method combines the advantages of DGPS and satellite remote sensing to provide more accurate geographic location and lay a solid foundation for the accurate prediction of the sun's trajectory.

[0144] The final result is Figure 2 The figure shows the calculation results after the fusion of three types of data. The figure shows the geographic coordinates of the differential GPS corrected position (red dot), the satellite image position (blue dot) and the final position after fusion (green dot). It can be seen from the figure that there is a certain difference between the position corrected by differential GPS and the position obtained by satellite image. The final position after data fusion is between the two, indicating that by combining DGPS and satellite remote sensing data, more accurate geographic location information can be obtained. This result provides a reliable data basis for the accurate prediction of the sun's trajectory, which helps to improve the energy capture efficiency and operation effect of photovoltaic systems.

[0145] S12, obtaining terrain undulation data. Different terrains affect the sunlight capture efficiency of the photovoltaic system. In order to ensure that the photovoltaic panels can capture sunlight at the best angle and position, it is necessary to obtain detailed terrain undulation data;

[0146] First, generate a contour map using terrain elevation data (such as digital elevation model DEM data):

[0147]

[0148] Among them, t represents the interpolation parameter, h represents the height of the contour line, and Z1 and Z2 represent the elevation of adjacent points respectively. Elevation data can be obtained through digital elevation models (DEM) or laser radar (LiDAR), which provide three-dimensional information of the terrain. DEM data is usually stored in the form of a raster, and each pixel represents the height value of a certain point on the ground. The goal of this step is to determine the location of the contour lines, that is, to mark all points on the topographic map whose elevation is equal to a specific height value (such as 100 meters). Through these points, contour lines representing the undulations of the terrain can be drawn.

[0149] Then perform the interpolation calculation:

[0150] x=(1-t)*x1+t*x2

[0151] y=(1-t)*y1+t*y2

[0152] Among them, x and y represent the coordinates of the interpolation points, x1 and x2 represent the horizontal coordinates of the adjacent points, and y1 and y2 represent the vertical coordinates of the adjacent points. The purpose of interpolation calculation is to find the exact position of the contour line between each pair of adjacent points. Through these interpolation points, smooth contour lines can be drawn. This step ensures the accuracy and continuity of the contour lines, so that the topographic map can accurately reflect the actual undulations of the terrain.

[0153] Next, the generated contour data is processed and analyzed:

[0154]

[0155] Among them, slope represents the slope, Δz represents the elevation difference, and Δd represents the horizontal distance. Slope calculation is used to evaluate the steepness of the terrain, which is very important for the layout and installation angle selection of the photovoltaic system. A steeper slope may make it impossible to install the photovoltaic panels stably, while a flatter terrain is conducive to the stability and sunlight capture of the photovoltaic panels. By analyzing the slope data, the best installation location for the photovoltaic panels can be determined to avoid shadows and insufficient light caused by the undulating terrain.

[0156] Terrain relief data such as Figure 3 The figure shows the calculation results of terrain undulation data. The figure shows the contour map generated by digital elevation model (DEM) data. The contour lines in the figure represent areas of different elevations, and the colors from blue to brown represent the changes from low to high. These contour lines can intuitively show the undulations of the terrain and provide important terrain information support for the subsequent layout and installation of photovoltaic systems. By analyzing these contour lines, the installation position and angle of photovoltaic panels can be optimized to ensure maximum sunlight capture and improve the overall efficiency of the system.

[0157] S13. Obtain vegetation coverage and building shading to ensure maximum sunlight capture by the photovoltaic system;

[0158] First, images of the study area were obtained through Google Maps to identify the distribution of vegetation and buildings.

[0159] Then, the image is preprocessed using a simple image processing algorithm:

[0160]

[0161] Among them, Igray represents a grayscale image, and R, G, and B represent the pixel values ​​of red, green, and blue, respectively.

[0162] Next, perform color threshold segmentation:

[0163]

[0164] Among them, Thr represents the binarized image, I represents the pixel value, μ represents the mean of the image, σ represents the standard deviation of the image, and T represents the threshold. This formula uses standardization to make the threshold segmentation more stable.

[0165] Then, calculate the density of vegetation cover and building occlusion:

[0166]

[0167] Where Den represents the density of the target area, represents the sum of the standardized pixel values ​​in the target area, and n represents the number of pixels in the target area.

[0168] Finally, a density detection result map is generated to visualize the density of vegetation coverage and building occlusion, such as Figure 4 As shown in the figure, this figure shows the distribution of vegetation coverage and building shading in the study area. The left figure shows the density of vegetation coverage, and the right figure shows the density of building shading. These density maps intuitively reflect the different types of shading in the study area. By analyzing these maps, the layout and installation of photovoltaic panels can be optimized to ensure maximum sunlight capture and improve the overall efficiency of the system.

[0169] In this implementation, S2 includes the following sub-steps:

[0170] First, collect image data to obtain clean images of photovoltaic panels through drone photography, fixed camera monitoring or manual photography:

[0171] Iclean(x,y)=255

[0172] Among them, Iclean(x,y) represents the value of the pixel (x,y) in the clean photovoltaic panel image, and 255 represents a pure white pixel. Compared with the traditional simple background image formula, this formula uses a fixed high value to represent the clean surface, which improves the standardization and consistency of image generation and innovatively solves the consistency problem of background photovoltaic panel image processing.

[0173] Then, collect images of photovoltaic panels with dust, and use image processing technology to annotate and identify dust-covered areas:

[0174]

[0175] Among them, Idusty(x,y) represents the value of the pixel (x,y) in the dusty photovoltaic panel image, 255 represents the initial white pixel, N and M represent the width and height of the image respectively, x0 and y0 are the dust center positions, and σ controls the range of dust distribution. Compared with the traditional simple binary division formula, this formula uses a combination of logarithmic and Gaussian distribution to improve the accurate simulation of dust distribution and innovatively solve the simulation problem of dust density changes at different positions.

[0176] Next, collect images of photovoltaic panels with dirt and use image processing technology to annotate and identify the dirt-covered areas:

[0177]

[0178] Among them, Idirty(x,y) represents the value of the pixel (x,y) in the dirty photovoltaic panel image, 255 represents the initial white pixel, and N and M represent the width and height of the image respectively. Compared with the traditional simple binary division formula, this formula uses a combination of square root function and sine cosine function to improve the multi-dimensional characteristic simulation of dirt distribution and innovatively solves the complexity of dirt being affected by environmental factors.

[0179] Finally, a cleanliness test result map is generated to visualize the classification results of photovoltaic panel images of different cleanliness states. Through color mapping technology and image processing technology, the identified dust and dirt areas are marked and a comparison map is generated to intuitively display the cleanliness test results of photovoltaic panels, such as Figure 5 As shown in the figure, the surface of photovoltaic panels in three different cleanliness states is shown. The left picture shows a clean photovoltaic panel, with a completely clean surface and no dust or dirt covering it; the middle picture shows a dusty photovoltaic panel, which shows the uneven distribution of dust. The dust density is low but still affects the light transmittance and power generation efficiency of the photovoltaic panel; the right picture shows a dirty photovoltaic panel, with evenly distributed dirt and high density, which significantly affects the light transmittance and power generation efficiency of the photovoltaic panel. Through these images, we can intuitively observe the impact of different cleanliness states on the surface of photovoltaic panels, providing an important reference for the cleaning, maintenance and optimization of photovoltaic systems.

[0180] In this implementation, S3 includes the following sub-steps:

[0181] S31. Obtain the inclination angle of the photovoltaic system to ensure the optimal inclination angle to maximize solar energy capture efficiency and system performance;

[0182] First, select and calibrate the laser rangefinder to ensure that it can accurately measure the height difference and horizontal distance. Then, select two measurement points at the bottom and top of the photovoltaic panel to measure the height difference and horizontal distance respectively. Next, place the laser rangefinder at the bottom measurement point of the photovoltaic panel, ensure that the device is stable and aligned with the top measurement point, measure and record the height difference (Δh), and take the average value of multiple measurements to improve accuracy. Then, place the laser rangefinder horizontally at the bottom measurement point, align it horizontally with the top measurement point, measure and record the horizontal distance (d), and also take the average value of multiple measurements. After that, apply the trigonometric function formula to calculate the inclination angle (θ) of the photovoltaic panel:

[0183]

[0184] Where Δh represents the height difference and d represents the horizontal distance. Compared with the traditional simple trigonometric function formula, this formula introduces a logarithmic function to solve the measurement error caused by the distance and height difference, thereby improving the measurement accuracy. If necessary, convert the inclination angle in radians to angle units:

[0185] θ degrees = θ × (180 / π)

[0186] Finally, the calculated inclination angle is recorded in a data table, including the measurement date, time, location and other information for subsequent analysis, as shown in Table 1:

[0187] Table 1

[0188]

[0189]

[0190] S32, obtaining the sunlight acquisition area of ​​the photovoltaic panel at different angles;

[0191] First, calculate the sun's altitude (hs). Use the geographic location (latitude, longitude), date, and time to calculate the sun's altitude and azimuth:

[0192] hs=arcsin(sin(φ)sin(δ)+cos(φ)cos(δ)cos(H))

[0193] Among them, φ is the geographic latitude, δ is the solar declination angle, and H is the hour angle.

[0194] Next, when the photovoltaic panel inclination angle (θ degrees) and the solar altitude angle (hs) are known, the effective sunlight acquisition area of ​​the photovoltaic panel at different inclination angles is calculated:

[0195] Aeffective=Acos(θdegrees-hs)

[0196] Among them, Aeffective is the effective sunlight acquisition area, A is the actual area of ​​the photovoltaic panel, θdegrees is the inclination angle of the photovoltaic panel, and hs is the solar altitude angle.

[0197] Then, a further correction is made to account for the difference between the azimuth angle of the PV panel (γp) and the azimuth angle of the sun (γs):

[0198] Aeffective=Acos(θ-hs)cos(γp-γs)

[0199] Among them, γp is the azimuth angle of the photovoltaic panel, and γs is the azimuth angle of the sun. Through these steps, the effective sunlight acquisition area of ​​the photovoltaic panel at different inclination angles can be gradually solved, and its capture efficiency can be approximately calculated under specific conditions.

[0200] The final result is Figure 6 As shown in the figure, it is found that as the inclination angle of the photovoltaic panel changes, the effective sunlight acquisition area also changes. When the inclination angle is 30°, the effective sunlight acquisition area is relatively large; when the inclination angle is 45°, the effective sunlight acquisition area is the largest, reaching the same as the actual area; when the inclination angle is 60°, the effective sunlight acquisition area decreases again. This shows that at a specific solar altitude angle, the optimal inclination angle of the photovoltaic panel is 45°, which can maximize the capture of solar energy and improve the efficiency of the photovoltaic system.

[0201] In this embodiment, S4 includes the following contents:

[0202] S41, using enhanced variational autoencoder (EVAE) processing: by introducing advanced regularization technology and dynamic adjustment mechanism, the model's ability to process complex geographic information data and photovoltaic panel cleanliness data is further improved to ensure high-precision prediction;

[0203] First, the geographic information data (including terrain information, vegetation cover, building occlusion) and photovoltaic panel cleanliness data are standardized and normalized to ensure that the data adapts to the input requirements of the enhanced variational autoencoder (EVAE). These preprocessing steps help improve the training effect and stability of the model.

[0204] Then, an enhanced variational autoencoder (EVAE) is constructed. The encoder part of EVAE maps the input data (such as the slope and elevation of the terrain, the density and type of vegetation cover, the height and distribution of buildings, the cleanliness of photovoltaic panels, etc.) to the latent variable space to generate representative latent variables z. These latent variables are not only low-dimensional representations of the data, but also contain the main features and patterns of the data, so that the essence of the data can be better captured in the latent variable space. The generation process of the latent variable z uses an approximate posterior distribution qφ(z|x), where φ(z|x) is the parameter of the encoder:

[0205] z~qφ(z∣x)

[0206] For example, the encoder integrates features such as the slope and elevation of the terrain, the density and type of vegetation cover, the height and distribution of buildings, and the cleanliness of photovoltaic panels to generate a latent variable z that condenses the key features of all the input data.

[0207] Next, advanced regularization techniques, including L2 regularization and Dropout, are introduced into the model to prevent overfitting and improve the generalization ability of the model. L2 regularization reduces the complexity of the model and avoids overfitting of the model on the training data by adding a weight decay term to the loss function. Dropout technology randomly discards a part of neurons during the training process to prevent the coordinated adaptation between neurons, thereby improving the robustness and generalization ability of the model.

[0208] The decoder then remaps the latent variable z back to the data space to generate a new data sample x". This process reversely restores the low-dimensional latent variable to high-dimensional data similar to the input data, allowing EVAE to not only perform data compression and dimensionality reduction, but also data generation and reconstruction:

[0209] x”~pθ(x|z)

[0210] Where pθ(x|z) represents the generated distribution, and its parameter is θ. Through the decoder, the latent variable z can be restored to data similar to the original input, including features such as terrain information, vegetation coverage, building occlusion, and photovoltaic panel cleanliness. Specifically, the latent variable z can decode the slope and elevation of the terrain, the density and type of vegetation coverage, the height and distribution of buildings, and the cleanliness data of photovoltaic panels, so that the generated data samples are similar to the original input data and contain all important features.

[0211] Finally, the total loss function for model training is constructed, combining reconstruction loss, KL divergence and regularization terms. The reconstruction loss measures the difference between the generated data x" and the input data x, the KL divergence measures the difference between the approximate posterior distribution qφ(z|x) and the prior distribution p(z), and the regularization term further prevents overfitting of the model by introducing additional constraints:

[0212] L(θ,φ;x)=Eqφ(z∣x)[logpθ(x∣z)]-DKL(qφ(z∣x)∥p(z))+Lreg(θ)

[0213] Among them, L(θ,φ;x) is the total loss function, Eqφ(z|x)[logpθ(x|z)] is the reconstruction loss, DKL(qφ(z|x)∥p(z)) is the KL divergence, and Lreg(θ) is the regularization loss.

[0214] Through the above process, the enhanced variational autoencoder (EVAE) generates high-quality latent variables and data representations. These latent variables and data representations reflect the complex characteristics of geographic information data (topographic information, vegetation coverage, building occlusion) and the cleanliness of photovoltaic panels, providing reliable basic data for subsequent steps and ensuring the maximization of photovoltaic efficiency at different angles, locations, times, and terrains.

[0215] S42, optimization using generative adversarial networks (GAN);

[0216] According to the results of S41, the hidden variables and data samples generated by EVAE are optimized through the enhanced generative adversarial network (EGAN), which improves the authenticity and diversity of the data and provides a reliable data basis for photovoltaic panel angle prediction.

[0217] First, the hidden variable z and data sample x" generated by S41 are used as the input of EGAN. EGAN consists of a generator, a discriminator, and a regulator. The generator G generates a new data sample The discriminator D evaluates the authenticity of the data sample, and the regulator R dynamically adjusts the output of the generator to improve the diversity and authenticity of the generated data. The generator G accepts the hidden variable z and generates new data samples:

[0218]

[0219] Next, the discriminator D evaluates the generated data and the authenticity of the real data x", the goal is to distinguish the real data x" from the generated data Then, the regulator R is introduced to dynamically adjust the output of the generator G so that the generated data is not only realistic but also diverse. The regulator achieves this goal by introducing a diversity constraint term in the loss function of the generator:

[0220] Ldiversity=Ez1,z2~pz(z)[∥G(z1)-G(z2)∥2]

[0221] Among them, z1 and z2 are different latent variables, and Ldiversity measures the diversity between data samples generated by different latent variables.

[0222] Then, the loss function of EGAN is constructed, including the losses of the generator, discriminator and regulator. The loss function LG of the generator, the loss function LD of the discriminator and the loss function LR of the regulator are:

[0223] LG=-Ez~pz(z)[logD(G(z))]+λdiversity*Ldiversity

[0224] LD=-Ex”~pdata(x”)[logD(x”)]-Ez~pz(z)[log(1-D(G(z)))]

[0225] LR=α(∣D(G(z))-0.5*∣+∣Ldiversity-β∣)

[0226] Among them, λdiversity is the weight of the diversity constraint, α is the weight of the regulator loss, and β is the target value of the diversity constraint. The goal of the generator is to generate realistic data samples and ensure the diversity of samples under the diversity constraint. The goal of the discriminator is to maximize its ability to discriminate real data while minimizing its ability to discriminate generated data. The goal of the regulator is to dynamically adjust the output of the generator so that the output of the discriminator is close to 0.5, indicating that the generated data and the real data are indistinguishable, and to ensure that the diversity of the generated data reaches the target value.

[0227] Finally, through repeated adversarial training and dynamic adjustment, the generator, discriminator, and regulator are continuously optimized. The generator gradually generates more realistic data samples, the discriminator improves its discrimination ability, and the regulator ensures the diversity and authenticity of the generated data. After multiple rounds of training, the generator is able to generate data samples that are highly similar and diverse to the real data distribution. These samples include features such as terrain information, vegetation coverage, building occlusion, and photovoltaic panel cleanliness. They are real and diverse, providing a reliable data basis for subsequent photovoltaic panel angle prediction.

[0228] S43. Based on the results of S42, a hybrid model is constructed by combining the latent variables generated by EVAE and the data samples optimized by EGAN to predict the illuminated area of ​​PV panels at different inclination angles and improve the efficiency of PV systems under various environmental conditions. As input, a multi-layer neural network model is constructed by combining terrain information (such as slope and elevation), vegetation cover (such as density and type), building shading (such as height and distribution), and photovoltaic panel cleanliness. The input layer of this model receives the optimized data samples The hidden layer extracts features through nonlinear activation functions (such as ReLU), and the output layer predicts the effective illumination area of ​​the photovoltaic panel at different inclination angles.

[0229] The goal of the model is to maximize the illumination area of ​​the photovoltaic panel, thereby improving the efficiency of the photovoltaic system. To this end, we introduce a loss function that combines the actual illumination data And the illumination area predicted by the model The mean square error (MSE) between them is optimized. The loss function is defined as:

[0230]

[0231] Where n is the number of samples, is the actual illumination area of ​​the ith sample, is the predicted illumination area of ​​the i-th sample.

[0232] Through repeated training and optimization, the model continuously adjusts parameters to make the predicted illumination area as close to the actual value as possible. During the training process, a gradient descent algorithm (such as Adam) is used to optimize the loss function and improve the prediction accuracy of the model. After multiple rounds of training, the model can accurately predict the illumination area of ​​photovoltaic panels at different inclinations, locations, times, and terrains.

[0233] S44, Adaptive Adjustment Strategy

[0234] According to the results of S43, the adaptive adjustment of the tilt angle of photovoltaic panels is achieved through reinforcement learning algorithm to maximize the illumination area and energy capture efficiency.

[0235] First, the hybrid model constructed by S43 is used as input for the prediction of the PV panel angle. The reinforcement learning model includes an intelligent agent that selects the best adjustment strategy to maximize the illuminated area based on the current PV panel inclination angle, environmental information (such as terrain, vegetation cover, building shading, and PV panel cleanliness), and historical data.

[0236] The intelligent agent uses the Q-learning algorithm in reinforcement learning to learn the optimal strategy through interaction with the environment. Specifically, the state s represents the current tilt angle of the photovoltaic panel and environmental information, the action a represents the action of adjusting the tilt angle of the photovoltaic panel, and the reward r represents the change in the illuminated area. The update formula of Q-learning is:

[0237] Q(s,a)←Q(s,a)+α(r+γa'maxQ(s',a')-Q(s,a))

[0238] Among them, α is the learning rate, γ is the discount factor, s′ is the new state after executing action a, and a′ is the action selected in the new state.

[0239] Through multiple training sessions, the intelligent agent gradually learns the optimal adjustment strategy, enabling the photovoltaic panels to always maintain the optimal tilt angle under different environmental conditions to maximize the illuminated area and energy capture efficiency.

[0240] After training, the intelligent agent's photovoltaic panel tilt adjustment strategy under different environmental conditions has been significantly optimized. The following is a part of the training results, showing the photovoltaic panel tilt adjustment strategy and the corresponding illumination area under different terrain, vegetation coverage, building shading and photovoltaic panel cleanliness conditions, as shown in Table 2.

[0241] Table 2

[0242]

[0243]

[0244] From the above training results, it can be seen that under different environmental conditions, the intelligent agent can adjust the inclination angle of the photovoltaic panel according to environmental information, thereby significantly improving the illumination area and energy capture efficiency.

[0245] It should be noted that the above content only illustrates the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.

Claims

1. An efficient photovoltaic power generation sun tracking method based on EVAE and EGAN network, characterized by: The method comprises the following steps: S1. Obtain geographic information data to ensure accurate geographic location, terrain undulations, vegetation coverage and building obstruction; S2. Obtain the cleanliness of photovoltaic panels to ensure the efficient operation of the photovoltaic system, avoid the impact of dust and dirt on the performance of photovoltaic panels, and improve energy capture efficiency; S3, obtaining the inclination angle of the photovoltaic system and the sunlight acquisition area of ​​the photovoltaic panel at different angles to optimize the solar energy capture efficiency and system performance; S4. Use the adversarial training mechanism of enhanced variational autoencoder EVAE and GANs to maximize photovoltaic efficiency at different angles, locations, times and terrains.

2. The high-efficiency photovoltaic power generation sun tracking method based on EVAE and EGAN network according to claim 1, characterized in that: The S1 related parameters include obtaining accurate geographic location: First, using the differential GPS system, the base station and the rover work together to obtain high-precision geographic location data. Δφ=φbase-φrover Δλ=λbase-λrover φcorrected=φrover+Δφ*log(Δφ+1) λcorrected=λrover+Δλ·log(Δλ+1) Where φbase represents the latitude of the base station, φrover represents the latitude of the rover station, λbase represents the longitude of the base station, and λrover represents the longitude of the rover station; Then, satellite remote sensing data is acquired, and the area where the photovoltaic system is located is accurately located through high-resolution satellite remote sensing images and image processing technology. Then, in the geographic information system (GIS) platform, differential GPS data and satellite remote sensing data are integrated to calculate the final precise location; Among them, φ 2 corrected represents the latitude after differential correction, φ 2 satellite represents the latitude of satellite remote sensing images, λ 2 corrected represents the longitude after differential correction, λ 2 Satellite represents the longitude obtained from satellite remote sensing images. This data fusion method combines the advantages of DGPS and satellite remote sensing, and can provide more precise geographic location, laying a solid foundation for the accurate prediction of the sun's trajectory.

3. The high-efficiency photovoltaic power generation sun tracking method based on EVAE and EGAN network according to claim 1, characterized in that: The S1 related parameters include terrain relief data: First, generate contour maps from terrain elevation data, including digital elevation model (DEM) data: Among them, t represents the interpolation parameter, h represents the height of the contour line, Z1 and Z2 represent the elevation of adjacent points respectively. The elevation data can be obtained through the digital elevation model DEM or LiDAR. These data provide three-dimensional information of the terrain and determine the location of the contour line, that is, mark all points with elevations equal to a specific height value on the topographic map. Through these points, contour lines representing the terrain undulations are drawn, and then interpolation calculations are performed: x=(1-t)*x1+t*x2 y=(1-t)*y1+t*y2 Among them, x and y represent the coordinates of the interpolation point, x1 and x2 represent the horizontal coordinates of the adjacent points, y1 and y2 represent the vertical coordinates of the adjacent points. The purpose of interpolation calculation is to find the exact position of the contour line between each pair of adjacent points, and draw smooth contour lines through these interpolation points; Next, the generated contour data is processed and analyzed: Among them, slope represents the slope, Δz represents the elevation difference, and Δd represents the horizontal distance. By analyzing the slope data, the optimal installation location of the photovoltaic panels can be determined to avoid shadows and insufficient light caused by undulating terrain.

4. The high-efficiency photovoltaic power generation sun tracking method based on EVAE and EGAN network according to claim 1, characterized in that: The S1 related parameters include the permission to obtain vegetation coverage and building occlusion: First, images of the study area were obtained through Google Maps to identify the distribution of vegetation and buildings; Then, the image is preprocessed using a simple image processing algorithm: Among them, Igray represents the grayscale image, R, G and B represent the pixel values ​​of red, green and blue respectively; Next, perform color threshold segmentation: Among them, Thr represents the binarized image, I represents the pixel value, μ represents the mean of the image, σ represents the standard deviation of the image, and T represents the threshold. This formula uses standardization to make the threshold segmentation more stable. Then, calculate the density of vegetation cover and building occlusion: Where Den represents the density of the target area, represents the sum of the normalized pixel values ​​in the target area, and n represents the number of pixels in the target area; Finally, a density detection result map is generated to visualize the density of vegetation cover and building occlusion respectively.

5. The high-efficiency photovoltaic power generation sun tracking method based on EVAE and EGAN network according to claim 1, characterized in that: The S2 related parameters include the cleanliness of the photovoltaic panels: First, collect image data to obtain clean images of photovoltaic panels through drone photography, fixed camera monitoring or manual photography: Iclean(x,y)=255 Among them, Iclean(x,y) represents the value of the pixel (x,y) in the clean photovoltaic panel image, and 255 represents a pure white pixel. A fixed high value is used to represent the clean surface, which improves the standardization and consistency of image generation and innovatively solves the consistency problem of background photovoltaic panel image processing; Then, collect images of photovoltaic panels with dust, and use image processing technology to annotate and identify dust-covered areas: Where Idusty(x,y) represents the value of the pixel (x,y) in the dusty photovoltaic panel image, 255 represents the initial white pixel, N and M represent the width and height of the image, x0 and y0 are the dust center positions, and σ controls the range of dust distribution; Next, collect images of photovoltaic panels with dirt and use image processing technology to annotate and identify the dirt-covered areas: Where Idirty(x,y) represents the value of the pixel (x,y) in the dirty photovoltaic panel image, 255 represents the initial white pixel, and N and M represent the width and height of the image, respectively; Finally, a cleanliness inspection result map is generated to visualize the classification results of photovoltaic panel images with different cleanliness states.

6. The high-efficiency photovoltaic power generation sun tracking method based on EVAE and EGAN network according to claim 1, characterized in that: The S3 includes: S31. Obtain the inclination angle of the photovoltaic system to ensure the optimal inclination angle to maximize solar energy capture efficiency and system performance; First, select and calibrate the laser rangefinder to ensure that it can accurately measure the height difference and horizontal distance. Then, select two measurement points at the bottom and top of the photovoltaic panel to measure the height difference and horizontal distance respectively. Among them, Δh represents the height difference, and d represents the horizontal distance. The logarithmic function is introduced to solve the measurement error caused by the distance and height difference, thereby improving the measurement accuracy. If necessary, the inclination angle in radians can be converted into angle units: θ degrees = θ × (180 / π) Finally, the calculated inclination angle is recorded in a data table, including the measurement date, time, and location information, for subsequent analysis; S32, obtaining the sunlight acquisition area of ​​the photovoltaic panel at different angles; First, calculate the sun's altitude angle hs, using the geographic location including latitude and longitude, date and time to calculate the sun's altitude angle and azimuth: hs=arcsin(sin(φ)sin(δ)+cos(φ)cos(δ)cos(H)) Among them, φ is the geographic latitude, δ is the solar declination angle, and H is the hour angle; Next, when the photovoltaic panel inclination angle θ degrees and the solar altitude angle hs are known, the effective sunlight acquisition area of ​​the photovoltaic panel at different inclination angles is calculated: Aeffective=Acos(θdegrees-hs) Among them, Aeffective is the effective sunlight acquisition area, A is the actual area of ​​the photovoltaic panel, θdegrees is the inclination angle of the photovoltaic panel, and hs is the solar altitude angle; Then, a further correction is made to account for the difference between the azimuth angle γp of the PV panel and the azimuth angle γs of the sun: Aeffective=Acos(θ-hs)cos(γp-γs) Among them, γp is the azimuth angle of the photovoltaic panel, and γs is the azimuth angle of the sun. Through these steps, the effective sunlight acquisition area of ​​the photovoltaic panel at different inclination angles is gradually solved, and its capture efficiency is approximately calculated under specific conditions.

7. The high-efficiency solar tracking system based on EVAE and EGAN network according to claim 1, characterized in that: The S4 includes: S41. Using enhanced variational autoencoder (EVAE) processing: By introducing advanced regularization technology and dynamic adjustment mechanism, the model's ability to process complex geographic information data and photovoltaic panel cleanliness data is further improved to ensure high-precision prediction; First, the geographic information data including terrain information, vegetation coverage, building shading and photovoltaic panel cleanliness data are standardized and normalized; Then, an enhanced variational autoencoder EVAE is constructed; The generation process of the latent variable z uses the approximate posterior distribution qφ(z|x), where φ(z|x) is the parameter of the encoder: z~qφ(z∣x) The encoder integrates features such as the slope and elevation of the terrain, the density and type of vegetation cover, the height and distribution of buildings, and the cleanliness of photovoltaic panels to generate a latent variable z, which condenses the key features of all input data; Next, we introduce advanced regularization techniques into the model, including L2 regularization and Dropout, to prevent overfitting and improve the generalization ability of the model. The decoder then remaps the latent variable z back to the data space to generate a new data sample x"; x”~pθ(x|z) Where pθ(x|z) represents the generated distribution, with parameter θ. Through the decoder, the latent variable z is restored to data similar to the original input, including terrain information, vegetation coverage, building occlusion, and photovoltaic panel cleanliness characteristics. Specifically, the latent variable z decodes the slope and elevation of the terrain, the density and type of vegetation coverage, the height and distribution of buildings, and the cleanliness data of photovoltaic panels, so that the generated data sample is similar to the original input data and contains all important features; Finally, the total loss function for model training is constructed L(θ,φ;x)=Eqφ(z∣x)[logpθ(x∣z)]-DKL(qφ(z∣x)∥p(z))+Lreg(θ) Among them, L(θ,φ;x) is the total loss function, Eqφ(z|x)[logpθ(x|z)] is the reconstruction loss DKL(qφ(z|x)||p(z)) is the KL divergence, and Lreg(θ) is the regularization loss; S42, using the generative adversarial network GAN optimization; First, the hidden variable z and data sample x″ generated by S41 are used as the input of EGAN. EGAN consists of a generator, a discriminator, and a regulator. The generator G generates a new data sample. The discriminator D evaluates the authenticity of the data samples, the regulator R dynamically adjusts the output of the generator to improve the diversity and authenticity of the generated data, and the generator G accepts the hidden variable z and generates new data samples: Next, the discriminator D evaluates the generated data and the authenticity of the real data x″. The goal is to distinguish the real data x″ from the generated data Then, the regulator R is introduced to dynamically adjust the output of the generator G so that the generated data is not only realistic but also diverse. The regulator achieves this goal by introducing a diversity constraint term in the loss function of the generator: Ldiversity=Ez1,z2~(pz(z)[||G(z1)-G(z2)||2] Among them, z1 and z2 are different latent variables, and Ldiversity measures the diversity between data samples generated by different latent variables. Then, the loss function of EGAN is constructed, including the losses of the generator, discriminator and regulator. The loss function LG of the generator, the loss function LD of the discriminator and the loss function LR of the regulator are: LG=-Ez~pz(z)[logD(G(z))]+λdiversity*Ldiversity LD=-Ex″~pdata(x″)[logD(x″)]-Ez~pz(z)[log1-D(G(z)))] LR=α(ID(G(z))-0.5*|+|Ldiversity-βI) Among them, λdiversity is the weight of the diversity constraint, α is the weight of the regulator loss, β is the target value of the diversity constraint, the goal of the generator is to generate realistic data samples and ensure the diversity of samples under the diversity constraint, the goal of the discriminator is to maximize its ability to discriminate real data while minimizing its ability to discriminate generated data, and the goal of the regulator is to dynamically adjust the output of the generator so that the output of the discriminator is close to 0.5, indicating that the generated data and the real data are indistinguishable, and to ensure that the diversity of the generated data reaches the target value; Finally, through repeated adversarial training and dynamic adjustment, the generator, discriminator, and regulator are continuously optimized; S43. Based on the results of S42, a hybrid model is constructed by combining the latent variables generated by EVAE and the data samples optimized by EGAN to predict the illuminated area of ​​PV panels at different inclination angles and improve the efficiency of PV systems under various environmental conditions. First, the data samples optimized by EGAN are As input, a multi-layer neural network model is constructed by combining terrain information, vegetation coverage, building shading, and photovoltaic panel cleanliness characteristics. The input layer of this model accepts optimized data samples. The hidden layer extracts features through nonlinear activation functions, and the output layer predicts the effective illumination area of ​​the photovoltaic panel at different inclination angles. The goal of the model is to maximize the illumination area of ​​the photovoltaic panel, thereby improving the efficiency of the photovoltaic system. A loss function is introduced, combined with actual illumination data And the illumination area predicted by the model The mean square error MSE between is optimized. The loss function is defined as: Where n is the number of samples, is the actual illumination area of ​​the ith sample, is the predicted illumination area of ​​the i-th sample; Through repeated training and optimization, the model continuously adjusts parameters to make the predicted illumination area as close to the actual value as possible. During the training process, the gradient descent algorithm is used to optimize the loss function to improve the prediction accuracy of the model. S44, adaptive adjustment strategy; Based on the results of S43, the adaptive adjustment of the photovoltaic panel tilt angle is achieved through a reinforcement learning algorithm to maximize the illumination area and energy capture efficiency; First, the result of the hybrid model constructed by S43 for PV panel angle prediction is used as input. The reinforcement learning model includes an intelligent agent, which selects the best adjustment strategy to maximize the illuminated area according to the current PV panel inclination angle, environmental information including terrain, vegetation coverage, building shading and PV panel cleanliness, and historical data; The intelligent agent uses the Q-learning algorithm in reinforcement learning to learn the optimal strategy through interaction with the environment. Specifically, the state s represents the current inclination angle of the photovoltaic panel and environmental information, the action a represents the action of adjusting the inclination angle of the photovoltaic panel, and the reward r represents the change in the illuminated area. The update formula of Q-learning is: Q(s,a)←Q(S,a)+α(r+γa′max Q(s',a')-Q(s,a)) Among them, α is the learning rate, γ is the discount factor, s′ is the new state after executing action a, and a′ is the action selected in the new state; Through multiple training sessions, the intelligent agent gradually learns the optimal adjustment strategy, enabling the photovoltaic panels to always maintain the optimal tilt angle under different environmental conditions to maximize the illuminated area and energy capture efficiency.

Citation Information

Cited By

  • Method, system and device for improving alignment precision of Beidou coordinate system and medium

    CN120044565A

  • A method, system, device and medium for improving alignment accuracy of a Beidou coordinate system

    CN120044565B

  • Planning construction method and device for photovoltaic panel in photovoltaic power station

    CN120706840A

  • A method and device for planning and constructing photovoltaic panels in a photovoltaic power station

    CN120706840B

  • Photovoltaic tracking support control method, device, system, equipment and program product

    CN120848597A