Autonomous unmanned aerial vehicle hyperspectral inspection method for photovoltaic power station anomaly detection

Abstract

The present application relates to a kind of unmanned aerial vehicle hyperspectral inspection methods, more particularly to a kind of full autonomous unmanned aerial vehicle hyperspectral inspection methods for photovoltaic power station anomaly detection.The method comprises the following steps:1, by unmanned airfield to the unmanned aerial vehicle task scheduling, the route and inspection cycle of unmanned aerial vehicle are planned;2, unmanned aerial vehicle flies to the target area to be detected according to the planned route and inspection cycle, and the spectral data of target area are collected;3, the collected spectral data are radiometrically corrected and normalized;4, based on the spectral data after radiometrically corrected and normalized, the method of spectral feature analysis and machine learning is used to detect dust, pollutants and vegetation on the surface of photovoltaic panel;5, based on the detection result, the influence degree of dust, pollutants and vegetation on the photoelectric conversion efficiency of photovoltaic panel is evaluated.The present application can provide scientific basis for the efficient operation of photovoltaic power station and the health management of ecological environment.

Description

Technical Field

[0001] This invention relates to a hyperspectral inspection method using unmanned aerial vehicles (UAVs), specifically a fully autonomous UAV hyperspectral inspection method for anomaly detection in photovoltaic power plants. Background Technology

[0002] With the large-scale application of photovoltaic power plants in special environments such as deserts, the power plant components are easily affected by environmental factors, such as internal vegetation growth, dust accumulation, and bird activity. These factors can affect the power generation efficiency of photovoltaic panels and the ecological stability of the power plant. Dust and pollutant accumulation on photovoltaic panels not only reduces photoelectric conversion efficiency but may also accelerate component aging and increase maintenance costs. Vegetation growth can cause shading, further reducing the output power of photovoltaic panels and also posing a fire hazard.

[0003] Traditional manual inspection methods are inefficient, labor-intensive, and difficult to achieve large-scale, precise monitoring, especially in complex environments such as deserts and mountains where inspections are even more challenging.

[0004] Therefore, there is an urgent need for a fully autonomous, efficient, and accurate inspection system to ensure the long-term efficient operation of photovoltaic power plants and the healthy management of the ecological environment. Summary of the Invention

[0005] The purpose of this invention is to solve the technical problems of low efficiency, high labor intensity, and difficulty in achieving large-scale and precise monitoring by manual inspection in the existing technology, especially in complex environments such as deserts and mountains. The invention provides a fully autonomous UAV hyperspectral inspection method for anomaly detection in photovoltaic power plants.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A fully autonomous UAV hyperspectral inspection method for anomaly detection in photovoltaic power plants is characterized by the following steps:

[0008] Step 1: Conduct task scheduling for drones through drone airports, and plan drone routes and inspection cycles;

[0009] Step 2: The UAV flies to the target area where the photovoltaic panels in the photovoltaic power station to be inspected are located according to the planned route and inspection cycle. Using the hyperspectral imaging sensor on the UAV, it collects spectral data of the photovoltaic panels and the surrounding environment in the target area.

[0010] Step 3: Perform radiometric correction and normalization on the collected spectral data to ensure the accuracy and consistency of the data;

[0011] Step 4: Based on the spectral data after radiometric correction and normalization, use spectral feature analysis and machine learning methods to detect dust, pollutants, and vegetation on the surface of the photovoltaic panel and obtain the detection results.

[0012] Step 5: Based on the detection results, assess the impact of dust accumulation, pollutants, and vegetation on the photovoltaic conversion efficiency of the photovoltaic panels, and issue a maintenance alarm when the impact of dust accumulation and / or pollutants and / or vegetation exceeds a preset threshold, thus completing the abnormal detection of the photovoltaic power station.

[0013] Furthermore, step 3 specifically involves:

[0014] Step 3.1: Use the following multinomial regression model to perform radiometric correction on the collected spectral data to compensate for spectral variations under different illumination conditions and ensure the accuracy and consistency of the data:

[0015] R cor =R raw -P(I)

[0016] In the formula: R cor R is the corrected spectral reflectance. raw P(I) represents the original spectral reflectance and is a polynomial regression model of the illuminance I.

[0017] Step 3.2: Perform min-max normalization on the radiometrically corrected spectral data to eliminate biases under different acquisition conditions. The mathematical expression for normalization is:

[0018]

[0019] In the formula: R norm R represents the normalized spectral reflectance. min and R max These represent the minimum and maximum reflectance values ​​in the spectral data, respectively.

[0020] Furthermore, step 4 specifically involves:

[0021] Step 4.1: Based on the spectral data after radiometric correction and normalization, perform spectral characteristic analysis on the vegetation and photovoltaic panels, extract the differences in spectral characteristics, and then, based on these differences, use a mapping index to enhance the spectral contrast between the vegetation and the photovoltaic panels. The formula for the mapping index is as follows:

[0022]

[0023] In the formula: ESVI is the cartographic index, R nir R represents the near-infrared reflectance. red R represents the red band reflectivity. swirR represents the reflectivity in the shortwave infrared band. green The green band reflectance is represented by α and β, which are both adjustment factors.

[0024] Step 4.2: Determine the optimal segmentation threshold for the mapping index ESVI using the Otsu method to achieve effective segmentation of vegetation areas.

[0025]

[0026]

[0027] In the formula: T * For the threshold, Let w1(T) be the inter-class variance and w2(T) be the threshold T. * The weights of the background and foreground are set, and μ1(T) and μ2(T) are the average gray values ​​of the background and foreground, respectively.

[0028] Step 4.3: Using machine learning methods, the spectral data after vegetation segmentation is detected to achieve the detection of dust, pollutants and vegetation obstruction on the surface of photovoltaic panels, obtain the detection results, and complete the anomaly detection of the photovoltaic power station.

[0029] Further, in step 4.3, the machine learning method calculates the area of ​​dust accumulation and pollutant shading on the photovoltaic panel surface using the following formula:

[0030]

[0031] The degree of vegetation shading is calculated using the following formula:

[0032]

[0033] In the formula: A is the area of ​​the photovoltaic panel surface where dust and pollutants are blocked, p i Let C represent the abnormal coverage of the i-th pixel, n be the total number of pixels, and R be the actual ground area represented by each pixel in the spectral data; veg A represents the percentage of vegetation cover. veg A represents the area covered by vegetation. total Total area of ​​photovoltaic panels.

[0034] Furthermore, in step 4.3, the abnormal coverage of the i-th pixel is obtained in the following way:

[0035] A1. By constructing a time-series-based spectral change detection model, the current spectral characteristics are compared with historical data to identify spectral changes caused by dust accumulation and pollutants. The spectral change detection model uses a dynamic time warping method, as shown in the following formula:

[0036]

[0037] In the formula: x i and y (i) All are time series signals, and π(i) is the alignment path;

[0038] A2. Based on the spectral changes caused by ash accumulation and pollutants, determine whether pollutants are present. If so, compare the extracted spectral features with existing spectral libraries and calculate spectral similarity using the spectral correlation coefficient, as shown in the following expression:

[0039]

[0040] In the formula: r is the correlation coefficient between the currently acquired spectral features and the reference spectrum, x i and y i These are the values ​​of the current spectral feature and the reference spectrum at the i-th pixel, respectively. and x i and y i The mean;

[0041] Identify the contaminant type based on the value of r; if not, proceed to step A3.

[0042] A3. The following support vector regression formula is used to perform regression calculations to determine the coverage degree of dust accumulation and contaminants, i.e., the abnormal coverage of the i-th pixel:

[0043]

[0044] In the formula: α i For the coefficients of the support vectors, K(x) i ,x) is the kernel function, b is the bias term, and x is the input spectral feature.

[0045] Furthermore, in step 5, the percentage of photoelectric conversion efficiency loss is calculated using the following formula:

[0046]

[0047] In the formula: η is the percentage of photoelectric conversion efficiency loss, P a P represents the photoelectric conversion efficiency under the current conditions. i This represents the photoelectric conversion efficiency under ideal conditions.

[0048] The beneficial effects of this invention are:

[0049] By collaborating with fully autonomous drones and unmanned aerial vehicles (UAVs), comprehensive autonomous inspections of both the inside and outside of photovoltaic power plants are achieved, reducing labor costs. Hyperspectral imaging technology, combined with spectral variation analysis and machine learning algorithms, enables precise monitoring of photovoltaic power plants, including dust accumulation on photovoltaic panels, pollutant buildup, and vegetation cover. The system provides a scientific basis for the efficient operation of photovoltaic power plants and the healthy management of the ecological environment, helping to improve the power generation efficiency of photovoltaic panels and reduce maintenance costs. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of a fully autonomous UAV hyperspectral inspection system for detecting anomalies in photovoltaic power plants;

[0051] Figure 2 This is a flowchart of a fully autonomous UAV hyperspectral inspection system for detecting anomalies in photovoltaic power plants. Detailed Implementation

[0052] To make the objectives, advantages, and features of this invention clearer, the following detailed description, in conjunction with the accompanying drawings and specific embodiments, provides a fully autonomous UAV hyperspectral inspection method for anomaly detection in photovoltaic power plants. The advantages and features of this invention will become clearer according to the following specific embodiments.

[0053] See Figure 2 This embodiment of a fully autonomous UAV hyperspectral inspection method for anomaly detection in photovoltaic power plants specifically includes the following steps:

[0054] (1) See Figure 1 The system manages and schedules fully autonomous drones through unmanned aerial vehicle (UAV) airports (tower-type or ground-based). These UAV airports have mission planning capabilities, automatically planning flight paths based on preset cycles or inspection needs, and assigning drones to perform inspection tasks. The fully autonomous drones require no human intervention; through automatic flight path planning, they autonomously fly to the area where the photovoltaic panels are located to carry out inspection tasks. The drones use hyperspectral imaging sensors to acquire spectral data of the photovoltaic panels and their surrounding environment, which is used to detect anomalies such as dust accumulation, pollutant buildup, and vegetation obstruction on the photovoltaic panel surface.

[0055] (2) The UAV flies to the target area where the photovoltaic panels in the photovoltaic power station to be inspected are located according to the planned route and inspection cycle, and uses the hyperspectral imaging sensor on the UAV to collect spectral data of the photovoltaic panels and their surrounding environment in the target area.

[0056] (3) The collected data is first radiometrically corrected to compensate for spectral changes under different lighting conditions and to ensure the consistency and accuracy of the data.

[0057] Radiometric correction uses a multinomial regression model to correct spectral biases under different light intensities. The specific formula is as follows:

[0058] R cor =R raw -P(I)

[0059] In the formula: R cor R is the corrected spectral reflectance. raw P(I) represents the original spectral reflectance and is a polynomial regression model of the illuminance I.

[0060] Next, the data is subjected to min-max normalization to eliminate biases caused by different acquisition conditions and improve the accuracy of subsequent analysis. The mathematical representation of normalization is as follows:

[0061]

[0062] In the formula: R norm R represents the normalized spectral reflectance. min and R max These are the minimum and maximum reflectance values ​​in the spectral data, respectively.

[0063] The preprocessed data can be used for subsequent anomaly detection and analysis.

[0064] (4) Based on remote sensing image data, the spectral reflectance characteristics of desert vegetation and photovoltaic panels were analyzed in depth, and typical spectral feature differences were extracted. In view of the high reflectance of vegetation and the low reflectance of photovoltaic panels, a new mapping index was designed to enhance the spectral contrast between vegetation signals and photovoltaic panels and suppress the interference of photovoltaic panels.

[0065] The formula for the cartographic index is as follows:

[0066]

[0067] In the formula: ESVI is the cartographic index, R nir R represents the near-infrared reflectance. red R represents the red band reflectivity. swir R represents the reflectivity in the shortwave infrared band. green α represents the green band reflectance, and β are both adjustment factors. α and β are used to control the suppression effect of the short-wave infrared band on the photovoltaic panel and to normalize the exponent, respectively, to prevent the denominator from approaching zero.

[0068] By statistically analyzing typical vegetation and background samples, the optimal segmentation threshold of the mapping index is determined using the Otsu method, thereby achieving effective segmentation of the vegetation area.

[0069] The formula for calculating the optimal threshold of the Otsu method is as follows:

[0070]

[0071]

[0072] In the formula: T * For the threshold, Let w1(T) be the inter-class variance and w2(T) be the threshold T. * The weights of the background and foreground are given by μ1(T) and μ2(T), which are the average gray values ​​of the background and foreground, respectively.

[0073] (5) Accurate detection of dust accumulation and pollutant identification in photovoltaic power plants can be achieved through machine learning methods.

[0074] Dust accumulation monitoring assesses the degree of dust accumulation by analyzing the spectral characteristics of the photovoltaic panel surface. Changes in spectral characteristics reflect the dust coverage on the photovoltaic panel. The impact of dust accumulation is quantified by calculating the characteristic spectral reflectance of the photovoltaic panel surface and comparing it with the baseline reflectance of a clean photovoltaic panel. A time-series-based spectral change detection model is constructed to compare current spectral characteristics with historical data to identify spectral changes caused by dust accumulation and contaminants. The change detection uses the Dynamic Time Warping (DTW) method, as shown in the following formula:

[0075]

[0076] In the formula: x i and y (i) All are time series signals, and π(i) is the alignment path. Differences in ash or pollutants are identified by calculating the minimum distance of spectral changes.

[0077] If contaminants are present, the extracted spectral features are compared with existing spectral libraries (whose spectral data serves as reference spectra), and spectral similarity is calculated using the spectral correlation coefficient, as shown in the following expression:

[0078]

[0079] In the formula: r is the correlation coefficient between the currently acquired spectral features and the reference spectrum, x i and y i These are the values ​​of the current spectral feature and the reference spectrum at the i-th pixel, respectively. and x i and y i The mean;

[0080] High correlation (r close to 1) indicates a match with a certain type in the pollutant spectral library.

[0081] To quantify the degree of ash accumulation and contaminants, support vector regression (SVR) can be used for regression estimation. The regression model is expressed as follows:

[0082]

[0083] In the formula: α i For the coefficients of the support vectors, K(x) i , x) is the kernel function, b is the bias term, and x is the input spectral feature; this model can accurately estimate the coverage of dust and pollutants, that is, the abnormal coverage of the i-th pixel.

[0084] Regarding the impact of vegetation, the degree of shading on photovoltaic panels and its impact on power generation efficiency are assessed by analyzing vegetation growth within the photovoltaic power station. Multi-temporal remote sensing data are used to analyze vegetation dynamics, and vegetation cover is calculated using the vegetation mapping index (ESVI from the previous step). The formula for calculating vegetation shading is as follows:

[0085]

[0086] In the formula: C veg A represents the percentage of vegetation cover. veg A represents the area covered by vegetation. total Total area of ​​photovoltaic panels.

[0087] The trained model is applied to remote sensing imagery of photovoltaic power plant areas to accurately extract anomalous regions, such as areas with dust accumulation, pollutants, and vegetation, and to assess their impact on power generation efficiency and safety. Based on the spatial resolution of the remote sensing imagery, the coverage area of ​​anomalous regions can be calculated using the following formula:

[0088]

[0089] In the formula: A represents the area of ​​the photovoltaic panel surface affected by accumulated dust, pollutants, and vegetation blockage; p i Let represent the abnormal coverage of the i-th pixel, n be the total number of pixels, and R be the size of the actual ground area represented by each pixel in the spectral data.

[0090] (6) The overall impact assessment of dust accumulation, pollutants, and vegetation can be conducted by examining changes in photoelectric conversion efficiency, as defined below:

[0091]

[0092] In the formula: η is the percentage of photoelectric conversion efficiency loss, P a P represents the photoelectric conversion efficiency under the current conditions. i This represents the photoelectric conversion efficiency under ideal conditions.

Claims

1. A fully autonomous UAV hyperspectral inspection method for anomaly detection in photovoltaic power plants, characterized in that, Includes the following steps: Step 1: Conduct task scheduling for drones through drone airports, and plan drone routes and inspection cycles; Step 2: The UAV flies to the target area where the photovoltaic panels in the photovoltaic power station to be inspected are located according to the planned route and inspection cycle. Using the hyperspectral imaging sensor on the UAV, it collects spectral data of the photovoltaic panels and the surrounding environment in the target area. Step 3: Perform radiometric correction and normalization on the collected spectral data to ensure the accuracy and consistency of the data; Step 4: Based on the spectral data after radiometric correction and normalization, use spectral feature analysis and machine learning methods to detect dust, pollutants, and vegetation on the surface of the photovoltaic panel and obtain the detection results. Step 4.1: Based on the spectral data after radiometric correction and normalization, perform spectral characteristic analysis on the vegetation and photovoltaic panels, extract the differences in spectral characteristics, and then, based on these differences, use a mapping index to enhance the spectral contrast between the vegetation and the photovoltaic panels. The formula for the mapping index is as follows: In the formula: ESVI is the cartographic index, R nir R represents the near-infrared reflectance. red R represents the red band reflectivity. swir R represents the reflectivity in the shortwave infrared band. green The green band reflectance is represented by α and β, which are both adjustment factors. Step 4.2: Determine the optimal segmentation threshold for the mapping index ESVI using the Otsu method to achieve effective segmentation of vegetation areas. In the formula: T * For the threshold, Let w1(T) be the inter-class variance and w2(T) be the threshold T. * The weights of the background and foreground are defined, where μ1(T) and μ2(T) are the average gray values ​​of the background and foreground, respectively. Step 4.3: Using machine learning methods, the spectral data after vegetation segmentation is detected to achieve the detection of dust, pollutants and vegetation obstruction on the surface of photovoltaic panels, obtain the detection results, and complete the anomaly detection of the photovoltaic power station. Step 5: Based on the detection results, assess the impact of dust accumulation, pollutants, and vegetation on the photovoltaic conversion efficiency of the photovoltaic panels, and issue a maintenance alarm when the impact of dust accumulation and / or pollutants and / or vegetation exceeds a preset threshold, thus completing the abnormal detection of the photovoltaic power station.

2. The fully autonomous UAV hyperspectral inspection method for anomaly detection in photovoltaic power plants according to claim 1, characterized in that, Step 3 specifically involves: Step 3.1: Use the following multinomial regression model to perform radiometric correction on the collected spectral data to compensate for spectral variations under different illumination conditions and ensure the accuracy and consistency of the data: R cor =R raw -P(I) In the formula: R cor R is the corrected spectral reflectance. raw Let P(I) be the original spectral reflectance, and let P(I) be the polynomial regression model of the light intensity I. Step 3.2: Perform min-max normalization on the radiometrically corrected spectral data to eliminate biases under different acquisition conditions. The mathematical expression for normalization is: In the formula: R norm R represents the normalized spectral reflectance. min and R max These represent the minimum and maximum reflectance values ​​in the spectral data, respectively.

3. The fully autonomous UAV hyperspectral inspection method for anomaly detection in photovoltaic power plants according to claim 1 or 2, characterized in that, In step 4.3, the machine learning method calculates the area of ​​dust accumulation and pollutant shading on the photovoltaic panel surface using the following formula: The degree of vegetation shading is calculated using the following formula: In the formula: A is the area of ​​the photovoltaic panel surface where dust and pollutants are blocked, p i Let C represent the abnormal coverage of the i-th pixel, n be the total number of pixels, and R be the actual ground area represented by each pixel in the spectral data; veg A represents the percentage of vegetation cover. veg A represents the area covered by vegetation. total Total area of ​​photovoltaic panels.

4. The fully autonomous UAV hyperspectral inspection method for anomaly detection in photovoltaic power plants according to claim 3, characterized in that, In step 4.3, the abnormal coverage status p of the i-th pixel is obtained in the following way. i : A1. By constructing a time-series-based spectral change detection model, the current spectral characteristics are compared with historical data to identify spectral changes caused by dust accumulation and pollutants. The spectral change detection model uses a dynamic time warping method, as shown in the following formula: In the formula: x i and y (i) All are time series signals, and π(i) is the alignment path; A2. Based on the spectral changes caused by ash accumulation and pollutants, determine whether pollutants are present. If so, compare the extracted spectral features with existing spectral libraries and calculate spectral similarity using the spectral correlation coefficient, as shown in the following expression: In the formula: r is the correlation coefficient between the currently acquired spectral features and the reference spectrum, x i and y i These are the values ​​of the current spectral feature and the reference spectrum at the i-th pixel, respectively. and x i and y i The mean; Identify the contaminant type based on the value of r; if not, proceed to step A3. A3. The following support vector regression formula is used to perform regression calculations to determine the coverage degree of dust accumulation and contaminants, i.e., the abnormal coverage of the i-th pixel: In the formula: α i For the coefficients of the support vectors, K(x) i ,x) is the kernel function, b is the bias term, and x is the input spectral feature.

5. The fully autonomous UAV hyperspectral inspection method for anomaly detection in photovoltaic power plants according to claim 1, characterized in that: In step 5, the percentage of photoelectric conversion efficiency loss is calculated using the following formula: In the formula: η is the percentage of photoelectric conversion efficiency loss, P a P represents the photoelectric conversion efficiency under the current conditions. i This represents the photoelectric conversion efficiency under ideal conditions.

Images