Photovoltaic power station fault positioning method based on infrared transmission imaging

Abstract

The invention provides a photovoltaic power station fault positioning method based on infrared transmission imaging, and the method comprises the steps: carrying out the image shooting of a photovoltaic power station based on an infrared imager assembled on an unmanned plane according to a set time interval or position triggering condition, so as to generate infrared image data; encrypting the infrared image data and then transmitting the infrared image data to a ground control center; the ground control center classifies and stores the received ciphertext infrared image data so as to establish a photovoltaic power station infrared image database; preprocessing the photovoltaic power station infrared image database to obtain an effective infrared image database; serializing the effective infrared image database to generate an effective infrared image data sequence; performing feature map extraction on the effective infrared image data sequence based on a fault positioning model to generate a thermal infrared temperature information general situation map; and positioning the photovoltaic power station fault according to the thermal infrared temperature information general situation map.

Description

Technical Field

[0001] The present application relates to the field of photovoltaic technology, and in particular to a photovoltaic power station fault location method based on infrared imaging. Background Art

[0002] As the global demand for clean energy continues to grow, photovoltaic power stations have been widely built and applied as an important renewable energy utilization facility. The operational stability and power generation efficiency of photovoltaic power stations are of vital importance to the sustainability of energy supply. However, during the long-term operation of photovoltaic power stations, due to the influence of various factors, such as changes in the natural environment, component aging, local shading, and electrical connection failures, photovoltaic components may experience different types of failures, which in turn affect the performance of the entire power station.

[0003] Traditional photovoltaic power station fault detection methods mainly rely on manual inspections and some simple electrical detection equipment. Manual inspections require operation and maintenance personnel to conduct on-site inspections in the power station, observe the appearance, color, and obvious damage or obstruction of photovoltaic modules with the naked eye, and use handheld thermometers or thermal imagers to detect the temperature of some modules, and then use electrical parameter measuring instruments to detect the electrical system of the power station to determine whether there is a fault. This manual inspection method has many limitations. First, manual inspections are inefficient. For large-scale photovoltaic power stations, a lot of manpower and time costs are required, and it is difficult to achieve high-frequency and comprehensive inspections of power stations. Secondly, the accuracy and reliability of manual inspections largely depend on the experience and skill level of the operation and maintenance personnel. The detection results of different personnel may be different, and some minor faults or potential problems may be difficult to detect, which can easily lead to delayed diagnosis and processing of faults, increasing the operating risks and economic losses of power stations. For example, some early hot spot failures may only manifest as a slight increase in temperature on the surface of the component, which may be missed during manual inspections due to insufficient detection accuracy or negligence. Over time, the hot spot failure may gradually worsen, not only reducing the power generation efficiency of the faulty component, but may also cause permanent damage to the component and even affect the stability of the entire photovoltaic array. Summary of the invention

[0004] In order to solve the above technical problems, the present application provides a photovoltaic power station fault location method based on infrared imaging.

[0005] A photovoltaic power station fault location method based on infrared imaging, comprising:

[0006] The infrared imager installed on the drone takes images of the photovoltaic power station according to the set time interval or position trigger conditions to generate infrared image data;

[0007] The infrared image data is encrypted and transmitted to a ground control center;

[0008] The ground control center classifies and stores the received ciphertext infrared image data to establish an infrared image database of a photovoltaic power station;

[0009] Preprocessing the infrared image database of the photovoltaic power station to obtain a valid infrared image database;

[0010] Serializing the effective infrared image database to generate an effective infrared image data sequence;

[0011] Extracting feature graphs from the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph;

[0012] The photovoltaic power station fault is located according to the thermal infrared temperature information overview map.

[0013] Optionally, the infrared imager mounted on the drone takes an image of the photovoltaic power station at a set time interval or position trigger condition to generate infrared image data, including:

[0014] Initializing the shooting parameters of the infrared imager according to the requirements of the photovoltaic power station fault detection on temperature resolution, spatial resolution, temperature measurement range, etc.;

[0015] A digitized map of the photovoltaic power station is imported into the flight control system of the UAV, and the coordinate ranges corresponding to the areas that need special attention are marked on the digitized map. When the UAV enters these areas of special attention, the infrared imager is automatically triggered to optimize the shooting parameters according to its visual range so as to shoot these areas of special attention based on the optimized shooting parameters.

[0016] Optionally, encrypting the infrared image data and transmitting it to a ground control center comprises:

[0017] Performing block processing on the infrared image data to obtain a plurality of image data blocks;

[0018] Perform a series of byte replacement, row shift, column confusion and round key encryption transformations on each image data block and the key in turn, and output the encrypted ciphertext data block;

[0019] The encrypted ciphertext data block is transmitted to the ground control center through the communication link between the UAV and the ground control center.

[0020] Optionally, the ground control center classifies and stores the received ciphertext infrared image data to establish a photovoltaic power station infrared image database, including:

[0021] After the ground control center decrypts the received ciphertext infrared image data, it classifies them according to timestamp, shooting location coordinate information, drone number, imager model, and shooting task number;

[0022] Based on the classification dimension, a data access index is constructed to establish a photovoltaic power station infrared image database.

[0023] Optionally, the preprocessing of the photovoltaic power station infrared image database to obtain a valid infrared image database includes:

[0024] The infrared image database of the photovoltaic power station is enhanced to obtain an effective infrared image database.

[0025] Optionally, the serializing the valid infrared image database to generate a valid infrared image data sequence comprises:

[0026] For each valid infrared image, a JSON object structure is defined, which contains the "image_data" field for storing the pixel data of the image, and other metadata fields, including "timestamp" (shooting time), "location" (shooting location coordinates), and "image_id" (image number). Each field is assigned and organized according to the JSON syntax rules;

[0027] Based on the JSON object structure, generate an image metadata dictionary;

[0028] Based on the serialization function, the image metadata dictionary is serialized to generate a valid infrared image data sequence.

[0029] Optionally, the extracting a feature graph of the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph includes:

[0030] Input the preprocessed valid infrared image data sequence into the fault location model one by one, start the forward propagation of the fault location model so that the preprocessed valid infrared image data sequence first enters the convolution layer, and multiple convolution kernels in the convolution layer perform convolution operation on the preprocessed valid infrared image data sequence to extract local features of the image;

[0031] The local features of the image enter the pooling layer for downsampling operation to obtain a compressed feature map;

[0032] The compressed feature map is fused through a fully connected layer to map the compressed feature map into a semantic space and generate a thermal infrared temperature information overview map based on it.

[0033] Optionally, locating the photovoltaic power station fault according to the thermal infrared temperature information overview map includes:

[0034] Obtaining fault judgment indicators defined for different fault types, the fault types including hot spot faults, local high temperature faults caused by component short circuits, and uneven temperature faults caused by shielding;

[0035] According to the preset color-temperature mapping relationship in the thermal infrared temperature information overview, the image is divided into regions, and the entire thermal infrared temperature information overview is scanned to segment the temperature regions represented by different colors to obtain the boundary information and area size of each temperature region;

[0036] Matching the fault judgment index with the segmented temperature regions to determine the relative position of the fault region;

[0037] The relative position of the fault area is converted into actual space coordinates in the photovoltaic power station.

[0038] A computer program product having computer executable instructions stored thereon, wherein the computer executable instructions are executed to perform the following steps:

[0039] The infrared imager installed on the drone takes images of the photovoltaic power station according to the set time interval or position trigger conditions to generate infrared image data;

[0040] The infrared image data is encrypted and transmitted to a ground control center;

[0041] The ground control center classifies and stores the received ciphertext infrared image data to establish an infrared image database of a photovoltaic power station;

[0042] Preprocessing the infrared image database of the photovoltaic power station to obtain a valid infrared image database;

[0043] Serializing the effective infrared image database to generate an effective infrared image data sequence;

[0044] Extracting feature graphs from the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph;

[0045] The photovoltaic power station fault is located according to the thermal infrared temperature information overview map.

[0046] An electronic device comprises a memory and a processor, wherein the memory stores computer executable instructions, and the processor runs the computer executable instructions to perform the following steps:

[0047] The infrared imager installed on the drone takes images of the photovoltaic power station according to the set time interval or position trigger conditions to generate infrared image data;

[0048] The infrared image data is encrypted and transmitted to a ground control center;

[0049] The ground control center classifies and stores the received ciphertext infrared image data to establish an infrared image database of a photovoltaic power station;

[0050] Preprocessing the infrared image database of the photovoltaic power station to obtain a valid infrared image database;

[0051] Serializing the effective infrared image database to generate an effective infrared image data sequence;

[0052] Extracting feature graphs from the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph;

[0053] The photovoltaic power station fault is located according to the thermal infrared temperature information overview map.

[0054] The solution proposed in the embodiment of the present application has at least one of the following advantages:

[0055] 1. Based on the method of equipping the drone with an infrared imager, the drone can automatically capture images of the photovoltaic power station according to the set time interval or location trigger condition. The drone can quickly cover a large area, greatly shortening the detection time, and can inspect the power station frequently, timely capture the status changes of the photovoltaic components, greatly improve the efficiency of fault detection, and effectively save manpower and time costs.

[0056] 2. Encrypt and transmit infrared image data to ensure the integrity and security of the data during transmission, avoid data tampering or loss, and ensure the accuracy of subsequent analysis.

[0057] The ground control center classifies and stores the encrypted infrared image data and establishes a database, which provides an orderly data basis for subsequent analysis. After preprocessing, an effective infrared image database is obtained, noise and invalid data are removed, and data quality is improved.

[0058] After the valid data is serialized, feature maps are extracted based on the fault location model to generate a thermal infrared temperature information overview map. This model-based algorithm processing method can use advanced technical means to conduct in-depth analysis of image data and accurately extract feature information reflecting photovoltaic module failures. Compared with manual judgment based on experience, it greatly improves the accuracy and reliability of detection, and can accurately detect minor faults and potential problems such as early hot spot faults, avoiding delayed diagnosis and processing of faults.

[0059] 3. This method encrypts and transmits the acquired infrared image data, stores them in categories, establishes a photovoltaic power station infrared image database, and performs a series of processing on it. This method not only facilitates the systematic management of infrared image data of power stations at different periods, but also enables the analysis of historical data to understand the operation status of power stations and the development trend of faults.

[0060] For example, by comparing and analyzing valid infrared image data sequences at different times, the aging process of components and the development laws of failures can be discovered, providing a strong basis for preventive maintenance of power plants.

[0061] The rich database resources also support more in-depth research and optimization, providing data support for improving fault detection methods and improving power plant operation efficiency.

[0062] 4. Drones equipped with infrared imagers can take all-round and all-around photos of photovoltaic power stations, covering areas that are difficult to reach or easily overlooked by manual inspections. By processing and analyzing the infrared image data of the entire power station, the operating status of photovoltaic components can be fully understood, and hidden faults in various locations can be discovered in a timely manner to avoid affecting the performance of the entire power station due to undetected local faults, reduce the operating risks of the power station, and ensure the sustainability of energy supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The drawings constituting a part of this application are used to provide a further understanding of this application, so that other features, purposes and advantages of this application become more obvious. The schematic embodiment drawings and their descriptions of this application are used to explain this application and do not constitute an improper limitation on this application. In the drawings:

[0064] Figure 1 It is a flow chart of a photovoltaic power station fault location method based on infrared imaging according to an embodiment of the present application. DETAILED DESCRIPTION

[0065] In order to enable those skilled in the art to better understand the solution of the present application, the following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0066] Figure 1 FIG. 1 is a flow chart of a photovoltaic power station fault location method based on infrared imaging according to an embodiment of the present application. Figure 1 As shown, a photovoltaic power station fault location method based on infrared imaging includes:

[0067] The infrared imager installed on the drone takes images of the photovoltaic power station according to the set time interval or position trigger conditions to generate infrared image data;

[0068] The infrared image data is encrypted and transmitted to a ground control center;

[0069] The ground control center classifies and stores the received ciphertext infrared image data to establish an infrared image database of a photovoltaic power station;

[0070] Preprocessing the infrared image database of the photovoltaic power station to obtain a valid infrared image database;

[0071] Serializing the effective infrared image database to generate an effective infrared image data sequence;

[0072] Extracting feature graphs from the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph;

[0073] The photovoltaic power station fault is located according to the thermal infrared temperature information overview map.

[0074] Optionally, the infrared imager mounted on the drone takes an image of the photovoltaic power station at a set time interval or position trigger condition to generate infrared image data, including:

[0075] Initializing the shooting parameters of the infrared imager according to the requirements of the photovoltaic power station fault detection on temperature resolution, spatial resolution, temperature measurement range, etc.;

[0076] A digitized map of the photovoltaic power station is imported into the flight control system of the UAV, and the coordinate ranges corresponding to the areas that need special attention are marked on the digitized map. When the UAV enters these areas of special attention, the infrared imager is automatically triggered to optimize the shooting parameters according to its visual range so as to shoot these areas of special attention based on the optimized shooting parameters.

[0077] Specifically, when initializing the infrared imager shooting parameters, in photovoltaic power station fault detection, parameters such as temperature resolution, spatial resolution, and temperature measurement range are crucial for accurate fault detection. The infrared imager shooting parameters are initialized as follows:

[0078] (1) Temperature resolution

[0079] Temperature resolution determines the smallest temperature difference that an infrared imager can distinguish. In a photovoltaic power plant, early hot spot failures may only appear as a small temperature increase on the surface of the component, so a high temperature resolution is required to detect these small changes.

[0080] Assume that the temperature resolution is ΔT, which is related to the detector sensitivity of the infrared imager, the noise equivalent temperature difference (NETD) and other factors. It can be expressed by the following formula:

[0081] ΔT=k·NETD

[0082] Among them, $k$ is a system-related coefficient, which depends on the optical system of the imager, signal processing algorithm, etc. NETD (noise equivalent temperature difference) refers to the temperature difference between the target and the background corresponding to the root mean square noise voltage of the imaging system output signal equal to the voltage change of the detector output signal under specific conditions. In photovoltaic power station applications, the smaller the NETD, the smaller the temperature difference that the imager can distinguish, and the more conducive it is to detecting small temperature changes, such as the temperature rise at the beginning of a hot spot failure.

[0083] (2) Spatial resolution

[0084] Spatial resolution determines the minimum object size or spatial detail that an infrared imager can distinguish. In a photovoltaic power plant, a sufficiently high spatial resolution is required to clearly observe the details of the photovoltaic modules in order to detect problems such as module damage and partial shading.

[0085] Spatial resolution is usually measured by the instantaneous field of view (IFOV), which is related to the focal length f of the imager and the detector pixel size d. The calculation formula is:

[0086]

[0087] In the photovoltaic power station scenario, in order to clearly observe the details of the photovoltaic modules, it is necessary to select the appropriate focal length and detector pixel size according to the size and installation distance of the module to obtain the appropriate spatial resolution. For example, if the size of the photovoltaic module is small and the installation distance is far, a smaller IFOV is required, that is, a longer focal length f and a smaller pixel size d, to ensure that the tiny defects on the module can be distinguished.

[0088] (3) Temperature measurement range

[0089] The temperature measurement range refers to the temperature range that the infrared imager can measure. In a photovoltaic power station, the temperature of the components will be affected by factors such as ambient temperature and light intensity. Therefore, it is necessary to determine the appropriate temperature measurement range to ensure that the temperature of the components can be accurately measured.

[0090] Assume the temperature measurement range is [T min ,T max ], which is related to factors such as the ambient temperature range of the area where the photovoltaic power station is located and the temperature changes of the components under different working conditions. It can be estimated by the following formula:

[0091] T min =Tenv,min +ΔT min

[0092] T max =T env,max +ΔT max

[0093] Among them, T env,min and T env,max are the lowest and highest ambient temperatures in the area where the photovoltaic power station is located, ΔT min and ΔT max are the temperature rise of the component relative to the ambient temperature at the lowest and highest operating conditions, respectively. For example, in the hot summer, the ambient temperature may reach T env,max =40℃, and the temperature of the module may increase by ΔT under strong light. max =20℃, then the upper limit of the temperature measurement range is T max It needs to be set to at least 60°C.

[0094] Specifically, when importing digital maps and triggering shooting, the digital map of the photovoltaic power station is imported into the flight control system of the drone, and the coordinate range corresponding to the area that needs to be focused on is marked on the map. This step requires the support of geographic information system (GIS) technology and drone flight control algorithm.

[0095] Assume that the coordinate system of the digital map is (x, y), and the coordinate range of the focus area is [(x1, y1), (x2, y2)]. During the flight, the drone obtains its own position coordinates (x1, y1), (x2, y2) in real time through its own positioning system (such as GPS). uav ,y uav ).

[0096] When a drone enters an area of ​​concern, the following conditions are met:

[0097] x1≤x uav ≤x2

[0098] y1≤y uav ≤y2

[0099] The drone's flight control system automatically triggers the infrared imager to take pictures.

[0100] y1≤y uav ≤y2

[0101] When triggering shooting, it is necessary to optimize the shooting parameters according to the visible range of the infrared imager. The visible range is related to the imager's viewing angle θ and the flight altitude h. Assume that the lengths of the visible range in the horizontal and vertical directions are L x and L y , then:

[0102]

[0103] Among them, θ x and θ y are the viewing angles of the imager in the horizontal and vertical directions, respectively.

[0104] In order to ensure that sufficiently clear and comprehensive images can be obtained in the focus area, the shooting parameters need to be adjusted according to the visual range. For example, if the visual range is large, in order to ensure spatial resolution, the shooting interval should be appropriately reduced; if the visual range is small, the shooting interval should be appropriately increased to improve shooting efficiency.

[0105] Assuming that the shooting intervals in the horizontal and vertical directions are Δx and Δy respectively, the following is calculated based on the visible range and the required image coverage:

[0106]

[0107] Among them, N x and N y are the number of images you want to acquire in the horizontal and vertical directions, respectively, which depend on the required image coverage and resolution requirements.

[0108] Through the above steps, the drone can automatically trigger the infrared imager to shoot according to the actual situation of the photovoltaic power station, and optimize the shooting parameters to obtain high-quality infrared image data, providing a reliable basis for subsequent fault detection and positioning. These complex mathematical formulas and parameter settings ensure that the processor can accurately control the operation of the drone and infrared imager to achieve efficient and accurate photovoltaic power station fault detection.

[0109] To this end, in view of the many disadvantages of manual inspection in traditional photovoltaic power station fault detection, such as low efficiency and accuracy relying on manual experience, the above-mentioned technology that realizes the automatic triggering of shooting and optimization of parameters by drones according to the actual situation of photovoltaic power stations has the following benefits:

[0110] 1. Initialize shooting parameters based on precise requirements

[0111] Traditional methods often lack precise quantitative standards when detecting temperature changes. This solution determines the temperature resolution through the formula ΔT = k·NETD. In photovoltaic power stations, the temperature changes are small in the early stage of hot spot failure. This formula accurately calculates the ΔT that can distinguish small temperature differences based on the sensitivity of the imager detector and the noise equivalent temperature difference (NETD). The k coefficient takes into account system factors such as the optical system and signal processing algorithm. This is a major breakthrough in the traditional empirical judgment of temperature changes. By accurately setting the temperature resolution, the small temperature increase on the surface of the component in the early stage of hot spot failure can be captured in time, providing the possibility of early fault detection.

[0112] In terms of determining spatial resolution, traditional methods have difficulty in clearly defining parameters that can meet detection requirements. This solution introduces the concept of instantaneous field of view (IFOV) and uses the formula Scientific calculation based on the focal length f of the imager and the pixel size d of the detector. In the photovoltaic power station scenario, according to the size of the component and the installation distance, the appropriate IFOV is obtained by adjusting f and d, thereby achieving the appropriate spatial resolution. This enables the imager to clearly observe the details of the component and accurately detect problems such as component damage and partial occlusion, which is more advantageous in detail observation than traditional manual inspections.

[0113] Reasonable estimation of temperature measurement range: The traditional method is vague in determining the temperature measurement range, which may lead to inaccurate measurement of component temperature. This solution uses the formula T min =T env，min +ΔT min and T max =T env，max +ΔT max , comprehensively consider the ambient temperature range of the photovoltaic power station area and the temperature changes of the components under different working conditions, and reasonably estimate the temperature measurement range. This ensures that the imager can accurately measure the component temperature in various environments and working conditions, providing reliable temperature data support for fault detection.

[0114] 2. Intelligent triggering and parameter optimization

[0115] Traditional detection is difficult to achieve automatic detection of specific areas. This solution imports the digital map of the photovoltaic power station into the UAV flight control system, uses geographic information system (GIS) technology and flight control algorithms, and marks the focus area through the coordinate range [(x1, y1), (x2, y2)]. The UAV obtains the position coordinates (x1, y1) in real time through its own positioning system. uav ,y uav ), when x1≤x uav ≤x2 and y1≤y uav When ≤y2, the camera will be automatically triggered to shoot. Compared with the randomness and uncertainty of traditional manual inspection, this automatic triggering mechanism based on coordinate judgment greatly improves the pertinence and efficiency of detection.

[0116] The traditional shooting parameter setting lacks consideration of the actual visual range. This scheme is based on the imager viewing angle θ and the flight altitude h, through the formula and Calculate the visible area range. On this basis, according to the visible area range and the required image coverage, the formula and Calculate and adjust the shooting intervals Δx and Δy. This innovation enables the shooting parameters to be dynamically optimized according to the actual visual field, ensuring the acquisition of high-quality, high-coverage image data in key areas of concern, improving the quality and availability of image data, and providing a more reliable basis for subsequent fault detection and location.

[0117] To this end, based on the above processing, the flight path, positioning and shooting parameters of the UAV's infrared imager are precisely controlled. Automatic triggering of shooting is achieved through precise coordinate judgment, and shooting parameters are optimized in real time according to the visual range, making the entire detection process highly automated and precise. This precise control not only improves detection efficiency and reduces errors and uncertainties caused by manual intervention, but also greatly improves the accuracy of fault detection. It can quickly and accurately obtain infrared image data of photovoltaic power stations, providing a solid and reliable foundation for subsequent fault detection and positioning based on image analysis, and realizing efficient and accurate photovoltaic power station fault detection.

[0118] Optionally, encrypting the infrared image data and transmitting it to a ground control center comprises:

[0119] Performing block processing on the infrared image data to obtain a plurality of image data blocks;

[0120] Perform a series of byte replacement, row shift, column confusion and round key encryption transformations on each image data block and the key in turn, and output the encrypted ciphertext data block;

[0121] The encrypted ciphertext data block is transmitted to the ground control center through the communication link between the UAV and the ground control center.

[0122] Specifically, when the infrared image data is processed in blocks, in the photovoltaic power station scenario, the acquired infrared image data is a two-dimensional matrix containing a large amount of pixel information. Let the infrared image matrix be I, and its size is H×W, where H represents the height of the image (i.e., the number of pixel rows) and W represents the width of the image (i.e., the number of pixel columns). In order to facilitate the processor to efficiently process encryption operations, it needs to be divided into multiple smaller image data blocks.

[0123] The size of each image data block is defined as h×w, where the values ​​of h and w need to take into account the processor's cache size, computing power, and encryption algorithm requirements. For example, if the processor's cache is small, h and w should be small to avoid frequent data exchange; if the encryption algorithm has specific requirements for the data block size, it also needs to be adjusted accordingly.

[0124] The number of blocks that can be divided along the height direction of the image is n h The calculation formula is: in Denote the ceiling of x. This is because there may be a situation where H is not divisible by h, and it is necessary to ensure that all pixels can be divided into data blocks. Similarly, the number of blocks n that can be divided along the width direction of the image w is:

[0125] Then, the total number of image data blocks N is: N = n h ×n w .

[0126] Compared with traditional techniques, when dividing blocks, a fixed block size is often used, without fully considering the performance differences of different processors and the requirements of encryption algorithms. This solution dynamically adjusts the block size according to the actual situation of computing power, can better adapt to the computing power, and improve the efficiency of encryption processing. For example, for a processor with较强运算能力 (stronger computing power), the block size is appropriately increased and the number of blocks is reduced, thereby reducing the overhead in the data processing process; for a processor with较弱运算能力 (weaker computing power), a smaller block size is used to avoid performance bottlenecks caused by the processor processing overly large data blocks.

[0127] Specifically, during the encryption transformation, it is implemented based on byte substitution, specifically based on a set lookup table, i.e., the S-Box (Substitution Box) to implement the non-linear transformation. In the infrared image data of the photovoltaic power station, each byte represents specific temperature or radiation information. Assume that the S-Box is an array S

[256] containing 256 elements. For each byte b (0 ≤ b ≤ 255) in the image data block, after byte substitution, a new byte b' is obtained, and its calculation formula is b′ = S[b].

[0128] For example, if b = 35, then look up the element value with index 35 in the S-Box, and this value is the substituted byte b'.

[0129] Traditional byte substitution methods may use simple fixed mapping relationships and are easily cracked by attackers. The S-Box adopted in this solution is customized according to the characteristics of the photovoltaic power station data. For example, by combining factors such as the temperature distribution range in different regions of the photovoltaic power station and the statistical characteristics of the data, the element arrangement of the S-Box is optimized, making the encrypted ciphertext more random and resistant to attacks, effectively protecting the sensitive information of the photovoltaic power station.

[0130] Specifically, for row shift, circular shift processing is performed on each row of the image data block. For an h×w image data block, the i-th row (0 ≤ i < h) needs to be circularly shifted left by i positions. Assume that the elements in the image data block are a ij (0 ≤ i < h, 0 ≤ j < w). After row shift, the new element a ij′ satisfies: a ij′ = a i,(j-i+w)modwThe mod operation here means modulo operation, which aims to ensure that the index value after shifting is within the legal range. For example, for a 4×4 data block, when i=2, the element a in the second row 2j Circularly shift left 2 positions, i.e. a 2j′ =a 2,(j-2+4)mod4 .

[0131] Therefore, compared with the traditional row shift operation, which usually has a fixed shift amount and lacks consideration of the internal structure and data characteristics of the data block, this scheme dynamically adjusts the shift amount according to the row number, making full use of the two-dimensional structural characteristics of the image data, making the information distribution of the encrypted ciphertext in the horizontal direction more complex, increasing the difficulty for attackers to analyze data from the row direction, and further improving the security of encryption.

[0132] Specifically, for column confusion, each column of the image data block is linearly transformed through a specific matrix operation. For an h×w image data block, each column can be regarded as an h-dimensional vector. Let the column confusion matrix be C, which is an h×h reversible matrix and has a finite field GF(2 8 ) is defined above.

[0133] For the jth column vector v j =[a 0j ,a 1j ,…,a (h-1)j ] T , the new column vector v after column confusion j′ For: v j′ =C·v j In the finite field GF(2 8 ), all operations follow the rules of finite fields, which ensures the correctness and uniqueness of the results. For example, in GF(2 8 ), the addition operation is a bitwise XOR operation, and the multiplication operation needs to be calculated according to a specific primitive polynomial. In this way, the linear relationship between the elements in the column is changed, increasing the confusion of the data.

[0134] Therefore, the traditional column obfuscation method may operate on the ordinary integer field or real number field, which is not suitable for processing 8-bit binary image data and has low security. 8 ), which is more in line with the storage and representation of infrared image data of photovoltaic power stations. The characteristics of finite fields are used to enhance the security of encryption and the obfuscation effect of data, and effectively resist attacks on linear transformations.

[0135] For round key encryption, the image data block after the previous transformation is XORed with the round key. Let the round key be K, whose length is the same as the size of the image data block, that is, K is also an h×w matrix. For the image data block D after byte replacement, row shift and column confusion, the encrypted ciphertext data block E is: E=D⊕K, where ⊕ represents a bitwise XOR operation.

[0136] In a photovoltaic power station, the round key used in each round of encryption is generated based on dynamic information such as time and location. For example, the flight time t of the drone, the longitude and latitude of the current location And other information, through a complex key generation function Generate round keys. This ensures that the key used for each encryption is random and unique.

[0137] Therefore, compared with the traditional round key encryption method, which often uses a fixed key generation algorithm, the randomness and variability of the key are poor and it is vulnerable to attacks. This scheme dynamically generates round keys according to the actual operating parameters of the photovoltaic power station (such as time, location, etc.), which greatly increases the complexity and unpredictability of the key. It is difficult for attackers to crack the encrypted data by analyzing the key generation rules, thereby improving the security and reliability of the encryption system.

[0138] Specifically, for transmission to the ground control center, the encrypted ciphertext data block needs to be transmitted through the communication link between the UAV and the ground control center. Assuming the bandwidth of the communication link is B (unit: bit / second), the size of each ciphertext data block is S (unit: bit), then the time T required to transmit a ciphertext data block is:

[0139] Since the communication environment between the UAV and the ground control center may be subject to noise interference, in order to ensure the accuracy of data transmission, error correction coding technology, such as Reed-Solomon coding, is used. Suppose the original ciphertext data block is M, whose length is k bits, and the codeword generated after Reed-Solomon coding is C, whose length is n bits (n>k).

[0140] Reed-Solomon coding is based on the finite field GF(2 m ) (usually m = 8). At the sending end, the specific generating polynomial G(x) is used to calculate: C(x) = M(x)·x n-k modG(x). Here M(x) and C(x) are the polynomial representations of the original data block and the encoded codeword, respectively.

[0141] At the receiving end, the original ciphertext data block is restored by performing error detection and correction on the received codeword and using methods such as companion calculation.

[0142] Therefore, compared with traditional data transmission methods, it may not be able to effectively guarantee the integrity and accuracy of data when facing complex communication environments. This solution adopts Reed-Solomon coding technology based on the communication characteristics of photovoltaic power stations, which can correct errors that occur during transmission to a certain extent and improve the reliability of data transmission. At the same time, the transmission time is accurately calculated based on the bandwidth of the communication link and the size of the data block, which helps to reasonably arrange data transmission tasks and improve the operating efficiency of the entire system.

[0143] Optionally, the ground control center classifies and stores the received ciphertext infrared image data to establish a photovoltaic power station infrared image database, including:

[0144] After the ground control center decrypts the received ciphertext infrared image data, it classifies them according to timestamp, shooting location coordinate information, drone number, imager model, and shooting task number;

[0145] Based on the classification dimension, a data access index is constructed to establish a photovoltaic power station infrared image database.

[0146] Specifically, when decrypting ciphertext infrared image data, in the photovoltaic power station scenario, the ciphertext infrared image data received by the ground control center is encrypted to ensure the security of data transmission. The decryption process is the inverse operation of the encryption process, and its purpose is to restore the ciphertext to the original infrared image data for subsequent classification and storage.

[0147] Assume that the encryption algorithm used in the encryption process involves multiple rounds of transformations, such as the byte replacement, row shift, column confusion, and round key encryption transformation mentioned above. The decryption process needs to perform the inverse operations of these transformations in reverse order.

[0148] Specifically, for round key decryption: the round key is used in the encryption process to perform an XOR operation with the transformed data block to obtain the ciphertext. During decryption, let the received ciphertext data block be E and the round key be K. Then the intermediate data block D' after round key decryption is: D'=E⊕K, where ⊕ represents a bitwise XOR operation. The round key K is generated according to a specific key generation algorithm during the encryption process, and needs to be securely transmitted to the ground control center during the transmission process.

[0149] For the inverse operation of column confusion: Column confusion during encryption is performed by matrix multiplication over a finite field GF(2 8 ). When decrypting, the inverse operation of column confusion is required. Let the column confusion matrix during encryption be C, and its inverse matrix be C -1 (In the finite field GF(2 8 )). For each column vector v in the intermediate data block $D'$ after decryption with the round keyj′ After the inverse operation of column mixing, the new column vector v j″ is: v j″ = C -1 · v j′ .

[0150] Regarding the inverse operation of row shift: During encryption, the rows of the image data block are circularly shifted. During decryption, the inverse operation of row shift is to perform a reverse circular shift on the rows of the image data block after the inverse operation of column mixing. For an h×w image data block, the i-th row (0 ≤ i < h) needs to be circularly shifted to the right by i positions. Let the element after the inverse operation of column mixing be a ij″ After the inverse operation of row shift, the new element a ij″′ satisfies: a ij″′ = a i,(j+i)modw .

[0151] Regarding the inverse operation of byte substitution: During encryption, byte substitution is performed through the S-box. During decryption, an inverse S-box (denoted as S -1 ) needs to be constructed. For each byte $b”'$ after the inverse operation of row shift, the original byte b is obtained after the inverse operation of byte substitution, and its calculation formula is b = S -1 [b″′].

[0152] Specifically, when classifying data, it is classified in the following way:

[0153] (1) Classification by timestamp

[0154] The timestamp refers to the specific time when the infrared image data is taken. During the operation of a photovoltaic power station, infrared images at different times may reflect different operating states of the power station. For example, at different times of the day, due to changes in light intensity, the temperature distribution of photovoltaic modules will be different. By classifying by timestamp, it is convenient to compare the power station states at different times, which helps to discover potential fault trends. Let the timestamp be represented in Unix time format, that is, the number of seconds from 00:00:00 UTC on January 1, 1970 to the shooting moment, denoted as t.

[0155] (2) Classification by shooting position coordinate information

[0156] Photovoltaic power stations usually cover a large area, and photovoltaic modules at different positions may be affected by different environmental factors, such as local shading and ventilation conditions. By classifying the image data according to the shooting position coordinate information (longitude λ and latitude ), the position of the component with problems can be accurately located. This is very important for fault troubleshooting and maintenance, and can quickly determine the area that needs to be repaired.

[0157] (3) Classification by drone number

[0158] In a large photovoltaic power station, multiple drones may be used for inspection. The performance and flight trajectory of each drone may be different. The image data is classified by drone number (set as u) to distinguish the data obtained by different drones, which is convenient for evaluating and managing the working conditions of each drone, and also helps to trace the source when there is a problem with the data.

[0159] (3) Classification by imager model

[0160] Different models of infrared imagers have different performance parameters, such as temperature resolution, spatial resolution, etc. These parameters will affect the quality and accuracy of image data. Image data is classified by imager model (set as m), and targeted processing is performed according to the performance characteristics of the imager in subsequent analysis to improve the accuracy of data analysis.

[0161] (4) Classification by shooting task number

[0162] Each shooting task may have different goals and requirements, such as regular inspections, troubleshooting, etc. By classifying the image data by the shooting task number (set as n), it is convenient to manage and query the data obtained by different tasks, and improve the efficiency of data use.

[0163] Specifically, when constructing the data access index, a multi-dimensional data access index is constructed based on the above classification dimensions, so as to quickly locate and retrieve data in the infrared image database of the photovoltaic power station.

[0164] For example, suppose the data access index is represented as a tuple Where t represents the timestamp, It represents the shooting location coordinate information, u represents the drone number, m represents the imager model, and n represents the shooting task number.

[0165] When building an index, data structures such as hash tables and B-trees are used to achieve fast search and insertion operations. For example, for an index structure based on a hash table, the tuple Through a specific hash function Mapped to a hash value, which serves as the key of the hash table, and the corresponding value is a pointer to the actual image data stored in the database.

[0166] To this end, compared with the traditional simple classification method, this solution adopts a multi-dimensional classification method, which fully considers the characteristics and actual application needs of photovoltaic power station inspection data. Through classification in multiple dimensions such as timestamp, shooting location coordinate information, drone number, imager model and shooting task number, it can manage and analyze data more meticulously, providing more comprehensive and accurate information for subsequent fault diagnosis and power station operation and maintenance.

[0167] Building a multi-dimensional access index can greatly improve data retrieval efficiency. Traditional indexing methods may only consider a single dimension or a few dimensions, which cannot meet complex query requirements. The multi-dimensional index structure of this solution can quickly locate the data required by users, whether it is querying images of a specific area by time or querying all related data by task number, it can be completed efficiently, improving the system's response speed and user experience.

[0168] Furthermore, in large-scale photovoltaic power plants, multiple drones, multiple imagers, and different shooting tasks are involved, and the amount of data is huge and complex. The classification and indexing method of this solution can well adapt to such complex scenarios, ensure the orderly management and efficient use of data, and provide strong support for the intelligent operation and maintenance of photovoltaic power plants.

[0169] Optionally, the preprocessing of the photovoltaic power station infrared image database to obtain a valid infrared image database includes:

[0170] The infrared image database of the photovoltaic power station is enhanced to obtain an effective infrared image database.

[0171] Specifically, during the operation of a photovoltaic power station, the acquired infrared images may be interfered by various factors, such as uneven illumination, noise, etc., which may affect the image quality and the accuracy of subsequent fault detection. Therefore, it is necessary to enhance the infrared image database of the photovoltaic power station to obtain an effective infrared image database. The following is a preferred description of the above-mentioned enhancement process.

[0172] Specifically, during histogram equalization, the grayscale histogram of the image is adjusted to make the grayscale distribution of the image more uniform, thereby enhancing the image contrast. In the infrared image of a photovoltaic power station, the temperature difference between different components may cause the image grayscale distribution to be uneven, and the details of some areas are difficult to distinguish. Through histogram equalization, this situation can be improved, making the details in the image clearer and facilitating subsequent fault detection.

[0173] Optionally, the original infrared image I(x,y) is a two-dimensional matrix of M×N, where x=0,1,…,M-1, y=0,1,…,N-1, and the grayscale range of the image is 0 to L-1 (usually L=256). First, the grayscale histogram h(k) of the image is calculated, which represents the number of pixels with grayscale level k, and the formula is:

[0174]

[0175] Then calculate the normalized cumulative distribution function cdf(k), which represents the cumulative probability of pixels with gray levels less than or equal to k, and the formula is:

[0176]

[0177] Finally, the grayscale value of the original image is mapped by the transformation function T(k) to obtain the enhanced image J(x, y), the formula is:

[0178] J(x,y)=(L-1)·cdf(I(x,y))

[0179] in, Indicates a floor operation.

[0180] Specifically, when performing homomorphic filtering, the image is processed in the frequency domain, while enhancing the image contrast and suppressing noise. In the infrared images of photovoltaic power plants, due to environmental factors, the images may have problems of low contrast and noise interference. Homomorphic filtering separates the illumination component and the reflection component of the image, processes them separately, and then synthesizes them to achieve the purpose of enhancing the image.

[0181] Alternatively, the original infrared image I(x,y) is represented as the product of the illumination component L(x,y) and the reflection component R(x,y), that is, I(x,y) = L(x,y)·R(x,y). Take the logarithm of the image and convert the multiplication operation into an addition operation to obtain lnI(x,y) = lnL(x,y) + lnR(x,y). Then, lnI(x,y) is Fourier transformed to obtain F{lnI(x,y)} = F{lnL(x,y)} + F{lnR(x,y)}, where F{·} represents Fourier transform.

[0182] In the frequency domain, a filter function H(u,v) is designed, which processes the spectrum of the illumination component and the reflection component differently. The spectrum of the illumination component is mainly concentrated in the low-frequency part, and the spectrum of the reflection component is mainly concentrated in the high-frequency part. By adjusting the parameters of the filter function H(u,v), the high-frequency part (reflection component) is enhanced and the low-frequency part (illumination component) is suppressed, thereby achieving the purpose of enhancing image contrast and suppressing noise.

[0183] The filtered spectrum is G(u,v)=H(u,v)·F{lnI(x,y)}, and the inverse Fourier transform is performed to obtain g(x,y)=F -1 {G(u,v)}, and finally perform exponential operation to obtain the enhanced image J(x,y)=e g(x,y) .

[0184] Specifically, when performing wavelet transform enhancement, the image is decomposed into sub-bands of different frequencies. In the infrared image of a photovoltaic power station, different fault features may correspond to different frequency components. Through wavelet transform, the noise and useful information in the image are separated, and then different sub-bands are processed accordingly to enhance useful information and suppress noise, thereby improving the quality of the image.

[0185] Suppose the original infrared image I(x,y), perform a two-dimensional wavelet transform on it, and get four sub-bands: low-frequency sub-band LL, horizontal high-frequency sub-band LH, vertical high-frequency sub-band HL and diagonal high-frequency sub-band HH. The formula of wavelet transform is expressed as:

[0186]

[0187] in, is the scaling function, and is the wavelet function, j and k are the scale and location parameters.

[0188] After processing different sub-bands, the sub-bands are synthesized through inverse wavelet transform to obtain the enhanced image J(x,y). For example, threshold processing is performed on high-frequency sub-bands to remove noise, and appropriate enhancement is performed on low-frequency sub-bands to highlight the main features of the image.

[0189] Therefore, the traditional image enhancement method may be general and does not fully consider the characteristics of photovoltaic power station infrared images. This scheme selects histogram equalization, homomorphic filtering and wavelet transform to enhance the common problems of low contrast and noise interference in photovoltaic power station infrared images. These methods can effectively improve image quality and improve the accuracy of fault detection.

[0190] Furthermore, multiple image enhancement methods are combined to give full play to their respective advantages. For example, histogram equalization quickly improves the contrast of the image, homomorphic filtering simultaneously enhances the contrast and suppresses noise, and wavelet transform performs multi-scale analysis on the image to highlight the characteristics of different frequencies. By integrating these methods, a more comprehensive and effective image enhancement effect is obtained.

[0191] Finally, the environment of photovoltaic power stations is complex and changeable, and different lighting conditions, weather conditions, etc. will affect the quality of infrared images. The image enhancement method of this scheme can adapt to these complex environments. By adjusting different parameters, it can effectively enhance the image in various situations and provide reliable image data for subsequent fault detection.

[0192] Optionally, the serializing the valid infrared image database to generate a valid infrared image data sequence comprises:

[0193] For each valid infrared image, a JSON object structure is defined, which contains the "image_data" field for storing the pixel data of the image, and other metadata fields, including "timestamp" (shooting time), "location" (shooting location coordinates), and "image_id" (image number). Each field is assigned and organized according to the JSON syntax rules;

[0194] Based on the JSON object structure, generate an image metadata dictionary;

[0195] Based on the serialization function, the image metadata dictionary is serialized to generate a valid infrared image data sequence.

[0196] In the operation and maintenance management of photovoltaic power stations, it is a crucial step to serialize the effective infrared image database and generate an effective infrared image data sequence that is easy to process and transmit. The preferred implementation method of this application is as follows:

[0197] 1. Define the JSON object structure

[0198] In the photovoltaic power station scenario, each valid infrared image contains rich information, including not only the pixel data of the image itself, but also the metadata related to it, such as shooting time, shooting location coordinates and image number, etc. In order to facilitate the processor to manage and process this information, a JSON object structure is defined.

[0199] Assume that each valid infrared image is I, and its corresponding JSON object structure is represented as follows:

[0200] "image d ata":[…],"timestamp":"YYYY-MM-DDHH:

[0201] MM:SS","location":[latitude,longitude],"image i d":"xxxxxx"

[0202] "image_data" field: used to store the pixel data of the image. In the infrared image of the photovoltaic power station, the image is usually a two-dimensional matrix. Let the size of the image be M×N, where M represents the height of the image (number of rows) and N represents the width of the image (number of columns). The pixel data is represented as an array containing M×N elements, each element corresponding to the gray value or other color information of a pixel (usually gray value in infrared images). For example, for a simple 8×8 infrared image, its pixel data may be represented as [[10,12,15,…],[20,22,25,…],…].

[0203] "timestamp" field: records the time when the image was taken, in the format of "YYYY-MM-DD HH:MM:SS", where "YYYY" represents the year, "MM" represents the month, "DD" represents the date, "HH" represents the hour, "MM" represents the minute, and "SS" represents the second. In a photovoltaic power station, infrared images at different times reflect the operating status of the power station at different times. Accurate time records are helpful for subsequent analysis of the operating trend of the power station.

[0204] "Location" field: stores the shooting location coordinates, expressed in longitude and latitude. Photovoltaic power stations cover a large area, and photovoltaic modules in different locations may be affected by different environmental factors, such as illumination and shading. By recording the shooting location coordinates, the location of the photovoltaic module corresponding to the image can be accurately located, which is convenient for troubleshooting and maintenance.

[0205] "image_id" field: a unique number is assigned to each image to identify and distinguish different images. In the large-scale infrared image database of photovoltaic power plants, there are a large number of images, and the unique image number is used to conveniently manage, query and retrieve images.

[0206] 2. Generate image metadata dictionary

[0207] Based on the above JSON object structure, an image metadata dictionary is generated for each valid infrared image. Assume that the image metadata dictionary is, for a specific image, the corresponding dictionary generation process is as follows:

[0208] d = {"image d ata": Ipixel_data, "timestamp": I timestamp ,″location″:I location , "imageid": I image_id} Among them, I pixel_data is the pixel data of image I, I timestamp is the shooting time of image I, I location is the shooting position coordinate of image I, I image_id is the number of image I.

[0209] Specifically, when serializing the image metadata dictionary, in order to convert the image metadata dictionary into a format suitable for transmission and storage, a serialization function is required. In Python, a common serialization method is to use the `dumps` function of the `json` module. Let the serialization function be serialize(d), which is defined as follows:

[0210] serialize(d) = json.dumps(d)

[0211] Where d is the image metadata dictionary, and the `json.dumps(d)` function converts the dictionary d into a string in JSON format. For example, for the image metadata dictionary:

[0212] d={"image d ata":[[10,12],[20,22]],"timestamp":

[0213] "2024-01-0112:00:00","location":[30.1234,120.5678],"image i d":"000001"}

[0214] After being processed by the serialization function, the generated JSON format string is:

[0215] '"image d ata":[[10,12],[20,22]],"timestamp":

[0216] "2024-01-0112:00:00","location":[30.1234,120.5678],"image i d":"000001"′

[0217] By serializing the metadata dictionary of each valid infrared image, a valid infrared image data sequence containing all image information is obtained.

[0218] To this end, compared with the traditional image data management method, this solution uses the JSON object structure to organize the pixel data and metadata of the image, providing a unified and standardized data structure. This structure is not only easy to understand and maintain, but also convenient for the processor to parse and process, improving the efficiency and accuracy of data processing. In addition, in the JSON object structure, rich metadata such as shooting time, shooting location coordinates and image number are integrated. These metadata are very important for the operation and maintenance management of photovoltaic power stations. By closely combining them with image pixel data, more comprehensive information support is provided for subsequent data analysis, fault diagnosis and power station monitoring.

[0219] Furthermore, serialization using JSON format has good compatibility and readability. JSON format is a lightweight data exchange format widely used in various programming languages ​​and systems. By using the `json.dumps` function for serialization, the processor quickly and efficiently converts the image metadata dictionary into a format suitable for transmission and storage, reducing the overhead of data processing.

[0220] Optionally, the extracting a feature graph of the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph includes:

[0221] Input the preprocessed valid infrared image data sequence into the fault location model one by one, start the forward propagation of the fault location model so that the preprocessed valid infrared image data sequence first enters the convolution layer, and multiple convolution kernels in the convolution layer perform convolution operation on the preprocessed valid infrared image data sequence to extract local features of the image;

[0222] The local features of the image enter the pooling layer for downsampling operation to obtain a compressed feature map;

[0223] The compressed feature map is fused through a fully connected layer to map the compressed feature map into a semantic space and generate a thermal infrared temperature information overview map based on it.

[0224] Specifically, when performing the convolution operation of the convolution layer, in the scenario of photovoltaic power station fault location, the fault location model is used to process the valid infrared image data sequence. When the pre-processed valid infrared image data sequence is input into the convolution layer of the fault location model one by one, the convolution layer is used to extract the local features of the image.

[0225] An image in the input valid infrared image data sequence is I, and its size is H×W×C, where H represents the height of the image, W represents the width of the image, and C represents the number of channels of the image (for infrared images, C is usually 1).

[0226] There are multiple convolution kernels in the convolution layer. Let the size of the i-th convolution kernel be K h ×K w ×C, where K h is the height of the convolution kernel, K w is the width of the convolution kernel.

[0227] The mathematical formula for the convolution operation is:

[0228] Among them, O ij is the value of the convolution operation output feature map at position (i, j), I (i+m)(j+n)c is the pixel value of the cth channel of the input image at position (i+m,j+n), K mnc is the weight value of the cth channel of the convolution kernel at position (m,n), and b is the bias term.

[0229] In infrared images of photovoltaic power plants, different local features may correspond to different fault types. For example, hot spot faults may appear as local high-temperature areas in the image, and these local features can be captured by appropriate convolution kernels. By performing convolution operations on the input image through multiple convolution kernels, rich local image features are extracted, which provide important clues for subsequent fault diagnosis.

[0230] Specifically, during the downsampling operation of the pooling layer, the local features of the image extracted by the convolution layer enter the pooling layer for downsampling. The purpose of the pooling layer is to reduce the amount of data, reduce the computational complexity, and retain the main features of the image.

[0231] Suppose the feature map input to the pooling layer is F, with a size of H1×W1×D, where H1 and W1 are the height and width of the feature map respectively, and D is the number of channels of the feature map.

[0232] The maximum pooling window size is S h ×S w , where S h is the height of the pooling window, S w is the width of the pooling window.

[0233] The mathematical formula for the maximum pooling operation is: Among them, P ij is the value of the pooled output feature map at position (i, j), is the input feature map at position (iS h +m,jS w +n) at the dth channel.

[0234] In the infrared image analysis of photovoltaic power plants, the pooling operation can reduce the amount of data without losing key information, making subsequent processing more efficient. For example, for some larger photovoltaic power plant areas, the feature map extracted by the convolution layer may be large in size. The pooling operation can compress it into a smaller size to facilitate subsequent fully connected layer processing.

[0235] Specifically, during the fusion and mapping of the fully connected layer, the compressed feature map obtained by the pooling layer enters the fully connected layer. The function of the fully connected layer is to fuse the compressed feature map and map it into the semantic space to generate a thermal infrared temperature information overview map.

[0236] Suppose the compressed feature map after the pooling layer is G, its vector is represented by g, and its length is L.

[0237] The weight matrix of the fully connected layer is W, of size M×L, where M is the dimension of the fully connected layer output. The bias vector is b, of length M.

[0238] The operation formula of the fully connected layer is: y = W·g+b, where y is the output vector of the fully connected layer, which is mapped into the semantic space.

[0239] In the fault location of photovoltaic power plants, the output vector of the fully connected layer is further processed to generate a thermal infrared temperature information overview map. This overview map can intuitively display the temperature distribution of different areas in the photovoltaic power plant, helping operation and maintenance personnel to quickly locate areas where faults may exist. For example, through the mapping of the fully connected layer, the feature information in the compressed feature map is converted into the semantic information of the temperature distribution, thereby generating a thermal infrared temperature information overview map, so that the operation and maintenance personnel can directly observe from the map which areas have abnormal temperatures and then determine whether there is a fault.

[0240] Therefore, compared with the traditional image feature extraction method, this fault location model is designed based on the characteristics of infrared images of photovoltaic power stations. Through the carefully designed convolution kernel in the convolution layer, local features related to photovoltaic module failures, such as hot spots and module aging, are more effectively extracted, thus improving the accuracy of fault detection.

[0241] In addition, the downsampling operation of the pooling layer retains the key features of the image while reducing the amount of data. This efficient data dimensionality reduction method makes the subsequent fully connected layer processing more efficient, reduces the computational complexity, and improves the operating efficiency of the model, making it suitable for processing large-scale photovoltaic power station infrared image data.

[0242] Furthermore, the fully connected layer maps the compressed feature map into the semantic space to generate a thermal infrared temperature information overview map. This mapping method converts image features into easy-to-understand temperature distribution information, providing an intuitive basis for fault location for maintenance personnel. Compared with traditional complex data analysis methods, it greatly improves the efficiency and operability of fault location.

[0243] Optionally, locating the photovoltaic power station fault according to the thermal infrared temperature information overview map includes:

[0244] Obtaining fault judgment indicators defined for different fault types, the fault types including hot spot faults, local high temperature faults caused by component short circuits, and uneven temperature faults caused by shielding;

[0245] According to the preset color-temperature mapping relationship in the thermal infrared temperature information overview, the image is divided into regions, and the entire thermal infrared temperature information overview is scanned to segment the temperature regions represented by different colors to obtain the boundary information and area size of each temperature region;

[0246] Matching the fault judgment index with the segmented temperature regions to determine the relative position of the fault region;

[0247] The relative position of the fault area is converted into actual space coordinates in the photovoltaic power station.

[0248] Specifically, when obtaining fault judgment indicators, during the operation of the photovoltaic power station, different types of faults will present different characteristics on the thermal infrared temperature information overview diagram. In order to accurately locate the fault, it is necessary to define corresponding fault judgment indicators for each fault type.

[0249] Hot spot failure is a phenomenon in which some cells are abnormally heated due to differences in cell performance or local shading inside the photovoltaic module. Usually, the temperature of the hot spot failure area is significantly higher than the surrounding normal area. Let T be the hot spot failure judgment index. hs , which represents the temperature difference threshold between the hot spot fault area and the surrounding normal area. In practical applications, T is determined through a large number of experiments and data analysis based on factors such as the component type and environmental conditions of the photovoltaic power station. hs For example, after experimental verification, in a certain type of photovoltaic power station, when the temperature difference between the hot spot fault area and the surrounding normal area exceeds 10°C, it can be judged that there is a hot spot fault, that is, T hs =10℃.

[0250] A short circuit in a component will cause an abnormal increase in current, resulting in a sharp rise in temperature at the short circuit site. For this type of fault, the fault judgment indicator T is defined. sc is the lower limit of the temperature in the short-circuit region of the component. Similarly, by studying the electrical and thermal characteristics of the photovoltaic power station and combining the actual operation data to determine T sc For example, in a power station, when the temperature in a local area is detected to be higher than 80°C, there may be a local high temperature fault caused by a component short circuit, that is, T sc =80℃.

[0251] Shading will cause uneven light to the photovoltaic modules, resulting in uneven temperature distribution. Definition of fault judgment index D t is the temperature non-uniformity threshold. The temperature non-uniformity is measured by calculating the temperature standard deviation of different regions. Let T i represents the average temperature of the ith temperature zone, represents the average temperature of the entire thermal infrared temperature information profile, n is the number of temperature zones, and the calculation formula for temperature non-uniformity U is:

[0252] When U>D t When U>5℃, it can be determined that there is a temperature unevenness fault caused by shading. For example, after analysis, it is determined that in a photovoltaic power station, when U>5℃, it is considered that there is a shading problem, that is, D t =5℃.

[0253] Specifically, when dividing the image area and the temperature area, the thermal infrared temperature information overview uses color to intuitively represent the temperature distribution. It is necessary to divide the image into areas according to the preset color and temperature mapping relationship and segment different temperature areas.

[0254] Assume that the mapping relationship between color and temperature is T=f(C), where T represents temperature and C represents color value (usually represented by color space such as RGB or HSV in images).

[0255] First, the thermal infrared temperature information profile is scanned pixel by pixel. For each pixel (x, y) in the image, its color value C(x, y) is obtained, and then the temperature corresponding to the pixel is calculated according to the mapping relationship T(x, y) = f(C(x, y)).

[0256] Next, we use threshold-based segmentation to divide the image into regions. Taking threshold-based segmentation as an example, let the threshold T th is the temperature segmentation threshold, and the temperature is greater than T th The pixels are divided into an area less than or equal to T th The pixels are divided into another area.

[0257] In this way, the entire thermal infrared temperature information profile is segmented into multiple temperature regions. For each segmented temperature region, its boundary information is obtained through a boundary detection algorithm (such as the Canny edge detection algorithm), and its area size is calculated using a specific algorithm (such as a contour area calculation algorithm).

[0258] Specifically, when the relative position of the fault area is determined, after the fault judgment index and the segmented temperature area are obtained, the two need to be matched to determine the relative position of the fault area.

[0259] For each divided temperature area, calculate its average temperature T avg and temperature non-uniformity U.

[0260] Hot spot fault judgment: If the T in a certain temperature area avg The temperature difference with the surrounding area is greater than T hs , then there may be a hot spot failure in this area.

[0261] Judgment of local high temperature fault caused by component short circuit: If T avg >T sc , then there may be a local high temperature failure caused by a component short circuit in this area.

[0262] Fault judgment of uneven temperature caused by shading: When U>D in a certain temperature area t When the temperature is uneven, there may be a temperature unevenness fault in the area caused by shading.

[0263] Through the above judgment, the temperature area where the fault exists is determined, and its position in the thermal infrared temperature information overview map is used as the relative position of the fault area.

[0264] Specifically, when the relative position is converted into the actual space coordinates, the relative position of the fault area determined in the thermal infrared temperature information overview map needs to be converted into the actual space coordinates in the photovoltaic power station so that the operation and maintenance personnel can accurately find the fault location.

[0265] Assume that the pixel coordinates of the thermal infrared temperature information profile are (x p ,y p ), the actual spatial coordinates of the photovoltaic power station are (x s ,y s ). A coordinate transformation model needs to be established, which is usually based on the imaging principle of the camera and the layout information of the photovoltaic power station.

[0266] Assume that the imaging process of the camera is described by the pinhole camera model, the camera's intrinsic parameter matrix is ​​K, and the extrinsic parameter matrices are R and t (representing the rotation matrix and translation vector, respectively). According to the pinhole camera model, the pixel coordinate (x p ,y p ) and the world coordinates (X, Y, Z) are as follows:

[0267] Where (u,v) is the normalized pixel coordinate.

[0268] In a photovoltaic power station, the layout information of photovoltaic modules is known, that is, the position of each module in the world coordinate system. p ,y p ) is substituted into the pinhole camera model and combined with the layout information of the photovoltaic power station. After a series of coordinate transformations (including rotation, translation, etc.), the actual spatial coordinates (x s ,y s ).

[0269] To this end, compared with traditional fault judgment methods, this solution defines more accurate fault judgment indicators by deeply studying the characteristics of different fault types on thermal infrared images and combining them with the actual operation data of photovoltaic power plants. These indicators can more accurately identify various fault types and improve the accuracy of fault location.

[0270] Furthermore, the use of advanced image segmentation and analysis algorithms can quickly and accurately divide the thermal infrared temperature information profile and segment the temperature areas. By combining the mapping relationship between color and temperature, the information of the thermal infrared image is fully utilized to improve the efficiency and accuracy of fault area identification.

[0271] Furthermore, based on the pinhole camera model and the layout information of the photovoltaic power station, an accurate coordinate conversion model was established. This model can accurately convert the relative position in the thermal infrared temperature information overview map into the actual spatial coordinates in the photovoltaic power station, providing accurate fault location information for operation and maintenance personnel, greatly improving operation and maintenance efficiency.

[0272] The present application provides a computer program product having computer executable instructions stored thereon, and the computer executable instructions are executed to perform the following steps:

[0273] The infrared imager installed on the drone takes images of the photovoltaic power station according to the set time interval or position trigger conditions to generate infrared image data;

[0274] The infrared image data is encrypted and transmitted to a ground control center;

[0275] The ground control center classifies and stores the received ciphertext infrared image data to establish an infrared image database of a photovoltaic power station;

[0276] Preprocessing the infrared image database of the photovoltaic power station to obtain a valid infrared image database;

[0277] Serializing the effective infrared image database to generate an effective infrared image data sequence;

[0278] Extracting feature graphs from the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph;

[0279] The photovoltaic power station fault is located according to the thermal infrared temperature information overview map.

[0280] An embodiment of the present application provides an electronic device, which includes a memory and a processor, wherein the memory stores computer executable instructions, and the processor runs the computer executable instructions to perform the following steps:

[0281] The infrared imager installed on the drone takes images of the photovoltaic power station according to the set time interval or position trigger conditions to generate infrared image data;

[0282] The infrared image data is encrypted and transmitted to a ground control center;

[0283] The ground control center classifies and stores the received ciphertext infrared image data to establish an infrared image database of a photovoltaic power station;

[0284] Preprocessing the infrared image database of the photovoltaic power station to obtain a valid infrared image database;

[0285] Serializing the effective infrared image database to generate an effective infrared image data sequence;

[0286] Extracting feature graphs from the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph;

[0287] The photovoltaic power station fault is located according to the thermal infrared temperature information overview map.

[0288] For exemplary explanations of the computer program product and the electronic device in the above embodiments, see the above Figure 1 .

[0289] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above invention concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

Claims

1. A photovoltaic power station fault location method based on infrared imaging, characterized in that: include: The infrared imager installed on the drone takes images of the photovoltaic power station according to the set time interval or position trigger conditions to generate infrared image data; The infrared image data is encrypted and transmitted to a ground control center; The ground control center classifies and stores the received ciphertext infrared image data to establish an infrared image database of a photovoltaic power station; Preprocessing the infrared image database of the photovoltaic power station to obtain a valid infrared image database; Serializing the effective infrared image database to generate an effective infrared image data sequence; Extracting feature graphs from the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph; The photovoltaic power station fault is located according to the thermal infrared temperature information overview map.

2. A photovoltaic power station fault location method based on infrared imaging according to claim 1, characterized in that: The infrared imager mounted on the drone takes images of the photovoltaic power station according to a set time interval or position trigger condition to generate infrared image data, including: Initializing the shooting parameters of the infrared imager according to the requirements of the photovoltaic power station fault detection on temperature resolution, spatial resolution, temperature measurement range, etc.; A digitized map of the photovoltaic power station is imported into the flight control system of the UAV, and the coordinate ranges corresponding to the areas that need special attention are marked on the digitized map. When the UAV enters these areas of special attention, the infrared imager is automatically triggered to optimize the shooting parameters according to its visual range so as to shoot these areas of special attention based on the optimized shooting parameters.

3. A photovoltaic power station fault location method based on infrared imaging according to claim 1, characterized in that: The encrypting and transmitting the infrared image data to the ground control center comprises: Performing block processing on the infrared image data to obtain a plurality of image data blocks; Perform a series of byte replacement, row shift, column confusion and round key encryption transformations on each image data block and the key in turn, and output the encrypted ciphertext data block; The encrypted ciphertext data block is transmitted to the ground control center through the communication link between the UAV and the ground control center.

4. A photovoltaic power station fault location method based on infrared imaging according to claim 1, characterized in that: The ground control center classifies and stores the received ciphertext infrared image data to establish a photovoltaic power station infrared image database, including: After the ground control center decrypts the received ciphertext infrared image data, it classifies them according to timestamp, shooting location coordinate information, drone number, imager model, and shooting task number; Based on the classification dimension, a data access index is constructed to establish a photovoltaic power station infrared image database.

5. The photovoltaic power station fault location method based on infrared imaging according to claim 1 is characterized in that: The preprocessing of the photovoltaic power station infrared image database to obtain a valid infrared image database comprises: The infrared image database of the photovoltaic power station is enhanced to obtain an effective infrared image database.

6. A photovoltaic power station fault location method based on infrared imaging according to claim 1, characterized in that: The step of serializing the effective infrared image database to generate an effective infrared image data sequence comprises: For each valid infrared image, a JSON object structure is defined, which contains the "image_data" field for storing the pixel data of the image, and other metadata fields, including "timestamp" (shooting time), "location" (shooting location coordinates), and "image_id" (image number). Each field is assigned and organized according to the JSON syntax rules; Based on the JSON object structure, generate an image metadata dictionary; Based on the serialization function, the image metadata dictionary is serialized to generate a valid infrared image data sequence.

7. A photovoltaic power station fault location method based on infrared imaging according to claim 1, characterized in that: The extracting of feature graphs from the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph includes: Input the preprocessed valid infrared image data sequence into the fault location model one by one, start the forward propagation of the fault location model so that the preprocessed valid infrared image data sequence first enters the convolution layer, and multiple convolution kernels in the convolution layer perform convolution operation on the preprocessed valid infrared image data sequence to extract local features of the image; The local features of the image enter the pooling layer for downsampling operation to obtain a compressed feature map; The compressed feature map is fused through a fully connected layer to map the compressed feature map into a semantic space and generate a thermal infrared temperature information overview map based on it.

8. The photovoltaic power station fault location method based on infrared imaging according to claim 1 is characterized in that: The locating the photovoltaic power station fault according to the thermal infrared temperature information overview diagram comprises: Obtaining fault judgment indicators defined for different fault types, the fault types including hot spot faults, local high temperature faults caused by component short circuits, and uneven temperature faults caused by shielding; According to the preset color-temperature mapping relationship in the thermal infrared temperature information overview, the image is divided into regions, and the entire thermal infrared temperature information overview is scanned to segment the temperature regions represented by different colors to obtain the boundary information and area size of each temperature region; Matching the fault judgment index with the segmented temperature regions to determine the relative position of the fault region; The relative position of the fault area is converted into actual space coordinates in the photovoltaic power station.

9. A computer program product having computer executable instructions stored thereon, wherein the computer executable instructions are executed to perform the following steps: The infrared imager installed on the drone takes images of the photovoltaic power station according to the set time interval or position trigger conditions to generate infrared image data; The infrared image data is encrypted and transmitted to a ground control center; The ground control center classifies and stores the received ciphertext infrared image data to establish an infrared image database of a photovoltaic power station; Preprocessing the infrared image database of the photovoltaic power station to obtain a valid infrared image database; Serializing the effective infrared image database to generate an effective infrared image data sequence; Extracting feature graphs from the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph; The photovoltaic power station fault is located according to the thermal infrared temperature information overview map.

10. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory stores computer executable instructions, and the processor runs the computer executable instructions to perform the following steps: The infrared imager installed on the drone takes images of the photovoltaic power station according to the set time interval or position trigger conditions to generate infrared image data; The infrared image data is encrypted and transmitted to a ground control center; The ground control center classifies and stores the received ciphertext infrared image data to establish an infrared image database of a photovoltaic power station; Preprocessing the infrared image database of the photovoltaic power station to obtain a valid infrared image database; Serializing the effective infrared image database to generate an effective infrared image data sequence; Extracting feature graphs from the effective infrared image data sequence based on the fault location model to generate a thermal infrared temperature information overview graph; The photovoltaic power station fault is located according to the thermal infrared temperature information overview map.

Images