Photovoltaic power station operation and maintenance management method

Abstract

The invention discloses a photovoltaic power station operation and maintenance management method, and relates to the technical field of photovoltaic power station management.The photovoltaic power station operation and maintenance management method includes the steps that a rail type unmanned aerial vehicle parking apron is installed over a power station, related unmanned aerial vehicles are arranged, the flight path and height are planned according to component layout, components are divided into small areas, priorities are determined, and fixed cameras are arranged for auxiliary monitoring; establishing a preliminary screening mechanism to realize image acquisition; after preprocessing and feature extraction are carried out on the collected image, a fault diagnosis model is input to judge the position and severity of a fault component, and a result is pushed; operation and maintenance personnel make a scheme according to the fault degree; after maintenance, images are collected again to verify the restoration effect, data are stored, and the fault diagnosis model is updated and optimized; the operation and maintenance management efficiency and accuracy are improved, the cost is reduced, the decision timeliness and the maintenance efficiency and quality are improved, refined operation and maintenance management is achieved, the economic benefits, safety and reliability of the power station are enhanced, and technical guarantee is provided for long-term stable operation of the power station.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power station management, and specifically to a method for operation and maintenance management of a photovoltaic power station. Background Art

[0002] In the existing operation and maintenance management of photovoltaic power stations, the detection and location of photovoltaic module failures mainly rely on manual inspections or simple electrical performance monitoring. Manual inspections are inefficient and greatly affected by subjective factors, easily missing some minor faults. Electrical performance monitoring can only detect the overall decline in power generation performance and is difficult to accurately determine the specific location of the faulty module. When the power station is large in scale, troubleshooting faulty modules is like looking for a needle in a haystack, consuming a large amount of human and time costs and seriously affecting the power generation efficiency and economic benefits of the power station. Therefore, the existing technology has obvious limitations in the operation and maintenance management of photovoltaic power stations, and a more efficient and accurate fault detection and location method is needed.

[0003] With the continuous expansion of the scale of photovoltaic power stations and the progress of technology, higher requirements are put forward for the operation and maintenance management of photovoltaic power stations. Traditional operation and maintenance methods can no longer meet the efficient and intelligent management needs of modern photovoltaic power stations. Especially in terms of fault detection and handling, it is necessary to quickly and accurately locate faulty modules and evaluate the severity of faults in order to take corresponding maintenance measures in a timely manner. In addition, with the development of unmanned aerial vehicle technology, image recognition technology, and deep learning algorithms, new technical means are provided for the operation and maintenance management of photovoltaic power stations. How to effectively integrate these advanced technologies to achieve the automation and intelligence of photovoltaic power station operation and maintenance management has become an important direction for technological development.

[0004] To solve the above defects, a technical solution is provided now. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems of low efficiency of manual inspections in the existing operation and maintenance management of photovoltaic power stations, inability of electrical performance monitoring to accurately locate faulty modules, and lack of efficient fault detection and handling means, and to propose a method for operation and maintenance management of a photovoltaic power station.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A method for operation and maintenance management of a photovoltaic power station includes the following steps:

[0008] S1: Build an image acquisition system. Install an orbital UAV apron over the photovoltaic power station and equip it with relevant UAVs. Plan the flight path and altitude according to the module layout, divide the power station modules into small areas to determine priorities, arrange fixed cameras for auxiliary monitoring, establish a preliminary screening mechanism, and achieve image acquisition of the power station;

[0009] S2: Image preprocessing and feature extraction. The images collected by the UAV and the fixed camera are transmitted to the preprocessing server in real time. After preprocessing such as denoising, the surface features of the visible light image and the temperature distribution features of the infrared thermal imaging image are extracted respectively using deep learning algorithms;

[0010] S3: Fault component identification and location. The image feature data is input into the fault diagnosis model with a CNN architecture. Combining the UAV positioning and component layout information, the position coordinates of the fault component are judged. The severity is judged according to the evaluation criteria of different fault types, and the results are pushed to the mobile terminal of the operation and maintenance personnel;

[0011] S4: Operation and maintenance strategy formulation and execution. The operation and maintenance personnel formulate a plan according to the fault degree, use AR assistance during maintenance, and conduct tests after completion;

[0012] S5: Verification after fault repair and data update. After repair, the UAV is used to collect images of the repaired components again to verify the repair effect. The fault-related data is stored in the database, and the fault diagnosis model is updated and optimized.

[0013] Furthermore, the specific process of S1 is as follows:

[0014] Install an orbital UAV apron over the PV power station, and be equipped with multiple UAVs with a camera resolution not lower than the preset pixels and infrared thermal imaging function; the UAVs perform full-coverage image collection of the PV power station at regular intervals according to the preset flight path and altitude, including visible light images and infrared thermal imaging images;

[0015] Identify and locate the obstacles inside and around the power station, including utility poles, substation equipment, buildings, and trees; in the flight path planning, use the obstacle avoidance sensor of the UAV and the pre-set geographic information system data to make the UAV automatically bypass these obstacles, ensuring flight safety while guaranteeing the integrity of image collection;

[0016] At the same time, establish a preliminary screening mechanism. After analyzing an anomaly, take detailed pictures of the anomaly point.

[0017] Furthermore, the planning process of the UAV flight path and altitude in S1 is as follows:

[0018] First, conduct a detailed surveying and mapping and analysis of the component layout of the PV power station to determine the arrangement, number of rows, number of columns of the components, as well as the interval and orientation information between the phalanxes;

[0019] Divide the power station components into several small areas, determine the priority for each small area according to the historical failure data, component aging degree, geographical location or environmental factors of the small area; conduct comprehensive analysis by collecting the level parameters of each small area, and the level parameters include: the number of historical failures, the average service time of area components, geographical location score, and environmental factor score;

[0020] Among them, the number of historical failures is determined by establishing a photovoltaic power station failure record database, recording the time and location of each failure occurrence. For each small area, count the number of failures that occurred in the past year;

[0021] The average service time of area components is determined by consulting the component installation records and procurement batch information of the power station, determining the installation time and initial service life data of the components in each small area, and calculating the average service time of the components in each small area;

[0022] The geographical location score is determined according to the geographical information system data of the power station to determine the positional relationship of each small area relative to geographical factors such as sand sources and obstacles; for small areas close to sand sources, trees, and building obstacles, if the distance from the sand source or obstacle is within 0 - 50 meters, give 90 points; if the distance is within 50 - 100 meters, give 70 points; if the distance is more than 100 meters, give 40 points;

[0023] The environmental factor score is obtained by installing multiple environmental monitoring sensors in the power station, including light intensity sensors, temperature sensors, etc., distributed in each small area to collect environmental data in real - time; for the situation of uneven light, count the light intensity data of each small area every day in the past month, and calculate its light uniformity index; for areas where the temperature difference is greater than the preset temperature difference, analyze the temperature change curve of each small area in a day, and calculate its temperature range; after normalizing the obtained light uniformity and temperature range, use them as the two right - angled sides of a right - angled triangle respectively, connect the other side of the right - angled triangle to form a right - angled triangle, and establish a triangular pyramid model with the correction coefficient as the height, calculate the volume of the triangular pyramid model, and record it as the environmental judgment value, and use this environmental judgment value as the standard for measuring the environmental factor score;

[0024] Then calibrate the obtained number of historical failures, average service time of area components, geographical location score, and environmental factor score as l h, t i, dp, and hu respectively. After normalization, substitute them into the following formula: To obtain the level judgment value JPZ, where ω1, ω2, ω3, and ω4 are the preset weight coefficients of the number of historical failures, average service time of area components, geographical location score, and environmental factor score respectively, and use the obtained level judgment value JPZ as the priority judgment standard for small areas;

[0025] Compare the obtained grading value JPZ with three preset grading value intervals, where the three preset grading value intervals are respectively set with the first priority level, the second priority level, and the third priority level. When it is determined which grading value interval the grading value JPZ belongs to, the priority level of this small area is determined;

[0026] Collect small areas with different priority levels at different frequencies and the number of shooting angles respectively, and ensure full coverage of all areas.

[0027] Further, the preliminary screening mechanism in S1 includes:

[0028] When the drone performs the first image acquisition according to the preset flight altitude, use the image acquisition system carried on the drone to simultaneously obtain visible light images and infrared thermal imaging images, and transmit the image data to the processing unit on the ground in real time;

[0029] For visible light images, perform preliminary analysis using an image recognition algorithm based on deep learning; the image recognition algorithm based on deep learning has been pre-trained with image samples of normal photovoltaic modules and modules with different defects to identify abnormal areas in the image and mark their positions and types;

[0030] For infrared thermal imaging images, detect areas with abnormal temperatures by comparing with a preset normal component temperature range model, and record their positions and temperature values;

[0031] Integrate the analysis results of visible light images and infrared thermal imaging images to generate a preliminary anomaly report, including the location information and anomaly types of the abnormal areas;

[0032] The drone automatically adjusts its flight path according to the location information in the preliminary anomaly report and flies towards the first anomaly point; when it is within 15 meters of the anomaly point, gradually reduce the flight altitude to the range of 5 - 10 meters, and at the same time adjust the angle and focal length of the camera to ensure that the detailed features of the abnormal area can be clearly photographed;

[0033] During the detailed shooting process, adopt a higher resolution image acquisition mode and shoot the abnormal area from multiple angles, including directly above and at a 45-degree angle, to obtain more comprehensive information; at the same time, use the positioning system of the drone to record the location information corresponding to each detailed image for subsequent maintenance personnel to locate the specific fault point;

[0034] After completing the detailed shooting of each anomaly point, the drone continues to fly towards other anomaly points in sequence according to the order in the anomaly report, repeating the detailed shooting process until all anomaly points have completed the shooting task;

[0035] After shooting, integrate all the detailed image data with the preliminary screening abnormal report, including the high-definition images and positions of each abnormal point.

[0036] Furthermore, the specific operation steps of S2 are as follows:

[0037] Transmit the image data collected by the drone and the fixed camera to the image preprocessing server in real time; first, perform preprocessing operations such as denoising, gray correction, and contrast enhancement on the images to improve the clarity and quality of the images.

[0038] Use an image feature extraction algorithm based on deep learning to extract features from the preprocessed images; for visible light images, extract the surface color, texture, and shape features of the components to determine whether there are physical damages to the components; for infrared thermal imaging images, extract the temperature distribution features of the components, and identify the areas with hot spot faults or abnormal temperatures by comparing with the component temperature distribution model under normal working conditions.

[0039] Furthermore, the specific operation steps of S3 are as follows:

[0040] Input the extracted image feature data into a pre-trained fault diagnosis model; the model adopts a convolutional neural network architecture and is trained with different photovoltaic component fault sample data to classify and locate the fault components.

[0041] When the model determines that there are fault components, combine the flight positioning data of the drone and the component layout information of the power station to determine the position coordinates of the fault components in the power station, and judge the severity of the fault in combination with the fault type.

[0042] And push the fault type, location information, and the evaluation result of the fault severity to the mobile devices of the operation and maintenance personnel in real time.

[0043] Furthermore, the specific operation steps of judging the severity of the fault in S3 are as follows:

[0044] Rupture degree evaluation: For the component rupture situation in physical damage, calculate the proportion of the rupture area to the total area of the component through image analysis; first, perform binary processing on the extracted visible light image, mark the component area and the rupture area with different values respectively, and use the adaptive threshold method for binary processing. The formula is where T is the threshold, W and H are the width and height of the image, and I(x, y) is the pixel value at the coordinate.

[0045] Then count the number of pixels N in the rupture area broken and the total number of pixels N of the component total , rupture ratio When P broken<0.1, it is judged as a minor fault; when 0.1 ≤ P broken <0.3, it is a moderate fault; when P broken ≥0.3, it is a severe fault;

[0046] Deformation degree evaluation: For the deformation of components, the geometric features in the image are used to measure; the contour information of the component edges is obtained through feature extraction, and the difference degree between the deformed component and the normal component contour is calculated. The Hough transform is used to detect the edge lines of the components. For a normal rectangular component, its edges are four straight lines;

[0047] Calculate the average of the angular deviation and displacement deviation between the deformed edge line and the normal position line; if the averages of both the angular deviation and displacement deviation are less than the preset threshold, it is judged as a minor deformation, that is, a minor fault; if one of the deviations exceeds the preset threshold but does not reach the preset upper limit, it is judged as a moderate fault; if the deviation exceeds the upper limit, it is judged as a severe fault;

[0048] Severity judgment of hot spot fault type: Using infrared thermal imaging images, first determine the hot spot area; by setting the temperature threshold T h , mark the area with a temperature higher than T h as the hot spot area;

[0049] Calculate the area S h of the hot spot area and the total area S t of the component, and the hot spot area ratio At the same time, calculate the average temperature T a of the hot spot area and the average temperature T n of the normal component, and the temperature difference ΔT = T n - T n ;

[0050] If P o <0.1 and ΔT<10 °C, it is judged as a minor fault; if 0.1 ≤ P o <0.3 or 10 °C ≤ ΔT<20 °C, it is judged as a moderate fault; if P o >0.3 or ΔT>20 °C, it is judged as a severe fault;

[0051] Severity judgment of short - circuit fault type: For short - circuit faults, in addition to judging from image features, electrical parameters also need to be combined; from the perspective of the image, a short - circuit will cause abnormal heating areas or color changes on the component surface; use the same method as the hot spot fault to determine the abnormal heating area, calculate the area ratio and temperature difference; at the same time, check whether there are charred and discolored areas, and determine the degree of color change through the analysis of the image color space;

[0052] If heat generation and discoloration with an area ratio less than 0.1 are found, and the electrical parameters show that the short-circuit current is less than 20% of the rated short-circuit current of the component, it is determined as a minor fault; if the area ratio is between 0.1 and 0.3, or the temperature difference of heat generation is between 10% and 20%, and the short-circuit current is between 20% and 50% of the rated short-circuit current, it is determined as a moderate fault; if the area ratio is greater than 0.3, the temperature difference is greater than 20%, and the short-circuit current is greater than 50% of the rated short-circuit current, it is determined as a severe fault;

[0053] For dust occlusion faults, it is judged by analyzing the gray histogram of visible light images; the gray histogram of normal components has a distribution pattern, and when there is dust occlusion, the histogram will change; calculate the entropy of the gray histogram where \(L\) is the gray level and \(p(i)\) is the probability that pixels with a gray value of \(i\) appear; the smaller the entropy value, the more concentrated the gray distribution of the image and the more serious the dust occlusion;

[0054] If the decrease amplitude of the entropy value is less than 10% compared with the entropy value of a normal component without dust occlusion, it is determined as a minor fault; if the decrease amplitude of the entropy value is between 10% and 30%, it is determined as a moderate fault; if the decrease amplitude of the entropy value is greater than 30%, it is determined as a severe fault.

[0055] Furthermore, the specific operation steps of S4 are as follows:

[0056] The operation and maintenance personnel formulate a targeted operation and maintenance plan according to the received fault information;

[0057] For minor faults, record the location, type, and discovery time of the faulty component, and mark it on the monitoring interface or map; maintain it within 3 - 5 days, prepare tools according to the fault; collect image and temperature data every 30 minutes and draw a trend curve; after an appearance inspection, handle the fault, conduct tests and comparisons, and record the details;

[0058] For moderate faults, a professional team will go to the site with tools within 1 hour and mark the fault condition; use a thermal imager and an electroluminescence detector to check the fault, analyze the cause and formulate a plan to upload to the system; repair according to the fault, monitor during the process to ensure the repair effect; comprehensively test the performance parameters and compare with the rated values, if qualified, mark it as normal operation and include it in the regular inspection;

[0059] For severe faults, remotely stop the relevant equipment, block the site, and set warning signs; a professional emergency repair team will arrive at the site with tools within 30 minutes to evaluate; disassemble and replace the damaged components; comprehensively test the electrical performance and safety after repair, if qualified, resume operation, record the data and update the status, and strengthen the inspection;

[0060] During the maintenance process, maintenance personnel use augmented reality technology on mobile devices to overlay and display detailed information of the faulty components, including internal structure, electrical connection method, and marked fault points, in the actual maintenance scenario to assist maintenance personnel in troubleshooting and repair operations, improving maintenance efficiency and quality.

[0061] Further, the specific operation steps of S5 are as follows:

[0062] After the maintenance is completed, use the drone again to collect images of the repaired components, and re-enter the collected image data into the fault diagnosis model for verification to ensure that the fault has been completely repaired;

[0063] At the same time, store the relevant data of this fault, including the original image of the faulty component, fault type, maintenance process record, and image after repair, in the operation and maintenance database of the power station, and update and optimize the fault diagnosis model.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] (1) In the present invention, by integrating the drone image acquisition system, deep learning algorithm, and convolutional neural network (CNN) fault diagnosis model, the efficiency and accuracy of photovoltaic power station operation and maintenance management are significantly improved; the full-coverage image acquisition of the drone combined with the auxiliary monitoring of fixed cameras ensures the continuity and integrity of power station monitoring. At the same time, the application of the deep learning algorithm makes the features extracted from the images more accurate, and the fault diagnosis model can quickly identify and locate the faulty components; not only reducing the need for manual inspections, lowering operation and maintenance costs, but also improving the accuracy and response speed of fault detection;

[0066] (2) In the present invention, it allows operation and maintenance personnel to formulate targeted operation and maintenance plans according to the severity of the fault, and receive fault information in real time through mobile devices, improving the timeliness of decision-making; during the maintenance process, the application of augmented reality technology further assists maintenance personnel in quickly and accurately troubleshooting and repairing operations, enhancing maintenance efficiency and quality. The verification step and data update mechanism after fault repair ensure the continuous optimization of the fault diagnosis model, providing more reliable technical support for subsequent operation and maintenance work;

[0067] (3) In the present invention, the photovoltaic power station can achieve more refined operation and maintenance management, reduce the power generation efficiency loss caused by faults, and enhance the economic benefits of the power station; at the same time, the fast and accurate fault detection and handling mechanism also reduces the safety risks of power station operation, improving the safety and reliability of the photovoltaic power station. It not only improves the operation and maintenance management level of the photovoltaic power station, but also provides a solid technical guarantee for the long-term stable operation of the power station. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] For the convenience of those skilled in the art to understand, the present invention will be further described below in conjunction with the accompanying drawings;

[0069] Figure 1 It is a flowchart of the method of the present invention. Specific embodiments

[0070] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0071] It should be understood that the terms "including" and "comprising" used in the specification and claims of this disclosure indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0072] It should also be understood that the terms used in this disclosure specification are only for the purpose of describing specific embodiments and are not intended to limit this disclosure. As used in this disclosure specification and claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms. It should be further understood that the term "and / or" used in this disclosure specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0073] As Figure 1 shown, a method for operation and maintenance management of a photovoltaic power station includes the following steps:

[0074] Building an image acquisition system, installing an orbital UAV apron over the photovoltaic power station and equipping it with relevant UAVs, planning flight paths and altitudes according to factors such as component layout, dividing the power station components into small areas to determine priorities, arranging fixed cameras for auxiliary monitoring, establishing a preliminary screening mechanism, and realizing image acquisition of the power station;

[0075] Installing an orbital UAV apron over the photovoltaic power station and equipping it with multiple UAVs with high-resolution cameras (resolution not lower than the preset pixels) and infrared thermal imaging functions; the UAVs perform full-coverage image acquisition of the photovoltaic power station at regular intervals according to the preset flight paths and altitudes, including visible light images and infrared thermal imaging images; the process of planning the flight paths and altitudes of the UAVs is as follows:

[0076] First, conduct a detailed surveying and mapping and analysis of the component layout of the photovoltaic power station to determine the arrangement, number of rows, number of columns, intervals between phalanxes, and orientation information of the components; divide the power station components into several small areas, determine the priority for each small area, and determine it based on the historical fault data, component aging degree, geographical location (such as near sand and dust sources, areas vulnerable to occlusion, etc.) or environmental factors (such as areas with uneven illumination, large temperature differences); conduct a comprehensive analysis by collecting the level parameters of each small area, and the level parameters include: historical fault times, average service time of area components, geographical location score, and environmental factor score;

[0077] Among them, the historical fault times are obtained by establishing a fault record database for the photovoltaic power station, recording the time and location (accurate to the specific small area) of each fault occurrence. For each small area, count the number of faults that occurred within the past year (or other set time periods, such as half a year, two years, etc., determined according to the actual operation situation and data accumulation degree of the power station); the average service time of area components is determined by referring to the component installation records and procurement batch information of the power station, determining the installation time and initial service life data of the components in each small area, and calculating the average service time of the components in each small area; the geographical location score is determined based on the geographical information system (GIS) data of the power station to determine the positional relationship of each small area relative to geographical factors such as sand and dust sources and occluders; for small areas near sand and dust sources, trees, buildings and other occluders, if the distance from the sand and dust source or occluder is within 0 - 50 meters, give 90 points; if the distance is within 50 - 100 meters, give 70 points; if the distance is more than 100 meters, give 40 points; the environmental factor score is obtained by installing multiple environmental monitoring sensors in the power station, including light intensity sensors, temperature sensors, etc., distributed in each small area to collect environmental data in real time; for the situation of uneven illumination, count the daily light intensity data of each small area in the past month and calculate its light uniformity index; for areas with large temperature differences, analyze the temperature change curve of each small area within a day and calculate its temperature range; after normalizing the obtained light uniformity and temperature range, use them as the two right-angled sides of a right-angled triangle respectively, connect the other side of the right-angled triangle to form a right-angled triangle, and establish a triangular pyramid model with the correction coefficient as the height, calculate the volume of this triangular pyramid model, and record it as the environmental judgment value, and use this environmental judgment value as the standard for measuring the environmental factor score;

[0078] Then, calibrate the obtained historical fault times, average service time of area components, geographical location score, and environmental factor score as l h, t i, dp, and hu respectively. After normalization, substitute them into the following formula: can To obtain the level judgment value JPZ, where ω1, ω2, ω3, and ω4 are the preset weight coefficients of the historical failure times, average usage time of regional components, geographical location score, and environmental factor score respectively, and use the obtained level judgment value JPZ as the priority judgment standard for the small area; compare the obtained level judgment value JPZ with three preset level judgment value intervals, and the three preset level judgment value intervals are respectively set with the first priority level, the second priority level, and the third priority level. After determining the level judgment value interval to which JPZ belongs, determine the priority level of the small area; collect different priority level small areas at different frequencies and the number of shooting angles respectively, and ensure full coverage of all areas.

[0079] According to the order of priority from high to low, plan the image acquisition path for each small area in turn. For the small areas of the first priority level, drones are preferably arranged to start from the edge of the power station or a convenient access position and shoot along a relatively dense path to ensure full coverage and shooting from multiple angles. For example, use a snake-shaped path or a spiral path (gradually advancing from the edge of the area to the center), and at the same time set multiple stop points on the path for multi-angle shooting (such as angles of 0°, 45°, 90°, etc.) to obtain component status information more clearly; for the small areas of the second priority level, after completing the shooting of the first priority areas, the path can be relatively sparse, but still ensure full coverage of the entire area. A zigzag-like path can be used, and the stop point angles are selected at key positions (such as the four corners and the center position of the area, etc.) to obtain representative image data while taking into account efficiency and comprehensiveness; the small areas of the third priority level are collected last, and the path design is mainly for efficient coverage. For example, use a simple straight back-and-forth path and perform a small number of multi-angle shootings (such as 0° and 90°) at key positions to quickly complete the inspection of the area while ensuring a certain image quality.

[0080] Identify and locate the obstacles inside and around the power station, including utility poles, substation equipment, buildings, and trees; in the flight path planning, use the obstacle avoidance sensor of the drone and the pre-set geographical information system (GIS) data to make the drone automatically bypass these obstacles to ensure flight safety while ensuring the integrity of image acquisition.

[0081] At the same time, establish a preliminary screening mechanism. After analyzing an anomaly, perform detailed shooting for the anomaly point. The specific process is as follows:

[0082] When the drone conducts the first image acquisition at the preset flight altitude, the image acquisition system carried on the drone is used to simultaneously obtain visible light images and infrared thermal imaging images, and the image data is transmitted to the ground processing unit in real time; for the visible light images, a preliminary analysis is carried out using an image recognition algorithm based on deep learning; this algorithm has been pre-trained with image samples of normal photovoltaic modules and modules with common defects (such as surface dust, stains, bird droppings, obvious physical damage, etc.), quickly identifying the abnormal areas in the images and marking their positions and approximate types;

[0083] For the infrared thermal imaging images, by comparing with the preset normal component temperature range model, the areas with abnormal temperatures are detected, and their positions and temperature values are recorded; integrating the analysis results of the visible light images and the infrared thermal imaging images, a preliminary abnormal report is generated, including the position information (represented by the row and column numbers or coordinates of the components) and abnormal types (physical damage, temperature anomaly) of the abnormal areas;

[0084] The drone automatically adjusts its flight path according to the position information in the preliminary abnormal report and flies towards the first abnormal point; when approaching the abnormal point, it gradually reduces the flight altitude to the range of 5 - 10 meters, and at the same time adjusts the angle and focal length of the camera to ensure that the detailed features of the abnormal area can be clearly photographed; during the detailed shooting process, a higher resolution image acquisition mode is adopted, and the abnormal area is photographed from multiple angles, including directly above and at a 45-degree angle, to obtain more comprehensive information; at the same time, using the positioning system of the drone, the position information corresponding to each detailed image is recorded so that the subsequent maintenance personnel can quickly locate the specific fault point; after completing the detailed shooting of each abnormal point, the drone continues to fly towards other abnormal points in the order in the abnormal report and repeats the above detailed shooting process until all abnormal points have been photographed; after the shooting is completed, all the detailed image data is integrated with the preliminary screened abnormal report, including the high-definition images and positions of each abnormal point.

[0085] A number of fixed cameras are reasonably arranged at high places around the photovoltaic power station for assisting in monitoring the flight area of the drone and for conducting preliminary image acquisition of the power station when the drone is unable to operate (such as during the no-fly period caused by bad weather), ensuring the continuity and integrity of image acquisition.

[0086] Image preprocessing and feature extraction, the images collected by the drone and the fixed cameras are transmitted to the preprocessing server in real time. After preprocessing such as denoising, deep learning algorithms are used to extract the surface features of the components in the visible light images and the temperature distribution features in the infrared thermal imaging images respectively;

[0087] Transmit the image data collected by the drone and fixed cameras to the image preprocessing server in real time; first, perform preprocessing operations such as denoising, gray correction, and contrast enhancement on the images to improve the clarity and quality of the images; use an image feature extraction algorithm based on deep learning to extract features from the preprocessed images; for visible light images, extract the surface color, texture, and shape features of the components to determine whether there are physical damages (such as cracks, deformations, obstacles, etc.) on the components; for infrared thermal imaging images, extract the temperature distribution features of the components, and identify the areas with hot spot faults or abnormal temperatures by comparing with the component temperature distribution model under normal working conditions.

[0088] Fault component identification and location: Input the image feature data into the fault diagnosis model with a CNN architecture, combine the drone positioning and component layout information to determine the position coordinates of the fault components, and judge the severity according to the evaluation criteria for different fault types (physical damage, hot spot, short circuit, dust occlusion, etc.), and push the results to the mobile terminals of the operation and maintenance personnel.

[0089] Input the extracted image feature data into a pre-trained fault diagnosis model; the model adopts a convolutional neural network (CNN) architecture and is trained with different photovoltaic component fault sample data (including fault images of different types and degrees and the corresponding actual fault situations) to classify and locate the fault components; when the model determines that there are fault components, combine the flight positioning data of the drone and the component layout information of the power station to determine the position coordinates of the fault components in the power station, and judge the severity of the fault according to the fault type. Specifically:

[0090] Crack degree evaluation: For the component crack situation in physical damage, calculate the proportion of the crack area in the total area of the component through image analysis. First, perform binary processing on the extracted visible light image, mark the component area and the crack area with different values (for example, the component area is 0 and the crack area is 1), and use the adaptive threshold method for binary processing. The formula is where T is the threshold, W and H are the width and height of the image, and I(x, y) is the pixel value at the coordinate.

[0091] Then count the number of pixels N in the crack area broken and the total number of pixels N of the component total , and the crack ratio When P broken < 0.1, it is judged as a minor fault; when 0.1 ≤ P broken < 0.3, it is a moderate fault; when P broken ≥ 0.3, it is a severe fault.

[0092] Deformation degree assessment: For the deformation of components, geometric features in the image are used to measure. By feature extraction, the contour information of the component edge is obtained, and the difference degree between the deformed component and the normal component contour is calculated. The Hough transform is used to detect the edge lines of the component. For a normal rectangular component, its edges should be four approximate straight lines; calculate the average value of the angle deviation and displacement deviation between the deformed edge line and the normal position line. If the average values of both the angle deviation and displacement deviation are less than the preset threshold, it is judged as slight deformation, that is, a slight fault; if one of the deviations exceeds the preset threshold but does not reach the preset upper limit, it is judged as a moderate fault; if the deviation exceeds the upper limit, it is judged as a severe fault;

[0093] Severity judgment of hot spot fault type: Using the infrared thermal imaging image, first determine the hot spot area; by setting the temperature threshold T h (This threshold is determined according to the normal operating temperature range of the component and empirical values. For example, it is 10°C higher than the normal temperature), mark the area with a temperature higher than T h as the hot spot area; calculate the area S h of the hot spot area and the total area S t of the component, and the proportion of the hot spot area Meanwhile, calculate the average temperature T a of the hot spot area and the temperature difference ΔT = T n - T n - T n with the average temperature T o of the normal component; if P o < 0.1 and ΔT < 10°C, it is judged as a slight fault; if 0.1 ≤ P o < 0.3 or 10°C ≤ ΔT < 20°C, it is judged as a moderate fault; if P o > 0.3 or ΔT > 20°C, it is judged as a severe fault;

[0094] Severity judgment of short - circuit fault type: For short - circuit faults, in addition to judging from image features, electrical parameters also need to be combined (if there is an electrical parameter monitoring system to cooperate). From the perspective of the image, a short - circuit may cause abnormal heating areas or color changes (such as local charred marks) on the component surface; use the same method as the hot spot fault to determine the abnormal heating area, calculate its area ratio and temperature difference; meanwhile, check if there are charred and discolored areas, and determine the degree of color change by analyzing the image color space (such as RGB or HSV color space);

[0095] If heat generation and discoloration with an area ratio less than 0.1 are found, and the electrical parameters show that the short-circuit current is less than 20% of the rated short-circuit current of the component, it is determined as a minor fault; if the area ratio is between 0.1 - 0.3, or the temperature difference of heat generation is between 10% - 20%, and the short-circuit current is between 20% - 50% of the rated short-circuit current, it is determined as a moderate fault; if the area ratio is greater than 0.3, the temperature difference is greater than 20%, and the short-circuit current is greater than 50% of the rated short-circuit current, it is determined as a severe fault;

[0096] For dust occlusion faults, it is judged by analyzing the gray histogram of visible light images. The gray histogram of a normal component has a certain distribution pattern. When there is dust occlusion, the histogram will change; calculate the entropy of the gray histogram where L is the number of gray levels (256), and p(i) is the probability that a pixel with gray value i appears. The smaller the entropy value, the more concentrated the gray distribution of the image, and the more serious the dust occlusion; if the decrease in the entropy value is less than 10% (compared with the entropy value of a normal component without dust occlusion), it is determined as a minor fault; if the decrease in the entropy value is between 10% - 30%, it is determined as a moderate fault; if the decrease in the entropy value is greater than 30%, it is determined as a severe fault;

[0097] And push the fault type (such as physical damage, hot spot fault, short-circuit fault, etc.), location information, and the evaluation result of the fault severity to the mobile devices of operation and maintenance personnel (such as mobile phones, tablets) in real time.

[0098] Formulation and execution of operation and maintenance strategies. Operation and maintenance personnel formulate plans according to the fault degree. For minor faults, conduct regular maintenance and remote monitoring. For moderate faults, the professional team responds quickly for detection and repair. For severe faults, shut down urgently and block, and then conduct professional emergency repair. Use AR assistance during the repair and conduct tests after completion;

[0099] Operation and maintenance personnel formulate targeted operation and maintenance plans according to the received fault information;

[0100] For minor faults, record the location (number, row and column, coordinates), type, and discovery time of the faulty component, and mark it on the monitoring interface or map; perform maintenance within 3 - 5 days, and prepare tools according to the fault (cleaning tools, heat sinks, etc.); collect image and temperature data every 30 minutes, draw a trend curve, and handle it in advance if it deteriorates; after visual inspection, handle the fault (cleaning, heat dissipation, sealing scratches), conduct tests and comparisons, and record the details; for moderate faults, a professional team should arrive at the scene within 1 hour with tools (detectors, welding equipment, etc.) and mark the fault condition; use a thermal imager, electroluminescence detector, etc. to check for faults, analyze the causes, formulate a plan, and upload it to the system; repair the fault (cleaning, replacing solar cells, repairing encapsulation), monitor during the process to ensure the repair effect; comprehensively test the performance parameters and compare them with the ratings, and mark normal operation if qualified, and include it in the regular inspection; for severe faults, remotely stop the relevant equipment, block the scene, and set warning signs; a professional emergency repair team (electrical, photovoltaic, fire protection experts) should arrive at the scene within 30 minutes with tools (welding, detection, fire extinguishing, replacement tools, etc.) to evaluate; disassemble and replace the damaged components (broken components, short - circuited lines, etc.), with comprehensive safety protection and precise operation throughout the process; after repair, comprehensively test the electrical performance and safety, and resume operation if qualified, record the data, update the status, and strengthen the inspection;

[0101] During the repair process, maintenance personnel use augmented reality (AR) technology on mobile devices to overlay the detailed information of the faulty component (including internal structure, electrical connection method, fault point marking) on the actual repair scene, assisting maintenance personnel to quickly and accurately conduct fault troubleshooting and repair operations, improving the repair efficiency and quality.

[0102] Verification and data update after fault repair: After repair, use a drone to collect images of the repaired component again to verify the repair effect, store the fault - related data in the database, and update and optimize the fault diagnosis model;

[0103] After the repair is completed, use a drone to collect image data of the repaired component again and input the collected image data back into the fault diagnosis model for verification to ensure that the fault has been completely repaired; at the same time, store the relevant data of this fault (including the original image of the faulty component, fault type, repair process record, image after repair, etc.) in the power station's operation and maintenance database, and update and optimize the fault diagnosis model to continuously improve the accuracy of its fault identification and location, providing more reliable technical support for subsequent operation and maintenance work.

[0104] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only the specific embodiments. Obviously, according to the content of this specification, many modifications and variations can be made. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A photovoltaic power station operation and maintenance management method, characterized in that: The following steps are involved: S1: Build an image acquisition system, install a rail-type drone landing pad and equip relevant drones above the photovoltaic power station, plan the flight path and altitude according to the component layout, divide the power station components into small areas to determine the priority, arrange fixed cameras for auxiliary monitoring, establish a preliminary screening mechanism, and realize image acquisition of the power station; S2: Image preprocessing and feature extraction: The images collected by the drone and fixed camera are transmitted to the preprocessing server in real time. After preprocessing such as denoising, the component surface features of the visible light image and the temperature distribution features of the infrared thermal imaging image are extracted using a deep learning algorithm. S3: Faulty component identification and location: input image feature data into the CNN architecture fault diagnosis model, combine the drone positioning and component layout information to determine the location coordinates of the faulty component, determine the severity based on the evaluation criteria of different fault types, and push the results to the mobile terminal of the operation and maintenance personnel; S4: Operation and maintenance strategy formulation and execution. Operation and maintenance personnel formulate plans based on the severity of the fault, use AR to assist in maintenance, and conduct tests after completion. S5: Verification and data update after fault repair. After maintenance, use the drone to collect images of the repaired components again to verify the repair effect, store fault-related data in the database, and update and optimize the fault diagnosis model.

2. A photovoltaic power station operation and maintenance management method according to claim 1, characterized in that: The specific process of S1 is as follows: Install a track-type drone landing pad above the photovoltaic power station and equip it with multiple drones with cameras with a resolution of no less than the preset pixels and infrared thermal imaging functions; the drones regularly collect full-coverage images of the photovoltaic power station according to the preset flight path and altitude, including visible light images and infrared thermal imaging images; Identify and locate obstacles in and around the power station, including electric poles, substation equipment, buildings, and trees; use the drone’s obstacle avoidance sensors and pre-set GIS data to automatically bypass these obstacles during flight path planning, ensuring flight safety while also ensuring the integrity of image acquisition; At the same time, a preliminary screening mechanism is established to analyze the abnormalities and take detailed photos of the abnormal points; Several fixed cameras are reasonably arranged at high locations around the photovoltaic power station to assist in monitoring the drone flight area and to perform preliminary image acquisition of the power station when the drone is unable to operate, ensuring the continuity and integrity of image acquisition.

3. A photovoltaic power station operation and maintenance management method according to claim 2, characterized in that: The planning process of the UAV flight path and altitude in S1 is as follows: First, the layout of the photovoltaic power station components is mapped and analyzed in detail to determine the arrangement of the components, the number of rows and columns, and the spacing and orientation between the arrays; Divide the power station components into several small areas and determine the priority of each small area based on the historical failure data of the small area, the degree of component aging, geographical location or environmental factors; Comprehensive analysis is performed by collecting level parameters of each small area, including: number of historical failures, average time of regional component use, geographical location score, and environmental factor score; The number of historical faults is recorded by establishing a photovoltaic power station fault record database, recording the time and location of each fault. For each small area, the number of faults that occurred in the past year is counted; The average usage time of regional components is determined by consulting the component installation records and purchase batch information of the power station, determining the installation time and initial service life data of the components in each small area, and calculating the average usage time of the components in each small area; The geographical location score is based on the power station's geographic information system data to determine the location relationship of each small area relative to the geographical factors of the sand source and the shelter. For small areas close to the sand source, trees, and building shelters, if the distance to the sand source or shelter is within 0-50 meters, 90 points will be given; if the distance is within 50-100 meters, 70 points will be given; if the distance is more than 100 meters, 40 points will be given; Environmental factor scoring is achieved by installing multiple environmental monitoring sensors in the power station, including light intensity sensors, temperature sensors, etc., distributed in various small areas, to collect environmental data in real time; for uneven lighting, statistics are collected on the daily light intensity data of each small area in the past month to calculate its light uniformity index; for areas with temperature differences greater than the preset temperature difference, the temperature change curve of each small area within a day is analyzed to calculate its temperature range; the obtained light uniformity and temperature range are normalized and used as the two right-angled sides of a right triangle respectively, and the other side of the right triangle is connected to establish a right triangle, and a triangular pyramid model is established with the correction coefficient as the height, the volume of the triangular pyramid model is calculated, and recorded as the ring judgment value, which is used as the standard for measuring environmental factor scoring; The historical failure times, average use time of regional components, geographical location score and environmental factor score are calibrated as lh, ti, dp and hu respectively, and then normalized and entered into the following formula: To obtain the level judgment value JPZ, where ω1, ω2, ω3, and ω4 are the preset weight coefficients of the number of historical failures, the average time of regional component use, the geographical location score, and the environmental factor score, and the obtained level judgment value JPZ is used as the priority judgment standard for the small area; The obtained level judgment value JPZ is compared with the three preset level judgment intervals, and the preset three level judgment intervals correspond to the first priority level, the second priority level and the third priority level, respectively. When the level judgment interval to which the level judgment value JPZ belongs is determined, the priority level of the small area is determined; Small areas of different priority levels are collected at different frequencies and numbers of shooting angles to ensure full coverage of all areas.

4. A photovoltaic power station operation and maintenance management method according to claim 2, characterized in that: The initial screening mechanism in S1 includes: When the UAV performs the first image acquisition at the preset flight altitude, the image acquisition system carried by the UAV is used to simultaneously acquire visible light images and infrared thermal imaging images, and the image data is transmitted to the processing unit on the ground in real time; For visible light images, a deep learning-based image recognition algorithm is used for preliminary analysis. The deep learning-based image recognition algorithm is pre-trained with samples of images of normal photovoltaic modules and modules with different defects to identify abnormal areas in the image and mark their locations and types. For infrared thermal imaging images, by comparing them with the preset normal component temperature range model, the area with abnormal temperature is detected and its location and temperature value are recorded; Combine the analysis results of visible light images and infrared thermal imaging images to generate a preliminary abnormality report, including the location information of the abnormal area and the type of abnormality; The drone automatically adjusts its flight path based on the location information in the preliminary anomaly report and flies to the first anomaly point. When the distance to the anomaly point is within 15 meters, the drone gradually lowers its flight altitude to 5-10 meters, and adjusts the angle and focal length of the camera to ensure that the detailed features of the anomaly area can be clearly captured. During the detailed shooting process, a higher-resolution image acquisition mode is used, and the abnormal area is photographed from multiple angles, including directly above and at a 45-degree angle, to obtain more comprehensive information. At the same time, the positioning system of the drone is used to record the location information corresponding to each detailed image, which is used by subsequent operation and maintenance personnel to locate the specific fault point. After completing the detailed photography of each abnormal point, the drone continues to fly to other abnormal points in the order in the abnormal report and repeats the detailed photography process until the photography task of all abnormal points is completed; After the shooting is completed, all the detailed image data are integrated with the abnormality report obtained from the preliminary screening, including the high-definition image and location of each abnormal point.

5. A photovoltaic power station operation and maintenance management method according to claim 1, characterized in that: The specific operation steps of S2 are as follows: The image data collected by the drone and fixed camera is transmitted to the image preprocessing server in real time; first, the image is preprocessed by denoising, grayscale correction, and contrast enhancement to improve the image clarity and quality; The image feature extraction algorithm based on deep learning is used to extract features from the preprocessed images. For visible light images, the surface color, texture, and shape features of the components are extracted to determine whether the components are physically damaged. For infrared thermal imaging images, the temperature distribution characteristics of the components are extracted and compared with the component temperature distribution model under normal working conditions to identify areas with hot spot failures or temperature anomalies.

6. A photovoltaic power station operation and maintenance management method according to claim 1, characterized in that: The specific operation steps of S3 are as follows: The extracted image feature data is input into a pre-trained fault diagnosis model; the model uses a convolutional neural network architecture and is trained with different photovoltaic module fault sample data to classify and locate faulty components; When the model determines that there is a faulty component, it combines the flight positioning data of the drone and the component layout information of the power station to determine the location coordinates of the faulty component in the power station, and determines the severity of the fault based on the fault type; The fault type, location information and fault severity assessment results are pushed to the operation and maintenance personnel's mobile devices in real time.

7. A photovoltaic power station operation and maintenance management method according to claim 6, characterized in that: The specific operation steps for determining the severity of the fault in S3 are as follows: Crack degree assessment: For component cracks in physical damage, the proportion of the cracked area to the total area of ​​the component is calculated through image analysis. First, the extracted visible light image is binarized, and the component area and the cracked area are marked with different values. The adaptive threshold method is used for binarization, and the formula is: Where T is the threshold, W and H are the width and height of the image, and I(x, y) is the pixel value at the coordinate; Then count the number of pixels in the rupture area N broken and the total number of pixels of the component N total , rupture ratio When P broken <0.1, it is judged as a minor fault; when 0.1≤P brooken <0.3, it is a moderate fault; when P broken ≥0.3, it is a severe fault; Deformation degree assessment: The deformation of components is measured using geometric features in the image; The contour information of the component edge is obtained through feature extraction, and the difference between the deformed component and the normal component contour is calculated. The Hough transform is used to detect the edge straight lines of the component. For a normal rectangular component, its edge is four straight lines. Calculate the average value of the angle deviation and displacement deviation between the deformed edge straight line and the normal position straight line; if the average values ​​of the angle deviation and displacement deviation are both less than the preset threshold, it is judged as slight deformation, that is, slight fault; if one of the deviations exceeds the preset threshold but does not reach the preset maximum limit, it is judged as moderate fault; if the deviation exceeds the maximum limit, it is judged as severe fault; Determine the severity of hot spot fault type: Use infrared thermal imaging images to first determine the hot spot area; then set the temperature threshold T h , the temperature is higher than T h The area is marked as the hot spot area; Calculate the area S of the hot spot area h and the total area of ​​the components S t , hot spot area ratio At the same time, calculate the average temperature T of the hot spot area a Compared with the normal component average temperature T n The temperature difference ΔT = T n -T n ; If P o <0.1 and ΔT<10℃, it is judged as a minor fault; if 0.1≤P o <0.3 or 10℃≤ΔT<20℃, it is judged as a moderate fault; if P o >0.3 or ΔT>20℃, it is judged as a serious fault; Determine the severity of short-circuit fault type: For short-circuit faults, in addition to judging from image features, electrical parameters are also needed. From the image perspective, short circuits can cause abnormal heating areas or color changes on the surface of components. The abnormal heating areas are determined using the same method as hot spot faults, and the area ratio and temperature difference are calculated. At the same time, check whether there are burnt or discolored areas, and determine the degree of discoloration by analyzing the image color space. If the area ratio is less than 0.1 and the heat and discoloration are found, and the electrical parameters show that the short-circuit current is less than 20% of the rated short-circuit current of the component, it is judged as a minor fault; if the area ratio is between 0.1-0.3, or the heat temperature difference is between 10%-20%, and the short-circuit current is between 20%-50% of the rated short-circuit current, it is judged as a moderate fault; if the area ratio is greater than 0.3, the temperature difference is greater than 20%, and the short-circuit current is greater than 50% of the rated short-circuit current, it is judged as a severe fault; For dust occlusion faults, the grayscale histogram of the visible light image is analyzed to make a judgment; the grayscale histogram of a normal component has a distribution pattern, and when there is dust occlusion, the histogram will change; the entropy of the grayscale histogram is calculated Where L is the gray level, p(i) is the probability of the pixel with gray value i appearing; the smaller the entropy value, the more concentrated the gray distribution of the image is, and the more serious the dust occlusion is; If the entropy value decreases by less than 10% compared with the entropy value of a normal component without dust obstruction, it is judged as a minor fault; if the entropy value decreases by between 10% and 30%, it is judged as a moderate fault; if the entropy value decreases by more than 30%, it is judged as a severe fault.

8. A photovoltaic power station operation and maintenance management method according to claim 1, characterized in that: The specific operation steps of S4 are as follows: The operation and maintenance personnel formulate targeted operation and maintenance plans based on the fault information received; For minor faults, record the location, type, and discovery time of the faulty component and mark it on the monitoring interface or map; perform maintenance within 3-5 days and prepare tools according to the fault; Collect images and temperature data every 30 minutes and draw trend curves; handle faults according to appearance inspection, test and compare, and record details; For moderate faults, a professional team will arrive at the site with tools within 1 hour to mark the fault condition; use thermal imagers and electroluminescent detectors to check the fault, analyze the cause, formulate a plan and upload it to the system; Repair according to faults and monitor the process to ensure the repair effect; comprehensively test performance parameters and compare them with rated values. If qualified, mark them as normal operation and include them in regular inspections; For serious faults, we remotely shut down related equipment, block the site, and set up warning signs; within 30 minutes, a professional repair team will arrive with tools for on-site assessment; Disassemble and replace damaged parts; After repair, conduct a comprehensive test on electrical performance and safety. If qualified, resume operation, record data update status, and strengthen inspections; During the maintenance process, maintenance personnel use augmented reality technology on mobile devices to superimpose detailed information of faulty components, including internal structure, electrical connection methods, and fault point markings, on the actual maintenance scenario, assisting maintenance personnel in troubleshooting and repair operations, and improving maintenance efficiency and quality.

9. A photovoltaic power station operation and maintenance management method according to claim 1, characterized in that: The specific operation steps of S5 are as follows: After the repair is completed, the drone is used again to collect images of the repaired components, and the collected image data is re-entered into the fault diagnosis model for verification to ensure that the fault has been completely repaired; At the same time, the relevant data of the fault, including the original image of the faulty component, the fault type, the maintenance process record, and the repaired image, are stored in the power station's operation and maintenance database, and the fault diagnosis model is updated and optimized.

Images