Photovoltaic power station cleaning method and system based on cooperative operation of unmanned aerial vehicle and robot

Abstract

The invention discloses a photovoltaic power station cleaning method and system based on cooperative operation of an unmanned aerial vehicle and a robot, and relates to the technical field of photovoltaic power generation operation and maintenance. Photovoltaic modules are partitioned, a reference image library is established, the unmanned aerial vehicle is utilized to inspect and obtain images under the same parameters, and a photovoltaic module pollutant identification model is combined to identify pollutants of the photovoltaic modules; identifying pollutant types and degrees, and generating structured data; the data processing platform constructs a dynamic priority model to determine a cleaning sequence, and plans a cleaning path according to different pollution characteristics; and the unmanned aerial vehicle automatically transfers the cleaning robot to a target area to execute operation, and automatically returns and warehouses after the task is completed. According to the scheme, the intelligent level, the automation level and the power generation efficiency of operation and maintenance of the photovoltaic power station are greatly improved.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation operation and maintenance technology, specifically to a method and system for cleaning photovoltaic power plants based on the collaborative operation of drones and robots. Background Technology

[0002] As photovoltaic (PV) power generation gains increasing importance in the national energy structure, the large-scale development of PV power plants places higher demands on operation and maintenance efficiency. The accumulation of pollutants such as dust, bird droppings, and fallen leaves on the surface of PV modules can lead to decreased light transmittance, hot spots, and reduced power generation efficiency, and in severe cases, can cause module aging and even safety hazards. Currently, PV power plants generally use manual cleaning, fixed cleaning equipment, or cleaning robots for regular or temporary cleaning operations. Some advanced systems have introduced environmental monitoring or power monitoring as auxiliary judgment methods, but cleaning decisions still largely rely on experience-based arrangements and lack precise perception and intelligent scheduling mechanisms. Existing technologies include applications that use drones to acquire image information for inspections, and solutions that utilize robots to perform cleaning tasks, but these typically operate independently, lacking an efficient collaborative mechanism, making it difficult to achieve intelligent closed-loop management of "on-demand cleaning."

[0003] The aforementioned existing technologies generally suffer from the problem of blind cleaning: on the one hand, uniform cycle cleaning can easily lead to over-cleaning of low-pollution areas, increasing equipment wear and energy consumption; on the other hand, sudden pollution (such as after a sandstorm) may not be detected in time, causing cleaning delays and affecting power generation revenue. Furthermore, traditional methods cannot implement differentiated cleaning strategies for different types of pollutants, and robot deployment is limited by terrain and path planning capabilities, resulting in low overall operational efficiency. Therefore, how to accurately identify the surface contamination status of photovoltaic modules, scientifically formulate cleaning priorities, and drive efficient collaborative operation between drones and robots has become an urgent technical challenge to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a photovoltaic power station cleaning method and system based on the collaborative operation of drones and robots, which can solve the problems of blind cleaning, over-cleaning or insufficient cleaning in existing photovoltaic power station cleaning methods, and improve the intelligence and precision of cleaning operations.

[0005] To achieve the above objectives, this application provides the following technical solution: In a first aspect, the present invention provides a method for cleaning a photovoltaic power station based on the collaborative operation of unmanned aerial vehicles (UAVs) and robots, comprising the following steps: S1, the photovoltaic modules of the photovoltaic power station are divided into several zones and numbered respectively. Visible light photos and infrared photos of each zone are taken by drones. At the same time, the altitude, latitude and longitude, heading angle and lens angle parameters of the drone are recorded to form a reference sample. S2, conduct regular inspections or temporary inspections triggered by preset conditions. During the inspection, the drone takes photos of each zone with reference to the parameters of the sample, and synchronizes the photos to the data processing platform for storage, thus obtaining the stored photos. S3 uses a photovoltaic module pollutant identification model to classify and identify pollutants in stored photos and quantify the degree of pollution. The identification results are then converted into structured data and synchronized to the data processing platform. S4, based on structured data, determines the execution order of cleaning zones through a dynamic priority evaluation model, then generates path parameters for the cleaning robot according to different pollution characteristics, and after completing multi-device status verification, sends out scheduling instructions in layers. S5: The drone receives the dispatch command, identifies and grabs the cleaning robot, and transports the cleaning robot to the target area according to the shortest planned transfer path and places it stably. After cleaning is completed, the cleaning robot is then transported to the next target area or the base station charging position. S6: After the cleaning robot arrives at the target zone, it matches the latitude and longitude coordinates of the zone and retrieves the path parameters, performs path cleaning operations, and conducts multi-dimensional cleaning effect verification. When it encounters obstacles, it executes the corresponding level of response mechanism. S7, after the cleaning plan is completed, the drones carry the cleaning robots back and put them into storage.

[0006] Furthermore, the specific process of S3 is as follows: When classifying and identifying pollutants, features are extracted from the stored photos, and the softmax classifier of the output layer of the photovoltaic module pollutant identification model classifies the pixels in the stored photos into dusty areas, hard pollutants, and light pollutants. Simultaneously, by combining infrared image identification with stored photos, hot spot areas with temperatures exceeding the surrounding average by more than 8°C are identified and marked as priority cleaning areas. When quantifying the degree of pollution, the dust coverage rate of the dust accumulation area in a single photovoltaic module is calculated. A coverage rate of <30% is considered as not requiring cleaning, 30% ≤ coverage rate <60% is considered as requiring routine cleaning, and a coverage rate of ≥60% is considered as requiring intensive cleaning. When the area of ​​a single hard contaminant is ≥4cm² or the number of hard contaminants on a single photovoltaic module is ≥5, it is considered as requiring targeted cleaning. When the area covered by light contaminants is ≥20% or when the junction box area of ​​the photovoltaic module is obstructed, it is considered as requiring immediate cleaning. The above identification and judgment results are converted into structured data containing partition number, pollution type, and pollution level and synchronized to the data processing platform.

[0007] Furthermore, the specific process in S4 of determining the cleaning zone execution order based on structured data and a dynamic priority evaluation model is as follows: When determining the cleaning partition execution order, the data processing platform constructs a dynamic priority evaluation model, which calculates the priority score by combining three types of parameters: Pollution urgency: 10 points are awarded for hot spot areas, junction boxes shielded by light pollutants, or light pollutant coverage areas ≥20%; 8 points are awarded for dust coverage ≥60% or areas with dense hard pollutants; 5 points are awarded for dust coverage ≤30% <60%. Environmental adaptability: Wind speed < 5m / s and no rainfall in the next 2 hours are considered adaptable; wind speed 5-8m / s or short-term rainfall warning are considered conditionally adaptable; wind speed ≥ 8m / s are considered unsuitable and the zone is locked. Route efficiency: The spatial distance between each zone is calculated using the digital twin model of the power station, and a clustering algorithm is used to group zones with a distance ≤ 50 meters into the same cleaning batch; The priority score calculation formula is: Priority Score = Pollution Urgency × 0.5 + Environmental Adaptability × 0.2 + Path Efficiency × 0.3, which generates the cleaning zone execution sequence.

[0008] Furthermore, the specific process of generating path parameters for the cleaning robot in S4 for different pollution characteristics is as follows: The dust accumulation area adopts an S-shaped reciprocating path with 100% coverage. When the dust accumulation coverage is ≥60%, a secondary reciprocating mode is activated. For areas with hard contaminants, a fixed-point cleaning subroutine is superimposed on the S-shaped path: a round-trip path with a side length of 10-20cm is planned with the center of the contaminant as the origin, the dwell time is ≥10 seconds, the brush head pressure is increased by 30% and adjusted in real time by the pressure sensor; For areas with light pollutants: When the cleaning robot is equipped with a dustbin, it uses an edge-gathering and center-cleaning path. First, it plans a closed rectangular path along the edge of the photovoltaic module to gather fallen leaves to the center and then cleans them in a straight line. When the cleaning robot is not equipped with a dustbin, it uses a directional pushing and edge-exiting path. The photovoltaic module is divided into three longitudinal areas along its length: left, middle, and right. It pushes and sweeps from the high side to the low side according to the installation tilt angle. The pollutants are pushed to the low-side outlet and then the robot performs 2-3 back-and-forth sweeps along the low-side edge. For pollutants in crevices, the robot is identified by an image sensor and then moves back and forth ±10cm around the crevices to perform point sweeping.

[0009] Furthermore, in step S6, after the robot vacuum arrives at the target zone, it matches the zone's latitude and longitude coordinates, retrieves the path parameters, and performs a precise path cleaning operation. The specific process is as follows: The positioning module matches the latitude and longitude coordinates of the partition, retrieves the path parameters in S4, and moves at a speed of 0.8-1.2m / s based on the tracked chassis differential drive system. The wheel speed encoder corrects the trajectory deviation in real time. When the distance is ≤8cm, the speed is reduced to 0.2m / s and the steering angle is adjusted.

[0010] Furthermore, the specific process for verifying the multi-dimensional cleaning effect in S6 is as follows: The roller brush shaft has a built-in pressure sensor that dynamically monitors the contact pressure with the photovoltaic module surface. When the pressure is >50N, the roller brush is automatically raised; when the pressure is <30N, the downward pressure is increased. The high-definition camera on the front of the drone takes images of the cleaned area at a set frequency, compares them with the photos stored in the S3, and calculates the cleaning coverage. If the coverage is ≥90%, the cleaning is considered to have met the standard; otherwise, local re-cleaning is triggered. If a hot spot with a temperature 5°C or higher than the surrounding average is found after scanning the photovoltaic module surface, the coordinates of the hot spot are immediately marked and uploaded to the data processing platform to trigger a potential fault warning.

[0011] Furthermore, the specific implementation of the corresponding level of response mechanism when encountering an obstacle in S6 is as follows: During the cleaning process, if the cleaning robot detects a sudden increase in obstacles, or if the sensors identify a path height difference or gap width exceeding the cleaning robot's obstacle-crossing capability, or if the cleaning robot's posture undergoes a significant change, a Level 3 response will be immediately executed: The three-level response is as follows: using the robot's own adaptive obstacle-crossing capability, it bypasses obstacles and automatically returns to the original path, cleaning without interruption throughout the process; The secondary response is as follows: if the width or height of the obstacle exceeds the obstacle-crossing capability of the cleaning robot, the cleaning robot stops working and takes an image of the obstacle, sends an alarm message containing coordinates to the data processing platform, and switches to a low-power waiting-for-command mode. The first-level response is as follows: if the robot's tilt angle is detected to be ≥15° or the brush is stuck, an emergency stop is immediately initiated, power output is cut off, and an audible and visual alarm is issued, awaiting manual intervention.

[0012] Furthermore, the process of S2 is specifically as follows: The types of inspections include regular inspections of the entire photovoltaic power station and temporary inspections triggered by preset conditions, such as sandstorm weather, abnormal photovoltaic array power generation efficiency, and manual power activation. During the inspection, the drone flew over the entire photovoltaic panel array area, referencing the parameters of the sample, and took photos of each photovoltaic section. During the inspection, the drone interacts with the data processing platform, synchronizing the photos of the photovoltaic zones it captures to the data processing platform in real time for storage, thus creating stored photos.

[0013] Furthermore, in S3, the photovoltaic module contaminant identification model is constructed based on a deep convolutional neural network.

[0014] Secondly, the present invention also provides a photovoltaic power station cleaning system based on the collaborative operation of drones and robots, employing the aforementioned cleaning method, including: Data collection module: used to divide the photovoltaic modules of the photovoltaic power station into several zones and number them respectively. The drone takes visible light and infrared photos of each zone, and records the drone's altitude, latitude and longitude, heading angle and lens angle parameters at the time of shooting to form a reference sample. Data synchronization module: used to carry out regular inspections or temporary inspections triggered by preset conditions. During the inspection, the drone takes photos of each zone with reference to the parameters of the sample, and synchronizes the photos to the data processing platform for storage, thus obtaining the stored photos. Identification and Judgment Module: Used to classify and identify pollutants in stored photos and quantify the degree of pollution using a photovoltaic module pollutant identification model, convert the identification results into structured data and synchronize them to the data processing platform; Path planning module: It is used to determine the execution order of cleaning zones based on structured data through a dynamic priority evaluation model, generate path parameters for cleaning robots according to different pollution characteristics, and issue scheduling instructions in layers after completing multi-device status verification. Cleaning plan specification module: This module enables the drone to receive scheduling instructions, identify and grab the cleaning robot, transport the cleaning robot to the target area according to the planned shortest transport path and place it stably, and then transport the cleaning robot to the next target area or base station charging position after cleaning is completed. Cleaning verification module: After the cleaning robot arrives at the target area, it matches the latitude and longitude coordinates of the area and retrieves the path parameters, performs precise path cleaning, and conducts multi-dimensional cleaning effect verification. When it encounters obstacles, it executes the corresponding level of response mechanism. End module: Used to return the drone carrying the cleaning robot and put them into storage after the entire cleaning plan is completed.

[0015] Compared with the prior art, the present invention has the following beneficial technical effects: This invention provides a photovoltaic power station cleaning method based on collaborative operation of drones and robots. By establishing a closed-loop process including drone inspection, AI image recognition, dynamic priority assessment, and precise robot cleaning, it achieves accurate perception of the contamination status of photovoltaic modules and on-demand cleaning. The method improves the accuracy of contamination identification by employing zonal modeling and image comparison with the same parameters; avoids blind scheduling by comprehensively considering contamination level, environmental conditions, and path efficiency through a dynamic priority model; improves cleaning efficiency and prevents module damage by customizing cleaning paths and intensity control based on different contaminant types; utilizes drones to enable cross-regional robot transport, overcoming terrain limitations and improving deployment flexibility; and combines a multi-dimensional verification mechanism to achieve closed-loop feedback of cleaning quality. Therefore, this solution effectively solves the problems of over-cleaning, under-cleaning, and disordered operation in traditional cleaning methods, significantly improving the intelligence and automation level of photovoltaic power station operation and maintenance, as well as power generation efficiency. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a photovoltaic power station cleaning method based on the collaborative operation of drones and robots in an embodiment of the present invention. Detailed Implementation

[0017] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0018] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0019] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, an electrical connection, or a communication connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0021] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0022] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0023] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0026] With the widespread application of photovoltaic power generation technology and the continuous expansion of photovoltaic power plant scale, the light transmittance of the module surface is decreasing due to the accumulation of pollutants such as dust, bird droppings, fallen leaves, and sand, leading to increasingly prominent problems such as reduced power generation efficiency and hot spot effects. Traditional cleaning methods mostly rely on manual inspections or fixed-cycle operations, lacking the ability to dynamically perceive the actual pollution status, which easily leads to wasted cleaning resources or untimely cleaning, affecting the overall operational efficiency of the power plant. In addition, the deployment of existing cleaning robots is limited by terrain and path planning accuracy, making it difficult to achieve flexible cross-regional scheduling and resulting in low levels of collaborative operation.

[0027] See Figure 1 This invention provides a method for cleaning photovoltaic power plants based on the collaborative operation of drones and robots, comprising the following steps: S1, the photovoltaic modules of the photovoltaic power station are divided into several zones and numbered respectively. Visible light photos and infrared photos of each zone are taken by drones. At the same time, the altitude, latitude and longitude, heading angle and lens angle parameters of the drone are recorded to form a reference sample. In this step, dividing the photovoltaic array into several zones refers to dividing it into multiple logical management units based on its spatial layout. For example, each zone may consist of 10×10 modules, and each zone is assigned a unique number to facilitate subsequent location and task assignment. The zones can be rectangular, hexagonal, or other regular geometric shapes, and the boundary information is entered into a digital twin model for path optimization and obstacle avoidance calculations.

[0028] Visible light photography is used to obtain visual information about the surface of photovoltaic modules, with a resolution of 0.5 mm / pixel or higher, which can clearly show details such as dust distribution, hard stains (such as bird droppings), and light debris (such as leaves); infrared photography uses thermal imaging technology to detect differences in the operating temperature of the modules and identify potential hot spot areas, usually using uncooled infrared sensors with a wavelength range of 8–14 μm.

[0029] S2, conduct regular inspections or temporary inspections triggered by preset conditions. During the inspection, the drone takes photos of each zone with reference to the parameters of the sample, and synchronizes the photos to the data processing platform for storage, thus obtaining the stored photos. In this embodiment, regular inspection refers to a full-coverage image acquisition task that is automatically started at a set time interval (such as once every 7 days); temporary inspection is a non-periodic inspection triggered by external events, such as receiving a sandstorm warning signal, detecting that the power generation of a certain array drops sharply beyond a threshold (such as 15%), or being started remotely and manually by maintenance personnel.

[0030] S3 uses a photovoltaic module pollutant identification model to classify and identify pollutants in stored photos and quantify the degree of pollution. The identification results are then converted into structured data and synchronized to the data processing platform. In this embodiment, the photovoltaic module contaminant identification model has the ability to distinguish between dust accumulation areas, hard contaminants (such as dried bird droppings and pebbles), and lightweight contaminants (such as fallen leaves and plastic bags). The contaminant classification and identification process first preprocesses the image, including noise reduction, white balance correction, and radiometric calibration (for infrared images). Then, multi-scale features are extracted, and finally, a softmax classifier is used to output the probability distribution of the category to which each pixel belongs. The quantitative determination of the degree of pollution is a numerical evaluation based on the classification: for dust accumulation areas, the proportion of the area of ​​a single photovoltaic module is calculated as the dust coverage rate; for isolated contaminants, their number and individual area size are counted; for hot spot areas, areas with a temperature difference ≥8℃ in the infrared image are marked and associated with the nearest physical partition.

[0031] S4, based on structured data, determines the execution order of cleaning zones through a dynamic priority evaluation model, then generates path parameters for the cleaning robot according to different pollution characteristics, and after completing multi-device status verification, sends out scheduling instructions in layers. S5: The drone receives the dispatch command, identifies and grabs the cleaning robot, and transports the cleaning robot to the target area according to the shortest planned transfer path and places it stably. After cleaning is completed, the cleaning robot is then transported to the next target area or the base station charging position. S6: After the cleaning robot arrives at the target zone, it matches the latitude and longitude coordinates of the zone and retrieves the path parameters, performs path cleaning operations, and conducts multi-dimensional cleaning effect verification. When it encounters obstacles, it executes the corresponding level of response mechanism. S7, after the cleaning plan is completed, the drones carry the cleaning robots back and put them into storage.

[0032] This invention, driven by data, leverages the high-altitude mobility of drones and the ground-based execution capabilities of robots to achieve efficient, precise, and safe cleaning of large-scale photovoltaic power plants. Compared to traditional methods, this solution significantly reduces the frequency of ineffective operations, extends the lifespan of components, and improves the overall power generation revenue of the power plant. It also boasts excellent scalability, adapting to different types of photovoltaic sites and climatic conditions.

[0033] In a more specific embodiment of the present invention, the specific process of S3 is as follows: The pollutant classification and identification process first preprocesses the image data based on visible light and infrared photographs acquired by the UAV during inspections. This preprocessing includes noise reduction, contrast enhancement, and geometric correction to improve the accuracy of subsequent feature extraction. Subsequently, a photovoltaic module pollutant identification model is constructed using a deep convolutional neural network (CNN) with pixel-level segmentation capabilities. This model has been trained on a large number of labeled samples, covering different lighting conditions, weather conditions, and typical pollution types (such as dust, bird droppings, leaves, dust, and oil stains), enabling fine-grained semantic segmentation of the photovoltaic module surface condition. In the output stage of the photovoltaic module pollutant identification model, a softmax classifier is used to assign a category probability distribution to each pixel, ultimately classifying it into one of three categories: dusty areas, hard pollutants, or light pollutants. Dust accumulation areas typically appear as large areas of uniform, low-reflectivity coverings with weak spectral response and blurred boundaries. Hard pollutants such as bird droppings, pebbles, and adhesive mud clumps exhibit localized high reflectivity or abrupt texture changes, with irregular shapes but clear boundaries. Light pollutants such as fallen leaves, grass clippings, and plastic films are discontinuous and mobile, often appearing as discrete patches or strip-like structures in images.

[0034] Because contaminant accumulation can lead to poor localized heat dissipation, creating areas of abnormally high temperature—a phenomenon known as hot spot effect—infrared image data is simultaneously retrieved during the analysis of stored photographs to detect whether the temperature difference between each area and the average temperature of the surrounding normal photovoltaic modules exceeds 8°C. If the temperature difference in a certain area reaches or exceeds this threshold, it is determined to have a significant risk of hot spots and is automatically marked as a priority cleaning area.

[0035] In terms of quantifying the degree of pollution, for areas with accumulated dust, the proportion of the area occupied by the dust accumulation on a single standard photovoltaic module is calculated, i.e., the dust coverage rate. When the coverage rate is less than 30%, the current pollution level is considered to have a small impact on power generation efficiency, and cleaning tasks are not initiated temporarily. When the coverage rate is between 30% and 60%, it is defined as a normal pollution state and included in the periodic cleaning plan. When the coverage rate is not less than 60%, it is considered severe pollution, requiring key cleaning operations, and if necessary, increasing the cleaning frequency or activating the enhanced cleaning mode. For hard pollutants, a dual judgment standard is set: if the projected area of ​​a single pollutant is greater than or equal to 4 square centimeters, or if five or more independent hard pollutant points are found on the same photovoltaic module, a fixed-point cleaning instruction is triggered, prompting the cleaning robot to add a local fine-tuning program on top of the general path. For light pollutants, if their total coverage area reaches more than 20% of the entire panel area, or if their physical location happens to block the junction box and its ventilation and heat dissipation area, it is judged to require immediate cleaning, as it not only reduces light transmittance but may also cause electrical connections to overheat, posing a safety hazard.

[0036] In a more specific embodiment of the present invention, based on the structured data output in S3, including information such as zone numbers, pollution types, and pollution levels, the data processing platform initiates a dynamic priority assessment model to generate the optimal cleaning operation sequence. This dynamic priority assessment model does not rely solely on a single pollution indicator for ranking; instead, it introduces a multi-dimensional weighted scoring mechanism, making comprehensive decisions based on pollution severity, environmental safety, and route economics. This avoids problems such as frequent inter-regional transport and high-risk weather operations caused by the traditional simple ranking based on pollution levels.

[0037] The pollution urgency level is used to characterize the urgency of the impact of different types of pollution on power generation performance. Its scoring rules are set according to the actual degree of harm of pollutants to the electrical performance and thermal stability of photovoltaic modules: When infrared images detect hot spot areas with temperatures more than 8°C higher than the surrounding average, or visible light images identify light pollutants (such as fallen leaves or plastic film) obstructing junction boxes, or when the coverage area reaches 20% or more of a single panel, it is judged as seriously affecting heat dissipation and current conduction, and is assigned the highest score of 10 points; for heavily dusty areas with a dust coverage rate of ≥60%, or situations where multiple hard pollutants (such as bird droppings or pebbles) form a dense pollution zone on a single photovoltaic module, although it does not immediately cause a failure, it significantly reduces light transmittance, and is assigned a score of 8 points; while for moderate dust accumulation (30% ≤ coverage rate <60%), its impact on power generation efficiency is within the tolerable range, but it still needs to be included in the plan, and is assigned a score of 5 points.

[0038] Environmental adaptability is used to assess whether current and short-term weather conditions are suitable for drone flight and ground cleaning robot operation. Wind speed < 5 m / s and no rainfall in the next 2 hours are considered adaptable. Wind speed 5-8 m / s or short-term rainfall warning are considered conditionally adaptable. Wind speed ≥ 8 m / s are considered unsuitable and the zone is locked.

[0039] Route efficiency reflects the spatial continuity of cleaning tasks and the energy consumption cost of transportation. The spatial distance of each zone is calculated by the digital twin model of the power station, and the clustering algorithm is used to group zones with a distance ≤ 50 meters into the same cleaning batch. The priority score calculation formula is: Priority Score = Pollution Urgency × 0.5 + Environmental Adaptability × 0.2 + Path Efficiency × 0.3, which generates the cleaning zone execution sequence.

[0040] Based on the scores of each zone, a cleaning sequence is generated to determine the order of execution for each zone. This sequence not only reflects the urgency of pollution control but also takes into account operational safety and resource utilization efficiency. For example, two heavily dusty zones, both scoring 8 points, may have a different execution order if one is located in a remote corner with no other task points nearby. This demonstrates the moderating effect of route efficiency on the overall scheduling.

[0041] Through the above-described steps, this application achieves dynamic sequencing of cleaning tasks under multi-factor coupling. By employing a triple assessment mechanism of pollution urgency, environmental adaptability, and route efficiency, it solves the problems of frequent relocation, high-risk operations, and energy waste caused by prior art that only schedules cleaning based on pollution levels. By using a weighted fusion approach to highlight the dominant role of pollution factors while retaining environmental safety constraints and route economic considerations, the entire collaborative operation system possesses greater robustness and practicality.

[0042] In a more specific embodiment provided by the present invention, the specific process of generating path parameters for the cleaning robot in S4 for different pollution characteristics is as follows: The dust accumulation area adopts an S-shaped reciprocating path with 100% coverage. When the dust accumulation coverage is ≥60%, a secondary reciprocating mode is activated. For areas with hard contaminants, a fixed-point cleaning subroutine is superimposed on the S-shaped path: a round-trip path with a side length of 10-20cm is planned with the center of the contaminant as the origin, the dwell time is ≥10 seconds, the brush head pressure is increased by 30% and adjusted in real time by the pressure sensor; For areas with light pollutants: When the cleaning robot is equipped with a dustbin, it uses an edge-gathering and center-cleaning path. First, it plans a closed rectangular path along the edge of the photovoltaic module to gather fallen leaves to the center and then cleans them in a straight line. When the cleaning robot is not equipped with a dustbin, it uses a directional pushing and edge-exiting path. The photovoltaic module is divided into three longitudinal areas along its length: left, middle, and right. It pushes and sweeps from the high side to the low side according to the installation tilt angle. The pollutants are pushed to the low-side outlet and then the robot performs 2-3 back-and-forth sweeps along the low-side edge. For pollutants in crevices, the robot is identified by an image sensor and then moves back and forth ±10cm around the crevices to perform point sweeping.

[0043] Specifically, the path parameter generation for dust accumulation areas employs a full-coverage S-shaped reciprocating path design to ensure that the roller brush or cleaning brush head thoroughly cleans the surface of the photovoltaic modules. This path uses the long side of the photovoltaic module as a reference and moves reciprocally along the width direction at a set step interval (typically 80%-90% of the effective width of the brush head), ensuring appropriate overlap between adjacent paths to achieve 100% coverage. When the system determines that the dust accumulation coverage of a certain area is ≥60%, a secondary reciprocating mode is activated, which repeats the entire S-shaped path cleaning process within the same area, increasing the cleaning efficiency by increasing the number of passes. This mode is suitable for heavily dusty scenarios, such as after a sandstorm or in areas that have not been cleaned for a long time, and can significantly improve the removal rate of stubborn dust. As an optional implementation, the secondary reciprocating mode can also employ a staggered scanning strategy, offsetting the second path laterally by a certain distance (e.g., 5-10cm) from the first path to enhance the disturbance effect on localized accumulation areas.

[0044] For areas with hard contaminants, a fixed-point cleaning subroutine is superimposed on the basic S-shaped path. This subroutine locates the center of the contaminant based on the coordinate information output by the contaminant identification model and constructs a localized, refined cleaning path around this center. Specifically, a 10-20cm long, round-trip straight path is planned, the direction of which can be adjusted according to the contaminant's extension trend to ensure the brush head fully engages with the contaminant. During this process, the cleaning robot reduces its travel speed to 0.3-0.5m / s and increases the pressure applied by the brush head by 30% compared to the conventional value to enhance mechanical friction and effectively remove highly adhesive stains (such as bird droppings and oil stains). Simultaneously, a built-in pressure sensor in the brush head monitors the contact pressure in real time. If the pressure exceeds a safe threshold (e.g., 60N), the automatic fine-tuning mechanism retracts to prevent damage to the photovoltaic glass surface; if the pressure is insufficient, the downward pressure is dynamically increased to ensure consistent cleaning results. As an alternative, the fixed-point cleaning path can also be designed as a small-range circular trajectory or a multi-directional intersecting path to further enhance the removal capability of irregularly shaped hard contaminants.

[0045] The path strategy for areas with light pollutants varies depending on whether the cleaning robot is equipped with a vacuuming function. When the robot has a vacuum box, a combined edge-gathering and center-cleaning path is used: First, a closed rectangular path is planned along the outer contour of the photovoltaic module. The roller brush rotates in a centripetal direction to push light debris such as fallen leaves and paper scraps towards the center of the panel. After the pollutants have gathered, the robot switches to a straight reciprocating mode for high-intensity cleaning of the central area, while the negative pressure vacuuming device is activated to suck the gathered material into the storage compartment. This method avoids pollutants being dispersed by the wind during cleaning and improves the one-time removal rate. When the robot does not have a vacuuming function, a directional pushing and edge-exiting path strategy is used: The entire photovoltaic module is divided into three longitudinal cleaning areas (left, center, and right) along its length. The cleaning sequence proceeds from the high side to the low side according to the installation angle of the photovoltaic module, using gravity to assist the sliding of pollutants. After the pushing and sweeping action is completed, the pollutants remaining at the bottom edge are guided to the preset outlet (width ≥ 10cm). The robot then performs 2-3 reciprocating sweeping operations along this edge to thoroughly remove the accumulated material. Furthermore, for small fallen leaves stuck in junction box gaps or bracket gaps, the system uses an image sensor to precisely locate the center of the gap and initiates a jog-sweep mode: centering on the gap, it performs high-frequency, small-amplitude reciprocating motions within a ±10cm range in the front-back direction, repeatedly impacting and loosening the obstruction. This jog mode can effectively reduce the gap residue rate to below 3%. As a variant embodiment, the jog-sweep mode can also be combined with a vibration module to further improve the removal efficiency of contaminants stuck in the gap by applying micro-amplitude high-frequency vibrations.

[0046] In a more specific embodiment of the invention, by matching the latitude and longitude coordinates of the partition and retrieving the path parameters in S4, the tracked chassis differential drive system travels at a speed of 0.8-1.2 m / s, and the wheel speed encoder corrects the trajectory deviation in real time. When the detected distance is ≤8cm, the speed is reduced to 0.2 m / s and the steering angle is adjusted. The technical solution of the cleaning robot performing precise path cleaning after arriving at the target partition aims to improve its path tracking accuracy and operational safety in complex photovoltaic array environments. This process involves the coordinated operation of multiple key technical features to ensure the efficient and stable completion of the cleaning task.

[0047] In a more specific embodiment of the present invention, the specific process of verifying the multi-dimensional cleaning effect in S6 is as follows: The roller brush shaft has a built-in pressure sensor that dynamically monitors the contact pressure with the photovoltaic module surface. When the pressure is >50N, the roller brush is automatically raised; when the pressure is <30N, the downward pressure is increased. The built-in pressure sensor on the roller brush shaft dynamically monitors the contact pressure with the photovoltaic module surface. This refers to the integration of a high-precision pressure sensor (such as a strain gauge or piezoresistive sensor) inside the roller brush drive shaft or on the support structure of the cleaning robot to collect the vertical force between the roller brush and the photovoltaic module surface in real time. This pressure sensor continuously outputs pressure data at a sampling frequency of no less than 10Hz to determine whether the currently applied cleaning force is within a safe and effective range. Under normal operating conditions, the ideal contact pressure range is set at 30–50N: when the detected pressure exceeds 50N, the roller brush assembly is automatically raised to prevent scratches on the glass surface or wear on the coating due to overpressure; when the pressure is below 30N, the downward pressure is adjusted in the opposite direction to ensure that cleaning efficiency is not affected.

[0048] The high-definition camera on the front of the drone takes images of the cleaned area at a set frequency, compares them with the photos stored in the S3, and calculates the cleaning coverage. If the coverage is ≥90%, the cleaning is considered to have met the standard; otherwise, local re-cleaning is triggered. When scanning the photovoltaic module surface, if a hot spot is found that remains at a temperature 5°C higher than the surrounding average after cleaning, the coordinates of the hot spot are immediately marked and uploaded to the data processing platform to trigger a potential fault warning. This refers to using an infrared thermal imaging sensor (such as an uncooled micro-bolometer array with a spatial resolution of 0.1°C) integrated into a cleaning robot or drone to perform a comprehensive scan of the cleaned photovoltaic modules. First, temperature differences caused by environmental factors such as shadows and edge effects are eliminated, and then the average temperature rise of each area relative to neighboring healthy units is calculated. Once an area is detected that is consistently higher than the surrounding average by more than 5°C, and there is no obvious pollution residue in the visible light images before and after cleaning, it is determined to be an abnormal heat point not induced by pollution.

[0049] In a more specific embodiment of the present invention, the execution of the corresponding level of response mechanism when encountering an obstacle in S6 is specifically as follows: During the cleaning process, if the cleaning robot detects a sudden increase in obstacles, or if the sensors identify a path height difference or gap width exceeding the cleaning robot's obstacle-crossing capability, or if the cleaning robot's posture undergoes a significant change, a Level 3 response will be immediately executed: The three-level response is as follows: using the robot's own adaptive obstacle-crossing capability, it bypasses obstacles and automatically returns to the original path, cleaning without interruption throughout the process; The secondary response is as follows: if the width or height of the obstacle exceeds the obstacle-crossing capability of the cleaning robot, the cleaning robot stops working and takes an image of the obstacle, sends an alarm message containing coordinates to the data processing platform, and switches to a low-power waiting-for-command mode. The first-level response is as follows: if the robot's tilt angle is detected to be ≥15° or the brush is stuck, an emergency stop is immediately initiated, power output is cut off, and an audible and visual alarm is issued, awaiting manual intervention.

[0050] In a more specific embodiment of the present invention, the process of S2 is as follows: The types of inspections include regular inspections of the entire photovoltaic power station and temporary inspections triggered by preset conditions, such as sandstorm weather, abnormal photovoltaic array power generation efficiency, and manual power activation. The inspection types include regular inspections and ad-hoc inspections. Regular inspections refer to full-coverage image acquisition tasks that are automatically initiated according to a preset cycle (such as every 7 days or every 14 days). They are used to systematically grasp the overall pollution status evolution trend of photovoltaic power plants and are suitable for normal operation phases without significant external disturbances. Ad-hoc inspections, on the other hand, are emergency or responsive inspections triggered at unplanned times. Their initiation depends on the occurrence of a specific event and is designed to quickly respond to sudden pollution or performance degradation. The preset conditions are specifically divided into three categories: The first category is triggered by environmental meteorological conditions, that is, when a sandstorm is detected or a sandstorm warning is issued in the nearby area, the inspection process is automatically activated to deal with the problem of large-area dust accumulation caused by high concentration of particulate matter deposition; the second category is triggered by electrical performance anomalies, that is, when the power monitoring system detects that the power generation efficiency of one or more photovoltaic arrays has dropped by more than 15% compared with the same period in history, and after excluding other factors such as shading and inverter failure, it is determined that there may be serious surface pollution, thus triggering drone inspection; the third category is triggered by manual enablement, which allows operation and maintenance personnel to manually issue inspection commands through remote control terminals, which is suitable for special working condition investigation, new equipment commissioning verification or system debugging scenarios.

[0051] During the inspection, the drone flies over the entire photovoltaic panel array area using the parameters of the reference sample and takes pictures of each photovoltaic section. The parameters of the reference sample refer to the imaging parameters such as the drone's flight altitude, latitude and longitude coordinates, heading angle, and lens pitch angle recorded in the initial modeling stage.

[0052] During the inspection, the drone interacts with the data processing platform, synchronizing the photos of the photovoltaic zones it captures to the data processing platform in real time for storage, thus creating stored photos.

[0053] In a more specific embodiment of the present invention, in step S3, the photovoltaic module contaminant identification model is constructed based on a deep convolutional neural network. As a core algorithm component of the image recognition and judgment module, the photovoltaic module contaminant identification model is used to jointly analyze visible light and infrared images collected by the UAV, achieving accurate classification of contaminant types and quantitative assessment of contamination levels on the photovoltaic module surface. This photovoltaic module contaminant identification model, based on a deep convolutional neural network (DCNN) architecture, possesses multi-level feature extraction capabilities and can automatically learn and abstract deep representation patterns of different contaminants in terms of spatial distribution, texture structure, edge contours, and thermal radiation characteristics.

[0054] In another embodiment of the present invention, a photovoltaic power station cleaning system based on the collaborative operation of drones and robots is also provided. The photovoltaic power station cleaning method based on the aforementioned collaborative operation of drones and robots includes: Data collection module: used to divide the photovoltaic modules of the photovoltaic power station into several zones and number them respectively. The drone takes visible light and infrared photos of each zone, and records the drone's altitude, latitude and longitude, heading angle and lens angle parameters at the time of shooting to form a reference sample. Data synchronization module: used to carry out regular inspections or temporary inspections triggered by preset conditions. During the inspection, the drone takes photos of each zone with reference to the parameters of the sample, and synchronizes the photos to the data processing platform for storage, thus obtaining the stored photos. Identification and Judgment Module: Used to classify and identify pollutants in stored photos and quantify the degree of pollution using a photovoltaic module pollutant identification model, convert the identification results into structured data and synchronize them to the data processing platform; Path planning module: It is used to determine the execution order of cleaning zones based on structured data through a dynamic priority evaluation model, generate path parameters for cleaning robots according to different pollution characteristics, and issue scheduling instructions in layers after completing multi-device status verification. Cleaning plan specification module: This module enables the drone to receive scheduling instructions, identify and grab the cleaning robot, transport the cleaning robot to the target area according to the planned shortest transport path and place it stably, and then transport the cleaning robot to the next target area or base station charging position after cleaning is completed. Cleaning verification module: After the cleaning robot arrives at the target area, it matches the latitude and longitude coordinates of the area and retrieves the path parameters, performs precise path cleaning, and conducts multi-dimensional cleaning effect verification. When it encounters obstacles, it executes the corresponding level of response mechanism. End module: Used to return the drone carrying the cleaning robot and put them into storage after the entire cleaning plan is completed.

[0055] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for cleaning photovoltaic power station based on cooperation between unmanned aerial vehicle and robot, characterized in that, Includes the following steps: S1, the photovoltaic modules of the photovoltaic power station are divided into several zones and numbered respectively. Visible light photos and infrared photos of each zone are taken by drones. At the same time, the altitude, latitude and longitude, heading angle and lens angle parameters of the drone are recorded to form a reference sample. S2, conduct regular inspections or temporary inspections triggered by preset conditions. During the inspection, the drone refers to the parameters of the sample, takes photos of each zone, and synchronizes the photos to the data processing platform for storage, thus obtaining the stored photos. S3 uses a photovoltaic module pollutant identification model to classify and identify pollutants in stored photos and quantify the degree of pollution. The identification results are then converted into structured data and synchronized to the data processing platform. S4 determines the cleaning zone execution order based on structured data through a dynamic priority evaluation model, then generates path parameters for the cleaning robot according to different pollution characteristics, and sends out scheduling instructions in layers after completing multi-device status verification. S5: The drone receives the dispatch command, identifies and grabs the cleaning robot, and transports the cleaning robot to the target area according to the shortest planned transfer path and places it stably. After cleaning is completed, the cleaning robot is then transported to the next target area or the base station charging position. S6: After the cleaning robot arrives at the target zone, it matches the latitude and longitude coordinates of the zone and retrieves the path parameters, performs path cleaning operations, and conducts multi-dimensional cleaning effect verification. When it encounters obstacles, it executes the corresponding level of response mechanism. S7, after the cleaning plan is completed, the drones carry the cleaning robots back and put them into storage. 2.The photovoltaic power station cleaning method based on cooperation between a UAV and a robot according to claim 1, characterized in that, The specific process of S3 is as follows: When classifying and identifying pollutants, features are extracted from the stored photos, and the softmax classifier of the output layer of the photovoltaic module pollutant identification model classifies the pixels in the stored photos into dusty areas, hard pollutants, and light pollutants. Simultaneously, by combining infrared image identification with stored photos, hot spot areas with temperatures exceeding the surrounding average by more than 8°C are identified and marked as priority cleaning areas. When quantifying the degree of pollution, the dust coverage rate of the dust accumulation area in a single photovoltaic module is calculated. A coverage rate of <30% is considered as not requiring cleaning, 30% ≤ coverage rate <60% is considered as requiring routine cleaning, and a coverage rate of ≥60% is considered as requiring intensive cleaning. When the area of ​​a single hard contaminant is ≥4cm² or the number of hard contaminants on a single photovoltaic module is ≥5, it is considered as requiring targeted cleaning. When the area covered by light contaminants is ≥20% or when the junction box area of ​​the photovoltaic module is obstructed, it is considered as requiring immediate cleaning. The above identification and judgment results are converted into structured data containing partition number, pollution type, and pollution level and synchronized to the data processing platform.

3. A photovoltaic power station cleaning method based on collaborative operation of unmanned aerial vehicles and robots according to claim 2, characterized in that, The specific process in S4 of determining the cleaning zone execution order based on structured data and a dynamic priority evaluation model is as follows: When determining the cleaning partition execution order, the data processing platform constructs a dynamic priority evaluation model, which calculates the priority score by combining three types of parameters: Pollution urgency: 10 points are awarded for hot spot areas, junction boxes shielded by light pollutants, or light pollutant coverage areas ≥20%; 8 points are awarded for dust coverage ≥60% or areas with dense hard pollutants; 5 points are awarded for dust coverage ≤30% <60%. Environmental adaptability: Wind speed < 5m / s and no rainfall in the next 2 hours are considered adaptable; wind speed 5-8m / s or short-term rainfall warning are considered conditionally adaptable; wind speed ≥ 8m / s are considered unsuitable and the zone is locked. Route efficiency: The spatial distance between each zone is calculated using the digital twin model of the power station, and a clustering algorithm is used to group zones with a distance ≤ 50 meters into the same cleaning batch; The priority score calculation formula is: Priority Score = Pollution Urgency × 0.5 + Environmental Adaptability × 0.2 + Path Efficiency × 0.3, which generates the cleaning zone execution sequence.

4. A photovoltaic power station cleaning method based on collaborative operation of unmanned aerial vehicles and robots according to claim 3, characterized in that, The specific process of generating path parameters for the cleaning robot in S4 for different pollution characteristics is as follows: The dust accumulation area adopts an S-shaped reciprocating path with 100% coverage. When the dust accumulation coverage is ≥60%, a secondary reciprocating mode is activated. For areas with hard contaminants, a fixed-point cleaning subroutine is superimposed on the S-shaped path: a round-trip path with a side length of 10-20cm is planned with the center of the contaminant as the origin, the dwell time is ≥10 seconds, the brush head pressure is increased by 30% and adjusted in real time by the pressure sensor; For areas with light pollutants: When the cleaning robot is equipped with a dustbin, it uses an edge-gathering and center-cleaning path. First, it plans a closed rectangular path along the edge of the photovoltaic module to gather fallen leaves to the center and then cleans them in a straight line. When the cleaning robot is not equipped with a dustbin, it uses a directional pushing and edge-exiting path. The photovoltaic module is divided into three longitudinal areas along its length: left, middle, and right. It pushes and sweeps from the high side to the low side according to the installation tilt angle. The pollutants are pushed to the low-side outlet and then the robot performs 2-3 back-and-forth sweeps along the low-side edge. For pollutants in crevices, the robot is identified by an image sensor and then moves back and forth ±10cm around the crevices to perform point sweeping.

5. A photovoltaic power station cleaning method based on collaborative operation of unmanned aerial vehicles and robots according to claim 1, characterized in that, The specific process of the robot vacuum cleaner in S6, after arriving at the target zone, matching the zone's latitude and longitude coordinates and retrieving path parameters to perform precise path cleaning, is as follows: By matching the latitude and longitude coordinates of the partition, the path parameters in S4 are retrieved. Based on the differential drive system of the tracked chassis, the trajectory deviation is corrected in real time by the wheel speed encoder at a speed of 0.8-1.2m / s. When the distance is detected to be ≤8cm, the speed is reduced to 0.2m / s and the steering angle is adjusted.

6. A photovoltaic power station cleaning method based on collaborative operation of unmanned aerial vehicles and robots according to claim 5, characterized in that, The specific process for verifying the multi-dimensional cleaning effect in S6 is as follows: The roller brush shaft has a built-in pressure sensor that dynamically monitors the contact pressure with the photovoltaic module surface. When the pressure is >50N, the roller brush is automatically raised; when the pressure is <30N, the downward pressure is increased. The high-definition camera on the front of the drone takes images of the cleaned area at a set frequency, compares them with the photos stored in the S3, and calculates the cleaning coverage. If the coverage is ≥90%, the cleaning is considered to have met the standard; otherwise, local re-cleaning is triggered. If a hot spot with a temperature 5°C or higher than the surrounding average is found after scanning the photovoltaic module surface, the coordinates of the hot spot are immediately marked and uploaded to the data processing platform to trigger a potential fault warning.

7. A photovoltaic power station cleaning method based on collaborative operation of unmanned aerial vehicles and robots according to claim 6, characterized in that, The specific implementation of the corresponding level of response mechanism when encountering an obstacle in S6 is as follows: During the cleaning process, if the cleaning robot detects a sudden increase in obstacles, or if the sensors identify a path height difference or gap width exceeding the cleaning robot's obstacle-crossing capability, or if the cleaning robot's posture undergoes a significant change, a Level 3 response will be immediately executed: The three-level response is as follows: using the robot's own adaptive obstacle-crossing capability, it bypasses obstacles and automatically returns to the original path, cleaning without interruption throughout the process; The secondary response is as follows: if the width or height of the obstacle exceeds the obstacle-crossing capability of the cleaning robot, the cleaning robot stops working and takes an image of the obstacle, sends an alarm message containing coordinates to the data processing platform, and switches to a low-power waiting-for-command mode. The first-level response is as follows: if the robot's tilt angle is detected to be ≥15° or the brush is stuck, an emergency stop is immediately initiated, power output is cut off, and an audible and visual alarm is issued, awaiting manual intervention.

8. A photovoltaic power station cleaning method based on collaborative operation of unmanned aerial vehicles and robots according to claim 1, characterized in that, The process of S2 is as follows: The types of inspections include regular inspections of the entire photovoltaic power station and temporary inspections triggered by preset conditions, such as sandstorm weather, abnormal photovoltaic array power generation efficiency, and manual power activation. During the inspection, the drone flew over the entire photovoltaic panel array area, referencing the parameters of the sample, and took photos of each photovoltaic section. During the inspection, the drone interacts with the data processing platform, synchronizing the photos of the photovoltaic zones it captures to the data processing platform in real time for storage, thus creating stored photos.

9. A photovoltaic power station cleaning method based on collaborative operation of unmanned aerial vehicles and robots according to claim 1, characterized in that, In S3, the photovoltaic module contaminant identification model is constructed based on a deep convolutional neural network.

10. A photovoltaic power station cleaning system based on collaborative operation of drones and robots, characterized in that, The cleaning method according to any one of claims 1-9 includes: Data collection module: used to divide the photovoltaic modules of the photovoltaic power station into several zones and number them respectively. The drone takes visible light and infrared photos of each zone, and records the drone's altitude, latitude and longitude, heading angle and lens angle parameters at the time of shooting to form a reference sample. Data synchronization module: used to carry out regular inspections or temporary inspections triggered by preset conditions. During the inspection, the drone takes photos of each zone with reference to the parameters of the sample, and synchronizes the photos to the data processing platform for storage, thus obtaining the stored photos. Identification and Judgment Module: Used to classify and identify pollutants in stored photos and quantify the degree of pollution using a photovoltaic module pollutant identification model, convert the identification results into structured data and synchronize them to the data processing platform; Path planning module: It is used to determine the execution order of cleaning zones based on structured data through a dynamic priority evaluation model, generate path parameters for cleaning robots according to different pollution characteristics, and issue scheduling instructions in layers after completing multi-device status verification. Cleaning plan specification module: This module enables the drone to receive scheduling instructions, identify and grab the cleaning robot, transport the cleaning robot to the target area according to the planned shortest transport path and place it stably, and then transport the cleaning robot to the next target area or base station charging position after cleaning is completed. Cleaning verification module: After the cleaning robot arrives at the target area, it matches the latitude and longitude coordinates of the area and retrieves the path parameters, performs precise path cleaning, and conducts multi-dimensional cleaning effect verification. When it encounters obstacles, it executes the corresponding level of response mechanism. End module: Used to return the drone carrying the cleaning robot and put them into storage after the entire cleaning plan is completed.

Images