A photovoltaic module intelligent robot autonomous cleaning system

Abstract

This invention discloses an intelligent robot autonomous cleaning system for photovoltaic modules, relating to the field of photovoltaic module cleaning technology. The system includes: an environmental data acquisition module for real-time acquisition of environmental data from photovoltaic modules in a photovoltaic module assembly via a sensor array; a frame positioning module for performing frame positioning operations based on the light intensity data; a reference point conversion module for fusing surface height difference and tilt angle information from the laser ranging data to convert the frame positioning results into three-dimensional spatial reference points; a fall-prevention zone generation module for generating a fall-prevention zone by shifting a dynamic safety distance inward; a cleaning path determination module for generating a cleaning path covering the photovoltaic module assembly by combining the fall-prevention zone with a pollution distribution heatmap; decomposing the cleaning path into atomic behavior chains; and a cleaning drive module for driving the cleaning robot to perform cleaning operations according to the atomic behavior chains.

Description

Technical Field

[0001] This application relates to the field of photovoltaic module cleaning technology, and more specifically, to a photovoltaic module intelligent robot autonomous cleaning system. Background Technology

[0002] Against the backdrop of the global energy transition, photovoltaic (PV) power generation, as a crucial component of clean and renewable energy, plays a vital role in ensuring energy supply through its efficient and stable operation. Dust, dirt, and other contaminants on the surface of PV modules significantly reduce light absorption efficiency, leading to a decrease in power generation. Statistics show that modules that have not been cleaned for a long time exhibit a significant drop in power generation efficiency. Therefore, timely cleaning and maintenance of PV modules are essential measures to ensure efficient power generation from PV systems and reduce the cost per kilowatt-hour.

[0003] Traditional manual cleaning methods not only face safety risks associated with working at heights, but also suffer from low efficiency, high costs, and environmental constraints, making them difficult to adapt to the operation and maintenance needs of large-scale photovoltaic power plants. Therefore, the automation and intelligentization of cleaning operations have become an inevitable trend in the industry.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides an intelligent robot autonomous cleaning system for photovoltaic modules to solve the above-mentioned technical problems.

[0006] This application provides an intelligent robot autonomous cleaning system for photovoltaic modules, comprising:

[0007] An environmental data acquisition module is used to collect environmental data of the photovoltaic modules in the photovoltaic module array in real time through a sensor array. The environmental data includes light intensity data and laser ranging data.

[0008] The border positioning module is used to perform border positioning operations based on the light intensity data;

[0009] The reference point conversion module is used to fuse the surface height difference and tilt angle information in the laser ranging data and convert the border positioning result into a three-dimensional spatial reference point.

[0010] The anti-fall zone generation module is used to generate an anti-fall zone by shifting a dynamic safety distance inward along the plane normal direction of the photovoltaic module, with the three-dimensional spatial reference point as the origin.

[0011] The cleaning path determination module is used to combine the anti-drop zone and the contamination distribution heat map to generate a cleaning path covering the photovoltaic module assembly; the cleaning path is decomposed into atomic behavior chains, which include edge movement, gap crossing, and single-row cleaning;

[0012] A cleaning drive module is used to drive the cleaning robot to perform cleaning operations according to the atomic behavior chain.

[0013] Furthermore, the environmental data also includes infrared thermal radiation data, magnetic field strength data, and visible light image data; the border positioning module performs border positioning operations based on the illumination intensity data, including:

[0014] When the light intensity is greater than the strong light threshold, the position of the metal frame is identified based on the temperature gradient abrupt change in the infrared thermal radiation data.

[0015] When the light intensity is less than the weak light threshold, the spatial peak location of the embedded magnetic track coordinates is based on the magnetic field intensity data.

[0016] When the light intensity is between the strong light threshold and the weak light threshold, the outline of the border is identified by the edge color difference of the visible light image data.

[0017] Furthermore, the reference point conversion module integrates the surface height difference and tilt angle information from the laser ranging data to convert the bounding box positioning result into a three-dimensional spatial reference point, including:

[0018] When the light intensity is greater than the strong light threshold, the position of the metal frame identified by the infrared thermal radiation data is used as the planar constraint boundary.

[0019] When the light intensity is less than the weak light threshold, the coordinates of the pre-embedded magnetic track located by the magnetic field intensity data are used as the planar constraint boundary.

[0020] When the light intensity is between the strong light threshold and the weak light threshold, the bounding contour identified by the visible light image data is used as the planar constraint boundary.

[0021] By integrating the surface height difference from the laser ranging data, the three-dimensional spatial vector of the splicing gap between adjacent photovoltaic modules is calculated;

[0022] The surface normal vector field of the component is constructed by combining the tilt angle information, and the planar distortion caused by installation deformation is compensated by non-uniform B-spline interpolation;

[0023] Using the planar constraint boundary as a reference, a set of three-dimensional spatial reference points with curvature features is established by extending along the three-dimensional spatial vector.

[0024] Furthermore, the drop protection zone generation module uses the three-dimensional spatial reference point as the origin and generates a drop protection zone by shifting inward by a dynamic safety distance along the plane normal direction of the photovoltaic module, including:

[0025] The curvature characteristics of the three-dimensional spatial reference point set are analyzed to identify geometric abrupt change regions at the edges of photovoltaic modules;

[0026] The pollutant humidity distribution is retrieved based on the infrared thermal radiation data, and the oil stain attachment area is extracted by combining the visible light image data.

[0027] A dynamic friction coefficient model based on pollutant type was established, including: the friction coefficient in high humidity areas decreases exponentially; the friction coefficient in oily areas decreases in a stepwise manner.

[0028] Using the geometric abrupt change region as the risk anchor point, the critical slip threshold is calculated according to the dynamic friction coefficient model;

[0029] By overlaying real-time wind speed sensor data, a composite safety distance that takes into account wind load is generated;

[0030] A non-linear offset with curvature adaptation is performed along the plane normal direction to form a drop-proof zone at the edge of the photovoltaic module.

[0031] Furthermore, the cleaning path determination module, in conjunction with the drop-proof zone and the contamination distribution heat map, generates a cleaning path covering the photovoltaic module assembly, including:

[0032] The pollution distribution heatmap was subjected to multifractal dimension decomposition to extract the Hurst index for each region;

[0033] Areas with a Hurst index greater than the pollution aggregation threshold are designated as priority cleaning core areas;

[0034] Using the anti-fall zone as a rigid motion boundary, a two-layer topology path including a contamination core connection layer and a gradient cover layer is constructed.

[0035] The pollution core connection layer uses an ant colony algorithm to generate the optimal energy consumption path connecting the priority cleaning core area, wherein the energy consumption assessment value of the photovoltaic panel gap crossing action is dynamically corrected by the modulus of the three-dimensional spatial vector.

[0036] The gradient overlay layer generates contamination diffusion paths with equal contamination lines based on the descent direction of the contamination gradient, and the density of path points is positively correlated with the contamination gradient value.

[0037] During the generation of the gradient overlay layer, an artificial potential field constraint mechanism is integrated, including: the contamination gradient field provides an gravitational potential field, the boundary of the anti-fall zone generates a repulsive potential field, and the normal vector field of the component surface constrains the degrees of freedom of movement.

[0038] Furthermore, the cleaning path determination module decomposes the cleaning path into a chain of atomic behaviors, including:

[0039] The two-layer topology path is compiled into a chain of atomic behaviors;

[0040] The movement speed along the edge is adaptively adjusted based on the curvature characteristics of the three-dimensional spatial reference point;

[0041] The pre-calibrated posture of the cleaning robot that crosses gaps is calculated based on the three-dimensional spatial vector of the splicing gaps;

[0042] The operating parameters of the cleaning mechanism for single-line cleaning are configured based on the pollution concentration level of the priority cleaning core area.

[0043] Furthermore, the cleaning drive module drives the cleaning robot to perform cleaning operations according to the atomic behavior chain, including:

[0044] When performing actions that cross gaps, the inertial load abrupt change range of the robotic arm joints of the cleaning robot is predicted based on the three-dimensional spatial vector of the splicing gap.

[0045] The deformation error at the gap edge is monitored in real time using the visible light image data;

[0046] Based on the coupling relationship between the abrupt change range of the inertial load and the deformation error, a fuzzy PID controller is used to dynamically allocate the torque of the joint motor to suppress vibration.

[0047] Furthermore, when performing a single-line cleaning action, the base pressure setting value is determined based on the degree of contamination accumulation in the priority cleaning core area;

[0048] The wind pressure disturbance compensation coefficient is generated by integrating real-time wind speed sensor data;

[0049] The brush speed compensation amount is derived based on the dynamic friction coefficient model.

[0050] Output cleaning execution commands with wind pressure compensation and speed correction;

[0051] When the edge-moving behavior is triggered, the magnetic field strength data is called in real time to correct the lateral offset of the cleaning robot relative to the photovoltaic frame.

[0052] Furthermore, after the gap-crossing behavior is completed, actual motion energy consumption data is collected and compared with the energy consumption assessment value of the contaminated core connection layer;

[0053] The mapping relationship between the magnitude of the three-dimensional spatial vector and the energy consumption assessment value is dynamically adjusted based on the comparison results.

[0054] After a single-line cleaning operation is completed, the rate of change of the infrared thermal radiation data is extracted to determine the cleaning effect;

[0055] The pollution distribution heat map is reconstructed based on the rate of change of the infrared thermal radiation data.

[0056] The cleaning path determination module is triggered to dynamically update the cleaning path for the unexecuted areas based on the reconstructed contamination distribution heatmap.

[0057] Furthermore, the photovoltaic module intelligent robot autonomous cleaning system also includes a communication interruption emergency module; wherein, the communication interruption emergency module is used for:

[0058] When a communication interruption with the control center is detected, the pre-existing local atomic behavior chain of the cleaning robot is activated as the execution benchmark.

[0059] Based on the set of three-dimensional spatial reference points, the positioning closed-loop control is achieved by integrating the inertial measurement unit built into the cleaning robot.

[0060] The laser ranging data is acquired in real time, and the boundary coordinates of the fall protection zone are dynamically corrected.

[0061] If, during a single-row cleaning operation, the local contamination intensity calculated from real-time infrared thermal radiation data exceeds the locally stored equivalent threshold, then the local cleaning operation is switched to the contamination line diffusion path of the gradient cover layer.

[0062] After communication is restored, the execution status of the cleaning path recorded locally will be synchronized with the cleaning path stored in the control center.

[0063] Based on the embodiments provided in this application, the safety and efficiency of photovoltaic module cleaning are significantly improved through multi-sensor collaboration and intelligent path planning technology. The environmental data acquisition module simultaneously acquires light intensity and laser ranging data, providing multi-dimensional environmental information for subsequent modules and avoiding the limitations of single-sensor data. The frame positioning module uses light intensity data to accurately identify the photovoltaic module frame, providing basic coordinates for subsequent path planning. The reference point conversion module innovatively integrates surface height difference and tilt angle information, converting the two-dimensional positioning result into a three-dimensional spatial reference point, solving the positioning problem caused by complex terrain and changes in module installation angle.

[0064] The anti-fall zone generation module dynamically generates a safe working area based on 3D benchmarks, effectively avoiding the risk of the cleaning robot falling while operating at edges and improving system reliability. The cleaning path determination module combines the anti-fall zone with a contamination heatmap to generate the optimal path and breaks it down into standardized atomic behavior chains. This enables the robot to efficiently complete complex actions such as edge movement and gap crossing, reducing blind spots and repetitive work, and improving cleaning coverage. Through the collaborative work of the modules, the overall system achieves full automation and intelligence in photovoltaic module cleaning, reducing the need for manual intervention and adapting to the operation and maintenance needs of large-scale photovoltaic power plants. Attached Figure Description

[0065] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0066] Figure 1 This is a structural diagram of an optional intelligent robot autonomous cleaning system for photovoltaic modules according to an embodiment of this application;

[0067] Figure 2 This is a flowchart illustrating an optional border positioning module performing a border positioning operation based on illumination intensity data according to an embodiment of this application.

[0068] Figure 3 This is a flowchart of an optional method for converting a bounding box positioning result into a three-dimensional spatial reference point according to an embodiment of this application.

[0069] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0071] Optionally, such as Figure 1 As shown, this application provides an intelligent robot autonomous cleaning system for photovoltaic modules, comprising:

[0072] The environmental data acquisition module 101 is used to collect environmental data of the photovoltaic modules in the photovoltaic module set in real time through a sensor array. The environmental data includes light intensity data and laser ranging data.

[0073] The border positioning module 102 is used to perform border positioning operations based on light intensity data;

[0074] The reference point conversion module 103 is used to fuse the surface height difference and tilt angle information in the laser ranging data and convert the bounding box positioning result into a three-dimensional spatial reference point.

[0075] The anti-fall zone generation module 104 is used to generate an anti-fall zone by offsetting a dynamic safety distance inward along the plane normal direction of the photovoltaic module with the three-dimensional space reference point as the origin.

[0076] The cleaning path determination module 105 is used to combine the anti-drop zone and the contamination distribution heat map to generate a cleaning path covering the photovoltaic module assembly; the cleaning path is decomposed into atomic behavior chains, which include edge movement, gap crossing and single-row cleaning;

[0077] The cleaning drive module 106 is used to drive the cleaning robot to perform cleaning operations according to the atomic behavior chain.

[0078] Based on the embodiments provided in this application, the safety and efficiency of photovoltaic module cleaning are significantly improved through multi-sensor collaboration and intelligent path planning technology. The environmental data acquisition module simultaneously acquires light intensity and laser ranging data, providing multi-dimensional environmental information for subsequent modules and avoiding the limitations of single-sensor data. The frame positioning module uses light intensity data to accurately identify the photovoltaic module frame, providing basic coordinates for subsequent path planning. The reference point conversion module innovatively integrates surface height difference and tilt angle information, converting the two-dimensional positioning result into a three-dimensional spatial reference point, solving the positioning problem caused by complex terrain and changes in module installation angle.

[0079] The anti-fall zone generation module dynamically generates a safe working area based on 3D benchmarks, effectively avoiding the risk of the cleaning robot falling while operating at edges and improving system reliability. The cleaning path determination module combines the anti-fall zone with a contamination heatmap to generate the optimal path and breaks it down into standardized atomic behavior chains. This enables the robot to efficiently complete complex actions such as edge movement and gap crossing, reducing blind spots and repetitive work, and improving cleaning coverage. Through the collaborative work of the modules, the overall system achieves full automation and intelligence in photovoltaic module cleaning, reducing the need for manual intervention and adapting to the operation and maintenance needs of large-scale photovoltaic power plants.

[0080] Furthermore, environmental data also includes infrared thermal radiation data, magnetic field strength data, and visible light image data; such as... Figure 2 As shown, the border positioning module performs border positioning operations based on illumination intensity data, including:

[0081] S201, When the light intensity is greater than the strong light threshold, the position of the metal frame is identified based on the temperature gradient change of infrared thermal radiation data.

[0082] When the ambient light intensity exceeds the strong light threshold, the strong light will interfere with the acquisition of visible light images. For example, the strong light threshold can be set to 8000 lux, 9000 lux, etc. On a sunny day at noon, the light intensity on the surface of photovoltaic modules often exceeds 10000 lux, triggering infrared positioning.

[0083] The metal frame and the photovoltaic panel surface have different thermal conductivity characteristics. Under strong light, the metal heats up faster. By collecting the temperature distribution using an infrared sensor, the boundary of the abrupt temperature gradient (i.e., the junction between the metal frame and the photovoltaic panel) can be identified. For example, when the light intensity is 12000 lux, the surface temperature of the photovoltaic panel glass is about 50°C, and the temperature of the aluminum alloy frame is about 65°C. If the infrared thermal imager detects an area with a temperature gradient exceeding 5°C / cm, it can be determined to be the frame location.

[0084] S202, When the light intensity is less than the weak light threshold, the spatial peak location of the embedded magnetic track coordinates is based on the magnetic field intensity data.

[0085] When the ambient light intensity is below the low light threshold, the visible light image lacks clarity, and magnetic field strength data can be used for bounding box localization. For example, the low light threshold can be set to 500 lux, 400 lux, etc. For instance, when the light intensity is below 300 lux on a cloudy day or in the early morning, magnetic field localization is triggered.

[0086] In practical implementation, magnetic tracks (such as magnetic metal strips) are pre-embedded during the installation of photovoltaic modules. In low-light environments, a magnetic field sensor detects the spatial magnetic field intensity distribution, and the peak position of the magnetic field corresponds to the coordinates of the magnetic track, which is used to determine the position of the module frame. For example, when the light intensity is 200 lux, the visible light camera image is blurry, and the magnetic field sensor detects that the magnetic field intensity in a certain area reaches 50 μT (the surrounding ambient magnetic field is about 30 μT). This peak point is the position coordinate of the pre-embedded magnetic track.

[0087] S203, when the light intensity is between the strong light threshold and the weak light threshold, the outline of the border is identified by the edge color difference of the visible light image data.

[0088] Under moderate illumination conditions, there is a color difference between the photovoltaic module frame (such as a black adhesive strip or metal frame) and the glass surface. Image processing algorithms are used to extract the contours of abrupt color differences at the edges to determine the frame location. For example, when the illumination intensity is 3000 lux, a visible light camera captures an image of the module and detects abrupt changes in the blue channel values ​​in the RGB color space (e.g., the frame's RGB value is [50,50,50], while the glass surface's is [200,200,200]). The area where the edge color difference exceeds a threshold is the frame contour.

[0089] Furthermore, such as Figure 3 As shown, the reference point conversion module integrates the surface height difference and tilt angle information from the laser ranging data to convert the bounding box positioning results into three-dimensional spatial reference points, including:

[0090] S301, when the light intensity is greater than the strong light threshold, the position of the metal frame identified by infrared thermal radiation data is used as the planar constraint boundary.

[0091] S302, when the light intensity is less than the weak light threshold, the coordinates of the pre-embedded magnetic track located by the magnetic field intensity data are used as the plane constraint boundary.

[0092] S303, when the light intensity is between the strong light threshold and the weak light threshold, the bounding contour identified by the visible light image data is used as the planar constraint boundary.

[0093] S304, integrates the surface height difference from laser ranging data to calculate the three-dimensional spatial vector of the splicing gap between adjacent photovoltaic modules;

[0094] In practical implementation, laser ranging sensors acquire surface height data of adjacent photovoltaic modules. By calculating the height difference and module tilt angle, a three-dimensional vector (containing X, Y, and Z axis components) is generated to describe the spatial position and direction of the splicing gap. This vector is used to characterize the width, height difference, and tilt angle of the gap. For example, for two adjacent photovoltaic modules, the surface height of the left module is 1.5 meters, and that of the right module is 1.45 meters, with tilt angles of 15° and 16° respectively. After fusing the laser ranging data, a three-dimensional vector (0.2 meters, 0.05 meters, -0.1 meters) is obtained, indicating that the gap is 0.2 meters wide in the horizontal direction, 0.05 meters high in the vertical direction, and tilts downward along the Z-axis.

[0095] S305, combined with tilt angle information, constructs the component surface normal vector field, and compensates for the planar distortion caused by installation deformation through non-uniform B-spline interpolation;

[0096] In practical implementation, based on the tilt angle data of each point (such as the surface slope obtained by laser ranging), the normal vector (a vector perpendicular to the surface) of each point is determined, forming a normal vector distribution field of the entire component surface, which is used to describe the spatial orientation of the surface. For example, the tilt angle of the central region of the component is 10°, and the normal vector is (0,0,cos10°); the tilt angle of the edge region is 12°, and the normal vector is (0.1,0,cos12°). The normal vectors of all points constitute a vector field characterizing the surface undulations.

[0097] S306 establishes a set of three-dimensional spatial reference points with curvature features by extending along three-dimensional spatial vectors, using planar constraint boundaries as a reference.

[0098] In practical implementation, photovoltaic modules may undergo deformation (such as bending) during installation. Non-uniform B-spline interpolation algorithms, by fitting the height data of measured points, smoothly compensate for local distortions, approximating the non-planar surface of the module as a regular curved surface, thus improving the accuracy of 3D modeling. For example, if three local protrusions (height difference of 0.3 cm) are measured on the surface of a module, non-uniform B-spline interpolation adds interpolation nodes to the protruding areas, generating a smooth curved surface model and controlling the distortion error within 0.1 cm.

[0099] Furthermore, the drop protection zone generation module uses a three-dimensional spatial reference point as the origin and, along the plane normal direction of the photovoltaic module, shifts inward by a dynamic safety distance to generate a drop protection zone, including:

[0100] Analyze the curvature characteristics of a three-dimensional spatial reference point set to identify geometric abrupt change regions at the edges of photovoltaic modules;

[0101] In practical implementation, the curvature of each point in the three-dimensional reference point set (such as the rate of change of the normal vector of adjacent points) is calculated to identify the locations of abrupt curvature changes (such as component edge corners and gap connections). These areas are prone to causing instability when the robot walks. For example, if the curvature of the reference point at the upper left corner of the component is 0.1 / m, and the curvature of the adjacent points suddenly increases to 0.5 / m, this area is determined to be a region of geometrical abrupt change (such as a right-angle corner), and the risk of falling should be given special attention.

[0102] The humidity distribution of pollutants is retrieved based on infrared thermal radiation data, and the oil stain attachment area is extracted by combining visible light image data.

[0103] Infrared thermal radiation data reflects surface temperature distribution; areas with high humidity dissipate heat slowly and have high temperatures, allowing humidity to be inferred from temperature. Visible light images extract areas with oil contamination based on color features (such as the reflective properties of oil stains), which is used to assess surface friction risk. For example, if an infrared image shows an area with a temperature 3°C higher than its surroundings, it is identified as a high-humidity area with 80% humidity. In a visible light image, this area appears shiny and reflective, indicating an area with oil contamination, requiring a reduction in the robot's movement speed in that area.

[0104] A dynamic friction coefficient model based on pollutant type was established, including: the friction coefficient in high humidity areas decreases exponentially; the friction coefficient in oily areas decreases in a stepwise manner.

[0105] Among them, a mathematical model was established based on the influence of different pollutants (such as high humidity and oil) on the surface friction coefficient: the friction coefficient in high humidity areas decreases exponentially with increasing humidity, while the friction coefficient in oily areas decreases abruptly, which is used to calculate the critical slip risk of the robot. For example, when the humidity increases from 50% to 80%, the friction coefficient decreases exponentially from 0.6 to 0.3; the friction coefficient in oily areas decreases abruptly from 0.6 to 0.2, and the robot's safe distance is adjusted accordingly.

[0106] Using the geometrically abrupt region as the risk anchor point, the critical slip threshold is calculated according to the dynamic friction coefficient model;

[0107] Specifically, regions with abrupt geometric changes (such as edge corners) are used as risk anchor points. By combining the dynamic friction coefficient and the robot's weight, the critical external force threshold that leads to slippage is calculated to determine the safe operating range. For example, at a corner with abrupt curvature changes, the friction coefficient is 0.3 and the robot weight is 10kg. The calculated critical slippage threshold is 3kgf. Based on this, the safe distance for this area needs to be offset inward by 0.2 meters.

[0108] By overlaying real-time wind speed sensor data, a composite safety distance that takes into account wind load is generated;

[0109] In one specific implementation, the composite safety distance is determined based on the following formula:

[0110]

[0111] in, The composite safety distance (unit: cm) is the dynamic distance offset inward along the plane normal of the photovoltaic module, used to generate the boundary of the drop protection zone; The basic safety distance (unit: cm) is set according to the standard size of photovoltaic modules. In this embodiment, it is set to 5cm (applicable to 1.6m×1m conventional modules). The geometric mutation risk coefficient is calculated based on the curvature characteristics of a three-dimensional spatial reference point set: smooth region (κ≤0.1rad / cm): =1.0; Moderate mutation region (0.1<κ≤0.3rad / cm): =1.3; Severe mutation region (κ>0.3rad / cm): =1.5; The static friction coefficient for a clean surface is 0.8 for the photovoltaic glass surface; The dynamic friction coefficient is determined by the type of contaminant: High humidity areas: = exp(-0.02 H) (H is the percentage of humidity); Oily areas: when the oil coverage is >15%, =0.3 (step descent); The weight of the friction coefficient can be set, for example, 1.2, 1.1, etc. (when photovoltaic modules are installed at an angle, the friction coefficient has a significant impact on drop prevention). Wind speed influence factor (unit: s) 2 / m 2 (Set according to wind speed level: wind speed v≤5m / s: Kw=0.005s) 2 / m 2 ;5 <v≤10m / s:Kw=0.01s 2 / m 2 v>10m / s: Kw=0.02s 2 / m 2 ; This is the curvature adaptive coefficient (unit: cm / rad), for example, it can be 0.3, 0.4, etc. (to balance the influence of geometric smoothness on the safety distance). The surface curvature of the component (unit: rad / cm) is calculated by fitting a curve using laser ranging data.

[0112] In practice, real-time wind speed generates wind loads (such as horizontal thrust). The impact of these wind loads on the robot is calculated based on the wind speed, and then superimposed on the friction coefficient model to dynamically adjust the safety distance (e.g., the higher the wind speed, the greater the safety distance offset). For example, at a wind speed of 10 m / s, the wind load generates a horizontal thrust of 5 N on the robot. Combined with a friction coefficient of 0.3, the calculated composite safety distance needs to be offset inward by 0.1 meters from the original distance.

[0113] A non-linear offset with curvature adaptation is performed along the plane normal direction to form a drop-proof zone at the edge of the photovoltaic module.

[0114] Specifically, based on the curvature of each point on the component surface, a non-linear offset is made along the normal vector direction (perpendicular to the surface): the offset is larger in areas with large curvature (such as corners) and smaller in areas with small curvature, forming an irregular anti-fall boundary. For example, the curvature of the straight edge of the component is close to 0, so it is offset by 0.1 meters along the normal; the curvature at the corner is 0.5 / m, so it is offset by 0.3 meters, forming a concave anti-fall zone to ensure the stability of the robot when working at the edge.

[0115] Furthermore, the cleaning path determination module combines the drop-proof zone and the contamination distribution heat map to generate a cleaning path covering the photovoltaic module assembly, including:

[0116] Multifractal dimension decomposition was performed on the pollution distribution heat map to extract the Hurst index for each region;

[0117] In practice, the pollution distribution heatmap uses color depth to represent the degree of pollution, multifractal dimension decomposition can analyze the complexity of the pollution distribution, and the Hurst index is used to characterize the pollution aggregation trend (the larger the Hurst index, the more concentrated the pollution). For example, if the pollution heatmap of a certain area shows a blocky dark color distribution, the Hurst index after decomposition is 0.8, indicating that the pollution in this area is highly concentrated and needs to be cleaned first.

[0118] Areas with a Hurst index greater than the pollution aggregation threshold are designated as priority cleaning core areas;

[0119] The pollution aggregation threshold can be set to, but is not limited to, 0.6, 0.7, etc. Areas exceeding this value have a high degree of pollution aggregation and a significant impact on power generation efficiency. These areas are marked as priority cleaning core areas, and cleaning paths are prioritized for them. For example, areas with a Hurst index of 0.8 have dense accumulation of pollution particles and a high degree of decline in power generation efficiency. These areas are marked as priority cleaning core areas, and cleaning is prioritized for them.

[0120] Using the fall-prevention zone as a rigid motion boundary, a two-layer topological path is constructed, consisting of a contamination core connection layer and a gradient cover layer.

[0121] The core contamination connection layer uses an ant colony algorithm to generate the optimal energy consumption path for connecting and prioritizing the cleaning of the core area. The energy consumption assessment value of the photovoltaic panel gap crossing action is dynamically corrected by the magnitude of the three-dimensional spatial vector.

[0122] In one specific implementation, the energy consumption assessment value for the gap crossing action is determined based on the following formula:

[0123]

[0124] in, The energy consumption assessment value (unit: J) for the photovoltaic panel gap crossing action is used for ant colony algorithm path optimization; The basic energy consumption coefficient (unit: J / cm) is determined by the robot motor power; in this embodiment, it is taken as 0.8 J / cm. The three-dimensional spatial vector magnitude (unit: cm) of the splicing seam is calculated from laser ranging data; The gap complexity coefficient is determined by the edge deformation error of visible light image recognition: flat gap (deformation error <2mm): =1.0; Moderately irregular gap (2mm ≤ error ≤ 5mm): =1.3; Severe distortion gap (error >5mm): =1.5; The pollution clustering degree is weighted and correlated with the Hurst index H: H > 0.7 (high clustering degree). =0.8; 0.4≤H≤0.7 (moderate clustering): =0.5; H<0.4 (low aggregation): =0.2; The component surface tilt angle (unit: °) is obtained from the reference point conversion module; The installation deformation correction factor (unit: 1 / °) reflects the effect of component planar distortion: Standard installation (distortion < 1°): δ = 0.02 / °; Moderate deformation (1° ≤ distortion ≤ 3°): δ = 0.03 / °; Severe deformation (distortion > 3°): δ = 0.05 / °; The angle between the plane normal vectors of adjacent components (unit: °) reflects the degree of installation misalignment.

[0125] The gradient overlay layer generates contamination diffusion paths along the descent direction of the contamination gradient, and the density of path points is positively correlated with the contamination gradient value.

[0126] Among these, the "fall-prevention zone" serves as a rigid motion boundary. This means that when planning a path, the robot must absolutely not enter the fall-prevention zone (i.e., the high-risk area where it might fall from the edge of the photovoltaic panel). This boundary is a hard constraint, and path planning must strictly avoid it.

[0127] Path planning is not simply a matter of scanning the entire area; instead, it's structured in two layers to address different problems. The core contamination connection layer focuses on quickly and efficiently connecting the most heavily contaminated areas (prioritizing core cleaning zones), aiming to find the most energy-efficient paths connecting these core areas. The gradient coverage layer focuses on meticulous, comprehensive cleaning of contaminated areas. It plans cleaning routes based on the gradient of contamination levels, much like drawing contour lines, with more intensive cleaning in more heavily contaminated areas.

[0128] In some embodiments of this application, the core connection layer is contaminated: the goal is to generate optimal paths connecting all priority cleaned core areas. An ant colony optimization algorithm is used, a biomimetic optimization algorithm that mimics ants finding food paths and excels at finding relatively optimal (usually the shortest distance or lowest cost) paths in complex environments. A key optimization factor is energy consumption: the primary metric for path "quality" is energy consumption. The path with the lowest energy consumption is considered optimal. One of the most energy-intensive actions in the path is crossing gaps between photovoltaic panels. Crossing gaps with different widths, height differences, and inclinations consumes different amounts of energy.

[0129] The correction is based on the magnitude of the three-dimensional spatial vector. This magnitude represents the actual "span" of the gap in three-dimensional space (a combination of distance and height difference). The larger the magnitude, the wider the gap or the greater the height difference, and the more energy the robot needs to cross it (such as lifting height, moving distance, and maintaining stability).

[0130] The dynamic correction process includes the following: When the ant colony algorithm calculates the path cost, if a path needs to cross a gap, the algorithm queries the magnitude of the 3D spatial vector corresponding to that gap. The larger the magnitude, the higher the energy consumption assessment value of the crossing action in the path cost calculation. The algorithm will tend to select gaps with lower energy consumption assessment values ​​(i.e., gaps that are easier to cross) to connect the polluted core area, thereby finding the connection path with the lowest overall energy consumption.

[0131] For example, suppose a robot needs to connect three heavily polluted core areas, A, B, and C, with multiple gaps to cross (Gap1, Gap2, Gap3). Laser ranging data calculates that: Gap1 has a small 3D vector magnitude (narrow gap, small height difference) and a low energy consumption assessment value. Gap2 has a medium 3D vector magnitude. Gap3 has a large 3D vector magnitude (wide gap or large height difference) -> a high energy consumption assessment value.

[0132] When the ant colony algorithm is searching for the optimal path connecting A, B, and C, it will prioritize trying to select paths that include Gap1 and Gap2, while trying to avoid paths that include the energy-intensive Gap3. In the end, it will generate a path that connects the three core areas with the lowest total energy consumption (e.g., A->cross Gap1->B->cross Gap2->C).

[0133] In some embodiments of this application, the gradient coverage layer aims to thoroughly and completely clean the entire photovoltaic panel area (especially contaminated areas) while avoiding drop-proof zones. The planning is based on the direction of the pollution gradient descent and the pollution gradient value. The pollution gradient refers to the rate of change of pollution levels (such as dust concentration or oil thickness) in space. High gradient areas indicate dramatic changes in pollution levels (such as the edge of a transition from heavily polluted to lightly polluted areas). Contamination contour lines are similar to contour lines on a map, connecting points with the same level of pollution.

[0134] Path generation methods include: Based on gradient descent direction: The robot's cleaning path generally follows the direction of the fastest decrease in pollution level (gradient descent direction). Imagine water flowing naturally from high (heavy pollution) to low (light pollution). Generating isocontamination diffusion paths: The robot moves along "lines" (isocontamination lines) with roughly the same pollution level. It gradually "diffused" from the high-pollution line to the low-pollution line. Path point density is positively correlated: The larger the pollution gradient value (the more drastic the pollution change), the denser the path points are set during path generation. This means that in areas with large changes in pollution level (such as the edge of a heavily polluted core area or the boundary between different types of pollution), the robot will move more slowly, cover more thoroughly, and perform more frequent or forceful cleaning actions to ensure complete removal of pollution. Conversely, in areas with uniform pollution or small gradients (such as the interior of a lightly polluted area), path points can be sparser, allowing the robot to move faster and improving efficiency.

[0135] In one specific implementation, suppose there is a circular area of ​​heavily contaminated oil in the center of a photovoltaic panel (prioritizing cleaning the core area), with the contamination gradually decreasing outwards. The contamination core connecting layer plans the path to this core area. The gradient cover layer is responsible for cleaning this core area and its surroundings.

[0136] Once the robot reaches the core area, it first performs circular or reciprocating cleaning along the innermost (most heavily contaminated) "isocontamination line" (the path points are very dense, and the cleaning is thorough). After completing one layer, it moves to the next "isocontamination line" with slightly less contamination, following the direction of the decreasing contamination gradient (outwards), to continue cleaning. As the contamination spreads outwards, the contamination gradient value gradually decreases (the contamination change tends to be gentler), and the path point density also decreases accordingly, allowing the robot's movement speed to be appropriately increased.

[0137] Ultimately, like ripples spreading across water, the cleaning process covered the entire area that needed to be cleaned from the inside out and from dense to sparse areas, with more cleaning resources invested in areas that were heavily polluted and subject to significant changes.

[0138] During the generation of the gradient overlay layer, an artificial potential field constraint mechanism is integrated, including: the pollution gradient field provides an gravitational potential field, the boundary of the anti-fall zone generates a repulsive potential field, and the component surface normal vector field constrains the degree of freedom of movement.

[0139] The pollution gradient field provides a gravitational potential field, transforming the pollution distribution heatmap (pollution severity map) into a virtual gravitational field. The more heavily polluted the area (high pollution value), the stronger the "attraction." This gravitational field "attracts" the cleaning robot to prioritize cleaning areas with high pollution levels. It guides the robot along the direction of the decreasing pollution gradient (i.e., from high-pollution areas to low-pollution areas), ensuring that the robot is always "pulled" towards the places that most need cleaning.

[0140] A repulsive potential field is generated at the boundary of the drop protection zone. A virtual repulsive field is set up at the boundary of the drop protection zone (i.e., the danger zone at the edge of the photovoltaic module). The closer to the boundary of the drop protection zone, the stronger the "repulsive force" generated. This repulsive field "pushes" the cleaning robot away, preventing it from approaching or entering the danger zone. It acts as a rigid safety boundary, ensuring that the generated path always remains within the safe working area.

[0141] The normal vector field of the module surface constrains the robot's degrees of freedom of movement, utilizing the photovoltaic module surface normal vector field (i.e., the direction vector perpendicular to the module surface at each point) previously calculated using laser ranging and tilt angle information. This field defines the theoretically safe directions for robot movement. This field constrains the robot's degrees of freedom of movement when generating a path. It ensures: Path conforms to the surface: The direction of robot movement (especially the direction of its cleaning mechanism) must be as perpendicular as possible to the local surface normal vector (i.e., movement within the plane of the module surface), preventing the robot from slipping, tipping over, or failing to clean due to incorrect path direction. Adaptation to curved surfaces / tilts: When the module surface has an installation tilt or slight deformation (not an absolute plane), the normal vector field provides a local coordinate system. Path planning is based on this coordinate system, ensuring that the robot's posture and trajectory always adapt to the surface geometry of its current location.

[0142] In one specific embodiment of this application, the cooperative working process of the three potential fields includes:

[0143] Initial guidance: The gravity field of the pollution gradient is the main driving force, guiding the path to converge towards the highly polluted area.

[0144] Safety avoidance: When the gravitational field attempts to pull the path closer to the edge (highly contaminated areas are sometimes near the edge), the repulsive field of the fall-prevention zone kicks in, pushing the path away from the danger boundary. Gravity and repulsion reach a dynamic equilibrium near the edge, forming a path that is both close to the contaminated area and maintains a safe distance.

[0145] Orientation constraints: Throughout the process, the normal vector field of the component surface continuously applies. This ensures that when calculating path points, the robot's movement direction vector must be projected onto the plane defined by the current point's normal vector. Path turns and curve changes must follow the local geometric features of the surface (reflected by changes in the normal vector). Especially when crossing seams or areas of surface distortion, the constraints of the normal vector field guarantee the physical feasibility and safety of movement commands (e.g., preventing wheels from dangling or getting stuck).

[0146] Path Generation: Path planning algorithms (such as gradient descent, potential field methods, or other optimization algorithms) calculate the final cleaning path (i.e., the gradient overlay path) under the combined influence of three "virtual forces": the attraction of the contamination gradient (primary objective), the anti-fall repulsion force (hard safety constraint), and the surface normal (kinematic constraint). This path will: prioritize covering heavily contaminated areas (gravity-dominated); strictly avoid anti-fall zones (repulsion guarantee); smoothly adhere to the actual surface of the component (normal constraint); and diffuse along the contamination gradient direction (gravity direction guidance).

[0147] Furthermore, the cleaning path determination module decomposes the cleaning path into a chain of atomic behaviors, including:

[0148] Compile the two-layer topology path into a chain of atomic behaviors;

[0149] In practice, the two-layer topology path is transformed into a sequence of atomic behaviors such as moving along edges, crossing gaps, and single-row cleaning.

[0150] The movement speed of the edge-moving behavior is adaptively adjusted based on the curvature characteristics of the three-dimensional spatial reference point;

[0151] In practice, the speed at which the cleaning robot moves along the edge is adjusted based on the curvature characteristics of the three-dimensional spatial reference point. When the curvature is large, the speed is reduced to ensure cleaning effectiveness and robot operation safety; when the curvature is small, the speed is appropriately increased to improve cleaning efficiency.

[0152] The pre-calibrated posture of the cleaning robot that crosses gaps is calculated based on the three-dimensional spatial vector of the splicing gaps;

[0153] In practice, before the robot crosses the gap between the photovoltaic modules, its posture is pre-adjusted according to the three-dimensional spatial vector of the splicing seam to ensure a smooth crossing process.

[0154] The working parameters of the cleaning agency for single-line cleaning operations are configured based on the pollution concentration level of the priority cleaning core area.

[0155] In practice, the operating parameters (such as pressure and speed) of the cleaning mechanism are set in stages according to the degree of contamination in the priority cleaning core area. In areas with high contamination, the pressure and speed are increased; in areas with low contamination, the pressure and speed are appropriately reduced.

[0156] Furthermore, the cleaning drive module drives the cleaning robot to perform cleaning operations according to the atomic behavior chain, including:

[0157] When performing actions that cross gaps, the inertial load change range of the robotic arm joints of the cleaning robot is predicted based on the three-dimensional spatial vector of the splicing gap.

[0158] In practical implementation, a dynamic model of the robotic arm is established based on the three-dimensional spatial vector of the splicing gap (generated by the reference point conversion module). By analyzing the relationship between the joint motion trajectory and the geometric characteristics of the gap, the time point and amplitude range of the inertial load mutation are calculated, providing a predictive basis for subsequent control.

[0159] The deformation error at the gap edge is monitored in real time using visible light image data;

[0160] In practice, visible light image data is used to extract gap edge feature points through an edge detection algorithm. The current frame is then matched with historical reference frames to calculate the displacement of these feature points as deformation error. Kalman filtering is then employed to smooth the error data.

[0161] Based on the coupling relationship between the sudden change range of inertial load and deformation error, a fuzzy PID controller is used to dynamically allocate the joint motor torque to suppress vibration.

[0162] Among them, joint motor torque refers to the torque output by the motor that drives mechanical joints, such as robot joints, to rotate, maintain posture, or overcome loads. It is a core parameter for measuring the power output capability of joint motors.

[0163] In practical implementation, a fuzzy control rule base is constructed, with the inertial load abrupt change range and deformation error as inputs and the joint motor torque as the output. Based on the magnitude and trend of the error, the parameters of the fuzzy PID controller are dynamically adjusted to achieve adaptive torque distribution. For example, when a large deformation error is detected, the proportional coefficient is increased for a faster response; when approaching the target position, the integral coefficient is increased to eliminate static errors.

[0164] Furthermore, when performing a single-line cleaning operation, the base pressure setpoint is determined based on the degree of contamination concentration in the priority cleaning core area;

[0165] The wind pressure disturbance compensation coefficient is generated by integrating real-time wind speed sensor data;

[0166] The pollution levels are determined based on the Hurst index in the pollution heat map. High-concentration areas (Hurst > 0.7) use a high-pressure cleaning mode, medium-concentration areas (0.4-0.7) use a medium-pressure mode, and low-concentration areas (< 0.4) use a low-pressure mode. Different modes correspond to different pressure settings for the cleaning mechanism to ensure a balance between pollution removal efficiency and energy consumption. Real-time wind speed data is collected, and the impact of wind pressure on the cleaning process is calculated using a wind speed-pressure conversion model. When strong winds (wind speed > 5 m / s) are detected, the cleaning pressure is automatically increased to offset the decrease in cleaning force caused by the wind pressure, maintaining a stable cleaning effect.

[0167] The brush speed compensation amount is derived based on the dynamic friction coefficient model;

[0168] In one specific implementation, brush rotation speed compensation is based on the following formula:

[0169]

[0170] in, The compensated brush rotation speed (unit: r / min) is used for single-row cleaning behavior control; Base rotational speed (unit: r / min), set according to component material: Glass component: =300r / min; Thin film module: =200r / min; The friction coefficient compensation factor is determined by the type of contaminant: High humidity areas: =1.2; Oily area: =1.8; The static friction coefficient of the clean surface can be taken as 0.8; The dynamic friction coefficient; The wind pressure compensation index (unit: s / m) reflects the effect of wind speed on rotational speed: High wind speed areas along the coast: =0.03s / m; Inland low wind speed areas: =0.01s / m; Real-time wind speed (unit: m / s), collected by a wind speed sensor; This is a correction factor for pollution aggregation, and is related to the Hurst index. Related: >0.7: =0.5; 0.4≤ ≤0.7: =0.3; <0.4: =0.1.

[0171] Among them, the items related to wind speed and items related to pollution concentration The impact on cleaning effectiveness is multiplicative. For example, strong winds weaken the brush pressure, requiring a proportional increase in rotation speed to maintain cleaning power; in highly polluted areas (Hur>0.7), the brush coverage density needs to be doubled (achieved by increasing rotation speed).

[0172] In actual implementation, the rotation speed needs to be set. To avoid motor overload. Among other things, Determined by motor performance and component load-bearing capacity, for example, .

[0173] Final output speed .

[0174] The system outputs cleaning commands that are compensated for wind pressure and corrected for rotation speed. The brush rotation speed is adjusted based on the type of contamination and surface humidity. In high-humidity areas, the rotation speed is reduced to avoid splashing, while in oily areas, the rotation speed is increased to enhance friction. Intelligent rotation speed compensation is achieved by inverting the humidity distribution of contaminants using infrared thermal radiation data and combining it with visible light images to identify oily areas.

[0175] When the edge-moving behavior is triggered, the magnetic field strength data is called in real time to correct the lateral offset of the cleaning robot relative to the photovoltaic frame.

[0176] During movement along the edge, a magnetic field sensor monitors the relative position of the robot to the pre-embedded magnetic track in real time. When lateral deviation is detected, a PID controller adjusts the speed difference of the drive wheels to achieve millimeter-level path correction, ensuring uniform cleaning coverage.

[0177] Furthermore, after the behavior of crossing the gap is completed, the actual motion energy consumption data is collected and compared with the energy consumption assessment value of the contaminated core connection layer;

[0178] The mapping relationship between the magnitude of the three-dimensional spatial vector and the energy consumption assessment value is dynamically adjusted based on the comparison results.

[0179] Specifically, by comparing the actual energy consumption of gap-crossing behavior with the pre-estimated value, when the deviation exceeds a certain threshold (e.g., ±5%), the mapping relationship between the three-dimensional spatial vector magnitude and energy consumption is corrected. For example, if the actual energy consumption is higher than the estimated value, the energy consumption weight of subsequent similar gap crossings is increased, making the path planning closer to actual working conditions.

[0180] After a single-row cleaning operation is completed, the rate of change of infrared thermal radiation data is extracted to determine the cleaning effect.

[0181] The cleaning effect is assessed by evaluating the rate of change in infrared thermal radiation data. Areas where the thermal radiation characteristics change due to contaminant removal, and the rate of change is below a certain threshold (e.g., 10%), are marked as substandard. The system automatically records the spatial coordinates of these areas, providing a basis for subsequent focused cleaning.

[0182] Reconstructing a pollution distribution heat map based on the rate of change of infrared thermal radiation data;

[0183] The trigger cleaning path determination module dynamically updates the cleaning path for unexecuted areas based on the reconstructed contamination distribution heatmap.

[0184] Based on the cleaning effectiveness assessment results, a new heat map of pollution distribution was constructed. Kriging interpolation was used to predict the pollution level in uncleaned areas, updating the priority cleaning core area. Using D... The Lite algorithm incrementally optimizes unexecuted paths, prioritizing the coverage of highly polluted residual areas.

[0185] Furthermore, the intelligent robot autonomous cleaning system for photovoltaic modules also includes a communication interruption emergency module; wherein, the communication interruption emergency module is used for:

[0186] When a communication interruption with the control center is detected, the pre-existing local atomic behavior chain of the cleaning robot is activated as the execution benchmark.

[0187] In the event of a communication interruption, the robot automatically switches to a locally pre-stored atomic behavior chain execution mode. The pre-interruption operational state is restored using timestamps, ensuring the continuity of the cleaning task. The local behavior chain includes preset safety boundaries and contingency strategies to prevent the risk of loss of control.

[0188] Based on a three-dimensional spatial reference point set, the positioning closed-loop control is achieved by integrating the inertial measurement unit built into the cleaning robot.

[0189] The system utilizes a set of three-dimensional spatial reference points and a built-in IMU to construct a relative positioning system. Positioning errors are corrected in real time using laser ranging data, and the RANSAC algorithm is employed to identify abnormal data points, ensuring positioning accuracy within ±5cm. When the accumulated positioning error exceeds a threshold, a repositioning process is automatically triggered.

[0190] Real-time acquisition of laser ranging data; dynamic correction of the boundary coordinates of the fall protection zone.

[0191] In practice, during communication interruptions, the laser ranging data is updated every 100ms to dynamically adjust the fall protection zone boundary. For newly discovered areas of sudden height changes (such as component edges lifting up), the safety buffer zone is automatically expanded to ensure that the robot always operates within the safe area.

[0192] If, during a single-row cleaning operation, the local contamination intensity calculated from real-time infrared thermal radiation data exceeds the equivalent threshold stored locally, then the local cleaning will be performed by switching to the contamination line diffusion path of the gradient overlay layer.

[0193] In some embodiments, when the system detects a communication interruption, it needs to determine whether the local pollution intensity exceeds an equivalent threshold using infrared thermal radiation data to decide whether to switch cleaning paths. For example, in a photovoltaic power station in the Northwest desert, sand and dust pollution is mainly composed of high-reflectivity particles. After cleaning, the surface of the photovoltaic modules will show a significant increase in infrared thermal radiation grayscale value due to the reduction of pollutants. The preset equivalent threshold for "infrared grayscale change rate" is 15%: if the grayscale change rate calculated by infrared data is less than 15% (e.g., only 10%) after a single row of cleaning in a certain area, it indicates severe pollution residue, triggering the iso-pollution diffusion path of the gradient cover layer and increasing the cleaning coverage density in that area. In a photovoltaic power station in the Southeast coastal region, salt spray pollution is accompanied by organic matter adhesion, and the pollutants have a high heat capacity, resulting in a more significant change in the surface temperature of the modules after cleaning. The equivalent threshold for "infrared temperature change rate" is set to 8℃: if the local temperature change after cleaning is less than 8℃ (e.g., only 5℃), it is determined that the cleaning is substandard, and the brush speed is automatically adjusted and the cleaning pressure is increased to perform a second cleaning for that area.

[0194] After crossing a gap, the actual energy consumption is compared with the pre-assessed value. When the deviation exceeds the equivalent threshold, the energy consumption mapping relationship is corrected. Taking a photovoltaic array arranged regularly in a plain area as an example, the gaps between the modules are uniform (about 20cm), and the preset energy consumption deviation threshold is ±7%. If the actual energy consumption of a crossing is 9% higher than the assessed value, it indicates that the mapping coefficient between the three-dimensional spatial vector magnitude and energy consumption needs to be adjusted. The system increases the energy consumption assessment coefficient for this type of gap from 0.8 to 0.9, making the path planning for subsequent gaps of the same type closer to the actual energy consumption. In mountainous photovoltaic power stations, the gaps between modules vary greatly due to terrain undulations. A dynamic threshold is used: the threshold for small gaps (<15cm) is set at ±5%, and the threshold for large gaps (>30cm) is relaxed to ±12%. For example, when crossing a 40cm gap, the actual energy consumption deviates from the assessed value by 11%, which is within a reasonable range and does not trigger correction. If the deviation reaches 15%, the function relationship between the vector magnitude and energy consumption is refitted using the least squares method to ensure the accuracy of energy consumption assessment under complex terrain.

[0195] When monitoring gap edge deformation using visible light images, an equivalent threshold for deformation error needs to be set based on the component installation method. For photovoltaic modules mounted on rigid supports (such as those fixed to concrete foundations), the module position stability is high, and the "absolute deformation" threshold is set to 2mm: when a gap edge displacement exceeding 2mm is detected (such as slight edge warping due to thermal expansion and contraction), the fuzzy PID controller automatically increases the damping coefficient of the robotic arm joint to suppress vibration and avoid collisions. For flexible supports or floating photovoltaic systems (such as water-based photovoltaics), the components allow for a certain degree of elastic deformation, and the "relative deformation" threshold is set to 0.5% (relative to the gap width): if the edge deformation of a 20cm wide gap reaches 1mm (0.5%), a compensation mechanism is triggered, adjusting the robotic arm's pre-calibrated attitude angle (such as increasing the pitch angle by 5°) to offset the impact of deformation on the crossing motion, ensuring the cleaning robot passes smoothly.

[0196] These equivalent thresholds are not fixed, but are dynamically optimized through on-site debugging, historical data training, or real-time sensor feedback, so that the system can accurately judge the working conditions and execute corresponding strategies in different environments, ensuring the safety and efficiency of cleaning operations.

[0197] In practice, during single-row cleaning, if the local contamination intensity exceeds the equivalent threshold, the system automatically switches to an isocontamination diffusion path. By calculating the contamination gradient direction, a spiral or serpentine local path is generated to improve the cleaning coverage of highly contaminated areas.

[0198] After communication is restored, the execution status of the cleaning path recorded locally will be synchronized with the cleaning path stored in the control center.

[0199] In practice, after communication is restored, an incremental synchronization protocol is used. The robot uploads information such as cleaning progress, energy consumption data, and updated pollution heatmaps, and the control center compares this data with local data to generate a patch to address any discrepancies. CRC checksums ensure the accuracy of data transmission, achieving seamless integration of cleaning tasks.

[0200] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A photovoltaic module intelligent robot autonomous cleaning system, characterized in that, include: An environmental data acquisition module is used to collect environmental data of the photovoltaic modules in the photovoltaic module array in real time through a sensor array. The environmental data includes light intensity data and laser ranging data, as well as infrared thermal radiation data, magnetic field strength data and visible light image data. The border positioning module is used to perform border positioning operations based on the light intensity data, including: when the light intensity is greater than the strong light threshold, identifying the position of the metal border based on the temperature gradient change of the infrared thermal radiation data; when the light intensity is less than the weak light threshold, locating the coordinates of the pre-embedded magnetic track based on the spatial peak of the magnetic field strength data; when the light intensity is between the strong light threshold and the weak light threshold, identifying the border outline through the edge color difference of the visible light image data. The reference point conversion module is used to fuse the surface height difference and tilt angle information in the laser ranging data and convert the border positioning result into a three-dimensional spatial reference point. The anti-fall zone generation module is used to generate an anti-fall zone by offsetting a dynamic safety distance inward along the plane normal direction of the photovoltaic module, with the three-dimensional spatial reference point as the origin. This includes: analyzing the curvature characteristics of the three-dimensional spatial reference point set to identify geometric abrupt change regions at the edge of the photovoltaic module; retrieving the pollutant humidity distribution based on the infrared thermal radiation data and extracting oil stain attachment areas using the visible light image data; establishing a dynamic friction coefficient model based on pollutant type, including: the friction coefficient in high humidity areas decays exponentially; the friction coefficient in oil stain areas decreases in a stepwise manner; using the geometric abrupt change region as a risk anchor point, calculating the critical slip threshold according to the dynamic friction coefficient model; superimposing real-time wind speed sensor data to generate a composite safety distance considering wind load; and performing a curvature-adaptive nonlinear offset along the plane normal direction to form an anti-fall zone at the edge of the photovoltaic module. The cleaning path determination module is used to generate a cleaning path covering the photovoltaic module assembly by combining the anti-drop zone and the pollution distribution heat map. This includes: performing multifractal dimension decomposition on the pollution distribution heat map to extract the Hurst index for each region; marking regions with Hurst indices greater than the pollution aggregation threshold as priority cleaning core areas; constructing a two-layer topology path including a pollution core connection layer and a gradient cover layer, using the anti-drop zone as a rigid motion boundary; the pollution core connection layer uses an ant colony algorithm to generate the optimal energy-consuming path connecting the priority cleaning core areas, wherein the energy consumption assessment value of the photovoltaic panel gap crossing action is dynamically corrected by the modulus of the three-dimensional spatial vector; the gradient cover layer generates an iso-pollution diffusion path based on the pollution gradient descent direction, with the path point density positively correlated with the pollution gradient value; and integrating an artificial potential field constraint mechanism during the generation of the gradient cover layer, including: the pollution gradient field providing an gravitational potential field, the anti-drop zone boundary generating a repulsive potential field, and the module surface normal vector field constraining the degrees of freedom of movement. The cleaning path is decomposed into atomic behavior chains, which include edge movement, gap crossing, and single-row cleaning. A cleaning drive module is used to drive the cleaning robot to perform cleaning operations according to the atomic behavior chain.

2. The photovoltaic module intelligent robotic autonomous cleaning system of claim 1, wherein, The reference point conversion module integrates the surface height difference and tilt angle information from the laser ranging data to convert the bounding box positioning result into a three-dimensional spatial reference point, including: When the light intensity is greater than the strong light threshold, the position of the metal frame identified by the infrared thermal radiation data is used as the planar constraint boundary. When the light intensity is less than the weak light threshold, the coordinates of the pre-embedded magnetic track located by the magnetic field intensity data are used as the planar constraint boundary. When the light intensity is between the strong light threshold and the weak light threshold, the bounding contour identified by the visible light image data is used as the planar constraint boundary. By integrating the surface height difference from the laser ranging data, the three-dimensional spatial vector of the splicing gap between adjacent photovoltaic modules is calculated; The surface normal vector field of the component is constructed by combining the tilt angle information, and the planar distortion caused by installation deformation is compensated by non-uniform B-spline interpolation; Using the planar constraint boundary as a reference, a set of three-dimensional spatial reference points with curvature features is established by extending along the three-dimensional spatial vector.

3. The photovoltaic module intelligent robotic autonomous cleaning system of claim 1, wherein, The cleaning path determination module decomposes the cleaning path into a chain of atomic behaviors, including: The two-layer topology path is compiled into a chain of atomic behaviors; The movement speed along the edge is adaptively adjusted based on the curvature characteristics of the three-dimensional spatial reference point; The pre-calibrated posture of the cleaning robot that crosses the gap is calculated based on the three-dimensional spatial vector of the splicing gap; The operating parameters of the cleaning mechanism for single-line cleaning are configured based on the pollution concentration level of the priority cleaning core area.

4. The photovoltaic module intelligent robotic autonomous cleaning system of claim 3, wherein, The cleaning drive module drives the cleaning robot to perform cleaning operations according to the atomic behavior chain, including: When performing actions that cross gaps, the inertial load abrupt change range of the robotic arm joints of the cleaning robot is predicted based on the three-dimensional spatial vector of the splicing gap. The deformation error at the gap edge is monitored in real time using the visible light image data; Based on the coupling relationship between the abrupt change range of the inertial load and the deformation error, a fuzzy PID controller is used to dynamically allocate the torque of the joint motor to suppress vibration.

5. The intelligent robot autonomous cleaning system for photovoltaic modules according to claim 4, characterized in that, When performing a single-line cleaning action, the base pressure setting value is determined based on the degree of contamination accumulation in the priority cleaning core area; The wind pressure disturbance compensation coefficient is generated by integrating real-time wind speed sensor data; The brush speed compensation amount is derived based on the dynamic friction coefficient model. Output cleaning execution commands with wind pressure compensation and speed correction; When the edge-moving behavior is triggered, the magnetic field strength data is called in real time to correct the lateral offset of the cleaning robot relative to the photovoltaic frame.

6. The intelligent robot autonomous cleaning system for photovoltaic modules according to claim 4, characterized in that, After the gap-crossing behavior is completed, the actual motion energy consumption data is collected and compared with the energy consumption assessment value of the pollution core connection layer. The mapping relationship between the magnitude of the three-dimensional spatial vector and the energy consumption assessment value is dynamically adjusted based on the comparison results. After a single-line cleaning operation is completed, the rate of change of the infrared thermal radiation data is extracted to determine the cleaning effect; The pollution distribution heat map is reconstructed based on the rate of change of the infrared thermal radiation data. The cleaning path determination module is triggered to dynamically update the cleaning path for the unexecuted areas based on the reconstructed contamination distribution heatmap.

7. The photovoltaic module intelligent robotic autonomous cleaning system of claim 3, wherein, The photovoltaic module intelligent robot autonomous cleaning system also includes a communication interruption emergency module; wherein, the communication interruption emergency module is used for: When a communication interruption with the control center is detected, the pre-existing local atomic behavior chain of the cleaning robot is activated as the execution benchmark. Based on the set of three-dimensional spatial reference points, the positioning closed-loop control is achieved by integrating the inertial measurement unit built into the cleaning robot. The laser ranging data is acquired in real time, and the boundary coordinates of the fall protection zone are dynamically corrected. If, during a single-row cleaning operation, the local contamination intensity calculated from real-time infrared thermal radiation data exceeds the locally stored equivalent threshold, then the local cleaning operation is switched to the contamination line diffusion path of the gradient cover layer. After communication is restored, the execution status of the cleaning path recorded locally will be synchronized with the cleaning path stored in the control center.

Images