Photovoltaic module self-cleaning control method, device, equipment, medium and program product

Abstract

The invention relates to the technical field of photovoltaic modules, and discloses a photovoltaic module self-cleaning control method, device and equipment, a medium and a program product, the method is applied to a decision-making end, and the method comprises the following steps: acquiring image data, actually measured output power data, actually measured irradiance data, battery piece temperature data and rated maximum power of a photovoltaic array; performing stain identification on the image data, and determining a stain heat value of at least one target photovoltaic module in the photovoltaic array; based on the stain heat value, the actually measured output power data, the actually measured irradiance data, the battery piece temperature data and the rated maximum power of the target photovoltaic module, recoverable energy loss is determined; determining the cleaning cost based on the stain heat value and the area information of the target photovoltaic module; and when the recoverable energy loss is larger than the cleaning cost, the cleaning robot is controlled to clean the target photovoltaic module, and unnecessary fresh water consumption and energy waste are reduced to the maximum extent.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic module technology, and specifically to self-cleaning control methods, devices, equipment, media, and program products for photovoltaic modules. Background Technology

[0002] In recent years, the "desertification and wasteland" new energy development strategy has been accelerated in arid and desert regions. Large-scale photovoltaic (PV) bases with an installed capacity exceeding ten million kilowatts have been built in specific desert areas, and are continuously expanding based on the 750 kV regional backbone grid. While the region enjoys abundant sunshine throughout the year, the average number of sandstorm days in spring reaches 32.7 days, significantly higher than the historical average. The long-term accumulation of fine sand, bird droppings, salt deposits, and other contaminants can reduce the light transmittance of PV modules by 15%-22%, resulting in annual power generation losses of 5%-18%, and in extreme cases, approaching 25%. Simultaneously, water resources are extremely scarce in the desert fringe areas. Traditional manual or timed high-pressure water jet cleaning consumes large amounts of fresh water and, due to its rigid schedule, fails to balance economic viability with power generation revenue. Therefore, there is an urgent need for a self-cleaning control method for PV modules suitable for arid and sandy environments to ensure the efficient, safe, and low-cost operation of such large-scale PV bases. Summary of the Invention

[0003] This invention provides a method, device, equipment, medium, and program product for controlling the self-cleaning of photovoltaic modules, in order to solve the problems in related technologies where manual or timed high-pressure water gun cleaning consumes a large amount of fresh water and cannot balance economic efficiency and power generation revenue due to rigid cycles.

[0004] In a first aspect, the present invention provides a self-cleaning control method for photovoltaic modules, applied at the decision-making end. The method includes: acquiring image data, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of a photovoltaic array; identifying stains in the image data to determine the stain heat value of at least one target photovoltaic module in the photovoltaic array, where the stain heat value characterizes the degree of influence of surface stains on power generation, and the target photovoltaic module is the photovoltaic module in the photovoltaic array containing stains; determining recoverable energy loss based on the stain heat value, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of the target photovoltaic module; determining the cleaning cost based on the stain heat value and area information of the target photovoltaic module; and controlling a cleaning robot to clean the target photovoltaic module when the recoverable energy loss exceeds the cleaning cost.

[0005] The self-cleaning control method for photovoltaic modules provided by this invention acquires key data such as images of the photovoltaic array, measured power, irradiance, and cell temperature. It accurately identifies target photovoltaic modules with stains by using stain recognition and quantifies their stain heat value, clearly understanding the impact of stains on power generation. This avoids the blind approach of traditional timed cleaning, which cleans regardless of the amount or cost of stains. Furthermore, it calculates recoverable energy loss by combining multi-dimensional operational data and rated parameters, and accurately calculates cleaning costs based on stain heat value and module area. The machine is only triggered when the recoverable energy loss exceeds the cleaning cost. Human cleaning ensures that each cleaning operation effectively recovers power generation losses and increases power plant revenue, while minimizing unnecessary freshwater consumption and energy waste, perfectly adapting to the water scarcity situation in arid and sandy regions. In addition, the automatic cleaning tasks performed by robots replace traditional manual climbing operations, reducing operation and maintenance safety risks. Moreover, the on-demand cleaning mode breaks the limitations of the rigid traditional cleaning cycle, reducing cleaning frequency and downtime while ensuring the efficient operation of large-scale photovoltaic bases. Ultimately, it meets the core requirements of efficient, safe, and low-cost operation of photovoltaic power plants in arid and sandy environments.

[0006] In one optional implementation, the step of determining recoverable energy loss based on the stain heat value of the target photovoltaic module, measured output power data, measured irradiance data, cell temperature data, and rated maximum power includes: determining a stain recoverability coefficient based on the stain heat value of the target photovoltaic module; determining the total energy loss of the photovoltaic array based on the measured output power data, measured irradiance data, cell temperature data, and rated maximum power; and determining recoverable energy loss based on the total energy loss and the stain recoverability coefficient.

[0007] In one optional implementation, the total energy loss of the photovoltaic array is determined based on measured output power data, measured irradiance data, cell temperature data, and rated maximum power. This includes: determining effective irradiance data based on measured irradiance data, a preset incident angle correction coefficient, and a spectral correction coefficient, wherein the incident angle correction coefficient and the spectral correction coefficient are fixed constants when no external sensing conditions are present; when a component attitude sensor and / or a meteorological acquisition device are configured, the correction coefficients are updated over time according to preset lookup table or interpolation rules based on real-time attitude parameters and meteorological parameters; determining instantaneous baseline power data based on effective irradiance data, cell temperature data, and rated maximum power; smoothing the instantaneous baseline power data based on a preset smoothing model to obtain target instantaneous baseline power data; determining instantaneous power loss data based on measured output power data and target instantaneous baseline power data, wherein the instantaneous power loss data is used to characterize the instantaneous power loss value at different times within the target time period; and determining the total energy loss based on the instantaneous power loss data.

[0008] In one optional implementation, the step of controlling a cleaning robot to clean a target photovoltaic module includes: constructing a directed topology graph with the photovoltaic module as a node and the guide rail connection relationship as an edge; planning a cleaning path for the cleaning robot based on the position of the cleaning robot, the target photovoltaic module, and the directed topology graph; and controlling the cleaning robot to clean the target photovoltaic module based on the cleaning path.

[0009] Secondly, the present invention also provides a self-cleaning control method for photovoltaic modules, applied to the control module of a cleaning robot. The cleaning robot is pre-configured with an image acquisition device. The method includes: when receiving a cleaning instruction sent by a decision-making end, controlling the cleaning robot to move to a preset position of the target photovoltaic module according to a cleaning path; controlling the image acquisition device to acquire target image data of the surface of the target photovoltaic module; inputting the target image data into a pre-constructed stain segmentation network so that the stain segmentation network outputs stain information of different areas on the surface of the target photovoltaic module; determining, based on the stain information of different areas, that there is at least one target area on the target photovoltaic module that needs to be cleaned; and controlling the cleaning robot to clean the target area.

[0010] The self-cleaning control method for photovoltaic modules provided by this invention allows the robot to accurately reach the target module location after receiving a cleaning command from the decision-making end, ensuring accurate positioning and avoiding unnecessary movement losses. Subsequently, the robot's built-in image acquisition device acquires images of the target module surface a second time, combining this with a pre-constructed stain segmentation network to accurately output stain information for different areas, achieving refined identification and positioning of stained areas and precisely locking onto the target area requiring cleaning. Based on this refined stain information, the robot only cleans the stained areas, eliminating the need for repeated work on clean areas. This significantly reduces freshwater and energy consumption during cleaning, perfectly adapting to the extremely scarce water resources in arid regions, while also improving cleaning efficiency and avoiding resource waste. Simultaneously, the real-time acquired target image data and stain segmentation results dynamically support precise adjustments to the cleaning actions, ensuring cleaning effectiveness and further guaranteeing the effective recovery of power generation efficiency after module cleaning.

[0011] Thirdly, the present invention provides a self-cleaning control device for photovoltaic modules, used to execute the method of the first aspect. The device includes: a first acquisition module for acquiring image data, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of a photovoltaic array; an identification module for identifying stains in the image data and determining the stain heat value of at least one target photovoltaic module in the photovoltaic array, wherein the stain heat value characterizes the degree of influence of stains on the surface of the photovoltaic module on power generation, and the target photovoltaic module is the photovoltaic module in the photovoltaic array that has stains; a first determination module for determining recoverable energy loss based on the stain heat value, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of the target photovoltaic module; a second determination module for determining cleaning cost based on the stain heat value and area information of the target photovoltaic module; and a first control module for controlling a cleaning robot to clean the target photovoltaic module when the recoverable energy loss is greater than the cleaning cost.

[0012] Fourthly, the present invention provides a photovoltaic module self-cleaning control device for executing the method of the second aspect. The device includes: a second control module for controlling a cleaning robot to move to a preset position of the target photovoltaic module according to a cleaning path when a cleaning instruction is received from a decision-making end; a third control module for controlling an image acquisition device to acquire target image data of the surface of the target photovoltaic module; a third determination module for inputting the target image data into a pre-constructed stain segmentation network so that the stain segmentation network outputs stain information of different areas on the surface of the target photovoltaic module; a fourth determination module for determining, based on the stain information of different areas, that there is at least one target area on the target photovoltaic module that needs to be cleaned; and a fourth control module for controlling the cleaning robot to clean the target area.

[0013] Fifthly, the present invention provides an electronic device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the photovoltaic module self-cleaning control method of the first aspect or any corresponding embodiment described above, or to perform the photovoltaic module self-cleaning control method of the second aspect described above.

[0014] In a sixth aspect, the present invention provides a computer-readable storage medium storing computer instructions, which are used to cause a computer to execute the photovoltaic module self-cleaning control method of the first aspect or any corresponding embodiment thereof, or to execute the photovoltaic module self-cleaning control method of the second aspect.

[0015] In a seventh aspect, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the photovoltaic module self-cleaning control method of the first aspect or any corresponding embodiment thereof, or to execute the photovoltaic module self-cleaning control method of the second aspect. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention;

[0018] Figure 2 This is a schematic diagram of the first process of a photovoltaic module self-cleaning control method according to an embodiment of the present invention; Figure 3 This is a schematic flowchart of another photovoltaic module self-cleaning control method according to an embodiment of the present invention; Figure 4 This is a structural block diagram of a first type of photovoltaic module self-cleaning control device according to an embodiment of the present invention; Figure 5 This is a structural block diagram of a second type of photovoltaic module self-cleaning control device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0021] 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 technical features indicated. 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.

[0022] In recent years, the "Sand and Desert" new energy development strategy has been accelerated in arid and desert regions. While these regions have abundant sunshine throughout the year, the average number of sandstorm days in spring reaches 32.7, significantly higher than the historical average. The long-term accumulation of fine sand, bird droppings, salt deposits, and other contaminants can reduce the light transmittance of photovoltaic modules by 15%-22%, resulting in annual power generation losses of 5%-18%, and in extreme cases, approaching 25%. Simultaneously, water resources are extremely scarce in the desert fringe areas. Current technologies typically involve manual or timed high-pressure water jet cleaning of photovoltaic modules, which consumes large amounts of fresh water and, due to its rigid schedule, fails to balance economic viability with power generation revenue.

[0023] In view of this, the embodiments of this application provide a self-cleaning control method for photovoltaic modules, which can be applied to the decision-making end to realize the self-cleaning control method for photovoltaic modules. The method provided in this application acquires key data such as images of the photovoltaic array, measured power, irradiance, and cell temperature. It then uses stain recognition to accurately locate target photovoltaic modules with stains and quantifies their stain heat values, clearly understanding the impact of stains on power generation. This avoids the blind approach of traditional timed cleaning, which cleans regardless of the amount or cost of the stains. Based on this, it calculates recoverable energy loss by combining multi-dimensional operational data and rated parameters. Simultaneously, it accurately calculates cleaning costs based on stain heat values ​​and module area, triggering robot cleaning only when the recoverable energy loss exceeds the cleaning cost. This ensures that each cleaning effectively recovers power generation losses and improves power plant revenue, while minimizing unnecessary freshwater consumption and energy waste, perfectly adapting to the water scarcity situation in arid and sandy regions. Furthermore, the automatic execution of cleaning tasks by robots replaces traditional manual climbing operations, reducing operational safety risks. The on-demand cleaning mode breaks the limitations of rigid traditional cleaning cycles, reducing cleaning frequency and downtime while ensuring the efficient operation of large-scale photovoltaic bases. Ultimately, it fulfills the core requirements of efficient, safe, and low-cost operation of photovoltaic power plants in arid and sandy environments.

[0024] According to an embodiment of the present invention, a self-cleaning control method for photovoltaic modules is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0025] This embodiment provides a self-cleaning control method for photovoltaic modules, applied to a decision-making end, where the decision-making process may include, but is not limited to, a server. Figure 1 This is a flowchart of a photovoltaic module self-cleaning control method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Acquire image data, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of the photovoltaic array.

[0026] For example, the image data of the photovoltaic array can be high-definition visual data of the surfaces of all components in the photovoltaic array. This high-definition visual data can accurately record the cleanliness of the component surfaces, clearly showing the distribution, shape, and coverage of stains such as dust, bird droppings, and salt frost. In this embodiment, the image data of the photovoltaic array is acquired by vision devices such as binocular industrial cameras deployed above the photovoltaic components. The measured output power data refers to the actual power output of the photovoltaic components (or the component cascade) measured in real time by an IV monitoring board deployed in the component string, typically measured in watts (W). It reflects the true power generation efficiency of the components under current stain coverage and environmental conditions, reflecting the current actual power generation level of the components. The measured irradiance data refers to the actual solar radiation intensity received by the surface of the photovoltaic components, collected in real time by a miniature weather probe, typically measured in watts per square meter (W / m²). 2 Irradiance is a core environmental factor affecting the power generation efficiency of photovoltaic (PV) modules, and its magnitude directly determines the module's power generation potential. Cell temperature data refers to the operating temperature of the core power generation component (cell) inside the PV module, measured in real time by sensors (such as those integrated into weather probes or the module itself), typically in degrees Celsius. Cell temperature affects the module's power generation performance; as temperature increases, power generation usually decreases, resulting in a negative power temperature coefficient. Rated maximum power refers to the maximum output power of the PV module as specified by the manufacturer under standard test conditions (STC), measured in watts (W). It is an inherent power generation capability parameter designed into the module itself and serves as the core benchmark for calculating the theoretical baseline power under ideal clean conditions, used to measure the difference between the module's current actual power generation capacity and its ideal state.

[0027] Step S102: Stain identification is performed on the image data to determine the stain heat value of at least one target photovoltaic module in the photovoltaic array. The stain heat value characterizes the degree of influence of stains on the surface of the photovoltaic module on power generation. The target photovoltaic module is the photovoltaic module in the photovoltaic array that has stains.

[0028] For example, in this embodiment, image data is input into a pre-constructed stain recognition network to identify stains on the surface of photovoltaic modules. This determines the shading area ratio and overall stain level of each photovoltaic module in the photovoltaic array. Based on the shading area ratio and overall stain level of each module, the stain heat value of the corresponding photovoltaic module is determined. Photovoltaic modules with stain heat values ​​greater than a preset threshold are selected as target photovoltaic modules. The specific content of the preset threshold can be determined according to requirements, and this embodiment does not impose specific limitations. The stain recognition network can be based on the SoilingNet-T stain recognition network. SoilingNet-T is a lightweight visual Transformer deployed at the edge, consisting of a convolutional feature extraction front-end, a hierarchical self-attention encoder, and a multi-class decoder. Through joint training on public datasets and power plant field samples, combined with knowledge distillation and pruning compression, the number of parameters is controlled within the trillions. The semantic segmentation steps include: S1, performing distortion correction, brightness normalization, and component grid line / cascade position calibration on the input image; S2, dividing the image into fixed-size patches and extracting multi-scale semantic features through a Transformer encoder; S3, outputting a category probability map of multiple types of stains such as "clean," "dust," "bird droppings," and "salt frost" for each pixel through a decoding head; S4, spatially aggregating the pixel results according to the actual arrangement of the photovoltaic array to form a two-dimensional matrix SoilingMap that corresponds one-to-one with the component grid lines or cascades.

[0029] Step S103: Based on the stain heat value of the target photovoltaic module, the measured output power data, the measured irradiance data, the cell temperature data, and the rated maximum power, determine the recoverable energy loss.

[0030] For example, in this embodiment of the application, the theoretical baseline power of the module under clean conditions is first calculated by combining the rated maximum power, the measured irradiance and the cell temperature; then the baseline power is compared with the measured output power to obtain the total energy loss; finally, the recoverable energy loss that can be recovered after cleaning is calculated by matching the corresponding recoverable coefficient according to the stain heat value.

[0031] Step S104: Determine the cleaning cost based on the stain heat value and area information of the target photovoltaic module.

[0032] For example, first, based on the area information of the target photovoltaic module, calculate the basic resource consumption required for cleaning (such as water consumption and operation time); then, determine the cleaning intensity according to the heat value of the stain (such as "spraying + brushing" for high heat value and dry brushing for low heat value), and match the corresponding water consumption and electricity consumption standards; finally, combine water price, electricity price, labor and equipment depreciation to calculate and summarize the various costs to obtain the total cleaning cost.

[0033] Step S105: When the recoverable energy loss is greater than the cleaning cost, control the cleaning robot to clean the target photovoltaic module.

[0034] For example, in this embodiment of the application, when the recoverable energy loss is greater than the cleaning cost, the location of the target photovoltaic module can be combined with the scheduling algorithm to plan a conflict-free cleaning path and sequence for the cleaning robot, and a targeted cleaning instruction can be issued to the cleaning robot. The robot can accurately reach the target module to complete the cleaning according to the instruction, and at the same time, the operation status can be fed back in real time.

[0035] The photovoltaic module self-cleaning control method provided in this embodiment acquires key data such as images of the photovoltaic array, measured power, irradiance, and cell temperature. It accurately identifies target photovoltaic modules with stains by using stain recognition and quantifies their stain heat value, clearly understanding the impact of stains on power generation. This avoids the blindness of traditional timed cleaning methods that indiscriminately clean regardless of the amount or cost of stains. Furthermore, it calculates recoverable energy loss by combining multi-dimensional operational data and rated parameters, and accurately calculates cleaning costs based on stain heat value and module area. The system only triggers a cleaning mechanism when the recoverable energy loss exceeds the cleaning cost. Robotic cleaning ensures that each cleaning operation effectively recovers power generation losses and increases power plant revenue, while minimizing unnecessary freshwater consumption and energy waste, perfectly adapting to the water scarcity situation in arid and sandy regions. In addition, by automatically performing cleaning tasks, robots replace traditional manual climbing operations, reducing operational safety risks. Furthermore, the on-demand cleaning model breaks the limitations of rigid traditional cleaning cycles, reducing cleaning frequency and downtime while ensuring the efficient operation of large-scale photovoltaic bases. Ultimately, it fulfills the core requirements of efficient, safe, and low-cost operation of photovoltaic power plants in arid and sandy environments.

[0036] This embodiment provides a self-cleaning control method for photovoltaic modules, applied to the decision-making end. Figure 2 This is a flowchart of a photovoltaic module self-cleaning control method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Acquire image data, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of the photovoltaic array. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0037] Step S202: Stain identification is performed on the image data to determine the stain heat value of at least one target photovoltaic module in the photovoltaic array. The stain heat value characterizes the degree of influence of stains on the surface of the photovoltaic module on power generation. The target photovoltaic module is the photovoltaic module in the photovoltaic array that has stains.

[0038] Step S203: Based on the stain heat value of the target photovoltaic module, the measured output power data, the measured irradiance data, the cell temperature data, and the rated maximum power, determine the recoverable energy loss.

[0039] Specifically, step S203 includes: Step S2031: Determine the stain recoverability coefficient based on the stain heat value of the target photovoltaic module.

[0040] For example, in this embodiment of the application, a mapping rule between the stain heat value and the actual cleaning recovery rate is first established using historical cleaning data of the power plant (the higher the heat value, the easier the stain is to clean, and the closer the recoverability coefficient is to 1); then, the stain heat value of the target component is substituted into this rule to match and obtain the corresponding stain recoverability coefficient. Specifically, the corresponding optical transmittance reduction coefficient can be calculated based on the stain heat value of the target photovoltaic module, and the stain recoverability coefficient k (with a value range of 0 to 1) is calibrated based on historical operating data to characterize the proportion of recoverable loss to total loss after cleaning operations.

[0041] Step S2032: Determine the total energy loss of the photovoltaic array based on the measured output power data, measured irradiance data, cell temperature data, and rated maximum power.

[0042] In some optional implementations, step S2032 above includes: Step a1: Determine the effective irradiance data based on the measured irradiance data, the preset incident angle correction coefficient, and the spectral correction coefficient. The incident angle correction coefficient and the spectral correction coefficient are fixed constants when there are no external sensing conditions. When a component attitude sensor and / or meteorological acquisition device are configured, the correction coefficient is updated over time according to the real-time attitude parameters and meteorological parameters according to the preset lookup table or interpolation rules.

[0043] For example, the effective irradiance data refers to the effective irradiance at different times within the target time period. The specific content of the target time period can be determined according to actual needs, and this application embodiment does not impose specific limitations. In this application embodiment, the effective irradiance data can be calculated using the following formula:

[0044] in, Represents the effective irradiance at time t. Represents the measured irradiance at time t; This indicates the preset incident angle correction factor. Indicates the spectral correction factor; This refers to the reference irradiance value of photovoltaic modules under standard test conditions (STC). The incident angle correction factor and spectral correction factor mentioned above are dimensionless coefficients; the units for effective irradiance and measured irradiance are consistent, both being W / m².2 To facilitate project implementation, the default setting will be... and The constants are set as fixed constants obtained from the field calibration. If on-site observation of component attitude or meteorological conditions are available, the data can also be dynamically updated over time: the incident angle is calculated based on the sun's position and array attitude, and the incident angle correction coefficient is obtained from a table; the spectral correction coefficient is obtained based on meteorological quantities such as air quality and aerosols. Both implementations can achieve the expected technical effects: the basic implementation ensures a simple and reliable computational link, while the enhanced implementation improves the estimation accuracy under different time periods and weather conditions.

[0045] Step a2: Determine the instantaneous baseline power data based on the effective irradiance data, cell temperature data, and rated maximum power.

[0046] For example, the instantaneous baseline power data can be the instantaneous baseline power at different times within the target time period. In this embodiment, the instantaneous baseline power can be calculated using the following formula:

[0047] in, This indicates the rated maximum power of the component cascade under standard test conditions (STC). Represents the effective irradiance at time t. This is the power temperature coefficient (a negative value, taken from the component nameplate or calibration). This represents the temperature of the solar cell at time t.

[0048] Step a3: Smooth the instantaneous baseline power data based on a preset smoothing model to obtain the target instantaneous baseline power data.

[0049] For example, the target instantaneous baseline power data refers to the target instantaneous baseline power at different times within the target time period. In this embodiment, the target instantaneous baseline power data can be calculated using the following formula:

[0050] in, This represents the target baseline power at time t. Represents the smoothing coefficient. This represents the target baseline power at the previous moment.

[0051] Step a4: Determine the instantaneous power loss data based on the measured output power data and the target instantaneous baseline power data. The instantaneous power loss data is used to characterize the instantaneous power loss value at different times within the target time period.

[0052] For example, in this embodiment of the application, the instantaneous power loss value can be calculated by the following formula:

[0053] in, This represents the instantaneous power loss at time t. This represents the measured output power value at time t; the meanings of the other variables will not be elaborated further.

[0054] Step a5: Determine the total energy loss based on the instantaneous power loss data.

[0055] For example, the total energy loss can be calculated using the following formula:

[0056] in, Indicates the target time period. This represents the total energy loss during the target time period. This is a time integration variable used to represent any moment within the target time period; Indicates the start time.

[0057] Step S2033: Determine the recoverable energy loss based on the total energy loss and the stain recoverability coefficient.

[0058] For example, in this embodiment of the application, the recoverable energy loss can be determined by the following formula:

[0059] in, This indicates that energy loss can be recovered. This represents the stain recoverability coefficient; the meanings of the other variables will not be elaborated further.

[0060] Step S204: Determine the cleaning cost based on the stain heat value and area information of the target photovoltaic module. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0061] Step S205: When the recoverable energy loss is greater than the cleaning cost, control the cleaning robot to clean the target photovoltaic module.

[0062] Specifically, step S205 includes: Step S2051: Construct a directed topology graph by using photovoltaic modules as nodes and rail connections as edges.

[0063] For example, in this embodiment of the application, each grid in the SoilingMap records the shading area ratio and overall contamination level of the corresponding component (or sub-component), which can be regarded as the contamination heat value at that location. The SoilingMap is mapped to the station geometry, guide rail connectivity, and restricted area constraints into a directed topology graph (node ​​= work unit / transfer point, edge = passable passage). Node weight calculation: weight = recoverable power generation loss - work cost (updated in real time); edge cost = path length + slope + number of turns + congestion risk.

[0064] Step S2052: Based on the location of the cleaning robot, the target photovoltaic module, and the directed topology graph, plan the cleaning path of the cleaning robot.

[0065] For example, in this embodiment of the application, the cleaning path can be planned through the following steps: (1) GNN extracts node / edge embedding vectors (reflecting reward density, risk, and accessibility); (2) The pointer-based strategy network selects the target node by autoregression based on "robot position - battery level - remaining time window - unfinished nodes"; (3) MARL adopts a "centralized training, decentralized execution" framework. During the execution phase, the robot makes independent decisions through local observation and lightweight broadcast (position, target, remaining power). When the remaining power is predicted to be insufficient, it automatically inserts a charging and replenishment node backtracking route.

[0066] If multiple robots compete for the same resource, priority should be given to robots with higher revenue density, lower power consumption, and greater offline penalties, and the target should be staggered or replaced; if the path is blocked, rolling replanning should be triggered.

[0067] Specifically, during path planning, the SoilingMap, along with the site geometry, guide rail connectivity, and restricted area constraints, is mapped into a directed topology graph at the scheduling time. Nodes represent cleanable standard work units or cross-row transfer points, while edges represent passable channels such as guide rails or bridges. The weight of each node is derived from the "recoverable power generation loss" and the operational cost, and is updated in real time based on irradiance, wind field, and traffic conditions. Edge costs are given by a combination of path length, gradient, number of turns, congestion, and risk information. Based on this, the scheduling module uses a graph neural network to learn the representations of nodes and edges, obtaining embedding vectors that reflect revenue density, risk, and reachability. Subsequently, a pointer-based policy network autoregressively selects the next target node and travel edge according to the state of "current robot position - power - remaining time window - set of unfinished nodes," while simultaneously performing a look-ahead check on energy and time feasibility. When multiple robots collaborate, a multi-agent reinforcement learning framework of "centralized training and decentralized execution" is adopted: during the training phase, global state and joint reward are used to optimize queue efficiency and avoid conflicts and deadlocks; during the execution phase, each robot makes independent decisions based solely on local observations and lightweight broadcasts (including current position, target node, expected time period, and remaining battery power). To eliminate resource conflicts and the risk of passing each other, the system introduces a rapid online allocation and conflict resolution process after the strategy output: if two or more robots plan to occupy the same work unit at the same time or enter the same narrow side passage from opposite directions, they will give way, stagger time slots, or replace target nodes according to preset priorities (e.g., those with higher reward density, lower battery power, or greater offline penalty). If the path is temporarily blocked or a sudden increase in wind and sand increases the risk, the corresponding edge will be immediately set as unreachable or its cost will be increased, triggering a rolling replanning. The system refreshes node weights and edge costs at fixed intervals or through event triggers, and obtains new work sequences and travel paths within the short horizon. When it is predicted that the remaining power is insufficient to safely complete the next work segment, the strategy automatically inserts a turnaround route to the nearest reachable charging / replenishment node, ensuring uninterrupted operation and continuous satisfaction of safety distance and positioning error constraints. This process, without altering the robot's body or communication architecture, achieves online optimization and dynamic rearrangement of cleaning sequences and travel paths, guaranteeing "priority cleaning of high-weight areas, energy and time feasibility, controllable conflicts, and recoverability under environmental disturbances."

[0068] Step S2053: Based on the cleaning path, control the cleaning robot to clean the target photovoltaic modules. For example, in this embodiment, the planned cleaning path is sent to the cleaning robot; the cleaning robot moves precisely along the path, and upon reaching the target module, initiates the corresponding cleaning action, transmitting its position and work progress in real time; the path is dynamically adjusted based on feedback until all target modules are cleaned.

[0069] This embodiment also provides a self-cleaning control method for photovoltaic modules, which can be used in the control module of a cleaning robot. The cleaning robot is pre-configured with an image acquisition device. Figure 3 This is a flowchart of a photovoltaic module self-cleaning control method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S301: When a cleaning instruction is received from the decision-making end, the cleaning robot is controlled to move to the preset position of the target photovoltaic module according to the cleaning path.

[0070] For example, the preset position can be any position of the target photovoltaic module. The specific content of the preset position is not limited in the embodiments of this application, as long as it is reasonable.

[0071] Step S302: Control the image acquisition device to acquire target image data of the target photovoltaic module surface.

[0072] For example, the image acquisition device may include, but is limited to, a high-definition camera. The target image data may be a 640×640 pixel image of the component surface.

[0073] Step S303: Input the target image data into the pre-constructed stain segmentation network so that the stain segmentation network outputs stain information for different areas on the surface of the target photovoltaic module.

[0074] For example, in this embodiment of the application, the stain information of different areas on the surface of the target photovoltaic module can be analyzed by the stain segmentation network, and the stain information can be the stain heat value.

[0075] Step S304: Based on the stain information of different areas, determine that there is at least one target area on the target photovoltaic module that needs to be cleaned.

[0076] For example, in this embodiment of the application, the area with a heat value greater than a preset threshold is taken as the target area. The preset threshold can be determined according to the requirements, and this embodiment of the application does not make specific limitations.

[0077] Step S305: Control the cleaning robot to clean the target area. For example, control the robot to move to the target area for cleaning.

[0078] The method provided in this application involves a robot receiving a cleaning instruction from the decision-making end. First, it accurately reaches the target component location using a preset cleaning path, ensuring accurate positioning and avoiding unnecessary movement losses. Then, it uses its built-in image acquisition device to acquire images of the target component surface, combining this with a pre-built stain segmentation network to accurately output stain information for different areas. This enables refined identification and positioning of stained areas, precisely locking onto the target area requiring cleaning. Based on this refined stain information, the robot cleans only the stained areas, eliminating the need for repeated cleaning of clean areas. This significantly reduces freshwater and energy consumption during cleaning, perfectly adapting to the extremely scarce water resources in arid regions, while also improving cleaning efficiency and avoiding resource waste. Simultaneously, the real-time acquired target image data and stain segmentation results dynamically support precise adjustments to the cleaning actions, ensuring cleaning effectiveness and further guaranteeing the effective recovery of power generation efficiency after component cleaning.

[0079] The self-cleaning control method for photovoltaic modules provided in this application will be specifically described below through a specific embodiment.

[0080] Example: This embodiment is structured around four key stages: perception, decision-making, execution, and feedback, and is deployed within a large, centralized photovoltaic array located on the edge of a desert. First, binocular industrial cameras and miniature weather probes are equidistantly installed on the aluminum alloy frame or tracker beams above every two rows of modules. The cameras are positioned facing the module surface, recording high-definition RGB images and providing depth information for subsequent obstacle-crossing by the robot using parallax ranging. All front-end devices are connected to an edge intelligent cabinet on the side of the array via PoE (Power over Ethernet) or Wi-Fi 6 / LoRa-Mesh wireless bridges. This cabinet houses a JetsonOrin NX edge computing module and a wide-temperature industrial switch, forming the system's "vision and decision-making" center.

[0081] In real-time operation, the 640×640 pixel component surface image sequence captured by the camera is first fed into the SoilingNet-T stain recognition network designed in this invention. SoilingNet-T is a lightweight visual Transformer deployed at the edge, consisting of a convolutional feature extraction front-end, a hierarchical self-attention encoder, and a multi-class decoding head. Through joint training on public datasets and power plant field samples, combined with knowledge distillation and pruning compression, the number of parameters is controlled within the megabyte range. Its semantic segmentation steps include: S1, performing distortion correction, brightness normalization, and component grid line / cascade position labeling on the input image; S2, dividing the image into fixed-size patches, and extracting multi-scale semantic features through the Transformer encoder; S3, outputting a category probability map of multiple types of stains such as "clean," "dust," "bird droppings," and "salt frost" for each pixel through the decoding head; S4, spatially aggregating the pixel results according to the actual arrangement of the photovoltaic array to form a two-dimensional matrix SoilingMap that corresponds one-to-one with the component grid lines or cascades. Each grid in the SoilingMap records the shading area ratio and overall soiling level of the corresponding component (or sub-component), which can be considered as the soiling heat value at that location. The edge processing unit synchronously reads the voltage, current, power, irradiance, and temperature data from the cascade IV monitoring board at the same timestamp. It calculates the short-time baseline power using an embedded photovoltaic equivalent circuit model and performs exponential smoothing. Based on the smoothed baseline power and the SoilingMap, it calculates the recoverable power generation loss benefit within a given time window and compares it with the economic function composed of the total cleaning cost calculated based on water consumption, electricity consumption, and labor. When the "recoverable benefit" is... When the "cleaning cost" is not less than the preset economic threshold (which is determined by a combination of electricity price, water price and operation and maintenance strategy), the system determines that the current condition of "immediate cleaning benefits are greater than costs" is met, and automatically enters the multi-robot collaborative cleaning scheduling stage.

[0082] In terms of short-time baseline power calculation, this embodiment uses a single-diode model as the theoretical basis and employs an engineering approximation method to construct a baseline model suitable for on-site online calculations. The theoretical maximum output power of the module under "ideal clean conditions" is used as a comparison reference. The edge processing unit synchronously acquires operating parameters such as the module's planar irradiance and module / cell temperature at each sampling moment, introducing an intermediate physical quantity called "effective irradiance": based on the measured planar irradiance, it is corrected according to pre-calibrated incident angle correction coefficients and spectral correction coefficients, used to characterize the equivalent irradiance level comparable to standard test conditions under the current geometric incident and spectral conditions. Based on known parameters such as the module's rated maximum power and power temperature coefficient under standard test conditions (STC), the system converts the effective irradiance and real-time temperature into the theoretical maximum output power value corresponding to the current moment, thereby obtaining a short-time baseline power sequence reflecting the output capability that the cascade should have under clean, unobstructed conditions. The aforementioned correction coefficients can be obtained through the module datasheet and on-site calibration at the power plant, possessing clear physical meaning and traceability. To mitigate the impact of factors such as irradiance fluctuations, measurement noise, and inverter dynamic adjustments, this embodiment performs exponential smoothing on the calculated theoretical baseline power sequence to obtain a continuous and stable smoothed baseline power curve. This smoothed baseline power serves as a unified reference for subsequent power generation loss estimation and cleanliness decision threshold determination, ensuring the accuracy and engineering feasibility of pollution-induced damage analysis and dispatch control.

[0083] In calculating power generation loss, the system uses the deviation between the actual output power obtained from the monitoring board and the aforementioned smoothed baseline power as the main characteristic of pollution-induced damage. Specifically, firstly, based on the short-time baseline power model and after exponential smoothing, a smoothed baseline power curve is obtained representing the output capacity of the cascade under ideal clean and unobstructed conditions. At the same timestamp, the smoothed baseline power is compared with the measured output power. When the baseline value is greater than the measured value, the difference is recorded as the instantaneous power generation loss at that moment. If the difference is negative or extremely small, it is treated as zero to avoid misjudging measurement errors and inverter scheduling factors as pollution losses. The system integrates or accumulates the instantaneous power loss within a preset judgment window to obtain the total power generation loss during that time period.

[0084] Combining the SoilingMap generated by the SoilingNet-T stain segmentation network, the system further calculates the corresponding optical transmittance reduction factor based on the occlusion area ratio, stain type, and level of each grid, and calibrates the "stain recoverability coefficient" k (ranging from 0 to 1) based on historical operating data to characterize the proportion of recoverable losses to total losses after cleaning operations. This allows for the separation of recoverable energy losses from total power generation losses, providing a basis for subsequent economic assessments. In application scenarios requiring reduced reliance on measured differences, this embodiment also supports a feedforward estimation method based on SoilingMap, which directly corrects the effective irradiance and baseline power using stain coverage and transmittance reduction factors to obtain recoverable power generation losses for comparison with cleaning costs and scheduling decisions.

[0085] The scheduling module maps the SoilingMap into topological graph nodes. Based on the node weights (higher weights indicate greater power generation losses), it calls algorithms based on Graph Neural Networks (GNNs) and Multi-Agent Reinforcement Learning (MARLs) to plan the cleaning sequence and travel paths for the tracked cleaning robot fleet. The tracked robot body adopts a carbon fiber integrated frame, equipped with rubber tracks, retractable atomizing spray booms, and high-speed nylon roller brushes. The robot's internal MCU (Micro-Controller Unit) manages the walking motors, cleaning actuators, millimeter-wave obstacle avoidance radar, and IMU (Inertial Measurement Unit) via a CAN-FD bus. Magnetic induction strips and vision-inertial fusion positioning jointly ensure that the robot's positioning error on the component row guide rails does not exceed 3 cm, maintaining stable obstacle crossing even in environments with 30° table angles, intersecting beams, and high winds. The robot transmits its pose, remaining power, and task completion status back to the scheduling server in real time. The scheduling algorithm dynamically rearranges tasks accordingly, avoiding operation interruptions due to insufficient power, path blockage, or sudden sandstorms.

[0086] In path planning, the system first maps the SoilingMap, station geometry, guide rail connectivity, and restricted area constraints into a directed topology graph at the scheduling time. Nodes represent cleanable standard work units or cross-row transfer points, and edges represent passable channels such as guide rails or bridges. The weight of each node is obtained by combining "recoverable power generation loss" and work cost, and is updated in real time with irradiance, wind field, and traffic status; the edge cost is given by combining path length, slope, number of turns, congestion, and risk information. Based on this, the scheduling module uses a graph neural network to learn the representation of nodes and edges, obtaining embedding vectors that reflect the benefit density, risk, and accessibility. Then, a pointer-based policy network autoregressively selects the next target node and travel edge according to the state of "current robot position - power - remaining time window - set of unfinished nodes," while performing a look-forward check on energy and time feasibility. When multiple robots collaborate, a multi-agent reinforcement learning framework of "centralized training and decentralized execution" is adopted: during the training phase, global state and joint reward are used to optimize queue efficiency and avoid conflicts and deadlocks; during the execution phase, each robot makes independent decisions based solely on local observations and lightweight broadcasts (including current position, target node, expected time period, and remaining battery power). To eliminate resource conflicts and the risk of passing each other, the system introduces a rapid online allocation and conflict resolution process after the strategy output: if two or more robots plan to occupy the same work unit at the same time or enter the same narrow side passage from opposite directions, they will give way, stagger time slots, or replace target nodes according to preset priorities (e.g., those with higher reward density, lower battery power, or greater offline penalty). If the path is temporarily blocked or a sudden increase in wind and sand increases the risk, the corresponding edge will be immediately set as unreachable or its cost will be increased, triggering a rolling replanning. The system refreshes node weights and edge costs at fixed intervals or through event triggers, and obtains new work sequences and travel paths within the short horizon. When it is predicted that the remaining power is insufficient to safely complete the next work segment, the strategy automatically inserts a turnaround route to the nearest reachable charging / replenishment node, ensuring uninterrupted operation and continuous satisfaction of safety distance and positioning error constraints. This process, without altering the robot's body or communication architecture, achieves online optimization and dynamic rearrangement of cleaning sequences and travel paths, guaranteeing "priority cleaning of high-weight areas, energy and time feasibility, controllable conflicts, and recoverability under environmental disturbances."

[0087] To adapt to extreme climates characterized by sandstorms, water scarcity, and large diurnal temperature variations, this embodiment employs two operating modes at the software level: during sandstorms or red / yellow alerts for strong winds, the system maintains only low-frame-rate camera inspections and increases the economic threshold to avoid secondary pollution during peak dust periods; on normal sunny days, it maintains high-frequency inspections and on-demand cleaning. At the hardware level, the robot's atomizing nozzles generate 60-80 micron fine mists at low pressures of 0.3–0.5 MPa, saving over 60% more fresh water than traditional high-pressure water guns while maintaining cleaning effectiveness; all drives and actuators utilize wide-temperature components ranging from -20℃ to 70℃ and are equipped with IP66 dustproof housings and NBR sealing rings to ensure reliability under varying diurnal temperature ranges.

[0088] The distinctive features of this embodiment are as follows: First, it is the first to transplant a lightweight Transformer stain segmentation network to the edge, achieving millisecond-level component-level stain recognition; second, it innovatively integrates real-time power generation loss assessment with a cleaning cost-benefit model to form an economic threshold decision, no longer relying on a fixed period; third, it introduces GNN-MARL multi-robot collaboration, enabling the number of robots to be expanded in a matrix with "plug and play," and automatically reconstructing paths when there are sudden obstacles or equipment shutdowns; fourth, the design takes into account the water scarcity, frequent sandstorms, and high-temperature impacts in southern Xinjiang, simultaneously reducing water consumption and failure rates through low-pressure atomization, wide-temperature devices, and intelligent tailwind window strategies. Alternative implementation methods include: adding an electrostatic dust removal or plasma deposition module between the camera and the robot to achieve "dry" waterless cleaning; replacing tracked robots with wheeled robots or drones equipped with spray booms to adapt to high-tilt or distributed rooftops; and further incorporating carbon emission or load peak shaving constraints into the threshold algorithm to meet carbon trading or energy storage interaction scenarios. All alternative approaches do not deviate from the core idea of ​​this invention, namely, to achieve on-demand, precise, and autonomous cleaning of photovoltaic modules through edge vision perception, economic decision-making, and multi-robot collaborative closed loop, thereby significantly improving power generation efficiency and reducing operation and maintenance costs in desert-arid large-scale base conditions.

[0089] In this embodiment, the system consists of four interconnected layers from bottom to top: a front-end perception layer, an edge intelligent processing layer, a multi-robot execution layer, and an energy consumption and safety management layer. Each layer is interconnected with a hybrid communication architecture of gigabit Ethernet and LoRa-Mesh according to the principle of layered decoupling, so that it can operate independently on site without relying on the public network or remote cloud computing.

[0090] At the front-end sensing layer, a pair of binocular industrial cameras are installed in the center of an aluminum alloy longitudinal beam shared by every two rows of components. The cameras are fixed to a stainless steel pan-tilt head with vibration-damping pads, with the lenses maintaining an approximately 15-degree downward angle relative to the component surface, allowing a single field of view to cover five rows and twelve columns of components. The cameras are connected to an industrial switch at the end of the row via PoE (Power over Ethernet), and the switch is then connected to an edge cabinet via fiber optic cable. Along the same sensing link, miniature weather probes (collecting irradiance, temperature, humidity, wind speed, and particulate matter) and several cascaded IV monitoring boards are also connected in series; both the probes and monitoring boards are connected to a serial port server via RS-485 bus, which then converts the data into TCP / IP frames, maintaining the same timestamp accuracy as the image stream. The camera pan-tilt head incorporates a low-power defogging and temperature control module, which can... The lens is kept clean under extreme temperature differences ranging from 25°C to 70°C, thus ensuring all-weather imaging quality.

[0091] All perceived data converges at the edge intelligent processing layer, which employs an embedded SoC (System-on-Chip) module with dedicated AI and acceleration units, running a lightweight Linux environment. First, RGB images from the camera undergo color correction and brightness adaptation processing before being fed into a lightweight vision Transformer network quantized at the edge. The network outputs a two-dimensional soiling heatmap aligned with the component arrangement, where the matrix elements correspond to the occlusion ratio of each component surface. The processor then reads IV monitoring data from the same timestamp, calculates the theoretical baseline power using an embedded photovoltaic equivalent model, and compares it with the actual power to obtain an estimate of instantaneous power loss. Power loss, real-time electricity price, and resource consumption per cleaning cycle are fed into an economic threshold decision logic; only when the condition "expected revenue exceeds the cost of the current cleaning cycle" is met does the system generate a task list, thus entering the multi-robot scheduling stage.

[0092] The scheduling algorithm maps the components to be cleaned in the SoilingMap to a weighted node graph. Node features include the degree of soiling, geographical coordinates, and the reachability time for each robot. The core scheduling mechanism uses a graph neural network to extract global relationships, combined with a multi-agent reinforcement learning algorithm, to quickly generate a conflict-free shortest cleaning path and corresponding job sequence for each robot. The scheduling results are distributed via a LoRa-Mesh network. To improve link reliability in windy and dusty environments, the network uses a spread spectrum configuration with high redundancy and supports multi-hop relays to ensure stable instruction reception even when the robot is deep within the component array.

[0093] The multi-robot execution layer consists of tracked cleaning robots. Each robot engages with a U-shaped guide rail on top of the component via a magnetic guide structure, allowing it to traverse the height differences between beams and supports with the cooperation of its track wheels and guide rails. The microcontroller unit (MCU) inside the robot manages the walking motor, atomizing pump, nylon roller brush, and lifting mechanism via a CAN-FD bus, while simultaneously collecting data from the inertial measurement unit, millimeter-wave radar, and magnetic induction strip sensors for centimeter-level positioning and obstacle crossing control. Upon receiving a scheduling command, the robot automatically moves towards the target component: if the command mode is "spray-brush," the atomizing pump outputs fine mist in the low-pressure range and simultaneously drives the roller brush; if the command only requires dry brushing, the pump is turned off to conserve water. After cleaning each component, the robot again uses a lightweight network on its body to quickly infer the local image, uploading the residual dirt score and its own pose, prompting the central SoilingMap to update in real time and guide subsequent decisions.

[0094] The energy and safety management layer monitors the system's total power, water tank level, airflow, and cabinet temperature within the edge cabinet. When the airflow exceeds a set threshold or the water level is too low, the management logic immediately sends a "return" command to the scheduler via gRPC. The robot stops cleaning, retracts its nozzles, and returns along the shortest path to the inductive charging and water replenishment base located in a corner of the array. The base simultaneously charges the battery and replenishes the spray water. Once external conditions return to a safe range, the edge processor re-enters the threshold judgment process, and the system automatically takes over the unfinished tasks without manual intervention.

[0095] To facilitate future expansion, the system writes all time-series data—including SoilingMap snapshots, power curves, meteorological information, and robot status—into a local database and provides a RESTful interface to connect with the power plant's SCADA or energy management platform. This allows for direct reuse of the same data output, whether for further overlaying energy storage scheduling or integration with carbon trading calculations, without requiring structural modifications to the clean energy subsystem.

[0096] Through the above integrated design, this implementation method clearly reveals the physical connection relationship of each hardware component, the logical calling order and dynamic data interaction of each software module, and how the entire closed loop achieves "on-demand triggering, on-site decision-making, collaborative execution, and real-time feedback" in extreme desert scenarios without referencing any unverified field test data.

[0097] This embodiment also provides a photovoltaic module self-cleaning control device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0098] This embodiment provides a self-cleaning control device for photovoltaic modules, such as... Figure 4 As shown, it includes: The first acquisition module 401 is used to acquire image data, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of the photovoltaic array. The identification module 402 is used to identify stains in image data and determine the stain heat value of at least one target photovoltaic module in the photovoltaic array. The stain heat value characterizes the degree of influence of stains on the surface of the photovoltaic module on power generation. The target photovoltaic module is the photovoltaic module in the photovoltaic array that has stains. The first determining module 403 is used to determine the recoverable energy loss based on the stain heat value of the target photovoltaic module, the measured output power data, the measured irradiance data, the cell temperature data, and the rated maximum power. The second determining module 404 is used to determine the cleaning cost based on the stain heat value and area information of the target photovoltaic module; The first control module 405 is used to control the cleaning robot to clean the target photovoltaic module when the recoverable energy loss is greater than the cleaning cost.

[0099] In some alternative implementations, the first determining module 403 includes: The first determining submodule is used to determine the stain recoverability coefficient based on the stain heat value of the target photovoltaic module; The second determination submodule is used to determine the total energy loss of the photovoltaic array based on measured output power data, measured irradiance data, cell temperature data, and rated maximum power. The third determination submodule is used to determine the recoverable energy loss based on the total energy loss and the stain recoverability coefficient.

[0100] In some optional implementations, the second determining submodule includes: The first determining unit is used to determine the effective irradiance data based on the measured irradiance data, the preset incident angle correction coefficient, and the spectral correction coefficient. The incident angle correction coefficient and the spectral correction coefficient are fixed constants when there are no external sensing conditions. When a component attitude sensor and / or a meteorological acquisition device are configured, the correction coefficient is updated over time according to the real-time attitude parameters and meteorological parameters according to the preset lookup table or interpolation rules. The second determining unit is used to determine instantaneous baseline power data based on effective irradiance data, cell temperature data, and rated maximum power. The processing unit is used to smooth the instantaneous baseline power data based on a preset smoothing model to obtain the target instantaneous baseline power data. The third determining unit is used to determine the instantaneous power loss data based on the measured output power data and the target instantaneous baseline power data. The instantaneous power loss data is used to characterize the instantaneous power loss value at different times within the target time period. The fourth determining unit is used to determine the total energy loss based on instantaneous power loss data.

[0101] In some alternative implementations, the first control module 405 includes: Construct a submodule to build a directed topology graph using photovoltaic modules as nodes and rail connections as edges; The planning submodule is used to plan the cleaning path of the cleaning robot based on the location of the cleaning robot, the target photovoltaic module, and the directed topology graph. The control submodule is used to control the cleaning robot to clean the target photovoltaic modules based on the cleaning path.

[0102] This embodiment provides another self-cleaning control device for photovoltaic modules, such as... Figure 5 As shown, it includes: The second control module 501 is used to control the cleaning robot to move to the preset position of the target photovoltaic module according to the cleaning path when it receives the cleaning instruction sent by the decision-making end. The third control module 502 is used to control the image acquisition device to acquire target image data of the target photovoltaic module surface; The third determining module 503 is used to input the target image data into a pre-constructed stain segmentation network so that the stain segmentation network outputs stain information of different areas on the surface of the target photovoltaic module; The fourth determining module 504 is used to determine, based on the stain information of different regions, that there is at least one target area on the target photovoltaic module that needs to be cleaned; The fourth control module 505 is used to control the cleaning robot to clean the target area.

[0103] The photovoltaic module self-cleaning control device provided in this embodiment of the invention can execute the photovoltaic module self-cleaning control method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0104] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0105] The following is a detailed reference. Figure 6 This diagram illustrates a suitable structural design for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 601, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 602 or a program loaded from memory 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of the electronic device. The processor 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0106] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0107] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a memory 608, or installed from a ROM 602. When the computer program is executed by the processor 601, it performs the functions defined in the photovoltaic module self-cleaning control method of the embodiments of the present invention.

[0108] Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0109] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the photovoltaic module self-cleaning control method shown in the above embodiments is implemented.

[0110] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0111] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A self-cleaning control method for photovoltaic modules, characterized in that, When applied to the decision-making process, the method includes: Acquire image data, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of the photovoltaic array; The image data is used to identify stains and determine the stain heat value of at least one target photovoltaic module in the photovoltaic array. The stain heat value characterizes the degree of influence of stains on the surface of the photovoltaic module on power generation. The target photovoltaic module is the photovoltaic module in the photovoltaic array that has stains. Based on the stain heat value, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of the target photovoltaic module, the recoverable energy loss is determined. The cleaning cost is determined based on the stain heat value and area information of the target photovoltaic module; When the recoverable energy loss exceeds the cleaning cost, the cleaning robot is controlled to clean the target photovoltaic module.

2. The method according to claim 1, characterized in that, The step of determining recoverable energy loss based on the stain heat value, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of the target photovoltaic module includes: The stain recoverability coefficient is determined based on the stain heat value of the target photovoltaic module; The total energy loss of the photovoltaic array is determined based on measured output power data, measured irradiance data, cell temperature data, and rated maximum power. The recoverable energy loss is determined based on the total energy loss and the stain recoverability coefficient.

3. The method according to claim 2, characterized in that, The determination of the total energy loss of the photovoltaic array based on measured output power data, measured irradiance data, cell temperature data, and rated maximum power includes: Effective irradiance data is determined based on the measured irradiance data, the preset incident angle correction coefficient, and the spectral correction coefficient. The incident angle correction coefficient and the spectral correction coefficient are fixed constants when there are no external sensing conditions. When a component attitude sensor and / or a meteorological acquisition device are configured, the correction coefficient is updated over time according to the real-time attitude parameters and meteorological parameters according to the preset lookup table or interpolation rules. The instantaneous baseline power data is determined based on the effective irradiance data, cell temperature data, and rated maximum power. The instantaneous baseline power data is smoothed based on a preset smoothing model to obtain the target instantaneous baseline power data. Instantaneous power loss data is determined based on the measured output power data and the target instantaneous baseline power data. The instantaneous power loss data is used to characterize the instantaneous power loss value at different times within the target time period. The total energy loss is determined based on the instantaneous power loss data.

4. The method according to claim 1, characterized in that, The step of controlling the cleaning robot to clean the target photovoltaic module includes: A directed topology graph is constructed by using photovoltaic modules as nodes and rail connections as edges. The cleaning path of the cleaning robot is planned based on the location of the cleaning robot, the target photovoltaic module, and the directed topology graph. The cleaning robot is controlled to clean the target photovoltaic module based on the cleaning path.

5. A self-cleaning control method for photovoltaic modules, characterized in that, The control module is applied to a cleaning robot, which is pre-configured with an image acquisition device. The method includes: When a cleaning instruction is received from the decision-making unit, the cleaning robot is controlled to move to the preset position of the target photovoltaic module according to the cleaning path. The image acquisition device is controlled to acquire target image data of the surface of the target photovoltaic module; The target image data is input into a pre-constructed stain segmentation network so that the stain segmentation network outputs stain information for different areas on the surface of the target photovoltaic module. Based on the stain information of the different regions, it is determined that there is at least one target area on the target photovoltaic module that needs to be cleaned; Control the cleaning robot to clean the target area.

6. A self-cleaning control device for photovoltaic modules, characterized in that, The apparatus for performing the method of claim 1, comprising: The first acquisition module is used to acquire image data, measured output power data, measured irradiance data, cell temperature data, and rated maximum power of the photovoltaic array. The identification module is used to identify stains in the image data and determine the stain heat value of at least one target photovoltaic module in the photovoltaic array. The stain heat value characterizes the degree of influence of stains on the surface of the photovoltaic module on power generation. The target photovoltaic module is the photovoltaic module in the photovoltaic array that has stains. The first determining module is used to determine the recoverable energy loss based on the stain heat value, measured output power data, measured irradiance data, cell temperature data and rated maximum power of the target photovoltaic module. The second determining module is used to determine the cleaning cost based on the stain heat value and area information of the target photovoltaic module; The first control module is used to control a cleaning robot to clean the target photovoltaic module when the recoverable energy loss is greater than the cleaning cost.

7. A self-cleaning control device for photovoltaic modules, characterized in that, For performing the method of claim 5, characterized in that the apparatus comprises: The second control module is used to control the cleaning robot to move to the preset position of the target photovoltaic module according to the cleaning path when it receives the cleaning instruction sent by the decision-making end. The third control module is used to control the image acquisition device to acquire target image data of the target photovoltaic module surface; The third determining module is used to input the target image data into a pre-constructed stain segmentation network so that the stain segmentation network outputs stain information of different areas on the surface of the target photovoltaic module; The fourth determining module is used to determine, based on the stain information of the different regions, that there is at least one target area on the target photovoltaic module that needs to be cleaned; The fourth control module is used to control the cleaning robot to clean the target area.

8. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the photovoltaic module self-cleaning control method according to any one of claims 1 to 4, or the photovoltaic module self-cleaning control method according to claim 5.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the photovoltaic module self-cleaning control method according to any one of claims 1 to 4, or to execute the photovoltaic module self-cleaning control method according to claim 5.

10. A computer program product, characterized in that, The method includes computer instructions for causing a computer to execute the photovoltaic module self-cleaning control method according to any one of claims 1 to 4, or to execute the photovoltaic module self-cleaning control method according to claim 5.

Images