Multifunctional integrated cleaning robot for photovoltaic module of power station

Abstract

The invention provides a multifunctional integrated cleaning robot for a photovoltaic module of a power station, which belongs to the technical field of photovoltaic module cleaning and comprises a control chip, a robot body device, an automatic cleaning device, a dust collection device, a water spraying system device, a solar charging device, an intelligent navigation device, an environment monitoring device, a communication interface device and a power management device. An intelligent cleaning control module is arranged in the control chip. The method comprises the following steps: constructing a three-dimensional map through environment perception; evaluating the pollution state of the photovoltaic module by using a sensor to generate a thermodynamic diagram; a double-layer optimization model is adopted for path planning and resource allocation, an upper layer adopts an improved ant colony algorithm to take charge of a global path, and a lower layer optimizes cleaning parameters based on a particle swarm algorithm; cleaning operation is executed, and real-time monitoring is carried out; evaluating the cleaning effect; water resource and energy intelligent management is carried out; multi-machine collaboration and task allocation are realized; processing an abnormal condition; the cleaning strategy is continuously optimized through machine learning, and efficient and accurate photovoltaic module cleaning is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic module cleaning, and in particular relates to a multifunctional integrated cleaning robot for photovoltaic modules in power stations. Background Art

[0002] As a crucial component of clean energy, photovoltaic power generation efficiency is directly impacted by the surface cleanliness of photovoltaic modules. Traditionally, photovoltaic module cleaning relies primarily on manual cleaning or simple mechanized equipment. These methods typically employ fixed parameters for uniform cleaning and are unable to adapt cleaning strategies based on contamination status. As photovoltaic power plants expand in size, this approach has become inadequate for efficient cleaning.

[0003] While currently available automated cleaning equipment offers basic cleaning capabilities, it generally lacks the ability to intelligently identify and assess the contamination status of PV panel surfaces. Cleaning parameter settings are often simplistic, often employing a single standard to treat areas with varying degrees of contamination. Furthermore, existing equipment exhibits significant limitations in cleaning path planning, resource allocation, and multi-machine collaboration, making it difficult to accurately allocate and optimize cleaning resources.

[0004] Therefore, existing technologies struggle to address technical challenges such as differential identification of contamination status on photovoltaic panels, precise configuration of cleaning parameters based on contamination status, and optimal allocation of cleaning resources. This results in low cleaning efficiency, wasted resources, and inconsistent cleaning quality. A solution that can intelligently identify contamination status and perform precise cleaning is urgently needed. This means that existing photovoltaic panel cleaning robot systems struggle to efficiently achieve precise cleaning based on contamination status and optimal resource allocation. Summary of the Invention

[0005] In view of this, the present invention provides a multifunctional integrated cleaning robot for photovoltaic modules in power stations, which can solve the technical problem in the prior art that photovoltaic module cleaning robot systems are difficult to efficiently achieve precise cleaning and optimal resource allocation based on pollution status.

[0006] The present invention is implemented as follows: The present invention provides a multifunctional integrated cleaning robot for photovoltaic modules in power stations. The intelligent cleaning control module is used to perform the following steps: system initialization and environmental perception; photovoltaic module pollution status assessment; cleaning path planning and resource allocation. Based on the photovoltaic module pollution status heat map and environmental map, a two-layer optimization model is constructed to perform cleaning path planning and resource allocation; cleaning operation execution and real-time monitoring; cleaning effect assessment and quality control; intelligent management and recycling of water resources; energy management and autonomous charging; multi-machine collaboration and task allocation; abnormal situation handling and self-protection; cleaning data analysis and optimization strategy.

[0007] Among them, the system initialization and environmental perception steps include: starting the lidar, millimeter-wave radar, ultrasonic sensor and visual camera through the intelligent navigation device to collect multi-source data, build a three-dimensional environmental map of the photovoltaic power station, mark the location, arrangement and cleaning channel width of the photovoltaic modules; at the same time, collect the current environmental temperature, humidity, wind speed and light intensity data through the environmental monitoring device to evaluate whether the current environment is suitable for cleaning operations.

[0008] Among them, the photovoltaic module pollution status assessment step includes: using the dirt detection sensor in the automatic cleaning device to scan the surface of the photovoltaic module, collecting surface reflectivity, transmittance and scattering rate data, combining with the high-definition images collected by the visual camera, and analyzing the type, distribution density and adhesion degree of dirt on the surface of the photovoltaic module through a deep learning algorithm; according to the analysis results, the photovoltaic modules are divided into heavily polluted areas, moderately polluted areas and lightly polluted areas, and a pollution status heat map is generated.

[0009] Among them, the two-layer optimization model includes: an upper-layer optimization algorithm that combines an improved ant colony algorithm with game theory is responsible for global cleaning path planning, and a lower-layer optimization algorithm based on a particle swarm algorithm is responsible for local resource allocation; the two-layer algorithms interact and iterate until convergence to obtain the globally optimal cleaning path and resource allocation plan.

[0010] In the upper-level optimization algorithm, the cleaning task is modeled as a weighted multi-objective traveling salesman problem, in which the location of the photovoltaic module is used as the node to be visited, the cost of moving between nodes is used as the edge weight, the degree of pollution is used as the node priority weight, and the remaining battery power is used as the path constraint condition; by introducing the Nash equilibrium strategy in game theory, ants can compete and cooperate with each other during the pheromone update process to avoid falling into local optimal solutions.

[0011] Among them, in the lower-level optimization algorithm, the specific cleaning parameters of each photovoltaic module are optimized, including the cleaning brush head speed, water spray volume and dust suction fan power; the resource allocation optimization equation group is used to achieve accurate allocation of cleaning resources; the resource allocation optimization equation group includes the cleaning efficiency equation, water resource consumption equation, energy consumption equation, brush head life equation and task completion time equation.

[0012] Among them, the cleaning operation execution and real-time monitoring steps include: controlling the robot body device to move to the target photovoltaic module position, starting the automatic cleaning device, and adjusting the working parameters of the main cleaning brush head and the auxiliary cleaning brush head according to the pollution status; synchronously starting the water spray system device and the dust suction device; monitoring the cleaning effect in real time through visual sensors. If the cleaning effect does not meet the standards, the cleaning parameters are automatically adjusted and the cleaning operation is repeated.

[0013] Among them, the cleaning effect evaluation and quality control steps include: after the cleaning operation is completed, the reflectivity sensor in the automatic cleaning device is used to scan the surface of the photovoltaic module again, and the changes in reflectivity, transmittance and scattering rate before and after cleaning are compared to calculate the cleaning efficiency; at the same time, a high-definition image of the photovoltaic module after cleaning is collected by a visual camera, and the proportion of residual dirt area is calculated using an image processing algorithm; if the cleaning efficiency is lower than 95% or the proportion of residual dirt area is higher than 5%, the area is marked as a substandard cleaning area and a second cleaning is arranged.

[0014] Among them, the robot body device is used to serve as the basic carrier of the entire system and provide a composite mobile mechanism, the automatic cleaning device is used to collect dirt type data on the surface of the photovoltaic module and perform physical cleaning operations, the dust suction device is used to collect dust particle distribution data and perform dust suction operations, the water spray system device is used to collect water quality parameters and perform water spray cleaning operations, the solar charging device is used to collect light intensity and battery charging status data, the intelligent navigation device is used to collect environmental map data and obstacle location information, the environmental monitoring device is used to collect temperature, humidity, wind speed and light intensity data, the communication interface device is used to realize data interaction with the remote monitoring center and multi-machine collaborative operation, and the power management device is used to monitor the battery voltage and control the system power supply.

[0015] Among them, the cleaning efficiency equation is used to calculate the cleaning efficiency under specific cleaning parameters. The input includes the degree of dirt adhesion, dirt type identification parameters, cleaning brush head speed, water spray volume and dust suction fan power, and the output is the expected cleaning efficiency; the water resource consumption equation is used to calculate the water resource consumption during the cleaning process; the energy consumption equation is used to calculate the energy consumption during the cleaning process; the brush head life equation is used to evaluate the impact of cleaning parameters on the life of the brush head; the task completion time equation is used to predict the time required to complete the cleaning task.

[0016] Compared to existing technologies, the present invention provides a multifunctional integrated cleaning robot for photovoltaic panels in power plants. This system utilizes an intelligent cleaning control module to accurately assess the contamination status of photovoltaic panels, establish a contamination status heat map, and implement differentiated cleaning strategies based on this assessment. This system utilizes a two-layer optimization model for cleaning path planning and resource allocation. The upper layer utilizes a modified ant colony algorithm and game theory for global path planning, while the lower layer utilizes a particle swarm algorithm for local resource optimization. These two layers interact and iterate to achieve a global optimal solution.

[0017] Compared to traditional technologies, this invention precisely adjusts cleaning parameters, including brush speed, water spray volume, and fan power, based on contamination type, distribution density, and adhesion level. This avoids the over- or under-cleaning problems often associated with traditional methods that use uniform parameters for all areas. Furthermore, through a set of resource allocation optimization equations, it achieves a multi-objective balance between water resources, energy, and equipment lifespan, significantly improving resource utilization efficiency.

[0018] The present invention successfully solves the technical problem in the prior art that photovoltaic module cleaning robot systems are difficult to efficiently achieve precise cleaning and optimal resource allocation based on pollution status. Through pollution status assessment, a two-layer optimization model and a multi-machine collaborative mechanism, precise cleaning and optimal resource allocation based on pollution status are achieved, which significantly improves cleaning efficiency and quality, while minimizing resource consumption and equipment loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of the method of the present invention;

[0020] Figure 2 is a schematic diagram of the cleaning robot of the present invention;

[0021] Figure 3 Schematic diagram of the self-cleaning device and the dust collection device of the present invention;

[0022] Figure 4 It is a bottom schematic diagram of the self-cleaning device and the dust collecting device of the present invention;

[0023] Figure 5 Schematic diagram of the water spray system device of the present invention;

[0024] In the accompanying drawings, the components represented by the reference numerals are as follows:

[0025] 1. Automatic cleaning device; 11. Main cleaning brush head; 12. Auxiliary cleaning brush head; 13. Dual drive motor; 14. Planetary gear reducer; 2. Dust suction device; 21. Main dust suction channel; 22. Auxiliary dust suction channel; 23. Dust collection box; 24. Centrifugal fan; 3. Water spray system device; 31. Water tank; 32. Spray nozzle; 33. Water purification device; 4. Solar charging device; 5. Intelligent navigation device; 6. Environmental monitoring device; 7. Communication interface device. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0027] like Figure 1FIG. 1 is a flow chart of a multifunctional integrated cleaning robot for photovoltaic modules in power plants provided by the present invention. The method includes the following steps:

[0028] A multifunctional integrated cleaning robot system for photovoltaic modules in power stations, comprising a control chip, a robot body device, an automatic cleaning device, a dust collection device, a water spraying system device, a solar charging device, an intelligent navigation device, an environmental monitoring device, a communication interface device, and a power management device. The control chip is electrically connected to the robot body device, the automatic cleaning device, the dust collection device, the water spraying system device, the solar charging device, the intelligent navigation device, the environmental monitoring device, the communication interface device, and the power management device, respectively. An intelligent cleaning control module is provided in the control chip. The robot body device is used as the basic carrier of the entire system and provides a composite mobile mechanism. The automatic cleaning device is used to collect dirt type data on the surface of the photovoltaic module and perform physical cleaning operations. The dust collection device is used to collect dust particle distribution data and perform dust collection operations. operation, the water spray system device is used to collect water quality parameters and perform water spray cleaning operations, the solar charging device is used to collect light intensity and battery charging status data, the intelligent navigation device is used to collect environmental map data and obstacle location information, the environmental monitoring device is used to collect temperature, humidity, wind speed and light intensity data, the communication interface device is used to realize data interaction with the remote monitoring center and multi-machine collaborative operation, the power management device is used to monitor the battery voltage and control the system power supply; the sampling frequency of the robot body device is 1 time per second, the sampling frequency of the automatic cleaning device is 10 times per second, the sampling frequency of the dust collection device is 5 times per second, the sampling frequency of the water spray system device is 2 times per second, the sampling frequency of the solar charging device is 1 time per minute, the sampling frequency of the intelligent navigation device is 20 times per second, and the sampling frequency of the environmental monitoring device is 1 time per minute.

[0029] The intelligent cleaning control module is used to perform the following steps:

[0030] S01. System initialization and environmental perception: The intelligent navigation device is used to activate the laser radar, millimeter-wave radar, ultrasonic sensor and visual camera to collect multi-source data, build a three-dimensional environmental map of the photovoltaic power station, and mark the location, arrangement and cleaning channel width of the photovoltaic modules; at the same time, the environmental monitoring device is used to collect the current environmental temperature, humidity, wind speed and light intensity data to evaluate whether the current environment is suitable for cleaning operations.

[0031] S02. Assessment of the pollution status of photovoltaic modules: The dirt detection sensor in the automatic cleaning device is used to scan the surface of the photovoltaic modules to collect surface reflectivity, transmittance and scattering data. Combined with the high-definition images captured by the visual camera, the type, distribution density and degree of adhesion of dirt on the surface of the photovoltaic modules are analyzed through a deep learning algorithm. Based on the analysis results, the photovoltaic modules are divided into heavily polluted areas, moderately polluted areas and lightly polluted areas to generate a pollution status heat map.

[0032] S03. Cleaning path planning and resource allocation: Based on the thermal map of the pollution status of photovoltaic modules and the environmental map, a two-layer optimization model is constructed to perform cleaning path planning and resource allocation. The two-layer optimization model adopts an upper-layer optimization algorithm that combines an improved ant colony algorithm with game theory to be responsible for global cleaning path planning, and a lower-layer optimization algorithm based on a particle swarm algorithm to be responsible for local resource allocation. In the upper-layer optimization algorithm, the cleaning task is modeled as a weighted multi-objective traveling salesman problem, in which the location of the photovoltaic module is used as the node to be visited, the moving cost between nodes is used as the edge weight, the pollution degree is used as the node priority weight, and the remaining battery power is used as the path constraint. Initialize the ant colony population, in which the path table of the α ant The β ant represents the global optimal cleaning path, the β ant represents the suboptimal cleaning path, and the δ ant represents the tertiary optimal cleaning path. By introducing the Nash equilibrium strategy in game theory, the ants compete and cooperate with each other during the pheromone update process to avoid falling into the local optimal solution. At the same time, the energy consumption prediction model is considered to dynamically predict energy consumption based on the path length, terrain complexity, and the degree of pollution of the photovoltaic components, ensuring that charging at the charging station can be planned in time when the power is insufficient. In the lower-level optimization algorithm, the specific cleaning parameters of each photovoltaic component are optimized, including the cleaning brush head speed, water spray volume, and dust suction fan power. Based on the equipment loss assessment model, the impact of different cleaning parameters on the equipment life is predicted, and the equipment wear is minimized while meeting the cleaning efficiency. The resource allocation optimization equation group realizes the precise allocation of cleaning resources; the resource allocation optimization equation group includes the cleaning efficiency equation, the water resource consumption equation, the energy consumption equation, the brush head life equation and the task completion time equation; the cleaning efficiency equation is used to calculate the cleaning efficiency under specific cleaning parameters, and the input includes the degree of dirt adhesion, dirt type identification parameters, cleaning brush head speed, water spray volume and dust suction fan power, and the output is the expected cleaning efficiency; the water resource consumption equation is used to calculate the water resource consumption during the cleaning process, and the input includes the photovoltaic module area, dirt coverage, dirt type identification parameters, water spray volume and cleaning duration, and the output is water resource consumption; the energy consumption equation is used to calculate the energy consumption during the cleaning process, and the input includes The equation includes the speed of the cleaning brush head, the energy consumption parameters of the water spray system, the power of the dust suction fan, the energy consumption parameters of the mobile system and the cleaning duration, and the output is the energy consumption; the brush head life equation is used to evaluate the impact of cleaning parameters on the life of the brush head, and the input includes the speed of the cleaning brush head, the pressure of the brush head, the dirt hardness parameter, the friction coefficient and the cleaning duration, and the output is the degree of wear of the brush head; the task completion time equation is used to predict the time required to complete the cleaning task, and the input includes the area of ​​the photovoltaic module, the dirt coverage rate, the speed of the cleaning brush head, the moving speed of the robot and the degree of dirt adhesion, and the output is the estimated task completion time; by iteratively updating the particle position and speed, the optimal cleaning parameter combination is searched to achieve a multi-objective balance between cleaning efficiency, resource consumption and equipment life;The cleaning path planning results of the upper-level algorithm serve as constraints for the lower-level algorithm. The resource allocation results of the lower-level algorithm are fed back to the upper-level algorithm for path adjustment. The two-level algorithm iterates alternately until convergence, achieving the globally optimal cleaning path and resource allocation solution. This method introduces a two-level optimization model based on an improved ant colony algorithm and game theory to better solve the global path planning problem under complex constraints. It also considers energy consumption prediction and charging station planning to ensure that the robot will not be interrupted due to insufficient power during operation. The interactive iteration of the upper and lower-level algorithms achieves the global optimal path planning and resource allocation.

[0033] S04. Execution and real-time monitoring of cleaning operations: Control the robot body device to move to the target photovoltaic module position, start the automatic cleaning device, and adjust the working parameters of the main cleaning brush head and the auxiliary cleaning brush head according to the pollution status; synchronously start the water spray system device, and use the array nozzle to spray water in a targeted manner on the target area, and adjust the water flow and spray angle in real time according to the pollution degree; start the dust suction device, and capture dust particles through the dual dust suction ducts and electrostatic adsorption net; monitor the cleaning effect in real time through the visual sensor. If the cleaning effect does not meet the standard, the cleaning parameters are automatically adjusted and the cleaning operation is repeated.

[0034] S05. Cleaning effect evaluation and quality control: After the cleaning operation is completed, the reflectivity sensor in the automatic cleaning device scans the surface of the photovoltaic module again, compares the changes in reflectivity, transmittance and scattering rate before and after cleaning, and calculates the cleaning efficiency. At the same time, a visual camera collects high-definition images of the photovoltaic module after cleaning, and uses an image processing algorithm to calculate the proportion of residual dirt area. If the cleaning efficiency is lower than 95% or the proportion of residual dirt area is higher than 5%, the area is marked as a substandard cleaning area and a second cleaning is arranged.

[0035] S06. Intelligent management and recycling of water resources: The water quality detection sensor in the water spray system device monitors the water quality parameters in the water tank in real time, including pH, hardness, turbidity and organic matter content; when the water quality does not meet the cleaning requirements, the built-in water purification device is activated to treat the water quality through multi-stage filtration and ultraviolet sterilization; based on the area of ​​the cleaning area and the degree of pollution, the estimated water consumption is calculated and water resources are reasonably allocated to ensure maximum water use efficiency.

[0036] S07. Energy management and autonomous charging: The power management device monitors the battery power in real time and calculates the remaining working time; the solar charging device monitors the current light intensity and charging efficiency, and automatically adjusts the angle of the solar charging panel to keep it perpendicular to the sunlight at all times, maximizing energy collection efficiency; when the battery power is lower than 30%, the shortest path is planned based on the environmental map to go to a sunny area, and the solar charging panel is deployed for autonomous charging; when the lighting conditions are insufficient, a path is planned to go to a fixed charging station for wired charging.

[0037] S08. Multi-machine collaboration and task allocation: Establish a point-to-point communication network with other cleaning robots through the communication interface device to exchange location information, cleaning progress and resource status; divide the tasks of the entire photovoltaic power station area based on a distributed algorithm to avoid multiple machines cleaning duplicate areas; when a robot is low on power or malfunctions, its responsible area is automatically allocated to other robots; coordinate the operation rhythm of multiple machines through real-time communication to ensure maximum overall cleaning efficiency.

[0038] S09. Abnormal situation handling and self-protection: The intelligent navigation device is used to detect changes in the surrounding environment in real time, identify moving obstacles and perform obstacle avoidance operations; monitor wind speed data, and when the wind speed exceeds 10 m / s, automatically stop the cleaning operation, retract the solar charging panel, and enter the wind protection state; monitor temperature data, when the ambient temperature is below 0°C, start the water tank heating device to prevent freezing; when the ambient temperature is above 45°C, start the cooling system to prevent hardware overheating; monitor the surface temperature of the photovoltaic module to avoid thermal stress damage caused by cleaning operations when the temperature is too high.

[0039] S10. Cleaning data analysis and optimization strategy: record complete data of each cleaning operation, including cleaning area, pollution level, cleaning parameters, cleaning efficiency and resource consumption; analyze data patterns through machine learning algorithms, and establish a mapping relationship between different pollution types and optimal cleaning parameters; identify pollution patterns, predict high-incidence areas and time periods, and formulate preventive cleaning plans; automatically adjust cleaning strategies and operation times according to different seasons and climatic conditions to ensure maximum cleaning efficiency throughout the year; wherein, the dirt type identification parameters include dirt material hardness index, dirt particle size distribution and dirt adhesion strength; the dirt adhesion level is expressed as a comprehensive parameter of the adhesion strength and coverage area of ​​the dirt layer to the surface of the photovoltaic module; the water resource consumption is expressed in liters to represent the amount of water used in the cleaning process; the energy consumption is expressed in watt-hours to represent the electrical energy consumed in the cleaning process; the brush head wear degree is expressed as the percentage loss of the cleaning brush head service life caused by the cleaning process; the friction coefficient represents the friction characteristic parameter between the cleaning brush head and the surface of the photovoltaic module; the dirt hardness parameter represents the Mohs hardness grade of the dirt material.

[0040] The specific implementation of the above steps is described in detail below.

[0041] Step S01 is specifically implemented by using an intelligent navigation device to initialize the system and perform environmental perception. This step first activates multi-source data collection using lidar, millimeter-wave radar, ultrasonic sensors, and visual cameras. Point cloud fusion technology is then used to align and fuse the data collected by these sensors to construct a three-dimensional environmental map of the photovoltaic power station. During the map construction process, a semantic segmentation fully convolutional neural network is used to classify and label photovoltaic modules, pathways, obstacles, and other objects in the scene, marking the location, arrangement, and width of the cleaning pathways. Simultaneously, an environmental monitoring device collects data on current ambient temperature, humidity, wind speed, and light intensity to construct an environmental parameter assessment model. When the temperature is below 5°C or above 40°C, the wind speed exceeds 8 m / s, or the humidity is below 20% or above 85%, the system determines that the environment is unsuitable for cleaning operations. This step aims to obtain comprehensive information about the operating environment, providing foundational data for subsequent cleaning path planning and preventing equipment damage or poor cleaning results from cleaning operations performed under unsuitable conditions.

[0042] The specific implementation of step S02 involves assessing the contamination status of photovoltaic modules. This step uses the dirt detection sensors in the automated cleaning device to scan the surface of the photovoltaic modules, collecting surface reflectance, transmittance, and scattering data at 10-centimeter intervals to ensure full coverage of the module surface. Simultaneously, a visual camera captures high-definition images with a resolution of at least 1080×720 pixels. A contamination detection convolutional neural network analyzes the captured images. A feature extraction layer identifies the characteristics of different types of contaminants, including dust, bird droppings, leaves, and industrial fallout. Combined with the reflectance data, the system determines the degree of contamination. Based on the analysis results, the system categorizes the photovoltaic modules into heavily contaminated, moderately contaminated, and lightly contaminated areas. Severely contaminated areas have a reflectance reduction of more than 15%, moderately contaminated areas have a reflectance reduction of 5% to 15%, and lightly contaminated areas have a reflectance reduction of less than 5%. Finally, a contamination status heat map is generated, with different colors used to identify areas of varying contamination levels. The purpose of this step is to accurately assess the contamination status of the photovoltaic modules and provide a basis for the formulation of subsequent cleaning strategies.

[0043] The specific implementation of step S03 involves cleaning path planning and resource allocation. This step constructs a two-layer optimization model based on the PV panel contamination status heat map and the environmental map. The upper-layer optimization algorithm utilizes an improved ant colony algorithm combined with game theory to perform global cleaning path planning. The cleaning task is modeled as a weighted multi-objective traveling salesman problem, where the PV panel location is the node to be visited, the inter-node movement cost is the edge weight, the pollution level is the node priority weight, and the remaining battery charge is the path constraint. The ant colony population is initialized to 50, with 5 ants (α-ants) representing the globally optimal cleaning path, 10 ants (β-ants) representing the second-best cleaning path, and 15 ants (δ-ants) representing the third-best cleaning path. By introducing the Nash equilibrium strategy from game theory, the ants compete and cooperate with each other during the pheromone update process, avoiding local optimal solutions. The pheromone volatility coefficient is set to 0.75, and the number of iterations is set to 100. Simultaneously, an energy consumption prediction model is constructed to dynamically predict energy consumption based on path length, terrain complexity, and PV panel contamination level. When the predicted remaining battery charge falls below 30%, a charging route to the charging station is automatically planned. The lower-level optimization algorithm, based on a particle swarm optimization algorithm, is responsible for local resource allocation, optimizing specific cleaning parameters for each PV panel, including brush head speed, water spray volume, and dust extraction fan power. The particle swarm size is set to 30, the maximum number of iterations is 50, the inertia weight is 0.8, the individual learning factor is 1.5, and the social learning factor is 1.5. Based on an equipment wear assessment model, the impact of different cleaning parameters on equipment life is predicted, minimizing equipment wear while maintaining a cleaning efficiency of at least 95%. Precise allocation of cleaning resources is achieved using a set of resource allocation optimization equations, including equations for cleaning efficiency, water consumption, energy consumption, brush head life, and task completion time. By iteratively updating particle positions and velocities, the optimal cleaning parameter combination is searched for, achieving a multi-objective balance between cleaning efficiency, resource consumption, and equipment life. The cleaning path planning results of the upper-level algorithm serve as constraints for the lower-level algorithm, and the resource allocation results of the lower-level algorithm are fed back to the upper-level algorithm for path adjustment. The two-level algorithm iterates alternately until convergence, with the path variation being less than 1% for 10 consecutive iterations. The purpose of this step is to obtain the global optimal cleaning path and resource allocation plan under limited resources to maximize cleaning efficiency.

[0044] The specific implementation method of step S04 is to perform cleaning operations and real-time monitoring. This step first controls the robot body device to move to the target photovoltaic module position according to the planned path, with a moving speed of 0.5 meters per second and a positioning accuracy controlled within ±2 centimeters. After reaching the target position, the automatic cleaning device is started to automatically adjust the working parameters of the main cleaning brush head and the auxiliary cleaning brush head according to the pollution status. For heavily polluted areas, the speed of the main cleaning brush head is set to 120 revolutions per minute and the pressure is 100 Newtons per square meter; for moderately polluted areas, the speed of the main cleaning brush head is set to 100 revolutions per minute and the pressure is 80 Newtons per square meter; for lightly polluted areas, the speed of the main cleaning brush head is set to 80 revolutions per minute and the pressure is 60 Newtons per square meter. The water spraying system device is started synchronously, and the target area is sprayed with water in a direction through the array nozzle. The water flow and spraying angle are adjusted in real time according to the degree of pollution. For heavily polluted areas, the water spray volume is set to 120 ml / m2 and the water spray angle is 45°; for moderately polluted areas, the water spray volume is set to 80 ml / m2 and the water spray angle is 30°; for lightly polluted areas, the water spray volume is set to 50 ml / m2 and the water spray angle is 20°. The dust suction device is started to capture dust particles through dual dust suction ducts and electrostatic adsorption nets. The fan power is automatically adjusted according to the distribution density of dust particles, ranging from 50 to 150 watts. The cleaning effect is monitored in real time by visual sensors with a sampling frequency of 5 frames per second. If the cleaning effect does not meet the standard, the cleaning parameters are automatically adjusted and the cleaning operation is repeated. The purpose of this step is to take appropriate cleaning measures for different pollution conditions to ensure the best cleaning effect.

[0045] The specific implementation method of step S05 is to evaluate the cleaning effect and conduct quality control. After the cleaning operation is completed, the reflectivity sensor in the automatic cleaning device scans the surface of the photovoltaic module again with a scanning interval of 10 cm. The changes in reflectivity, transmittance and scattering rate before and after cleaning are compared to calculate the cleaning efficiency. The cleaning efficiency calculation formula is: Cleaning efficiency = (reflectivity after cleaning - reflectivity before cleaning) / (standard reflectivity - reflectivity before cleaning) × 100%. At the same time, a high-definition image of the cleaned photovoltaic module is collected by a visual camera with a resolution of not less than 1080×720 pixels, and the image processing algorithm is used to calculate the residual dirt area ratio. The residual dirt area ratio calculation formula is: Residual dirt area ratio = number of residual dirt pixels / total number of pixels × 100%. If the cleaning efficiency is lower than 95% or the residual dirt area ratio is higher than 5%, the area will be marked as a substandard cleaning area and a secondary cleaning will be arranged. The secondary cleaning will increase the cleaning brush head speed by 20% and the water spray volume by 25% to ensure that the cleaning effect meets the standard. The purpose of this step is to quantitatively evaluate the cleaning effect to ensure that dirt on the surface of photovoltaic modules is effectively removed and to improve photovoltaic power generation efficiency.

[0046] The specific implementation of step S06 involves intelligent water resource management and recycling. This step uses water quality sensors within the sprinkler system to monitor water quality parameters within the water tank in real time, including pH, hardness, turbidity, and organic matter content, with a sampling frequency of once per minute. Water quality assessment criteria are: pH should be between 6.5 and 7.5, hardness should be less than 120 mg / L, turbidity should be less than 5 NTU, and organic matter content should be less than 10 mg / L. If water quality parameters exceed these ranges, the built-in water purification device is activated, processing the water through a multi-stage filtration system. This multi-stage filtration system includes a primary mechanical filter to remove large particles, a secondary activated carbon filter to absorb organic matter, a high-grade ion exchange resin to soften the water, and finally, a UV sterilizer for disinfection. A water consumption prediction model is developed based on the size of the clean area and the degree of pollution, calculating estimated water consumption. For heavily polluted areas, the estimated water consumption is 120 ml / m2; for moderately polluted areas, it is 80 ml / m2; and for lightly polluted areas, it is 50 ml / m2. The system allocates water resources based on a comparison of estimated water usage and the remaining water in the tank. When the remaining water in the tank is less than 1.2 times the estimated water usage, the system prioritizes cleaning lightly polluted areas to maximize water efficiency. This step aims to achieve intelligent water management and recycling, reduce water waste, and improve the sustainability of cleaning operations.

[0047] The specific implementation of step S07 is to perform energy management and autonomous charging. This step monitors the battery power in real time through the power management device, with a sampling frequency of 1 time per minute, and calculates the remaining working time. The remaining working time is calculated based on the current power, average power consumption and working mode, and a weighted average algorithm is used to consider historical energy consumption data. The current light intensity and charging efficiency are monitored by the solar charging device, and the sensitivity of the light intensity sensor is 10 to 2000W / m 2 , resolution is 1W / m 2 The system automatically adjusts the angle of the solar charging panel to keep it perpendicular to the sunlight at all times. It uses a dual-axis tracking system with a horizontal rotation range of 0° to 360° and a vertical pitch range of 0° to 90°. The angle adjustment accuracy is ±1° to maximize energy collection efficiency. When the battery power is less than 30%, the system plans the shortest path to the sunny area based on the environmental map and deploys the solar charging panel for autonomous charging. If the current light intensity is less than 200W / m 2 If the system determines that the lighting conditions are insufficient, it will plan a route to the nearest fixed charging station for wired charging at 500 watts. Once the battery reaches 80%, the robot can resume its cleaning mission. This step aims to achieve intelligent energy management, ensuring the robot can continue working and avoiding mission interruptions due to insufficient power.

[0048] The specific implementation of step S08 is to achieve multi-machine collaboration and task allocation. This step establishes a point-to-point communication network based on a wireless local area network (WLAN) via a communication interface device. The communication frequency is 2.4 GHz, the communication range is 100 meters, and the data transmission rate is 54 Mbps. Robots exchange location information, cleaning progress, and resource status data, with data updated every 10 seconds. Based on an improved K-means clustering algorithm, tasks are divided across the entire photovoltaic power station area. Areas are first categorized into high, medium, and low pollution levels, and then further divided based on the number and location of robots to ensure balanced distribution of tasks. A market-based task allocation algorithm is employed, with each robot bidding for tasks based on its own status. Bidding factors include distance from the target area, battery level, water level, and current load. If a robot's battery level drops below 20%, water level drops below 15%, or a malfunction occurs, the system automatically triggers a task reallocation process, allocating its assigned area to other robots based on the principle of minimum increment. Real-time communication coordinates the work of multiple robots, using a time-window-based conflict avoidance algorithm. When two robots' predicted trajectories intersect, the system automatically adjusts the speed or path of one robot to ensure maximum efficiency. This step aims to improve overall cleaning efficiency through multi-robot collaboration, reduce individual machine operating time, and increase the cleaning coverage of the photovoltaic power plant.

[0049] The specific implementation of step S09 is to handle abnormal situations and achieve self-protection. This step uses an intelligent navigation device to detect changes in the surrounding environment in real time. It then uses an improved Kalman filter algorithm to fuse multi-sensor data, identify moving obstacles, and predict their trajectory. When a moving obstacle is detected, the system automatically calculates a safe distance and generates an obstacle avoidance path, with obstacle avoidance taking priority over cleaning tasks. The system continuously monitors wind speed data, sampling once per second. When wind speeds exceed 10 m / s for more than 10 seconds, the system automatically stops cleaning, retracts the solar panels, and enters windbreak protection mode. Temperature data is also monitored, sampling once per minute. When the ambient temperature falls below 0°C, the water tank heater activates at 200 watts to prevent freezing and maintain the water temperature above 5°C. When the ambient temperature exceeds 45°C, the cooling system activates to prevent hardware overheating. The cooling system includes an air-cooled radiator and a thermal management control module to ensure that the temperature of key electronic components does not exceed 65°C. The surface temperature of the PV panels is monitored using an infrared temperature sensor with a measurement range of -20°C to 100°C and an accuracy of ±0.5°C. When the surface temperature of the PV panels exceeds 60°C, the system automatically suspends cleaning operations to prevent thermal stress damage. This step ensures that the robot can take appropriate protective measures in various abnormal situations, extending the equipment's service life and preventing safety accidents.

[0050] The specific implementation of step S10 involves performing cleaning data analysis and optimizing strategies. This step records complete data for each cleaning operation, including the coordinates of the cleaning area, pollution level index, cleaning parameter settings, cleaning efficiency results, and resource consumption. A random forest algorithm is used to analyze data patterns and construct a mapping relationship between different pollution types and optimal cleaning parameters. The random forest model consists of 100 decision trees with 15 feature dimensions, taking into account factors such as dirt type, adhesion level, ambient temperature, and humidity. Time series analysis is used to identify pollution patterns, and an ARIMA model is used to predict high-pollution areas and time periods with a temporal granularity of days and a spatial granularity of 10 square meters. The prediction accuracy reaches over 85%. Based on the predicted results, a preventive cleaning plan is formulated, increasing the cleaning frequency in high-pollution areas, shifting from passive cleaning to active prevention. An adaptive cleaning strategy library is established based on different seasons and climatic conditions, focusing on cleaning pollen pollution in spring, bird droppings pollution in summer, fallen leaves pollution in autumn, and dust pollution in winter. The system also automatically adjusts the optimal operation time based on historical data, avoiding periods of condensation and strong sunlight to ensure maximum cleaning efficiency year-round. This step aims to continuously optimize cleaning strategies through data analysis, improve cleaning efficiency, extend the service life of PV modules, and ultimately increase the power generation efficiency of PV power plants.

[0051] The neural network model involved in the present invention mainly includes a convolutional neural network for dirt detection and a semantic segmentation full convolutional neural network for environmental perception. First, the dirt detection convolutional neural network adopts a residual network structure, which includes an initial convolution layer, 5 residual blocks and a final fully connected layer. The initial convolution layer uses 64 7×7 convolution kernels with a step size of 2, and the input is a 1080×720×3 RGB image. Each residual block consists of two 3×3 convolution layers and a shortcut connection. The number of convolution kernels is 64, 128, 256, 512 and 512 respectively. The first layer step size of the first three residual blocks is 2 for downsampling. The residual block is followed by a global average pooling layer, which compresses the feature map into a one-dimensional vector, and then connects two fully connected layers with 256 and 8 neurons respectively. The last layer uses a softmax activation function to output the probability distribution of dirt types (dust, bird droppings, leaves, industrial sediment, etc.), and at the same time predicts the degree of dirt adhesion through a parallel regression head. The network's training dataset consists of 10,000 images of dirt on the surfaces of various photovoltaic modules. These images were collected under varying lighting, weather, and pollution conditions, and were annotated by professionals using the dirt type and degree of adhesion. Data augmentation techniques include random rotation (±15°), horizontal flipping, brightness adjustment (±20%), contrast variation (±15%), and random cropping. Training uses mini-batch gradient descent with a batch size of 32, an initial learning rate of 0.001, and a learning rate decay strategy that reduces the learning rate to 90% of its original value every 10 epochs, for a total of 100 epochs. The cross-entropy loss function is used for classification, and the mean absolute error loss function is used for regression. The two are weighted in a 7:3 ratio to form the final loss function.

[0052] The fully convolutional neural network for context-aware semantic segmentation employs an encoder-decoder architecture. The encoder uses a pretrained VGG16 network backbone with the final fully connected layer removed. The decoder consists of five upsampling blocks, each consisting of a transposed convolutional layer, two standard convolutional layers, and a skip connection to recover detailed features from the corresponding encoder layer. The final output layer uses a 1×1 convolution to transform the feature map into a per-pixel class probability distribution (PV panel, path, obstacle, etc.) at the same resolution as the input image. The network's training dataset consists of 5,000 images of PV power plant scenes, comprised of multimodal data collected by LiDAR, millimeter-wave radar, and visual cameras. Each image is pixel-wise annotated, classifying the scene into categories such as PV panel, clean path, obstacle, and background. Training is performed using mini-batch stochastic gradient descent with a batch size of 8, an initial learning rate of 0.0001, and a cosine annealing learning rate scheduling strategy for 200 epochs. The loss function uses a combination of weighted cross entropy and Dice loss to address class imbalance, with the former weighted at 0.7 and the latter at 0.3. To improve model generalization, various regularization techniques were applied during training, including weight decay (with a coefficient of 0.0005), dropout (with a probability of 0.5), and batch normalization. Training progress for both neural network models was monitored through performance evaluation on a validation set, with an early stopping strategy used to prevent overfitting. Performance was ultimately validated on a test set with varying lighting conditions, soil types, and PV plant layouts to ensure robustness in real-world applications.

[0053] The mathematical model or calculation process involved in the present invention is described in detail below.

[0054] In step S03, cleaning path planning and resource allocation involve multiple calculation processes. First, the upper-level optimization algorithm adopts the improved ant colony algorithm, and its path selection probability calculation is specifically expressed as follows:

[0055]

[0056] Where, is the probability that the kth ant moves from node i to node j at time t; τ ij (t) is the pheromone concentration from node i to node j; d ij is the Euclidean distance from node i to node j; p j is the pollution degree weight of node j; α is the pheromone importance coefficient, ranging from 1 to 5; β is the heuristic information importance coefficient, ranging from 1 to 5; γ is the pollution degree weight coefficient, ranging from 1 to 3; allowed k is the set of nodes to be visited by the kth ant.

[0057] The pheromone update rule is combined with the Nash equilibrium strategy of game theory, which is specifically expressed as follows:

[0058]

[0059] Where, τ ij (t+1) is the updated pheromone concentration; ρ is the pheromone volatility coefficient, which is 0.75; τ ij (t) is the original pheromone concentration; The pheromone increment of elite ants (including α, β, and δ ants); is a game theory adjustment term. The pheromone increment of elite ants is calculated as follows:

[0060]

[0061] Where, ω α 、ω β 、ω δ are the weight coefficients of α, β, and δ ants, with values ​​of 0.5, 0.3, and 0.2 respectively; is the pheromone increment of the corresponding ant, and the calculation formula is:

[0062]

[0063] Where, Q is the pheromone intensity coefficient, which is 100; L α / β / δ is the path length of the corresponding ant.

[0064] Game Theory Adjustments The calculation of takes into account the competition and cooperation among ants, which is specifically expressed as follows:

[0065]

[0066] Where λ is the game adjustment coefficient, ranging from 0.1 to 0.5; m is the number of ants participating in the game; U k (s k , s -k ) is the utility function of the kth ant, indicating that under its own strategy s k and other ant strategies -k The following income; is an indicator function, which takes the value 1 when the edge (i, j) is on the path of the kth ant, and 0 otherwise. k (s k , s -k ) is calculated as follows:

[0067]

[0068] Where, L kis the path length of the kth ant; μ is the pollution degree weight coefficient, ranging from 0.1 to 1; p j is the pollution degree of node j; Path k is the path set of the kth ant; v is the path overlap penalty coefficient, ranging from 0.1 to 1; |Path k ∩Path l | represents the overlapping part of the paths of the kth ant and the lth ant; |Path k | represents the total length of the k-th ant's path.

[0069] The energy consumption prediction model calculation is specifically expressed as follows:

[0070] E pred =∑ (i，j)∈Path [e base ·d ij +e slope ·h ij +e clean ·p j ·a j +e turn ·θ ij ];

[0071] Where, E pred is the predicted energy consumption, in watt-hours; Path is the planned path; e base is the basic moving energy consumption coefficient, which is 0.05 watt-hours / m; d ij is the distance from node i to node j, in meters; e slope is the slope energy consumption coefficient, which is 0.02 watt-hours / m·degree; h ij is the slope from node i to node j, in degrees; e clean is the clean energy consumption coefficient, which is 0.1 watt-hour / square meter·pollution index; p j is the pollution degree of node j; a j is the photovoltaic module area corresponding to node j, in square meters; e turn is the steering energy consumption coefficient, which is 0.01 watt-hours / degree; θ ij is the steering angle from node i to node j, in degrees.

[0072] In the lower-level optimization algorithm, the position and velocity update equations of the particle swarm optimization algorithm are specifically expressed as follows:

[0073]

[0074]

[0075] Where, and are the d-th dimension velocity of the i-th particle in the t-th and t+1-th iterations respectively; and are the positions of the dth dimension of the i-th particle in the t-th and t+1-th iterations respectively; w is the inertia weight, which is 0.8; c1 is the individual learning factor, which is 1.5; c2 is the social learning factor, which is 1.5; r1 and r2 are random numbers between [0, 1]; p id is the historical optimal position of the i-th particle in the d-th dimension; p gd is the historical optimal position of the group in the dth dimension.

[0076] The cleaning efficiency equation in the resource allocation optimization equation group is specifically expressed as follows:

[0077]

[0078] Where, E clean is the cleaning efficiency, ranging from 0 to 1; k1 is the maximum cleaning efficiency coefficient, ranging from 0.9 to 1; k2 is the influence coefficient of the brush head speed, ranging from 0.01 to 0.1; R brush T is the speed of the cleaning brush head, in revolutions per minute; clean is the cleaning duration in minutes; k3 is the water spraying coefficient, ranging from 0.01 to 0.1; W spray is the water spraying volume, in milliliters per square meter; k4 is the dust collection power influence coefficient, with a value range of 0.01 to 0.1; P suction is the power of the dust suction fan, in watts; f type (D type ) is the dirt type influence function, and its value range is 0.6~1; D type Identify parameters for the type of fouling; f adhere (D adhere ) is the influence function of the degree of dirt adhesion, and its value range is 0.5~1; D adhere is the degree of dirt adhesion. The dirt type influence function and the dirt adhesion degree influence function are expressed as:

[0079]

[0080] Where a1, a2, a3, and a4 are the parameters of the function affecting the type of dirt, which are obtained by fitting the experimental data; b1, b2, b3, and b4 are the parameters of the function affecting the degree of dirt adhesion, which are obtained by fitting the experimental data.

[0081] The water resource consumption equation is specifically expressed as follows:

[0082] W consume =A panel ·C dirt Wbase ·g type (D type )·g adhere (D adhere )·T clean ;

[0083] Where W consume is the water consumption, in liters; A panel is the area of ​​PV modules, in square meters; C dirt is the dirt coverage rate, ranging from 0 to 1; W base is the basic water spray rate, in liters / square meter·minute; g type (D type ) is the function of the influence of dirt type on water consumption, ranging from 0.8 to 1.5; g adhere (D adhere ) is the function of the influence of the degree of dirt adhesion on water consumption, and its value range is 1 to 2; T cleam The cleaning duration is in minutes.

[0084] The energy consumption equation is specifically expressed as follows:

[0085]

[0086] Where, E consume is the energy consumption, in watt-hours; P brush is the brush head power coefficient, which is R brush is the speed of the cleaning brush head, in revolutions per minute; P spray is the power coefficient of the water spray system, which is 0.05 watt-minute / ml; W spray is the water spraying volume, in ml / min; P suction is the power coefficient of the dust collection system, which is P suction P is the power of the dust suction fan, in watts; move is the power coefficient of the mobile system, which is V robot is the robot's moving speed, in meters per minute; P base The basic power consumption is 5 watts.

[0087] The brush head life equation is specifically expressed as follows:

[0088]

[0089] Where, L wearis the degree of brush head wear, expressed as the percentage loss of cleaning brush head service life caused by the cleaning process; c1 is the basic wear coefficient, ranging from 0.001 to 0.01; R bursh is the cleaning brush head speed, in revolutions per minute; c2 is the speed influence index, ranging from 1.2 to 1.8; P pressure is the brush head pressure, in Newtons per square meter; c3 is the pressure impact index, ranging from 1.1 to 1.5; H dirt is the dirt hardness parameter, Mohs hardness grade; c4 is the hardness influence index, ranging from 0.8 to 1.2; μ is the friction coefficient; c5 is the friction influence index, ranging from 0.9 to 1.3; T clean The cleaning duration is in minutes.

[0090] The task completion time equation is specifically expressed as follows:

[0091]

[0092] Where, T complete is the estimated task completion time, in minutes; A panel is the area of ​​PV modules, in square meters; C dirt is the dirt coverage rate, ranging from 0 to 1; V clean is the cleaning speed coefficient; W brush k is the width of the cleaning brush head, in meters; T is the adhesion degree influence coefficient, ranging from 0 to 2; D adhere The degree of dirt adhesion.

[0093] Cleaning speed coefficient V clean The calculation is as follows:

[0094]

[0095] Where V base R is the basic cleaning speed, which is 0.1 m / s; brush R is the speed of the cleaning brush head, in revolutions per minute; ref is the reference speed, which is 100 rpm.

[0096] In step S05, the calculation formula for cleaning efficiency and residual dirt area ratio is given in the solution. The cleaning efficiency calculation formula is specifically expressed as follows:

[0097]

[0098] Where η clean is the cleaning efficiency, in %; R after is the reflectivity after cleaning; R before is the reflectivity before cleaning; Rstd The reflectivity is the standard reflectivity, which is the reflectivity value of the photovoltaic module when it leaves the factory. The reflectivity is measured using a reflectivity sensor with a measurement wavelength range of 300 to 1100 nanometers and a measurement accuracy of ±1%.

[0099] The calculation formula for the residual dirt area ratio is as follows:

[0100]

[0101] Where R remain is the ratio of residual dirt area, in %; N dirt is the number of residual dirt pixels; N total The identification of residual dirt pixels uses image segmentation algorithm to convert the image into HSV color space and extract the dirt area through color threshold and texture features.

[0102] In step S07, the remaining working time is calculated based on the current power, average power consumption and working mode, as shown below:

[0103]

[0104] Where, T remain is the remaining working time, in hours; C current is the current remaining battery capacity in ampere hours; V battery is the nominal voltage of the battery in volts; η discharge is the discharge efficiency, ranging from 0.85 to 0.95; P avg is the average power consumption in watts, calculated using the weighted average algorithm:

[0105] P avg =α·P current +(1-α)·P history ;

[0106] Where α is the current power consumption weight, ranging from 0.3 to 0.7; P current is the current power consumption in watts; P history is the historical average power consumption in watts.

[0107] In step S08, the task division is based on the improved K-means clustering algorithm, which is mathematically expressed as follows:

[0108]

[0109] Where J is the optimization objective function; k is the number of clusters, which is equal to the number of robots; S i is the i-th cluster; x is the region coordinate; μ iis the center of the ith cluster; w1 is the distance weight, ranging from 0.4 to 0.6; w2 is the pollution degree weight, ranging from 0.2 to 0.4; w3 is the workload weight, ranging from 0.1 to 0.3; D i is the average pollution level of the i-th cluster; L i is the workload of the i-th cluster, which is determined by the area and pollution level.

[0110] In the task allocation algorithm based on the market mechanism, the robot's bidding for the task is calculated as follows:

[0111]

[0112] Where Bid i (j) is the bid of the ith robot for task j; w d is the distance weight, ranging from 0.3 to 0.5; d ij is the normalized distance from robot i to task j; w e is the power weight, ranging from 0.2 to 0.4; e i is the normalized power of robot i; w w is the water weight, ranging from 0.1 to 0.3; w i is the normalized water volume of robot i; w l is the load weight, ranging from 0.1 to 0.2; l i is the normalized current load of robot i.

[0113] In step S10, the random forest algorithm is used to analyze the mapping relationship between different pollution types and optimal cleaning parameters, and its decision tree splitting criterion adopts the Gini index:

[0114]

[0115] Where Gini(D) is the Gini index of the dataset D; |y| is the number of categories; p i is the sample proportion of the i-th category. Time series analysis uses AR I MA model predicts high-pollution areas and time periods, AR I The expression of the MA model is:

[0116] φ(B)(1-B) d X t =θ(B)ε t ;

[0117] Where φ(B) is the autoregressive polynomial; B is the lag operator; d is the difference order; X t is the time series; θ(B) is the moving average polynomial; ε t is a white noise sequence.I The parameters of the MA model are obtained by fitting historical pollution data, and the maximum likelihood estimation method is used to determine the optimal parameter combination.

[0118] The design and selection of the above formulas take into account the requirements and constraints of actual application scenarios. The ant colony algorithm formula incorporates pollution level weights and game theory adjustments, making path planning more intelligent, prioritizing cleaning of heavily polluted areas while avoiding overlapping paths for multiple robots. The energy consumption prediction model comprehensively considers the impact of factors such as travel distance, terrain slope, cleaning task, and turning on energy consumption, improving the accuracy of energy consumption predictions and facilitating timely planning of charging routes. The resource allocation optimization equations utilize appropriate functional forms and parameter settings to accurately describe the impact of various factors on cleaning efficiency, resource consumption, and equipment lifespan, achieving multi-objective balanced optimization. The multi-machine collaborative task allocation algorithm considers multiple factors, including distance, resource status, and load, ensuring optimal task allocation. Appropriate machine learning and time series models are used for data analysis to uncover data patterns and optimize cleaning strategies. The application of these formulas results in higher cleaning efficiency, more efficient resource utilization, and longer equipment lifespan.

[0119] Specifically, the core technology of this invention lies in the construction of a complete closed-loop control system, encompassing environmental perception, pollution assessment, and cleaning execution. First, the system uses multi-sensor fusion technology to construct a three-dimensional environmental map of the photovoltaic power station, providing basic data for path planning. Second, a deep learning algorithm is used to assess the contamination status of the photovoltaic module surface. By analyzing reflectivity, transmittance, and scattering data, combined with high-definition imagery, the system identifies dirt type, distribution density, and adhesion level, generating a heat map of the pollution status and accurately quantifying the degree of contamination.

[0120] This invention innovatively proposes a two-tiered optimization model for cleaning path planning and resource allocation. The upper-tier optimization model models the cleaning task as a weighted, multi-objective traveling salesman problem. By combining an improved ant colony algorithm with game theory, it avoids falling into local optimal solutions while also considering energy consumption forecasts to ensure operational continuity. The lower-tier optimization, however, targets specific cleaning parameters and uses a set of resource allocation optimization equations to find the optimal balance between cleaning efficiency, resource consumption, and equipment lifespan. The two-tier algorithm interacts and iterates to achieve a globally optimal cleaning path and resource allocation solution.

[0121] At the execution level, the system precisely controls brush head speed, water spray volume, and suction power based on optimization results, achieving differentiated cleaning results. Real-time monitoring and quality assessment ensure that cleaning results meet standards. Functional modules such as intelligent water resource management and recycling, energy management and autonomous charging, and multi-machine collaboration and task allocation further support the system's efficient operation.

[0122] The logic behind this technical solution lies in: providing a decision-making basis through precise pollution status assessment, employing a two-tier optimization model to ensure a globally optimal solution, and ensuring closed-loop control through precise execution and real-time monitoring. Ultimately, this system achieves precise cleaning and optimized resource allocation based on pollution status. Information flows smoothly between the system's modules, forming a complete technical closed loop: environmental perception → status assessment → planning optimization → precise execution → effect evaluation → strategy adjustment, fundamentally resolving the core issues inherent in traditional technologies.

[0123] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0124] The specific implementation of steps S01-S02 in this embodiment is the same as above and will not be repeated here.

[0125] The specific implementation method of step S03 is to perform cleaning path planning and resource allocation. This step constructs a two-layer optimization model based on the thermal map of the pollution status of photovoltaic modules and the environmental map. The upper-level optimization algorithm adopts a combination of improved ant colony algorithm and game theory to be responsible for global cleaning path planning, and models the cleaning task as a weighted multi-objective traveling salesman problem, in which the location of the photovoltaic module is used as the node to be visited, the moving cost between nodes is used as the edge weight, the pollution level is used as the node priority weight, and the remaining battery power is used as the path constraint. The number of initialized ant colony populations is 50, of which 5 ants are α ants representing the global optimal cleaning path, 10 ants are β ants representing the suboptimal cleaning path, and 15 ants are δ ants representing the third-optimal cleaning path. The calculation formula for the ant's path selection probability is:

[0126]

[0127] Where, is the probability that the kth ant moves from node i to node j at time t; τ ij (t) is the pheromone concentration from node i to node j; d ij is the Euclidean distance from node i to node j; p j is the pollution degree weight of node j; α is the pheromone importance coefficient, ranging from 1 to 5; β is the heuristic information importance coefficient, ranging from 1 to 5; γ is the pollution degree weight coefficient, ranging from 1 to 3; allowed k is the set of nodes to be visited by the kth ant. By introducing the Nash equilibrium strategy from game theory, ants can compete and cooperate with each other during the pheromone update process, avoiding falling into local optimal solutions. The pheromone update rule combines the Nash equilibrium strategy from game theory and is specifically expressed as follows:

[0128]

[0129] Where, τ ij (t+1) is the updated pheromone concentration; ρ is the pheromone volatility coefficient, which is 0.75; τ ij (t) is the original pheromone concentration; Increase the pheromone of elite ants; is a game theory adjustment term. The pheromone increment of elite ants is calculated as follows:

[0130]

[0131] Where, ω α 、ω β 、ω δ are the weight coefficients of α, β, and δ ants, with values ​​of 0.5, 0.3, and 0.2 respectively; is the pheromone increment of the corresponding ant. The game theory adjustment term uses the utility function:

[0132]

[0133] Where, L k is the path length of the kth ant; μ is the pollution degree weight coefficient, ranging from 0.1 to 1; p j is the pollution degree of node j; Path k is the path set of the kth ant; ν is the path overlap penalty coefficient, which ranges from 0.1 to 1. At the same time, an energy consumption prediction model is constructed, and the energy consumption prediction calculation formula is:

[0134] E pred =∑ (i，j)∈Path [e base ·d ij +e slope ·h ij +e clean ·p j ·a j +e turn ·θ ij ];

[0135] Where, E pred is the predicted energy consumption, in watt-hours; Path is the planned path; e base is the basic moving energy consumption coefficient, which is 0.05 watt-hours / m; d ij is the distance from node i to node j, in meters; e slope is the slope energy consumption coefficient, which is 0.02 watt-hours / m·degree; h ij is the slope from node i to node j, in degrees; e clean is the clean energy consumption coefficient, which is 0.1 watt-hour / square meter·pollution index; p j is the pollution degree of node j; a jis the photovoltaic module area corresponding to node j, in square meters; e turn is the steering energy consumption coefficient, which is 0.01 watt-hours / degree; θ ij is the turning angle from node i to node j, in degrees. The lower optimization algorithm is based on the particle swarm algorithm to allocate local resources. The particle position and velocity update equations are:

[0136]

[0137] Where, and are the d-th dimension velocity of the i-th particle in the t-th and t+1-th iterations respectively; and are the positions of the dth dimension of the i-th particle in the t-th and t+1-th iterations respectively; w is the inertia weight, which is 0.8; c1 is the individual learning factor, which is 1.5; c2 is the social learning factor, which is 1.5; r1 and r2 are random numbers between [0, 1]; p id is the historical optimal position of the i-th particle in the d-th dimension; p gd is the historical optimal position of the group in the dth dimension. The cleaning efficiency equation in the resource allocation optimization equation group is:

[0138]

[0139] Where, E clean is the cleaning efficiency, ranging from 0 to 1; k1 is the maximum cleaning efficiency coefficient, ranging from 0.9 to 1; k2 is the influence coefficient of the brush head speed, ranging from 0.01 to 0.1; R brush T is the speed of the cleaning brush head, in revolutions per minute; clean is the cleaning duration in minutes; k3 is the water spraying coefficient, ranging from 0.01 to 0.1; W spray is the water spraying volume, in milliliters per square meter; k4 is the dust collection power influence coefficient, with a value range of 0.01 to 0.1; P suction is the power of the dust suction fan, in watts; f type (D type ) and f adhere (D adhere ) are the influence functions of dirt type and adhesion degree respectively. The water resource consumption equation is:

[0140] W consume =A panel ·C dirt W base ·g type (D type )·g adhere (D adhere )·Tclean ;

[0141] Where W consume is the water consumption, in liters; A panel is the area of ​​PV modules, in square meters; C dirt is the dirt coverage rate, ranging from 0 to 1; W base is the basic water spray rate, in liters / square meter·minute; g type (D type ) and g adhere (D adhere ) are the influence functions of dirt type and adhesion degree on water consumption. The energy consumption equation is:

[0142]

[0143] Where, E consume is the energy consumption, in watt-hours; P brush is the brush head power coefficient; P spray is the power coefficient of the sprinkler system; P suction is the power coefficient of the dust collection system; P move is the power coefficient of the mobile system; P base is the basic power consumption. The brush head life equation is:

[0144]

[0145] Where, L wear is the degree of brush head wear; c1 is the basic wear coefficient; P pressure H is the brush head pressure, in Newtons / square meters; dirt is the dirt hardness parameter; v is the friction coefficient; c2 to c5 are the influence indexes of each factor. The task completion time equation is:

[0146]

[0147] Where, T complete is the estimated task completion time in minutes; V clean is the cleaning speed coefficient; W brush k is the width of the cleaning brush head, in meters; T is the adhesion influence coefficient. The cleaning path planning results of the upper-level algorithm serve as constraints for the lower-level algorithm. The resource allocation results of the lower-level algorithm are fed back to the upper-level algorithm for path adjustment. The two-level algorithm iterates alternately until convergence. The convergence condition is that the path change is less than 1% after 10 consecutive iterations. The goal of this step is to obtain the globally optimal cleaning path and resource allocation solution to maximize cleaning efficiency under limited resources.

[0148] The specific implementation of step S04 in this embodiment is the same as that described above and will not be repeated here.

[0149] The specific implementation of step S05 is to evaluate the cleaning effect and perform quality control. After the cleaning operation is completed, the reflectivity sensor in the automatic cleaning device scans the surface of the photovoltaic module again at a scanning interval of 10 cm. The changes in reflectivity, transmittance, and scattering before and after cleaning are compared to calculate the cleaning efficiency. The cleaning efficiency calculation formula is:

[0150]

[0151] Where η clean is the cleaning efficiency, in %; R after is the reflectivity after cleaning; R before is the reflectivity before cleaning; R std The standard reflectivity is the reflectivity value of the photovoltaic module when it leaves the factory. At the same time, a visual camera is used to collect high-definition images of the cleaned photovoltaic module with a resolution of no less than 1080×720 pixels, and the image processing algorithm is used to calculate the residual dirt area ratio. The formula for calculating the residual dirt area ratio is:

[0152]

[0153] Where R remain is the ratio of residual dirt area, in %; N dirt is the number of residual dirt pixels; N total = total number of pixels. If the cleaning efficiency falls below 95% or the residual dirt area exceeds 5%, the area is marked as substandard and a second cleaning is scheduled. During the second cleaning, the brush speed is increased by 20% and the water volume is increased by 25% to ensure that the cleaning effect meets the standard. This step aims to quantitatively evaluate the cleaning effect and ensure that dirt on the PV module surface is effectively removed, thereby improving PV power generation efficiency.

[0154] The specific implementation of step S06 in this embodiment is the same as that described above and will not be repeated here.

[0155] The specific implementation of step S07 is to perform energy management and autonomous charging. This step uses the power management device to monitor the battery power in real time, with a sampling frequency of once per minute, and calculate the remaining working time. The remaining working time calculation formula is:

[0156]

[0157] Where, T remain is the remaining working time, in hours; C current is the current remaining battery capacity in ampere hours; V battery is the nominal voltage of the battery in volts; ηdischarge is the discharge efficiency, ranging from 0.85 to 0.95; P avg is the average power consumption in watts, calculated using the weighted average algorithm:

[0158] P avg =α·P current +(1-α)·P history ;

[0159] Where α is the current power consumption weight, ranging from 0.3 to 0.7; P current is the current power consumption in watts; P history The unit is watt. The current light intensity and charging efficiency are monitored by the solar charging device. The sensitivity of the light intensity sensor is 10-2000W / m 2 , resolution is 1W / m 2 The system automatically adjusts the angle of the solar charging panel to keep it perpendicular to the sunlight at all times. It uses a dual-axis tracking system with a horizontal rotation range of 0° to 360° and a vertical pitch range of 0° to 90°. The angle adjustment accuracy is ±1° to maximize energy collection efficiency. When the battery power is less than 30%, the system plans the shortest path to the sunny area based on the environmental map and deploys the solar charging panel for autonomous charging. If the current light intensity is less than 200W / m 2 If the system determines that the lighting conditions are insufficient, it will plan a route to the nearest fixed charging station for wired charging at 500 watts. Once the battery reaches 80%, the robot can resume its cleaning mission. This step aims to achieve intelligent energy management, ensuring the robot can continue working and avoiding mission interruptions due to insufficient power.

[0160] The specific implementation of step S08 is to achieve multi-machine collaboration and task allocation. This step establishes a point-to-point communication network based on a wireless local area network through a communication interface device. The communication frequency is 2.4GHz, the communication range is 100 meters, and the data transmission rate is 54Mbps. The robots exchange location information, cleaning progress, and resource status data, and the data is updated once every 10 seconds. The entire photovoltaic power station area is divided into tasks based on the improved K-means clustering algorithm, and the optimization objective function is:

[0161]

[0162] Where J is the optimization objective function; k is the number of clusters, which is equal to the number of robots; S i is the i-th cluster; x is the region coordinate; μ iis the center of the ith cluster; w1 is the distance weight, ranging from 0.4 to 0.6; w2 is the pollution degree weight, ranging from 0.2 to 0.4; w3 is the workload weight, ranging from 0.1 to 0.3; D i is the average pollution level of the i-th cluster; L i is the workload of cluster i, which is determined by the area and pollution level. A task allocation algorithm based on a market mechanism is used, and each robot bids for tasks according to its own status. The bidding calculation formula is:

[0163]

[0164] Where Bid i (j) is the bid of the ith robot for task j; w d is the distance weight, ranging from 0.3 to 0.5; d ij is the normalized distance from robot i to task j; w e is the power weight, ranging from 0.2 to 0.4; e i is the normalized power of robot i; w w is the water weight, ranging from 0.1 to 0.3; w i is the normalized water volume of robot i; w l is the load weight, ranging from 0.1 to 0.2; l i is the normalized current load of robot i. When a robot's battery level drops below 20%, its water level drops below 15%, or a fault occurs, the system automatically triggers task reallocation, assigning its assigned area to another robot based on the principle of minimum increment. Real-time communication coordinates the work pace of multiple robots, and a time-window-based conflict avoidance algorithm is employed. When two robots' predicted trajectories intersect, the system automatically adjusts the speed or path of one robot to ensure maximum efficiency. This step aims to improve overall cleaning efficiency through multi-robot collaboration, reduce individual robot operation time, and increase the cleaning coverage of the photovoltaic power plant.

[0165] The specific implementation of steps S09-S10 in this embodiment is the same as above and will not be repeated here.

[0166] In addition, such as Figure 2-Figure 5 As shown, in this embodiment, other components of the robot are described in detail as follows.

[0167] Robot body device: Made of high-strength, lightweight polycarbonate and nano-titanium dioxide composite materials, with self-cleaning properties; multiple independent and interconnected functional partitions are constructed internally; equipped with a composite mobile mechanism with four-wheel independent drive and tracked assisted movement, each of which is driven by an independent high-precision brushless motor, each with a power of 200 watts and a speed range of 0 to 2000 rpm; the tracked assisted movement device is hidden at the bottom of the robot and is controlled by a hydraulic system to unfold and fold; the posture sensor collects the robot's inclination, acceleration and angular velocity data in real time to ensure stability on uneven terrain.

[0168] Automatic cleaning device: installed at the bottom of the robot body, it includes a main cleaning brush head and an auxiliary cleaning brush head; the main cleaning brush head is disc-shaped, with a diameter of 25 cm, and has multiple layers of cleaning bristles of different hardness distributed on the surface; the auxiliary cleaning brush head is arranged around the main cleaning brush head, in a long strip shape, and can be flexibly adjusted in angle; the two brush heads are connected by a combination of magnetic clips and mechanical locks; the dual-drive motor provides power, and different speeds and torque outputs are achieved through a planetary gear reducer; the built-in pressure sensor monitors the contact pressure between the brush head and the surface of the photovoltaic module to prevent excessive wiping from damaging the photovoltaic module; the dirt detection sensor identifies the type of dirt and the degree of adhesion through multi-wavelength spectral analysis.

[0169] Dust collection device: adopts a dual dust collection duct design, the main dust collection duct is responsible for collecting large particles of dust and debris, and the auxiliary dust collection duct is responsible for absorbing fine dust; the high-efficiency centrifugal fan provides strong suction, and the power can be adjusted within the range of 100 to 200 watts; the dust collection box adopts a dual filtration method combining cyclone separation and filter filtration, and the filter filtration accuracy reaches 0.1 microns; an electrostatic adsorption net is installed in the air duct, which is made of special conductive fiber material. When powered on, it generates an electrostatic field to enhance the dust collection effect; the dust particle counter monitors the number and size distribution of inhaled dust particles in real time to evaluate the cleaning effect.

[0170] Water spray system device: adopts a flexible foldable water tank design, made of high-strength and chemical-resistant rubber composite material, with a water storage capacity of 8 liters; the intelligent variable frequency water pump can automatically adjust the water flow according to cleaning needs, with an adjustment range of 0.3 to 2.5 liters / minute; the array nozzles are distributed in front of the cleaning brush head, and each nozzle can independently control the water spray angle (30° to 150°) and water spray volume; the water quality detection sensor monitors the water quality parameters in real time; the water pressure sensor monitors the pipeline pressure to prevent blockage or leakage; the water level sensor monitors the remaining water in the water tank; the built-in water purification device processes the recycled water through multi-stage filtration and ultraviolet sterilization technology.

[0171] Solar charging device: It adopts a foldable and angle-adjustable solar charging panel assembly composed of high-efficiency perovskite solar cells with a photoelectric conversion efficiency of more than 25%; the surface of the charging panel is covered with a nano-coating with self-repairing and anti-reflection functions, composed of silicone polymer and nano-silicon dioxide; the electric push rod controls the angle of the charging panel, and the light sensor detects the position of the sun in real time; the charging controller adopts maximum power point tracking technology; the battery uses a new solid-state lithium battery with an energy density of 450 watt-hours / kilogram; current and voltage sensors monitor the charge and discharge status; the temperature sensor monitors the temperature of the battery pack to prevent overcharging and over-discharging.

[0172] Intelligent navigation device: integrates lidar, millimeter-wave radar, ultrasonic sensor and visual camera; lidar has a ranging accuracy of ±2 cm, a scanning range of 360° and a scanning frequency of 10 Hz; millimeter-wave radar is used to monitor moving obstacles with a detection range of 0 to 100 meters; ultrasonic sensor is used for close-range precise ranging with a range of 0 to 2 meters; high-definition fisheye camera provides 1080P resolution panoramic images; inertial measurement unit provides acceleration, angular velocity and direction data; high-precision global positioning system receiver provides centimeter-level positioning accuracy; edge computing module processes sensor data in real time to build a three-dimensional environment map.

[0173] Environmental monitoring device: The built-in temperature sensor measures the ambient temperature in the range of -40℃ to 85℃ with an accuracy of ±0.5℃; the humidity sensor measures the relative humidity in the range of 0% to 100% with an accuracy of ±3%; the wind speed sensor measures the wind speed in the range of 0 to 60 m / s with an accuracy of ±5%; the light intensity sensor measures the light intensity in the range of 0 to 200,000 lux; the air pressure sensor measures the atmospheric pressure; the raindrop sensor detects precipitation; the photovoltaic module surface temperature sensor measures the module surface temperature using infrared temperature measurement technology.

[0174] Communication interface device: supports 5G and Wi-Fi dual-mode communication; the 5G communication module supports a maximum downlink rate of 1.5 Gbit / s and an uplink rate of 500 Mbit / s; the Wi-Fi module supports the IEEE 802.11ax standard with a transmission rate of up to 3.5 Gbit / s; the low-power Bluetooth module is used for short-range communication; the LoRa long-range communication module is used to maintain connections in areas with poor signal coverage; the antenna array uses MIMO technology to improve communication reliability; the encryption chip ensures data transmission security; the edge server provides local computing and storage capabilities.

[0175] Power management device: monitors battery voltage, current, temperature and remaining capacity; intelligent power distribution system dynamically distributes power according to the needs of each functional module; overcharge protection circuit prevents battery overcharging; over-discharge protection circuit prevents battery over-discharge; short-circuit protection circuit prevents short-circuit damage; temperature protection circuit monitors battery temperature; power conversion efficiency exceeds 95%; power status indication system displays the current power status through LED lights; fault diagnosis system monitors power system abnormalities in real time.

[0176] Control chip: Adopts a high-performance multi-core processor with a main frequency of 2.5 GHz, including 8 high-performance cores and 4 energy-efficient cores; built-in neural network processing unit with a computing power of 15 trillion operations per second; memory capacity is 16 GB and storage capacity is 512 GB; supports real-time operating system; integrates a high-precision real-time clock; supports multi-threaded parallel processing; low-power design, dynamic power consumption adjustment; watchdog timer ensures stable system operation; redundant design improves system reliability.

[0177] It should be noted that the variables involved in the present invention are explained in detail in Table 1(a) and Table 1(b) below.

[0178] Table 1(a) Variable Explanation Table

[0179]

[0180]

[0181] Table 1(b) Variable Explanation Table

[0182] <![CDATA[P brush :Brush head power coefficient]]> <![CDATA[P spray :Sprinkler system power coefficient]]> <![CDATA[P move :Mobile System Power Factor]]> <![CDATA[V robot :Robot moving speed<!-- 19 --> ]]> <![CDATA[P base :Basic Power Consumption]]> <![CDATA[L wear :Brush head wear degree]]> <![CDATA[c1: Basic wear coefficient]]> <![CDATA[P pressure :Brush head pressure]]> <![CDATA[H dirt :Dirt Hardness Parameter]]> μ: Friction coefficient <![CDATA[c2, c3, c4, c5: Influence indices of various factors]]> <![CDATA[T complete :Estimated task completion time]]> <![CDATA[V clean :Cleaning speed factor]]> <![CDATA[W brush :Cleansing brush head width]]> <![CDATA[k T :Adhesion degree influence coefficient]]> <![CDATA[η clean :Cleaning Efficiency]]> <![CDATA[R after :Reflectivity after cleaning]]> <![CDATA[R before :Reflectivity before cleaning]]> <![CDATA[R std :Standard reflectivity]]> <![CDATA[R remain :Ratio of residual dirt area]]> <![CDATA[N dirt :Number of pixels with residual dirt]]> <![CDATA[N total :Total number of pixels]]> <![CDATA[T remain :Remaining working time]]> <![CDATA[C current :Current remaining battery capacity]]> <![CDATA[V battery :Battery nominal voltage]]> <![CDATA[η discharge :Discharge efficiency]]> <![CDATA[P avg :Average power consumption]]> α: current power consumption weight <![CDATA[P current :Current power consumption]]> <![CDATA[P history :Historical average power consumption]]> J: Optimize objective function k: number of clusters <![CDATA[S i :i-th cluster]]> x: area coordinates <![CDATA[μ i :The center of the i-th cluster]]> <![CDATA[w1: Distance weight]]> <![CDATA[w2: Pollution degree weight]]> <![CDATA[w3: Workload weight]]> <![CDATA[D i : The average pollution level of the i-th cluster]]> <![CDATA[L i : Workload of cluster i]]> <![CDATA[Bid i (j): The bid of the i-th robot for task j]]> <![CDATA[w d :Distance Weight]]> <![CDATA[d ij : Normalized distance from robot i to task j]]> <![CDATA[w e :Power Weight]]> <![CDATA[e i :Normalized power of robot i]]> <![CDATA[w w :Water weight]]> <![CDATA[w i :Normalized water volume of robot i]]> <![CDATA[w l :Load Weight]]> <![CDATA[l i : Normalized current load of robot i]]> Gini(D): Gini index of data set D |y|: number of categories <![CDATA[p i : The proportion of samples in the i-th category]]> φ(B): Autoregressive polynomial B: lag operator d: difference order <![CDATA[X t :Time Series]]> θ(B): Moving average polynomial <![CDATA[ε t :white noise sequence]]>

[0183] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A multifunctional integrated cleaning robot system for photovoltaic modules in power plants, comprising a control chip, a robot body, an automatic cleaning device, a dust collection device, a water spray system, a solar charging device, an intelligent navigation device, an environmental monitoring device, a communication interface device, and a power management device. The control chip is electrically connected to the other devices and is provided with an intelligent cleaning control module. The system is characterized in that: The intelligent cleaning control module is used to perform the following steps: system initialization and environmental perception; photovoltaic module pollution status assessment; cleaning path planning and resource allocation, based on the photovoltaic module pollution status heat map and environmental map, a two-layer optimization model is constructed for cleaning path planning and resource allocation; cleaning operation execution and real-time monitoring; cleaning effect evaluation and quality control; intelligent water resource management and recycling; energy management and autonomous charging; multi-machine collaboration and task allocation; abnormal situation handling and self-protection; cleaning data analysis and optimization strategy.

2. The multifunctional integrated cleaning robot system according to claim 1, characterized in that: The system initialization and environmental perception steps include: using the intelligent navigation device to activate the laser radar, millimeter wave radar, ultrasonic sensor and visual camera to collect multi-source data, construct a three-dimensional environmental map of the photovoltaic power station, and mark the location, arrangement and cleaning channel width of the photovoltaic modules; at the same time, the environmental monitoring device collects the current environmental temperature, humidity, wind speed and light intensity data to evaluate whether the current environment is suitable for cleaning operations.

3. The multifunctional integrated cleaning robot system according to claim 2, characterized in that: The photovoltaic module pollution status assessment step includes: using the dirt detection sensor in the automatic cleaning device to scan the surface of the photovoltaic module, collecting surface reflectivity, transmittance and scattering rate data, combining with the high-definition images collected by the visual camera, and analyzing the type, distribution density and adhesion degree of dirt on the surface of the photovoltaic module through a deep learning algorithm; based on the analysis results, the photovoltaic module is divided into heavily polluted areas, moderately polluted areas and lightly polluted areas to generate a pollution status heat map.

4. The multifunctional integrated cleaning robot system according to claim 3, characterized in that: The two-layer optimization model includes: an upper-layer optimization algorithm that combines an improved ant colony algorithm with game theory is responsible for global cleaning path planning, and a lower-layer optimization algorithm based on a particle swarm algorithm is responsible for local resource allocation; the two-layer algorithms interactively iterate until convergence to obtain the globally optimal cleaning path and resource allocation plan.

5. The multifunctional integrated cleaning robot system according to claim 4, characterized in that: In the upper-level optimization algorithm, the cleaning task is modeled as a weighted multi-objective traveling salesman problem, where the location of the photovoltaic modules is used as the node to be visited, the cost of moving between nodes is used as the edge weight, the degree of pollution is used as the node priority weight, and the remaining battery power is used as the path constraint. By introducing the Nash equilibrium strategy from game theory, ants can compete and cooperate with each other during the pheromone update process to avoid falling into local optimal solutions.

6. The multifunctional integrated cleaning robot system according to claim 5, characterized in that: In the lower-level optimization algorithm, the specific cleaning parameters of each photovoltaic module are optimized, including the cleaning brush head speed, water spray volume and dust suction fan power; the resource allocation optimization equation group is used to achieve accurate allocation of cleaning resources; the resource allocation optimization equation group includes the cleaning efficiency equation, water resource consumption equation, energy consumption equation, brush head life equation and task completion time equation.

7. The multifunctional integrated cleaning robot system according to claim 6, characterized in that: The steps of executing and real-time monitoring the cleaning operation include: controlling the robot body device to move to the target photovoltaic module position, starting the automatic cleaning device, and adjusting the working parameters of the main cleaning brush head and the auxiliary cleaning brush head according to the pollution status; synchronously starting the water spray system device and the dust suction device; and monitoring the cleaning effect in real time through a visual sensor. If the cleaning effect does not meet the standard, the cleaning parameters are automatically adjusted and the cleaning operation is repeated.

8. The multifunctional integrated cleaning robot system according to claim 7, characterized in that: The cleaning effect evaluation and quality control steps include: after the cleaning operation is completed, the reflectivity sensor in the automatic cleaning device is used to scan the surface of the photovoltaic module again, and the changes in reflectivity, transmittance and scattering rate before and after cleaning are compared to calculate the cleaning efficiency; at the same time, a high-definition image of the photovoltaic module after cleaning is collected by a visual camera, and the residual dirt area ratio is calculated using an image processing algorithm; if the cleaning efficiency is lower than 95% or the residual dirt area ratio is higher than 5%, the area is marked as a substandard cleaning area and a second cleaning is arranged.

9. The multifunctional integrated cleaning robot system according to claim 8, characterized in that: The robot body device is used to serve as the basic carrier of the entire system and provide a composite mobile mechanism; the automatic cleaning device is used to collect dirt type data on the surface of the photovoltaic module and perform physical cleaning operations; the dust suction device is used to collect dust particle distribution data and perform dust suction operations; the water spray system device is used to collect water quality parameters and perform water spray cleaning operations; the solar charging device is used to collect light intensity and battery charging status data; the intelligent navigation device is used to collect environmental map data and obstacle location information; the environmental monitoring device is used to collect temperature, humidity, wind speed and light intensity data; the communication interface device is used to realize data interaction with the remote monitoring center and multi-machine collaborative operation; the power management device is used to monitor the battery voltage and control the system power supply.

10. The multifunctional integrated cleaning robot system according to claim 9, characterized in that: The cleaning efficiency equation is used to calculate the cleaning efficiency under specific cleaning parameters. The input includes the degree of dirt adhesion, dirt type identification parameters, cleaning brush head rotation speed, water spray volume and dust suction fan power, and the output is the expected cleaning efficiency; the water resource consumption equation is used to calculate the water resource consumption during the cleaning process; the energy consumption equation is used to calculate the energy consumption during the cleaning process; the brush head life equation is used to evaluate the impact of cleaning parameters on the life of the brush head; the task completion time equation is used to predict the time required to complete the cleaning task.

Images