A gantry type synchronous double-brush head photovoltaic panel intelligent cleaning robot and a cleaning method

Abstract

This invention discloses a gantry-type synchronous dual-brush head intelligent cleaning robot for photovoltaic panels and a cleaning method, belonging to the field of photovoltaic panel cleaning technology. It includes a gantry-type frame, a main control box, an STM32 main control board, an AI intelligent control unit, and a storage unit. The gantry-type frame spans above the photovoltaic panels, with a linear guide rail at its top and drive wheels rotatably mounted at its bottom. A walking motor is fixedly mounted at the top of the gantry-type frame, driving the drive wheels to move the frame across the photovoltaic panels. The main control box is fixed to the gantry-type frame, and the STM32 main control board, AI intelligent control unit, and storage unit are all integrated inside the main control box. A moving component is also located on the gantry-type frame. The robot uses AI to identify stains in different zones and performs differentiated cleaning, improving cleaning efficiency and reducing energy consumption. It utilizes reinforcement learning to adapt to complex outdoor conditions, reducing slippage and jamming. Simultaneously, it achieves autonomous closed-loop maintenance of hot spots without human intervention, making it more intelligent overall.

Description

Technical Field

[0001] This invention relates to an intelligent cleaning robot, specifically a gantry-type synchronous dual-brush head intelligent cleaning robot for photovoltaic panels and a cleaning method thereof, belonging to the field of photovoltaic panel cleaning technology. Background Technology

[0002] Solar photovoltaic panels are exposed to the outdoor environment for a long time, and their surfaces are easily contaminated with pollutants such as dust, mud, bird droppings, and scale. This not only significantly reduces the power generation efficiency of photovoltaic modules, but also easily causes local hot spot effects, accelerates module aging, and even causes irreversible damage. Therefore, regular and efficient cleaning is required. Existing photovoltaic cleaning equipment mainly consists of gantry-type or cross-rail mobile robots, which mostly adopt a synchronous belt drive dual-brush head structure. The cleaning operation is achieved by relying on the walking motor to drive the whole machine to move and the brush head motor to drive the disc brush to rotate. This type of equipment is widely used in the automated operation and maintenance of photovoltaic power plants.

[0003] Existing synchronous dual-brush photovoltaic cleaning robots have several shortcomings in practical applications. First, they lack the ability to detect stains in different zones, relying solely on a uniform cleaning strategy. This fails to identify differences in stain type and severity between the left and right areas of the photovoltaic panel, leading to incomplete cleaning of heavily soiled areas and over-cleaning of lightly soiled areas. Second, they lack sensorless warp adaptation, relying mostly on fixed parameter control. This prevents them from automatically recognizing panel warp and brush head adhesion based on brush head load changes, resulting in cleaning blind spots due to loose adhesion or damage to the photovoltaic panel coating due to excessive tight adhesion. Third, they have poor adaptability to complex outdoor conditions. In slippery, windy, sandy, resonant, and jamming conditions, traditional PID control struggles to achieve dynamic adjustment, leading to issues such as robot slippage, resonance damage, and mechanical jamming. Fourth, the hot spot maintenance process is fragmented, relying on drone inspections and manual cleaning intervention. This fails to achieve integrated closed-loop maintenance encompassing hot spot alarms, autonomous targeted cleaning, and power generation data verification, resulting in a low level of automation and intelligence. Summary of the Invention

[0004] The purpose of this invention is to provide a gantry-type synchronous dual-brush head photovoltaic panel intelligent cleaning robot and cleaning method to solve the above problems. It adopts AI to identify stains in different areas and clean them differently, thereby improving the cleaning effect and reducing energy consumption. It uses reinforcement learning to adapt to complex outdoor working conditions and reduce slippage and jamming. At the same time, it realizes autonomous closed-loop operation and maintenance of hot spots without human intervention, making it more intelligent overall.

[0005] The present invention achieves the above objectives through the following technical solution: a gantry-type synchronous dual-brush head photovoltaic panel intelligent cleaning robot and cleaning method, comprising a gantry-type frame, a main control box, an STM32 main control board, an AI intelligent control unit, and a storage unit; The gantry frame is straddling the photovoltaic panel. A linear guide rail is provided at the top of the gantry frame, and a drive wheel is rotatably installed at the bottom of the gantry frame. A walking motor is fixedly installed at the top of the gantry frame, and the walking motor drives the drive wheel to move the frame on the photovoltaic panel. The main control box is fixed on the gantry frame, and the STM32 main control board, AI intelligent control unit and storage unit are all integrated inside the main control box; A movable component is mounted on the gantry frame. The dual-brush head cleaning assembly is connected to the moving assembly. Driven by the moving assembly, the dual-brush head cleaning assembly cleans the photovoltaic panel. The storage unit is used to store convolutional neural network, Transformer model, DDPG reinforcement learning model, baseline load parameters, stain feature library and cleaning strategy parameters; The AI ​​intelligent control unit includes an adaptive leveling module, an AI zoned stain recognition module, a hot spot targeted maintenance module, and a reinforcement learning adaptive operating condition module.

[0006] Preferably, the moving component includes a drive motor, a ball screw, and a positioning plate. The drive motor is positioned on the gantry frame, the ball screw is connected to the drive motor, the positioning plate is threaded to the ball screw, and one end of the positioning plate is slidably supported on the linear guide rail. The drive motor drives the positioning plate to move laterally along the linear guide rail through the ball screw.

[0007] Preferably, the dual-brush head cleaning assembly includes a brush head motor and two nylon disc brushes. The brush head motor is fixed to a positioning plate, and the two nylon disc brushes are arranged side by side and are synchronously driven to rotate by the brush head motor through a synchronous pulley and a synchronous belt.

[0008] Preferably, a sampling resistor is connected in series in the brush head motor circuit. The STM32 main control board collects the voltage across the sampling resistor through the built-in ADC and calculates the real-time total load current of the brush head motor. The adaptive leveling module identifies the warping of the photovoltaic panel and the bonding state of the nylon disc brush based on the change in the total load of the brush head motor. When the load decreases, it is determined that the bonding is too loose, so the speed of the walking motor is increased. When the load increases, it is determined that the bonding is too tight, so the walking motor is stopped and an alarm is triggered. After the load returns to normal, the preset walking speed is restored.

[0009] Preferably, the AI ​​partition stain recognition module is based on convolutional neural network and Transformer model. It decouples and separates the load characteristics of the left and right nylon disc brushes from the total load current of the brush head motor. Combined with the real-time power generation data of each area of ​​the photovoltaic panel, it identifies the stain type and severity of the corresponding area. For dusty areas, it adopts a low-energy strategy of fast walking and single cleaning. For stubborn stain areas, it adopts a strategy of slow walking and repeated cleaning.

[0010] Preferably, the hot spot targeted maintenance module can receive hot spot alarm information sent by the photovoltaic string inverter, which includes the battery string number and location information. The STM32 main control board autonomously plans the walking path to the hot spot area based on the positioning information of the walking motor encoder, controls the dual brush head cleaning component to perform targeted enhanced cleaning on the target area, and obtains the current, voltage and power generation data of the photovoltaic panel before and after cleaning for comparison and verification. If the data does not return to normal, it sends an early warning information to the maintenance platform.

[0011] Preferably, the reinforcement learning adaptive working condition module has a built-in DDPG model, which identifies slippery, resonant, windy and sandy, and jamming working conditions based on the load of the walking motor, the load of the brush head motor, and the encoder deviation, and dynamically adjusts the walking speed, driving torque, and step distance to achieve anti-slip, resonance suppression, and fault self-recovery.

[0012] A gantry-type synchronous dual-brush head intelligent cleaning method for photovoltaic panels, comprising the following steps: The S1 and STM32 main control boards obtain the real-time total load current through the sampling resistor in the brush head motor circuit, and at the same time collect the encoder position signal of the walking motor and the power generation data of the photovoltaic panel area. S2. The adaptive leveling module identifies the bonding status between the nylon disc brush and the photovoltaic panel based on the change in the total load of the brush head motor, and accordingly executes actions such as increasing the walking motor speed, stopping alarm, or restoring normal speed. S3, the AI ​​partition stain recognition module calls convolutional neural network and Transformer model to decouple the independent load characteristics of the left and right brush heads from the total load current of the brush head motor, and combine the power generation data to identify the type and severity of stains. For floating dust areas, it adopts fast walking and single cleaning, and for stubborn stain areas, it adopts slow walking and repeated cleaning. S4. The hot spot targeted maintenance module receives hot spot alarms from the cascade inverter, autonomously goes to the target area to strengthen cleaning, compares the power generation data before and after cleaning to verify the effect, and reports an alarm if there is an abnormality. S5. The reinforcement learning adaptive working condition module uses the DDPG model to identify the working condition in real time based on the load of the walking motor, the load of the brush head motor, and the encoder deviation, and dynamically adjusts the walking parameters to achieve anti-slip, resonance suppression, and fault self-recovery. S6. Repeat the above steps to complete the intelligent cleaning and maintenance of the photovoltaic panel array.

[0013] The beneficial effects of this invention are as follows: By detecting the total load of the brush head motor through sampling resistance, and cooperating with the adaptive leveling module to identify the warping of the photovoltaic panel and the adhesion state of the nylon disc brush, cleaning blind spots and damage to the photovoltaic panel coating are avoided; the AI ​​partition stain recognition module can decouple the load characteristics of the left and right nylon disc brushes from the total load current, and combine power generation data to achieve partitioned differentiated cleaning, solving the problems of incomplete cleaning of heavy dirt and over-cleaning of light dirt; the reinforcement learning working condition adaptive module can identify complex working conditions such as outdoor wet and slippery conditions, wind and sand, and resonance, and dynamically adjust the walking parameters to avoid robot slipping, deviation and jamming; the hot spot targeted operation and maintenance module can receive alarms from the photovoltaic cascade inverter and autonomously complete targeted cleaning and effect verification, realizing closed-loop automated operation and maintenance of hot spots, and improving the intelligence, reliability and environmental adaptability of the cleaning robot as a whole. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0015] In the diagram: 1. Gantry frame; 2. Main control box; 3. Linear guide rail; 4. Drive wheel; 5. Walking motor; 6. Moving component; 601. Drive motor; 602. Ball screw; 603. Positioning plate; 7. Dual brush head cleaning component; 701. Brush head motor; 702. Nylon disc brush. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1As shown, a gantry-type synchronous dual-brush head photovoltaic panel intelligent cleaning robot and cleaning method include a gantry frame 1, a main control box 2, an STM32 main control board, an AI intelligent control unit, and a storage unit. The gantry frame 1 spans above the photovoltaic panel. A linear guide rail 3 is provided on the top of the gantry frame 1, and drive wheels 4 are rotatably installed on the bottom of the gantry frame 1. A walking motor 5 is fixedly installed on the top of the gantry frame 1. The walking motor 5 drives the drive wheels 4 to move the frame on the photovoltaic panel. The main control box 2 is fixed on the gantry frame 1. The STM32 main control board and AI intelligent control unit are also included. The unit and storage unit are both integrated inside the main control box 2; the moving component 6 is set on the gantry frame 1; the dual-brush head cleaning component 7 is connected to the moving component 6, and the dual-brush head cleaning component 7 cleans the photovoltaic panel under the drive of the moving component 6; the storage unit is used to store the convolutional neural network, Transformer model, DDPG reinforcement learning model, baseline load parameters, stain feature library and cleaning strategy parameters; the AI ​​intelligent control unit includes an adaptive leveling module, an AI zone stain recognition module, a hot spot targeted maintenance module and a reinforcement learning working condition adaptive module.

[0018] A gantry frame 1 is mounted above the photovoltaic panel. A walking motor 5 drives the frame to move. The dual-brush head cleaning assembly 7 performs cleaning operations under the drive of the moving assembly 6. The STM32 main control board obtains the total load current of the brush head motor 701 by collecting the sampling resistor voltage. Combined with the model and parameters in the storage unit, it realizes working condition identification and control adjustment to ensure stable and efficient cleaning operations.

[0019] The moving component 6 includes a drive motor 601, a ball screw 602, and a positioning plate 603. The drive motor 601 is positioned on the gantry frame 1. The ball screw 602 is connected to the drive motor 601. The positioning plate 603 is threadedly connected to the ball screw 602, and one end of the positioning plate 603 is slidably supported on the linear guide rail 3. The drive motor 601 drives the positioning plate 603 to move laterally along the linear guide rail 3 through the ball screw 602.

[0020] The drive motor 601 drives the positioning plate 603 to move laterally along the linear guide rail 3 via the ball screw 602, which enables the dual brush head cleaning assembly 7 to maintain high-precision and stable displacement, effectively reducing mechanical shaking and positioning deviation, and ensuring stable and reliable acquisition of the total load current of the brush head motor 701. This provides a precise mechanical motion basis for the adaptive leveling module, the AI ​​zone stain recognition module, and the reinforcement learning working condition adaptive module, ensuring that each module accurately executes its judgment and control actions on load, working condition, and stains.

[0021] The dual-brush head cleaning assembly 7 includes a brush head motor 701 and two nylon disc brushes 702. The brush head motor 701 is fixed to the positioning plate 603, and the two nylon disc brushes 702 are arranged side by side and are synchronously driven to rotate by the brush head motor 701 through a synchronous pulley and a synchronous belt.

[0022] The brush head motor 701 is fixed to the positioning plate 603, ensuring that the dual-brush head cleaning assembly 7 is firmly installed and operates without looseness. The two nylon disc brushes 702 are arranged side by side, providing uniform and complete coverage without dead corners. They are synchronously driven to rotate by the brush head motor 701 through the synchronous pulley and synchronous belt, ensuring that the left and right nylon disc brushes 702 rotate at the same speed and move synchronously. This makes the cleaning force on both sides balanced, and the resistance changes in the left and right areas can be accurately and completely reflected in the total load current of the brush head motor 701. This provides a stable and accurate data foundation for the subsequent AI partition stain recognition module to decouple and separate the independent load characteristics of the left and right nylon disc brushes 702.

[0023] A sampling resistor is connected in series in the brush head motor 701 circuit. The STM32 main control board collects the voltage across the sampling resistor through the built-in ADC and calculates the real-time total load current of the brush head motor 701. The adaptive leveling module identifies the warping of the photovoltaic panel and the bonding state of the nylon disc brush 702 based on the change in the total load of the brush head motor 701. When the load decreases, it is determined that the bonding is too loose, so the speed of the walking motor 5 is increased. When the load increases, it is determined that the bonding is too tight, so the walking motor 5 is controlled to stop and alarm. After the load returns to normal, the preset walking speed is restored.

[0024] The adaptive leveling module identifies the photovoltaic panel warping and the contact state of the nylon disc brush 702 based on the total load change of the brush head motor 701. It utilizes the characteristic that the physical resistance change generated by the contact between the nylon disc brush 702 and the photovoltaic panel surface is directly reflected in the load current of the brush head motor 701. This eliminates the need for additional pressure or distance sensors; the contact state can be determined simply by monitoring and analyzing the current and load data of the brush head motor 701. As the robot cleans along the planned path, the STM32 main control board continuously collects the real-time current of the brush head motor 701. Because the two nylon disc brushes 702 are driven by the same brush head motor 701 and contact the left and right areas of the photovoltaic panel respectively, the warping... When the surface of the photovoltaic panel is curved, the two nylon disc brushes 702 on the left and right will experience load changes one after another. The STM32 main control board, combined with the lateral position feedback from the encoder of the walking motor 5, analyzes the waveform and duration of the load change, and then compares the real-time current with the reference load of the brush head motor 701 to determine whether the adhesion is too tight or too loose, or whether it is suspended. After identifying the abnormality, the STM32 main control board performs adjustments according to the warping position and degree. When the adhesion is too loose, the speed of the walking motor 5 is increased to enhance the cleaning effect. When the adhesion is too tight, the walking motor 5 is stopped and an alarm is triggered. Once the load of the brush head motor 701 returns to the normal range, the system resumes the preset walking speed to ensure that there are no blind spots in the cleaning and that the photovoltaic panel coating is not damaged.

[0025] The AI-based zoned stain recognition module, based on convolutional neural networks and the Transformer model, decouples and separates the load characteristics of the two nylon disc brushes 702 from the total load current of the brush head motor 701. Combined with real-time power generation data of each area of ​​the photovoltaic panel, it identifies the type and severity of stains in the corresponding area. For dusty areas, it adopts a low-energy strategy of fast walking and single cleaning, while for stubborn stain areas, it adopts a strategy of slow walking and repeated cleaning.

[0026] The AI-based stain recognition module, based on convolutional neural networks and the Transformer model, decouples and separates the load characteristics of the left and right nylon disc brushes 702 from the total load current of the brush head motor 701. When the robot moves stably in the cleaning area, the STM32 main control board acquires continuous total load current signals of the brush head motor 701 at a high sampling frequency. This time-series signal is input into a pre-trained convolutional neural network model to extract key local waveform patterns reflecting stain characteristics. The feature sequence is then fed into the Transformer model. Since the left and right nylon disc brushes 702 are side-by-side and driven by the same synchronous belt, their resistance changes are correlated and differ over time. When the left nylon disc brush 702 encounters stubborn stains... The load characteristics of the stains appear periodically, while the right nylon disc brush 702 does not have this characteristic. The Transformer model calculates the correlation of different parts of the signal through a self-attention mechanism, accurately separating the total load characteristics into the load characteristics of the left nylon disc brush 702 and the right nylon disc brush 702. The system compares the separated load characteristics with the stain feature library in the storage unit to identify the stain type and severity in the left and right areas of the photovoltaic panel. At the same time, it combines the attenuation degree of real-time power generation data of each area of ​​the photovoltaic panel to verify the identification accuracy. Based on the identification results, a differentiated cleaning strategy is formulated. For the dusty area, a low-energy strategy of fast walking and single cleaning is adopted, while for the stubborn stain area, a stronger strategy of slow walking and repeated cleaning is adopted.

[0027] The hot spot targeted maintenance module can receive hot spot alarm information sent by the photovoltaic string inverter, which includes the battery string number and location information. The STM32 main control board autonomously plans the walking path to the hot spot area based on the positioning information of the encoder of the walking motor 5, controls the dual brush head cleaning component 7 to perform targeted enhanced cleaning on the target area, and obtains the current, voltage and power generation data of the photovoltaic panel before and after cleaning for comparison and verification. If the data does not return to normal, it sends an early warning information to the maintenance platform.

[0028] The photovoltaic cascade inverter continuously monitors the current and voltage of each string of solar cells. When the current or voltage of a string of cells drops abnormally and falls below the normal threshold, the inverter immediately determines that a hot spot effect may occur in that area. It then pushes the abnormal data, including the abnormal string number and location information, to the robot's STM32 main control unit in real time via a wireless network. Upon receiving the hot spot alarm and coordinate information, the robot immediately suspends its current routine cleaning task. Based on its precise position determined by the encoder and the target coordinates of the hot spot, it calculates the shortest and most efficient path to the hot spot area using a built-in path planning algorithm. After reaching the target area, the robot performs targeted and intensive cleaning. After cleaning, it obtains the real-time current and voltage data of the area from the cascade inverter via a wireless communication module. The STM32 main control unit compares the power generation data before and after cleaning. If the data returns to normal, the hot spot is determined to be eliminated. If the data is still abnormal, it indicates that there is serious physical damage. The robot automatically sends an early warning message to the operation and maintenance platform and notifies human intervention, thereby realizing a complete unmanned closed-loop operation and maintenance from problem discovery, handling to effect verification.

[0029] The reinforcement learning adaptive working condition module has a built-in DDPG model. Based on the load of the walking motor 5, the load of the brush head motor 701, and the encoder deviation, it identifies slippery, resonant, windy and sandy, and jamming working conditions, and dynamically adjusts the walking speed, drive torque and step distance to achieve anti-slip, resonance suppression and fault self-recovery.

[0030] This module employs the DDPG reinforcement learning model, which does not rely on fixed control rules. It can autonomously learn the optimal control strategy through real-time feedback data. During robot movement, the STM32 main control board continuously collects load information from the walking motor 5, the brush head motor 701, and encoder deviation information, comparing this data with the system's calibrated normal operating condition data to determine the operating status. When the encoder speed of the walking motor 5 fluctuates abnormally under normal control output and does not match the control commands, it is identified as a slippery slip condition. When the load of the walking motor 5 or the encoder feedback shows regular, large-scale periodic fluctuations, it is determined to be a resonance condition. When the load of the walking motor 5 or the brush head motor 701 experiences an irregular, abnormal increase, it is identified as a sandstorm cover or mechanical overload jamming condition. The model takes the above working conditions as input and outputs the optimal control parameters in real time. For slippery working conditions, it appropriately reduces the driving torque of the walking motor 5 and optimizes the walking speed to prevent the driving wheel 4 from spinning and slipping off the track. For resonance working conditions, it continuously fine-tunes the walking speed and step distance to break the inherent resonance frequency of the system and achieve resonance suppression. For jamming working conditions, it autonomously gets out of the jamming state through trial and error strategies such as short-term reverse of the walking motor 5 and changing the driving torque and direction of movement, and completes fault self-recovery, thereby maintaining stable operation in complex outdoor environments.

[0031] A gantry-type synchronous dual-brush head intelligent cleaning method for photovoltaic panels, comprising the following steps: S1 and STM32 main control boards obtain the real-time total load current through the sampling resistor in the 701 circuit of the brush head motor, and at the same time collect the encoder position signal of the walking motor 5 and the power generation data of the photovoltaic panel area. S2. The adaptive leveling module identifies the bonding status between the nylon disc brush 702 and the photovoltaic panel based on the total load change of the brush head motor 701, and accordingly executes actions such as increasing the speed of the walking motor 5, stopping the alarm, or restoring the normal speed. S3, the AI ​​partition stain recognition module calls convolutional neural network and Transformer model to decouple the independent load characteristics of the left and right brush heads from the total load current of the brush head motor 701, and combine the power generation data to identify the type and severity of stains. For the dusty area, it adopts fast walking and single cleaning, and for the stubborn stain area, it adopts slow walking and repeated cleaning. S4. The hot spot targeted maintenance module receives hot spot alarms from the cascade inverter, autonomously goes to the target area to strengthen cleaning, compares the power generation data before and after cleaning to verify the effect, and reports an alarm if there is an abnormality. S5, the reinforcement learning working condition adaptive module uses the DDPG model to identify the working condition in real time based on the load of the walking motor 5, the load of the brush head motor 701 and the encoder deviation, and dynamically adjusts the walking parameters to achieve anti-slip, resonance suppression and fault self-recovery. S6. Repeat the above steps to complete the intelligent cleaning and maintenance of the photovoltaic panel array.

[0032] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0033] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A gantry type intelligent cleaning robot for a dual brush head photovoltaic panel, characterized in that, Includes a gantry frame (1), a main control box (2), an STM32 main control board, an AI intelligent control unit, and a storage unit; The gantry frame (1) spans above the photovoltaic panel. A linear guide rail (3) is provided on the top of the gantry frame (1). A drive wheel (4) is rotatably installed on the bottom of the gantry frame (1). A walking motor (5) is fixedly installed on the top of the gantry frame (1). The walking motor (5) drives the drive wheel (4) to move the frame on the photovoltaic panel. The main control box (2) is fixed on the gantry frame (1), and the STM32 main control board, AI intelligent control unit and storage unit are all integrated inside the main control box (2); A movable component (6) is mounted on the gantry frame (1); The dual-brush head cleaning assembly (7) is connected to the moving assembly (6). Driven by the moving assembly (6), the dual-brush head cleaning assembly (7) cleans the photovoltaic panel. The storage unit is used to store convolutional neural network, Transformer model, DDPG reinforcement learning model, baseline load parameters, stain feature library and cleaning strategy parameters; The AI ​​intelligent control unit includes an adaptive leveling module, an AI zoned stain recognition module, a hot spot targeted maintenance module, and a reinforcement learning adaptive operating condition module.

2. The gantry type synchronous dual-brush head photovoltaic panel intelligent cleaning robot according to claim 1, characterized in that: The moving component (6) includes a drive motor (601), a ball screw (602), and a positioning plate (603). The drive motor (601) is positioned on the gantry frame (1). The ball screw (602) is connected to the drive motor (601) in a transmission connection. The positioning plate (603) is threadedly connected to the ball screw (602). One end of the positioning plate (603) is slidably supported on the linear guide rail (3). The drive motor (601) drives the positioning plate (603) to move laterally along the linear guide rail (3) through the ball screw (602).

3. The gantry type synchronous dual-brush head photovoltaic panel intelligent cleaning robot according to claim 2, characterized in that: The dual-brush head cleaning assembly (7) includes a brush head motor (701) and two nylon disc brushes (702). The brush head motor (701) is fixed to the positioning plate (603). The two nylon disc brushes (702) are arranged side by side and are synchronously driven to rotate by the brush head motor (701) through a synchronous wheel and a synchronous belt.

4. The gantry type synchronous dual-brush head photovoltaic panel intelligent cleaning robot according to claim 3, characterized in that: A sampling resistor is connected in series in the brush head motor (701) circuit. The STM32 main control board collects the voltage across the sampling resistor through the built-in ADC and calculates the real-time total load current of the brush head motor (701). The adaptive leveling module identifies the warping of the photovoltaic panel and the bonding state of the nylon disc brush (702) based on the change of the total load of the brush head motor (701). When the load decreases, it is determined that the bonding is too loose, so the speed of the walking motor (5) is increased. When the load increases, it is determined that the bonding is too tight, so the walking motor (5) is stopped and alarmed. After the load returns to normal, the preset walking speed is restored.

5. The gantry-type synchronous dual-brush head photovoltaic panel intelligent cleaning robot according to claim 3, characterized in that: The AI-based stain identification module is based on convolutional neural networks and the Transformer model. It decouples and separates the load characteristics of the two nylon disc brushes (702) from the total load current of the brush head motor (701). It combines the real-time power generation data of each area of ​​the photovoltaic panel to identify the stain type and severity of the corresponding area. For the dusty area, it adopts a low-energy strategy of fast walking and single cleaning. For the stubborn stain area, it adopts a strategy of slow walking and repeated cleaning.

6. The gantry-type synchronous dual-brush head photovoltaic panel intelligent cleaning robot according to claim 1, characterized in that: The hot spot targeted maintenance module can receive hot spot alarm information sent by the photovoltaic string inverter, which includes the battery string number and location information. The STM32 main control board autonomously plans the walking path to the hot spot area according to the positioning information of the encoder of the walking motor (5), controls the dual brush head cleaning component (7) to perform targeted enhanced cleaning on the target area, and obtains the current, voltage and power generation data of the photovoltaic panel before and after cleaning for comparison and verification. If the data does not return to normal, it sends a warning message to the maintenance platform.

7. The gantry-type synchronous dual-brush head photovoltaic panel intelligent cleaning robot according to claim 3, characterized in that: The reinforcement learning adaptive module has a built-in DDPG model. Based on the load of the walking motor (5), the load of the brush head motor (701), and the encoder deviation, it identifies slippery, resonant, windy and sandy, and jamming conditions, and dynamically adjusts the walking speed, driving torque and moving step distance to achieve anti-slip, resonance suppression and fault self-recovery.

8. A gantry-type synchronous dual-brush head intelligent cleaning method for photovoltaic panels, implemented using a gantry-type synchronous dual-brush head intelligent cleaning robot for photovoltaic panels according to any one of claims 1-7, characterized in that: The steps are as follows: S1, the STM32 main control board obtains the real-time total load current through the sampling resistor in the brush head motor (701) circuit, and at the same time collects the encoder position signal of the walking motor (5) and the power generation data of the photovoltaic panel area; S2. The adaptive leveling module identifies the bonding status between the nylon disc brush (702) and the photovoltaic panel based on the total load change of the brush head motor (701), and performs corresponding actions such as increasing the speed of the walking motor (5), stopping alarm, or restoring the normal speed. S3, the AI ​​partition stain recognition module calls the convolutional neural network and the Transformer model to decouple the independent load characteristics of the left and right brush heads from the total load current of the brush head motor (701), and combine the power generation data to identify the type and severity of stains. For the floating dust area, it adopts fast walking and single cleaning, and for the stubborn stain area, it adopts slow walking and repeated cleaning. S4. The hot spot targeted maintenance module receives hot spot alarms from the cascade inverter, autonomously goes to the target area to strengthen cleaning, compares the power generation data before and after cleaning to verify the effect, and reports an alarm if there is an abnormality. S5. The reinforcement learning working condition adaptive module uses the DDPG model to identify the working condition in real time and dynamically adjust the walking parameters based on the load of the walking motor (5), the load of the brush head motor (701) and the encoder deviation, so as to achieve anti-slip, resonance suppression and fault self-recovery. S6. Repeat the above steps to complete the intelligent cleaning and maintenance of the photovoltaic panel array.

Images