Intelligent photovoltaic cleaning method for air-ground cooperation of unmanned aerial vehicle and cleaning robot

Abstract

The invention relates to an intelligent photovoltaic cleaning method for air-ground cooperation of an unmanned aerial vehicle and a cleaning robot, and belongs to the technical field of photovoltaic cleaning. The method comprises the steps that the real-time pose of the unmanned aerial vehicle is determined in a visual positioning and laser positioning fusion mode; establishing a wind speed estimation model by taking the wind speed as a predictable state quantity; solving an optimal control sequence of the unmanned aerial vehicle in a prediction time domain based on a wind speed estimation model so as to compensate wind disturbance in advance; the track of the unmanned aerial vehicle is corrected to achieve accurate butt joint of the unmanned aerial vehicle and the cleaning robot; and an intelligent decision-making method is used for realizing the timely transfer of the cleaning robot by the unmanned aerial vehicle and the cleaning work of the photovoltaic panel of the cleaning robot. Through centralized control of the central platform, a single unmanned aerial vehicle assists multiple cleaning robots, dynamic task allocation and path planning are achieved, and the problems that desert photovoltaic cleaning efficiency is low and the coverage range is limited are solved.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic cleaning technology, and in particular to an intelligent photovoltaic cleaning method for air-ground collaboration between a drone and a cleaning robot. Background Art

[0002] With the rapid development of renewable energy, photovoltaic power generation, as an important clean energy source, has gained widespread application. my country's desert regions, rich in solar energy resources, are key areas for the development of the photovoltaic industry. The large-scale construction of photovoltaic power stations in desert regions not only promotes regional economic development but also provides important support for the transformation of the energy structure.

[0003] However, the unique natural environment of desert regions, such as frequent wind and dust, can easily lead to heavy dust accumulation on the surface of solar photovoltaic panels, which in turn reduces photovoltaic power generation efficiency, affecting power generation performance and economic benefits. Therefore, surface cleaning of solar photovoltaic panels has become a critical step in ensuring the normal operation of photovoltaic power plants.

[0004] Currently, the primary method for cleaning photovoltaic panels is manual cleaning. While simple and straightforward, manual cleaning methods suffer from low efficiency, high labor intensity, and high costs. In desert regions, integrating photovoltaic power plant construction with ecological restoration has become an inevitable trend in economic and industrial development. Therefore, developing efficient, energy-saving, and environmentally friendly photovoltaic panel cleaning technologies suitable for desert environments is of great significance. Summary of the Invention

[0005] To solve the above problems, an embodiment of the present invention aims to provide an intelligent photovoltaic cleaning method for air-ground collaboration between a drone and a cleaning robot.

[0006] An intelligent photovoltaic cleaning method using drones and cleaning robots in air-to-ground collaboration, comprising:

[0007] Step 1: Determine the real-time position of the drone using a fusion of visual positioning and laser positioning;

[0008] Step 2: Establish a wind speed estimation model by taking wind speed as a predictable state variable;

[0009] Step 3: Solve the optimal control sequence of the UAV in the prediction time domain based on the wind speed estimation model to compensate for wind disturbance in advance;

[0010] Step 4: Correct the trajectory of the drone to achieve docking between the drone and the cleaning robot;

[0011] Step 5: Use intelligent decision-making methods to control the drone to transport the cleaning robot and the cleaning robot to clean the photovoltaic panels.

[0012] Preferably, the step 1: determining the real-time position and posture of the UAV by integrating visual positioning and laser positioning includes:

[0013] Step 1.1: Use feature point matching and the PnP algorithm to solve the UAV's pose in visual positioning to obtain the first position estimate and the first covariance;

[0014] Step 1.2: Convert the drone's laser positioning signal into a second position estimate and a second covariance through range and angle data;

[0015] Step 1.3: Use Kalman filtering to fuse the first position estimate, the first covariance, the second position estimate, and the second covariance to determine the real-time pose of the UAV.

[0016] Preferably, the step 2: establishing a wind speed estimation model using wind speed as a predictable state quantity includes:

[0017] The wind speed estimation model is formed by predicting the wind speed in real time using the onboard barometer / IMU data; wherein the wind speed estimation model is:

[0018]

[0019] in, represents the wind speed estimation vector at time k, represents the wind speed vector estimated or measured at the last moment (k-1 moment), K represents the Kalman gain coefficient, z k represents the actual observation value at the kth moment, and H represents the observation matrix.

[0020] Preferably, step 3: solving the optimal control sequence of the UAV in the prediction time domain based on the wind speed estimation model to compensate for wind disturbance in advance includes:

[0021] The cost function is used to solve the optimal control sequence of the UAV in the prediction time domain to compensate for wind disturbance in advance. The cost function is:

[0022]

[0023] The constraints are:

[0024] x k+1 =Ax k +Bu k +Ew k

[0025] u min ≤u≤u max

[0026]

[0027] Among them, xk+i|k It represents the predicted value of the system state at the current time k for the future time k+i, u k+i represents the predicted control input at the future i-th time step (i.e. k+i moment), x k+N|K represents the state of the drone at the nth time step (i.e., k+n time) predicted at the current time k, x ref represents the ideal state of the UAV in docking or flight trajectory, N represents the prediction time domain, x k+1 represents the predicted state of the next discrete time step (k+1 moment), A represents the state transfer matrix, B represents the control matrix, E represents the wind disturbance coupling matrix, which describes the effect of wind on the UAV, and w k represents the airflow disturbance (wind speed / wind pressure) vector at time k, u min represents the minimum allowable value of the control input, u represents the control input vector, u max The maximum allowable value of the control input is expressed as u ff represents the feedforward control quantity, represents the pseudo-inverse of the control matrix, w k|k Represents the wind speed sequence predicted in the future time domain, B T represents the transpose of the control matrix, and B represents the control matrix.

[0028] Preferably, the step 4: correcting the trajectory of the drone to achieve docking between the drone and the cleaning robot includes:

[0029] Correct the target path based on predicted wind disturbances:

[0030]

[0031] Combined with PID feedback, the optimal control sequence of the drone is output:

[0032]

[0033] in, represents the dynamically corrected reference trajectory, x ref (t) represents the uncorrected ideal state, that is, the original planned flight trajectory of the UAV at a specific time t, w pred (τ) represents the predicted wind speed vector at the future time τ, d τ Indicates the differential time variable in the integral operation, u k represents the actual control input vector at discrete time step k, represents the optimal control input obtained by MPC optimization at time k, K p represents the LQR gain matrix, x fused represents the fusion estimation value of the drone’s position, speed, attitude and other state quantities, x refIt represents the ideal state of the UAV during docking or flight trajectory.

[0034] Preferably, the step 5: using an intelligent decision-making method to control the drone and the cleaning robot to clean the photovoltaic panels includes:

[0035] A digital twin model of the target photovoltaic field is constructed, and the optimal cleaning plan is automatically generated using Monte Carlo tree search.

[0036] The present invention also provides an intelligent photovoltaic cleaning system for air-ground collaboration between a drone and a cleaning robot, comprising:

[0037] The pose determination module is used to determine the real-time pose of the UAV by combining visual positioning and laser positioning;

[0038] A wind speed estimation module is used to establish a wind speed estimation model by taking wind speed as a predictable state quantity;

[0039] The wind disturbance compensation module is used to solve the optimal control sequence of the UAV in the prediction time domain based on the wind speed estimation model to compensate for wind disturbance in advance;

[0040] The docking module is used to correct the trajectory of the drone and achieve docking between the drone and the cleaning robot;

[0041] The photovoltaic cleaning module is used to control drones and cleaning robots to clean photovoltaic panels using intelligent decision-making methods.

[0042] The present invention also provides an electronic device, comprising a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein the transceiver, the memory, and the processor are connected via the bus. The electronic device is characterized in that when the computer program is executed by the processor, the steps in the above-mentioned intelligent photovoltaic cleaning method for air-ground collaboration between a drone and a cleaning robot are implemented.

[0043] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the above-mentioned intelligent photovoltaic cleaning method for air-ground collaboration between a drone and a cleaning robot are implemented.

[0044] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0045] The present invention relates to an intelligent photovoltaic cleaning method and device for air-to-ground collaboration between an unmanned aerial vehicle (UAV) and a cleaning robot. Compared with the existing technology, the present invention uses centralized control via a central platform, enabling the UAV to control multiple cleaning robots and achieve dynamic task allocation and path planning, thus solving the problems of low efficiency and limited coverage of photovoltaic cleaning in the desert.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 A flow chart of an intelligent photovoltaic cleaning method for air-ground collaboration between a drone and a cleaning robot provided by the present invention;

[0049] Figure 2 The multi-objective optimization flow chart provided by the present invention;

[0050] Figure 3 A schematic diagram of the dynamic game process provided by the present invention;

[0051] Figure 4 Schematic diagram of the intelligent decision-making process constructed for the present invention. DETAILED DESCRIPTION

[0052] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0053] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0054] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediary; or internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0055] See also Figure 1-2 , an intelligent photovoltaic cleaning method for air-ground collaboration between a drone and a cleaning robot, comprising:

[0056] Step 1: Determine the real-time position of the drone using a fusion of visual positioning and laser positioning;

[0057] Step 2: Establish a wind speed estimation model by taking wind speed as a predictable state variable;

[0058] Step 3: Solve the optimal control sequence of the UAV in the prediction time domain based on the wind speed estimation model to compensate for wind disturbance in advance;

[0059] Step 4: Correct the trajectory of the drone to achieve docking between the drone and the cleaning robot;

[0060] In the present invention, the intelligent photovoltaic cleaning system for air-ground collaboration between drones and cleaning robots includes: an environmental perception module for calling the cameras, lidars, and sensors on the drones and cleaning robots to collect environmental information of the photovoltaic power station in real time, including terrain, obstacles, and weather conditions;

[0061] The collaborative communication module is used to establish a communication channel between the drone and the cleaning robot based on the 5G and LoRa communication protocols. Through the multimodal information complementary mechanism and fault self-healing control strategy, it ensures that when an abnormality occurs in the robot, the drone can autonomously take over the task and trigger collaborative compensation with neighboring robots, thereby ensuring the overall operational continuity of the system.

[0062] Furthermore, steps 1-4 include:

[0063] The specific steps and key formulas for calculating the wind-resistant path in advance and achieving precise docking through the Predictive Prediction Algorithm (MPC) are divided into four core links: perception fusion, disturbance modeling, predictive control, and wind-resistant path generation:

[0064] 1. Multi-sensor fusion positioning (perception layer)

[0065] Objective: To output the real-time position of the drone with centimeter-level accuracy by fusing visual (camera) and laser (LiDAR) data.

[0066] Steps and formula:

[0067] Visual positioning: Use feature point matching (such as ORB-SLAM) and PnP algorithm to solve the pose and obtain the position estimate P cam and covariance ∑cam.

[0068] Laser positioning: LiDAR uses distance and angle data (d LiDAR ,θ LiDAR ) is converted to position P cam , covariance∑cam

[0069] Kalman filter fusion: P fused =∑LiDAR(∑cam+∑LiDAR) -1 P LiDAR

[0070] Output: fused state (position, velocity, attitude).

[0071] 2. Airflow disturbance modeling (dynamic expansion)

[0072] Objective: To model wind disturbance as a predictable state variable and embed it into the MPC prediction model.

[0073] Steps and formula:

[0074] Wind field state expansion: wind speed w k As a state quantity, the extended dynamics model,

[0075] w k+1 =A ω w k +ξ ω (ξ ω ~N(0,Q ω ))

[0076] A ω is the wind disturbance dynamic matrix (usually simplified to a random walk model).

[0077] Online estimation: Real-time update of w through onboard barometer / IMU data k , for example based on Kalman filtering,

[0078]

[0079] z k It is an indirect observation of wind disturbance measured by air pressure difference or accelerometer.

[0080] 3. Prediction algorithm (MPC rolling optimization)

[0081] Objective: To solve the optimal control sequence in the forecast time domain and compensate for wind disturbances in advance.

[0082] Steps and formula:

[0083] Cost function design:

[0084]

[0085] Q, R, and P are the weight matrices of state, control input, and terminal state, respectively.

[0086] Constraints:

[0087] Dynamic constraints (discretized model):

[0088] x k+1 =Ax k +Bu k +Ew k

[0089] E is the wind disturbance coupling matrix, which describes the effect of wind on the UAV.

[0090] Actuator saturation constraint: u min ≤u≤u max

[0091] Wind disturbance feedforward compensation: explicitly add wind disturbance compensation term to the control input,

[0092]

[0093] 4. Wind-resistant path generation and tracking (execution layer)

[0094] Objective: Dynamically adjust reference trajectory to achieve wind-resistant tracking.

[0095] Steps and formula:

[0096] Reference trajectory correction: Based on the predicted wind disturbance w k+i|k Correct the target path,

[0097]

[0098] Real-time feedback control: MPC outputs the optimal control sequence Combined with PID feedback,

[0099]

[0100] Summarize:

[0101] By integrating a closed-loop process of positioning → wind disturbance modeling → MPC rolling optimization → trajectory correction, centimeter-level precision docking is achieved, resistant to airflow interference. The core of the formula lies in the state-expanded MPC cost function and the wind disturbance feedforward compensation term, which together ensure system stability in strong winds. The photovoltaic panel of this invention features hidden "electromagnetic locks" on the edge, automatically aligning it with the magnetic charging head. Through electromagnetic guidance and a flexible buffer design, equipped with a buffer device, it can achieve a "soft landing" even with deviations, effectively increasing the success rate of aerial docking to over 95%.

[0102] Step 5: Use intelligent decision-making methods to control drones and cleaning robots to clean photovoltaic panels.

[0103] Furthermore, step 5 includes:

[0104] Through a distributed intelligent architecture, drones with different capabilities (global vision, rapid maneuverability) and cleaning robots (photovoltaic panel surface operations, fine operations) can autonomously share information, dynamically allocate tasks, and collaboratively complete complex tasks such as regional cleaning through local interactions. This intelligent decision-making method enables the cleaning of photovoltaic panels to proceed in an orderly manner.

[0105] like Figure 3 As shown in the figure, a dynamic game model for heterogeneous air-ground systems is established. A dynamic game algorithm automatically allocates tasks, such as prioritizing photovoltaic units with a clean power generation efficiency drop of more than 15%. A task allocation algorithm based on federated reinforcement learning is designed to achieve dynamic matching of drone and robot resources. A fault self-healing control strategy is also developed. When a robot malfunctions, the drone can autonomously take over the task and trigger coordinated compensation from neighboring robots, ensuring overall system operational continuity.

[0106] 1. Federated reinforcement learning-based task allocation algorithm to achieve dynamic matching of drone-robot resources

[0107] This solution uses a hierarchical Federated Reinforcement Learning (FRL) architecture combined with a heterogeneous graph attention network (HGAN) to achieve dynamic task allocation. The specific process is as follows:

[0108] (1) Federated reinforcement learning framework design

[0109] Local training layer: Each unmanned aerial vehicle (UAV) and ground robot (UGV) maintains a deep Q network (DQN) to calculate the task allocation strategy:

[0110]

[0111] in:

[0112] Q i is the Q-value function of agent i;

[0113] γ i It is a local reward (such as photovoltaic clean income, energy consumption cost, etc.);

[0114] γ is the discount factor;

[0115] θ i is the local network parameter;

[0116] Global aggregation layer: The cloud server regularly aggregates the policy parameters of each agent, using weighted federated averaging:

[0117]

[0118] Among them, ω i It is the trust weight of the agent, which prevents malicious or inefficient nodes from affecting the global strategy.

[0119] (2) Heterogeneous Graph Attention Network (HGAN) for dynamic matching

[0120] Node characteristics (UAV / UGV status):

[0121] h i =[position (x, y), power (E), failure probability (P f ), task load (L)]∈R 6

[0122] Edge weight calculation (task matching):

[0123] e ij =softmax(MLP([h i ||h j ||d ij ]))

[0124] Among them, d ij is the Euclidean distance between agent i and task j.

[0125] Global task allocation strategy:

[0126]

[0127] This strategy ensures that PV units with high power generation efficiency degradation (e.g., efficiency degradation ≥ 15%) are allocated preferentially while optimizing overall energy consumption.

[0128] 2. Fault self-recovery control strategy triggers collaborative compensation of neighboring robots

[0129] When a robot (UGV) fails, the system triggers the UAV to take over and coordinate compensation with neighboring robots through Bayesian game decision-making + spatiotemporal Petri net. The process is as follows:

[0130] (1) Anomaly detection (based on STL logic)

[0131] Define signal sequential timing logic (STL) to detect faults:

[0132] STL rules:

[0133] in is the predicted state, and δ is the fault tolerance threshold.

[0134] (2) Drone takeover decision (Bayesian game)

[0135] Profit function (whether to take over):

[0136] U UAV =α·mission urgency -β·flight energy consumption +γ·system overall benefit

[0137] α represents the weight coefficient of mission urgency, β represents the weight coefficient of flight energy consumption, and γ represents the weight coefficient of the overall benefit of the system.

[0138] Nash equilibrium solution:

[0139]

[0140] Represents the optimal strategy of the drone, argmax π represents the strategy that maximizes the drone's profit, E represents the expected value, and π represents the Nash equilibrium strategy.

[0141] If U UAV >threshold, the takeover is triggered.

[0142] (3) Collaborative compensation of neighboring robots (dynamic task redistribution)

[0143] Task redistribution optimization problem:

[0144]

[0145] λ represents the weight coefficient of the total moving distance.

[0146] in, is the neighbor set of the faulty robot i, T j is the original task, T new It is an adjusted task.

[0147] Dynamic working radius adjustment (to avoid task conflicts):

[0148] R' j =R j ·(1+η·E),

[0149] Among them, η is the elasticity coefficient, ensuring that the task coverage is complete, R' j Represents the adjusted power generation efficiency, R j represents the current power generation efficiency, ΔE represents the ratio of the benchmark efficiency to the actual loss, and E represents the current power generation efficiency loss.

[0150] A collaborative scheduling platform with a "physical-virtual" bidirectional mapping was constructed. This platform synchronizes the status of real-world desert photovoltaic panels in real time, rehearses various task scenarios on a computer, and automatically selects the optimal scheduling plan, achieving a cleaning efficiency of ≥98% and a 30% increase in energy efficiency. This process automatically selects the optimal scheduling plan.

[0151] 1. Digital twin modeling and real-time data synchronization

[0152] Build a high-fidelity virtual map of the photovoltaic field and synchronize the physical world status in real time:

[0153] Physical → Virtual Data Flow (Sensor → Digital Twin):

[0154]

[0155] Virtual → physical control flow (scheduling instructions → execution end):

[0156]

[0157] Where J(π) is the optimization objective function.

[0158] 2. Optimal scheduling plan generation (two stages)

[0159] Phase 1: Static task allocation based on mixed integer programming (MIP)

[0160] Objective function (minimize total energy consumption + maximize cleaning coverage):

[0161]

[0162] Constraints:

[0163] Among them, x ij =1 means robot j cleans photovoltaic panel i, 1 means clean, 0 means unclean, y j Indicates whether robot j is selected to perform the task, c ij is the energy consumption of movement, e j is the standby energy consumption of robot j, and λ is the energy efficiency trade-off coefficient.

[0164] Phase 2: Dynamic Adjustment Based on Deep Reinforcement Learning (DRL)

[0165] State Space:

[0166] s t =[E i (t),D i (t),S j (t), weather forecast W(t)]

[0167] Among them, E i (t) represents the remaining energy or energy state of photovoltaic panel i at time t, D i (t) represents the cleaning requirement of photovoltaic panel i at time t, S j (t) represents the state of robot j at time t, such as position, power, etc., and W(t) represents the weather forecast at time t, which affects the cleaning efficiency and energy consumption.

[0168] Action space (robot scheduling instructions):

[0169] a t ={route update, charging scheduling, emergency takeover}

[0170] Reward function (guides the strategy to meet cleaning efficiency and energy efficiency goals):

[0171]

[0172] Where N represents the number of cleaning plates, E saved represents the energy saved, and maxE represents the maximum energy.

[0173] 3. Dynamic conflict resolution and flexible execution

[0174] When the actual environment deviates from the virtual preview (such as a sudden sandstorm):

[0175] Rescheduling trigger conditions:

[0176]

[0177] in, represents the actual measured value at time t, Represents the virtual predicted value at time t, ∈ represents the threshold, which is used to determine whether the difference between the actual value and the virtual value meets the rescheduling condition.

[0178] Fast local adjustment (based on auction algorithm):

[0179]

[0180] 4. Energy efficiency and cleanliness rate guarantee mechanism

[0181] Photovoltaic efficiency dynamic prediction model:

[0182] E i (t+1)=E i (t)·exp(-αD i (t)-βW(t))

[0183] Among them, E i (t+1) represents the photovoltaic efficiency at time t+1, E i (t) represents the photovoltaic efficiency at time t, α is the parameter that affects the photovoltaic decline and is related to the amount of dust accumulation, D i (t) represents the amount of dust accumulation at time t, β is another parameter that affects the PV decline, which may be related to weather conditions (such as humidity), and W(t) represents some quantitative expression of the weather conditions at time t or the cleaning frequency.

[0184] Maintain E by adjusting cleaning frequency in real time i (t)≥0.98E0

[0185] Energy consumption optimization strategy:

[0186]

[0187] Among them, P max is the maximum power of the motor, ρ is the air density, C d is the drag coefficient.

[0188] Specific as Figure 4 As shown, a hybrid communication network of "5G high-speed channel + LoRa IoT private network" was built to build a low-latency (<50ms) air-ground IoT network. A digital twin simulation engine was developed:

[0189] ①Integrated meteorological satellite data (can predict sandstorms 2 hours in advance)

[0190] ② AI diagnosis of photovoltaic panel cleanliness (identification accuracy reaches 97%)

[0191] ③Equipment digital twin (real-time display of remaining power / working status)

[0192] The intelligent decision-making system uses Monte Carlo tree search to automatically generate the optimal cleaning plan, achieving a cleaning task completion rate of ≥98% and improving the system's overall energy efficiency by 30%.

[0193] The present invention uses centralized control from a central platform to enable a drone to control multiple cleaning robots and realize dynamic task allocation and path planning, thus solving the problems of low efficiency and limited coverage of desert photovoltaic cleaning.

[0194] The present invention also provides an intelligent photovoltaic cleaning system for air-ground collaboration between a drone and a cleaning robot, comprising:

[0195] The pose determination module is used to determine the real-time pose of the UAV by combining visual positioning and laser positioning;

[0196] A wind speed estimation module is used to establish a wind speed estimation model by taking wind speed as a predictable state quantity;

[0197] The wind disturbance compensation module is used to solve the optimal control sequence of the UAV in the prediction time domain based on the wind speed estimation model to compensate for wind disturbance in advance;

[0198] The docking module is used to correct the trajectory of the drone and achieve docking between the drone and the cleaning robot;

[0199] The photovoltaic cleaning module is used to control drones and cleaning robots to clean photovoltaic panels using intelligent decision-making methods.

[0200] The present invention also provides an electronic device, comprising a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein the transceiver, the memory, and the processor are connected via the bus. The electronic device is characterized in that when the computer program is executed by the processor, the steps in the above-mentioned intelligent photovoltaic cleaning method for air-ground collaboration between a drone and a cleaning robot are implemented.

[0201] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the above-mentioned intelligent photovoltaic cleaning method for air-ground collaboration between a drone and a cleaning robot are implemented.

[0202] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technical solution that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. An intelligent photovoltaic cleaning method using drones and cleaning robots in air-to-ground collaboration, characterized in that: include: Step 1: Determine the real-time position of the drone using a fusion of visual positioning and laser positioning; Step 2: Establish a wind speed estimation model by taking wind speed as a predictable state variable; Step 3: Solve the optimal control sequence of the UAV in the prediction time domain based on the wind speed estimation model to compensate for wind disturbance in advance; Step 4: Correct the trajectory of the drone to achieve docking between the drone and the cleaning robot; Step 5: Use intelligent decision-making methods to control drones and cleaning robots to clean photovoltaic panels.

2. The intelligent photovoltaic cleaning method of the air-ground collaboration of a drone and a cleaning robot according to claim 1 is characterized in that: Step 1: Determine the real-time position of the UAV by integrating visual positioning and laser positioning, including: Step 1.1: Use feature point matching and the PnP algorithm to solve the UAV's pose in visual positioning to obtain the first position estimate and the first covariance; Step 1.2: Convert the drone's laser positioning signal into a second position estimate and a second covariance through range and angle data; Step 1.3: Use Kalman filtering to fuse the first position estimate, the first covariance, the second position estimate, and the second covariance to determine the real-time pose of the UAV.

3. The intelligent photovoltaic cleaning method of the air-ground collaboration of a drone and a cleaning robot according to claim 2 is characterized in that: Step 2: establishing a wind speed estimation model using wind speed as a predictable state variable, including: The wind speed estimation model is formed by predicting the wind speed in real time using the onboard barometer / IMU data; wherein the wind speed estimation model is: in, represents the wind speed estimation vector at time k, represents the wind speed vector estimated or measured at time k-1, K represents the Kalman gain coefficient, z k represents the actual observation value at the kth moment, and H represents the observation matrix.

4. The intelligent photovoltaic cleaning method of drone and cleaning robot in air-ground collaboration according to claim 3 is characterized in that: Step 3: solving the optimal control sequence of the UAV in the prediction time domain based on the wind speed estimation model to compensate for wind disturbance in advance, including: The cost function is used to solve the optimal control sequence of the UAV in the prediction time domain to compensate for wind disturbance in advance. The cost function is: The constraints are: x k+1 =Ax k +Bu k +Ew k in min Oh, oh. max Among them, x k+i|k It represents the predicted value of the system state at the current time k for the future time k+i, u k+i represents the predicted control input at the future i-th time step, i.e., k+i, x k+N|K represents the state of the drone at the nth time step in the future, that is, time k+n, predicted at the current time k, x ref represents the ideal state of the UAV in docking or flight trajectory, N represents the prediction time domain, x k+1 represents the predicted state at the next discrete time step, i.e., k+1, A represents the state transfer matrix, B represents the control matrix, E represents the wind coupling matrix, which describes the effect of wind on the UAV, and w k represents the airflow disturbance vector at time k, u min represents the minimum allowable value of the control input, u represents the control input vector, u max The maximum allowable value of the control input is expressed as u ff represents the feedforward control quantity, represents the pseudo-inverse of the control matrix, w k|k Represents the wind speed sequence predicted in the future time domain, B T represents the transpose of the control matrix, and B represents the control matrix.

5. The intelligent photovoltaic cleaning method of the air-ground collaboration of a drone and a cleaning robot according to claim 4 is characterized in that: Step 4: Correcting the trajectory of the drone to achieve docking between the drone and the cleaning robot, including: Correct the target path based on predicted wind disturbances: Combined with PID feedback, the optimal control sequence of the drone is output: in, represents the dynamically corrected reference trajectory, x ref (t) represents the uncorrected ideal state, that is, the original planned flight trajectory of the UAV at a specific time t, w pred (τ) represents the predicted wind speed vector at the future time τ, d τ Indicates the differential time variable in the integral operation, u k represents the actual control input vector at discrete time step k, represents the optimal control input obtained by MPC optimization at time k, K p represents the LQR gain matrix, x fused represents the fusion estimation value of the drone’s position, velocity and attitude state, x ref It represents the ideal state of the UAV during docking or flight trajectory.

6. The intelligent photovoltaic cleaning method of claim 5 using drones and cleaning robots in air-to-ground collaboration, characterized in that: Step 5: Using an intelligent decision-making method to control the drone and the cleaning robot to clean the photovoltaic panels, including: A digital twin model of the target photovoltaic field is constructed, and the optimal cleaning plan is automatically generated using Monte Carlo tree search.

Images