Walking control method and system of photovoltaic robot

Abstract

The present invention relates to the field of robot walking control technology, and more particularly to a walking control method and system for a photovoltaic robot. The method comprises collecting point cloud data from two photovoltaic panels, fitting the surface equations of the photovoltaic panels based on the collected point cloud data, and calculating a first height difference between the two photovoltaic panels based on the fitted surface equations; obtaining a second height difference between adjacent photovoltaic panels based on the three-dimensional coordinates of target pixels acquired by a depth camera, obtaining spatial posture data of the photovoltaic panels and two-axis balance data of a cleaning robot, inputting the data into a pre-built first machine learning model, and outputting an adjustment strategy for the cleaning robot. The walking control method for a photovoltaic robot provided by the present invention simultaneously utilizes lidar point cloud data and depth camera data. After weighted fusion of the two sets of data, the method can effectively compensate for the errors of a single sensor under complex lighting and light pollution conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of walking control of cleaning robots, and in particular to a walking control method and system for photovoltaic robots. Background Art

[0002] Traditional photovoltaic panel cleaning relies primarily on manual labor, which is not only inefficient and labor-intensive, but also significantly increases the difficulty and cost of manual cleaning in large photovoltaic power plants or areas with complex terrain. Furthermore, improper manual cleaning can damage photovoltaic panels, affecting their power generation efficiency and service life. To address these issues, photovoltaic cleaning robots have emerged.

[0003] When existing photovoltaic cleaning robots walk between photovoltaic panels to clean dust, there may be height differences between adjacent photovoltaic panels. It is necessary to obtain the height difference to control the movement of the cleaning robot. In the existing technology, a single depth camera or camera is used to collect data to obtain the height difference. Due to the complex lighting and the widespread light pollution of photovoltaic panels, the collection accuracy is low, which affects subsequent processing. Summary of the Invention

[0004] In order to solve the above problems, the present invention provides a walking control method and system for a photovoltaic robot.

[0005] The present invention adopts the following technical solutions, the walking control method of the photovoltaic robot, including:

[0006] Install the LiDAR on the cleaning robot, scan the surfaces of two adjacent photovoltaic panels in the forward direction, and collect point cloud data of the two photovoltaic panels respectively;

[0007] The surface equation of the photovoltaic panels is fitted based on the collected point cloud data of the two photovoltaic panels, and the first height difference between the two photovoltaic panels is calculated according to the fitted surface equation of the photovoltaic panels;

[0008] An RGB-D depth camera is installed in front of the robot to obtain depth images of the surfaces of two adjacent photovoltaic panels in the forward direction.

[0009] Calculate the depth gradient of the photovoltaic panel surface through the depth map, obtain the gradient matrix, find the pixel with the largest gradient value, mark it as the target pixel, and obtain the 3D coordinates of the target pixel through the depth camera;

[0010] The second height difference between adjacent photovoltaic panels is obtained based on the three-dimensional coordinates of the target pixel point obtained by the depth camera. The first height difference and the second height difference are weighted and summed to obtain the standard height difference, and the inclination angle of the two photovoltaic panels is collected;

[0011] The spatial posture data of the photovoltaic panel and the dual-axis balance data of the cleaning robot are obtained and input into a pre-built first machine learning model to output an adjustment strategy for the cleaning robot. The adjustment strategy for the cleaning robot includes controlling the height and angle adjustment of the wheels. The height adjustment is to adjust the height of the wheels to keep the robot chassis level. The angle adjustment is to adjust the inclination angle of the wheels to adapt to the angle change of the photovoltaic panel surface.

[0012] As a further description of the above technical solution: the method for fitting the photovoltaic panel surface equation based on the collected point cloud data of two photovoltaic panels includes:

[0013] Centralize the point cloud data of the first photovoltaic panel, calculate the mean of the data points, and construct a centralized point matrix;

[0014] Calculate the covariance matrix, perform eigenvalue decomposition on the covariance matrix, take the eigenvector corresponding to the minimum eigenvalue, and record the eigenvector as the normal vector of the fitting plane;

[0015] The plane offset is generated by calculating the normal vector of the fitted plane using a formula;

[0016] The fitted photovoltaic panel surface equation of the first photovoltaic panel is obtained based on the normal vector and the plane offset of the fitted plane. Similarly, the fitted photovoltaic panel surface equation of the second photovoltaic panel is obtained.

[0017] As a further description of the above technical solution: the method for calculating the first height difference between the two photovoltaic panels based on the fitted photovoltaic panel surface equation is:

[0018] Get the absolute value of the plane offset difference between the two photovoltaic panels;

[0019] The first height difference is obtained by dividing the absolute value of the obtained plane offset difference by the modulus of the normal vector of the second photovoltaic panel.

[0020] As a further description of the above technical solution: the spatial posture data of the photovoltaic panels includes the standard height difference and tilt angle of the two photovoltaic panels;

[0021] The dual-axis balance data of the cleaning robot includes the wheelbase and center of gravity distribution of the cleaning robot, wherein the wheelbase includes the front and rear wheelbase and the left and right wheelbase.

[0022] As a further description of the above technical solution: the method for obtaining the center of gravity distribution includes:

[0023] Install pressure sensors on the four wheels of the cleaning robot to obtain the load borne by each wheel;

[0024] With the center of the chassis of the cleaning robot as the origin and the length of the cleaning robot as the horizontal axis, a rectangular coordinate system is established to obtain the coordinates of the four wheels, which are expressed as ( , )、( , )、( , )、( , );

[0025] Based on the load borne by the wheels and the coordinates of the four wheels, the horizontal and vertical coordinates of the center of gravity are calculated;

[0026] The inclination angle of the cleaning robot is detected by an inertial measurement unit, and the vertical coordinate is calculated. The inertial measurement unit includes an accelerometer and a gyroscope, and the distribution of the center of gravity of the cleaning robot is represented by the horizontal coordinate, the vertical coordinate and the vertical coordinate.

[0027] As a further description of the above technical solution: the method of detecting the tilt angle of the robot by the inertial measurement unit and calculating the vertical coordinate includes:

[0028] The acceleration of the cleaning robot in the horizontal, vertical and vertical directions is measured by the accelerometer and recorded as ;

[0029] The angular velocity of the cleaning robot around the horizontal, longitudinal and vertical axes is measured by the gyroscope and recorded as ;

[0030] Calculate and obtain the inclination angles around the horizontal and vertical axes;

[0031] The inclination angles around the horizontal axis and the vertical axis and the three-dimensional model data of the cleaning robot are input into the constructed second machine learning model, and the vertical coordinate of the center of gravity distribution of the robot is output.

[0032] As a further description of the above technical solution: the method for collecting the second machine learning model training data includes:

[0033] When the cleaning robot is in different working states and postures, an inclination sensor is installed on the cleaning robot body to collect inclination data around the horizontal and vertical axes in real time. Different inclination combinations are collected, including the inclination of the robot on flat ground, climbing, and turning, to cover various possible postures of the robot. 3D modeling software or laser scanning technology is used to obtain the 3D model data of the cleaning robot. The vertical coordinates of the center of gravity distribution of the cleaning robot in different working states and postures are obtained through actual measurement or calculation based on physical principles. A mapping relationship is established based on the collected inclination data around the horizontal and vertical axes and the 3D model data and the corresponding vertical coordinates of the center of gravity distribution as training data.

[0034] As a further description of the above technical solution: the training method of the second machine learning model includes:

[0035] Use convolutional neural network model;

[0036] Divide the collected data into training set, validation set and test set according to the preset ratio;

[0037] Use the training set to train the model, set the optimizer and learning rate, update the model parameters through the backpropagation algorithm, minimize the loss function, and verify it through the validation set to achieve the best performance of the model on the validation set;

[0038] Use the test set to evaluate the trained model, calculate the mean square error index of the model, and evaluate the performance of the model. After the performance evaluation of the model meets the standards, the trained second machine learning model will be deployed and applied.

[0039] As a further description of the above technical solution: the method for calculating the abscissa of the center of gravity is: multiplying the abscissa of each wheel of the cleaning robot by the load it bears, and dividing the sum by the total load, where the total load is the sum of the loads borne by each wheel;

[0040] The method for calculating the ordinate of the center of gravity is as follows: multiply the ordinate of each wheel of the cleaning robot by the load it bears, and then divide the sum by the total load.

[0041] A walking control system for a photovoltaic robot, which is used to implement the walking control method of the photovoltaic robot, comprises:

[0042] The first data acquisition module installs a laser radar on the cleaning robot, scans the surfaces of two adjacent photovoltaic panels in the forward direction, and collects point cloud data of the two photovoltaic panels respectively;

[0043] A first data processing module is configured to fit a surface equation of the two photovoltaic panels based on the collected point cloud data of the two photovoltaic panels, and calculate a first height difference between the two photovoltaic panels according to the fitted surface equation of the photovoltaic panels;

[0044] The second data acquisition module installs a depth camera in front of the robot to obtain the depth map of the surfaces of two adjacent photovoltaic panels in the forward direction;

[0045] The second data processing module calculates the depth gradient of the photovoltaic panel surface through the depth map, obtains the gradient matrix, finds the pixel position with the maximum gradient value, marks it as the target pixel, and obtains the three-dimensional coordinates of the target pixel through the depth camera;

[0046] The data analysis module obtains the second height difference between adjacent photovoltaic panels based on the three-dimensional coordinates of the target pixel point obtained by the depth camera, performs a weighted sum of the obtained first height difference and second height difference to obtain the standard height difference, and collects the inclination angle of the two photovoltaic panels;

[0047] The walking control module obtains the spatial posture data of the photovoltaic panel and the dual-axis balance data of the cleaning robot, inputs them into the pre-built first machine learning model, and outputs the cleaning robot adjustment strategy, including controlling the height and angle adjustment of the wheels.

[0048] Beneficial effects:

[0049] The walking control method of the photovoltaic robot provided by the present invention can obtain height information from both global (fitting the photovoltaic panel surface equation to obtain the first height difference) and local (depth gradient analysis to obtain the three-dimensional coordinates of the target pixel point and calculate the second height difference) perspectives by simultaneously utilizing lidar point cloud data and RGB-D depth camera data. After weighted fusion of the two sets of data, it can effectively compensate for the error of a single sensor under complex lighting and light pollution conditions, thereby obtaining a more accurate and robust standard height difference.

[0050] Furthermore, by accurately obtaining the standard height difference and the tilt angle of the photovoltaic panel, combined with the robot's center of gravity distribution and wheelbase, an adjustment strategy is generated through the first machine learning model to control and adjust the wheel angle and height so that the force on each wheel of the robot is balanced when crossing the height difference. This not only prevents the wheel on one side from being suspended or overloaded, but also ensures that the cleaning components always maintain good contact with the surface of the photovoltaic panel, thereby improving cleaning efficiency and reducing the risk of damage to the photovoltaic panel due to uneven force. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The present invention will be further explained below in conjunction with the accompanying drawings and examples:

[0052] Figure 1 This is a flow chart of the walking control method of the photovoltaic robot provided in Example 1 of the present invention;

[0053] Figure 2 A flowchart of a method for obtaining a first height difference between two photovoltaic panels is provided for embodiment 1 of the present invention;

[0054] Figure 3 A flowchart of a method for obtaining center of gravity distribution provided in Example 2 of the present invention;

[0055] Figure 4 This is a module connection diagram of the walking control system of the photovoltaic robot provided in Example 3 of the present invention;

[0056] Figure 5 A schematic structural diagram of a photovoltaic cleaning device provided in Example 4 of the present invention;

[0057] Figure 6 Schematic diagram of the walking assembly structure provided in Example 4 of the present invention Figure 1 ;

[0058] Figure 7Schematic diagram of the walking assembly structure provided in Example 4 of the present invention Figure 2 .

[0059] Figure numerals: 1. Main frame; 2. Arc-shaped cleaning frame; 21. Bearing seat; 22. Arc-shaped movable hole; 31. Base; 32. Travel wheel; 33. Rotating shaft; 34. Limiting nut; 35. Driving pulley; 36. Limiting guide wheel; 37. Protective hook; 4. Driving shaft; 5. LiDAR and depth camera. DETAILED DESCRIPTION

[0060] In order to make the technical means, creative features, objectives and effects of the present invention easier to understand, the present invention is further described below with reference to specific diagrams. It should be noted that the embodiments and features in the embodiments of this application can be combined with each other unless they conflict.

[0061] Example 1:

[0062] See also Figure 1 and Figure 2 , an embodiment of the present invention provides a technical solution: a walking control method for a photovoltaic robot.

[0063] The laser radar is installed on the cleaning robot, and the surfaces of two adjacent photovoltaic panels in the forward direction are scanned. The point cloud data of the two photovoltaic panels are collected respectively. The point cloud data on the first photovoltaic panel is expressed as: , =1, 2, ..., N; where Indicates the Point cloud data, For the Point cloud data horizontal coordinates, For the The vertical coordinate of the point cloud data, For the The Z coordinate of the point cloud data;

[0064] Based on the collected point cloud data of the two photovoltaic panels, the photovoltaic panel surface equation is fitted, and the first height difference between the two photovoltaic panels is calculated according to the fitted photovoltaic panel surface equation. ;

[0065] The method for fitting the photovoltaic panel surface equation based on the collected point cloud data of the two photovoltaic panels includes:

[0066] The point cloud data of the first photovoltaic panel is centralized, the mean of the data points is calculated, and the central point matrix H is constructed; ,in 、 and is the average value of the horizontal coordinate, vertical coordinate and Z coordinate of all point cloud data, and the covariance matrix is calculated. The calculation formula of the covariance matrix is: ,in, is the transpose of the central point matrix H, and the covariance matrix Perform eigenvalue decomposition, take the eigenvector (a1, b1, c1) corresponding to the minimum eigenvalue, and record this eigenvector as the normal vector of the fitting plane;

[0067] Generate plane offset based on the normal vector (a1, b1, c1) of the fitted plane , the plane offset The calculation formula is: , among which is 、 and The mean of all point cloud data, based on the normal vector of the fitted plane and plane offset The fitted photovoltaic panel surface equation of the first photovoltaic panel is obtained, which is expressed as: , similarly, the fitting photovoltaic panel surface equation of the second photovoltaic panel is obtained, which is expressed as: Where The normal vector coordinates of the second photovoltaic panel based on the fitting plane.

[0068] The first height difference between the two photovoltaic panels is obtained by calculating the fitted photovoltaic panel surface equation. The method is:

[0069] Get the absolute value of the plane offset difference between the two photovoltaic panels;

[0070] The first height difference is obtained by dividing the absolute value of the obtained plane offset difference by the modulus of the normal vector of the second photovoltaic panel. .

[0071] In this embodiment, the first height difference calculation formula is: ;in and is the plane offset of the two photovoltaic panels, is the normal vector of the second photovoltaic panel, This is to normalize the normal vector so that the scale of the height calculation remains consistent.

[0072] It should be noted that the height difference perpendicular to the plane is calculated by combining the offset difference of the two planes with the direction of the normal vector of the second plane. Because the modulus of the normal vector will affect the coefficient of the plane equation, normalization can eliminate this influence and ensure that the height difference is based on the unit normal vector. This way, the calculation result is more accurate and is not affected by the scaling of the plane equation coefficients.

[0073] It's important to understand that in three-dimensional space, the height difference between planes depends not only on the offset difference but also on the plane's orientation (normal vector). Normalizing the normal vector eliminates the scale changes caused by differences in normal vector length, ensuring that the height difference is calculated based on the actual vertical distance in the geometric sense, rather than a value affected by equation coefficients. For example, if the normal vector is scaled, the coefficients in the plane equation will also change, but the actual geometric relationship remains unchanged. Normalization ensures consistency in the calculation.

[0074] In this embodiment, the photovoltaic panel surface equation is fitted based on the collected point cloud data of the two photovoltaic panels, and the height difference of the photovoltaic panel surface is calculated by the photovoltaic panel surface equation instead of directly using discrete point data, thereby improving the calculation accuracy, avoiding the influence of noise points on the height calculation, and making the height difference calculation smoother and more stable.

[0075] An RGB-D depth camera (such as Intel RealSense or ZED) is installed in front of the robot to obtain the depth map of the surfaces of two adjacent photovoltaic panels in the forward direction. The depth map provided by the depth camera is expressed as: , is the depth map pixel coordinate, is the height of the photovoltaic panel surface relative to the depth camera;

[0076] The depth gradient of the photovoltaic panel surface is calculated through the depth map, the gradient matrix is obtained, the pixel position with the maximum gradient value is found, marked as the target pixel, and the three-dimensional coordinates of the target pixel are obtained through the depth camera, which is expressed as: , . is the three-dimensional coordinate of the target pixel point of the first photovoltaic panel, namely the horizontal coordinate, vertical coordinate and Z coordinate, is the three-dimensional coordinates of the target pixel point of the second photovoltaic panel, namely the horizontal coordinate, vertical coordinate and Z coordinate;

[0077] The depth gradient of the photovoltaic panel surface is calculated using depth information. The method for obtaining the gradient matrix includes:

[0078] The depth gradient The calculation formula is: ; midpoint The gradient strength at , which indicates the degree of depth change, is the depth map pixel coordinate, that is, the depth value of each pixel, represents the depth change rate along the x direction, represents the gradient in the horizontal direction, It represents the rate of change of depth along the y direction and the gradient in the vertical direction.

[0079] Specifically, it should be noted that in a depth image, edge points are usually accompanied by large gradient changes. Since there is an obvious change in depth value between a photovoltaic panel and the background or between adjacent photovoltaic panels, the depth gradient can be used to find the edge.

[0080] Obtain the second height difference of adjacent photovoltaic panels based on the three-dimensional coordinates of the target pixel point obtained by the depth camera , the second height difference The calculation formula is: .

[0081] The first height difference will be obtained and the second height difference , perform weighted summation to obtain the standard height difference;

[0082] Preferably, the first height difference to be obtained and the second height difference , and the calculation formula for the weighted sum to obtain the standard height difference is:

[0083] ;

[0084] Where, is the standard height difference, is the weight factor, optionally When the LiDAR signal is stable or the illumination is complex and affects the depth camera, assign 0.8> >0.7, on the contrary, when the LiDAR signal is unstable or the light adjustment is good, give 0.7> >0.6.

[0085] Specifically, by dynamically adjusting the size of the weight factor based on the real-time environmental status and automatically adjusting the weight under different lighting conditions, the system is made more robust, avoiding the long-term impact of erroneous camera data on the fusion results, allowing more reliable data sources to dominate the final calculation and improve overall measurement accuracy.

[0086] In this embodiment, the collected point cloud data of the two photovoltaic panels is used to fit their respective surface equations using the least squares method, thereby calculating the first height difference. This method is not affected by a single lighting condition and can more accurately reflect the true geometric shape of the photovoltaic panels. The RGB-D depth camera is then used to obtain a depth map in the forward direction, calculate the depth gradient, and find the pixel with the largest gradient value. The three-dimensional coordinates of this pixel are then obtained to obtain the second height difference. That is, the use of local depth change information can supplement and verify the results of the point cloud fitting. By simultaneously acquiring two sets of height difference data and then using a weighted summation, data deviations caused by environmental changes or local errors can be compensated. This redundant design makes the overall system less sensitive to external interference and ensures the accuracy of the collected data. This overcomes the fact that relying solely on depth cameras or cameras is easily affected by complex lighting and light pollution (such as solar reflections and shadows), resulting in high data noise and low collection accuracy. By fusing the results of the two different data collection methods (point cloud and depth map), they correct each other to a certain extent, reducing the impact of single sensor errors.

[0087] Example 2:

[0088] See Figure 3 Based on the above embodiments, this embodiment discloses a method for generating a cleaning robot adjustment strategy:

[0089] Collecting the tilt angles of the two photovoltaic panels, and marking the standard height difference and the tilt angle of the two photovoltaic panels as spatial posture data;

[0090] The wheelbase LJ and center of gravity distribution of the cleaning robot are obtained, and the wheelbase LJ and center of gravity distribution of the cleaning robot are recorded as dual-axis balance data; the wheelbase of the cleaning robot includes the front and rear wheelbases and the left and right wheelbases.

[0091] The method for obtaining the center of gravity distribution includes:

[0092] Pressure sensors are installed on the four wheels of the cleaning robot to obtain the loads F1, F2, F3 and F4 borne by each wheel;

[0093] With the center of the cleaning robot chassis as the origin and the length of the cleaning robot as the horizontal axis, a rectangular coordinate system is established to obtain the coordinates of the four wheels, which are expressed as ( , )、( , )、( , )、( , );in 、 、 and is the horizontal coordinate of the four wheels, 、 、 and are the vertical coordinates of the four vehicles.

[0094] Based on the load borne by the wheels and the coordinates of the four wheels, the horizontal and vertical coordinates of the center of gravity are calculated;

[0095] The inclination angle of the cleaning robot is detected by an inertial measurement unit, and the vertical coordinate is calculated. The inertial measurement unit includes an accelerometer and a gyroscope, and the distribution of the center of gravity of the cleaning robot is represented by the horizontal coordinate, the vertical coordinate and the vertical coordinate.

[0096] Specifically, in this embodiment, the purpose of obtaining the center of gravity distribution of the cleaning robot is that according to the changes in the center of gravity distribution, the system can adjust the angles of the suspension system and the wheels to ensure that the robot always maintains balance when crossing height differences or inclined photovoltaic panels to prevent overturning or wheels hanging in the air, and the center of gravity distribution data can reflect the force conditions of each wheel. If a wheel bears too much load, it may cause slipping or insufficient grip. The control system can adjust the driving force distribution and wheel height based on this data to make the force on each wheel more even, thereby improving passability and stability.

[0097] Based on the loads borne by the wheels and the coordinates of the four wheels, the method for calculating the abscissa and ordinate of the center of gravity includes:

[0098] The method for calculating the horizontal coordinate of the center of gravity is: multiplying the horizontal coordinate of each wheel of the cleaning robot by the load it bears, and dividing the sum by the total load. The total load is the sum of the loads borne by each wheel.

[0099] The method for calculating and obtaining the longitudinal coordinate of the center of gravity is: multiplying the longitudinal coordinate of each wheel of the cleaning robot by the load it bears, and dividing the sum by the total load.

[0100] The calculation formula for the horizontal coordinate of the center of gravity is: ;

[0101] The formula for calculating the vertical coordinate of the center of gravity is: .

[0102] The method of detecting the tilt angle of the robot by an inertial measurement unit and calculating and obtaining the vertical coordinate includes:

[0103] The acceleration of the cleaning robot in the horizontal, vertical and vertical directions is measured by the accelerometer and recorded as ;

[0104] The angular velocity of the cleaning robot around the horizontal, longitudinal and vertical axes is measured by the gyroscope and recorded as ;

[0105] Calculate and obtain the inclination angles around the horizontal and vertical axes;

[0106] Inclination angle around the horizontal axis The calculation formula is: ;

[0107] Inclination angle around the longitudinal axis The calculation formula is: ;

[0108] The inclination angles around the horizontal axis and the vertical axis and the three-dimensional model data of the cleaning robot are input into the constructed second machine learning model, and the vertical coordinate of the center of gravity distribution of the robot is output.

[0109] The method for collecting the second machine learning model training data includes:

[0110] When the cleaning robot is in different working states and postures, an inclination sensor is installed on the cleaning robot body to collect inclination data around the horizontal and vertical axes in real time (it should be noted that high-precision inclination sensors are expensive. If high-precision inclination sensors are directly installed on each cleaning robot for inclination data collection, the cost is too high. They can only be installed and used when experimental data is obtained). A variety of inclination combinations are collected, including the robot's inclination in various situations such as on flat ground, climbing, and turning, to cover various possible postures of the robot. The reason for this is that different working scenarios and actions will cause the robot's inclination to change, and these inclination changes are closely related to the robot's center of gravity distribution. By widely collecting inclination data in different working states and postures, the model can learn the complex relationship between inclination and center of gravity distribution.

[0111] Use 3D modeling software or laser scanning technology to obtain accurate 3D model data of the cleaning robot. This includes the size, shape, and mass distribution of each robot component. It should be noted that the structure and mass distribution of the robot are important factors in determining the location of its center of gravity. Accurate 3D model data can provide the model with accurate information about the robot's physical properties, helping the model better understand the inherent connection between the robot's structure and center of gravity distribution.

[0112] The vertical coordinates of the center of gravity distribution of the cleaning robot in different working states and postures can be obtained through actual measurement or calculation based on physical principles. For example, a weighing sensor can be used to measure the force of the robot at different support points, and the vertical coordinates of the center of gravity can be calculated in combination with the geometric parameters of the robot. Alternatively, the position of the center of gravity can be directly measured by performing a balance test on the robot on a specific experimental device. A mapping relationship is established based on the collected inclination data around the horizontal and vertical axes and the three-dimensional model data and the corresponding vertical coordinates of the center of gravity as training data.

[0113] The training method of the second machine learning model includes:

[0114] Use convolutional neural network model;

[0115] The collected data is divided into a training set, a validation set, and a test set according to a preset ratio (for example, 7:2:1). The training set is used to train the model, allowing it to learn the patterns and characteristics of the data. The validation set is used to monitor the model's performance during training, adjust hyperparameters, and prevent overfitting. The test set is used to ultimately evaluate the model's generalization ability and accuracy. This division of the data set ensures that the model can fully learn the characteristics of the data during training. At the same time, the validation set and test set are used to evaluate the model's performance and generalization ability, preventing the model from overfitting to the training data and performing poorly in actual applications.

[0116] Train the model using the training set. Set the optimizer (such as Adam, SGD, etc.) and the learning rate. Use the backpropagation algorithm to update the model parameters and minimize the loss function. The mean squared error (MSE) loss function is used to measure the difference between the model's predicted value and the true value. During training, regularly record the model's training loss and validation loss to monitor model convergence. If the model overfits, regularization methods can be used to address this. By continuously adjusting hyperparameters and optimizing the model structure, and validating the model using the validation set, the model achieves optimal performance on the validation set.

[0117] Specifically, this training model allows the model to gradually learn the mapping relationship between input data and output data, and continuously adjust the model parameters through the optimization algorithm, so that the model can accurately predict the vertical coordinates of the center of gravity distribution of the cleaning robot. At the same time, various strategies are used to prevent the model from overfitting and improve the model's generalization ability and stability.

[0118] Use the test set to evaluate the trained model, calculate the model's mean squared error (MSE) metric, and assess the model's performance. Once the model meets the criteria, it can be used. If the model's performance does not meet the requirements, further analyze the model's prediction results to identify any problems with the model, such as in which cases the model's predictions are inaccurate, whether they are due to issues with the data features or the model's structure. Based on the analysis results, further optimize and adjust the model, such as reselecting features, adjusting the model structure, and increasing the amount of data, until the model's performance reaches a satisfactory level. By evaluating and optimizing the model, the accuracy and generalization capabilities of the model can be continuously improved, enabling it to be better applied in real-world scenarios, accurately predicting the center of gravity distribution of the cleaning robot, and providing strong support for the robot's control and stability.

[0119] Inputting the spatial posture data and the two-axis balance data into a pre-built first machine learning model and outputting an adjustment strategy for the cleaning robot;

[0120] It should be noted that the cleaning robot's adjustment strategy includes wheel height and angle adjustment; height adjustment: the adaptive suspension system adjusts the wheel height in real time by receiving adjustment instructions to keep the robot chassis level, ensure that the cleaning components fit tightly against the surface of the photovoltaic panel, and make the force on the photovoltaic panel more uniform; angle adjustment: when the surface of the photovoltaic panel is tilted or uneven, the cleaning robot control system will adjust the wheel's tilt angle (camber or inclination) to adapt to the angle change of the surface, ensuring that the robot remains stable during driving and the cleaning components apply force evenly.

[0121] The method for collecting the first machine learning model training data includes:

[0122] At each PV plant, representative areas are selected and the height difference between adjacent PV panels is measured using high-precision measuring instruments (such as laser rangefinders). The tilt angle of the panels is also measured using an inclinometer. The tilt angle of PV panels is designed based on the local latitude and lighting conditions, and varies from region to region.

[0123] Measure and obtain the wheelbase and center of gravity distribution of the cleaning robot. For example, measure the center of gravity position of the cleaning robot when it is empty, fully loaded, and performing different cleaning tasks. Alternatively, establish a physical model of the cleaning robot, calculate the center of gravity distribution based on the mass and layout of its components, and obtain the center of gravity distribution data under different working conditions.

[0124] The collected data on the height difference and tilt angle of the photovoltaic panels is correlated with the wheelbase and center of gravity distribution data of the cleaning robot. For each set of data, robot control experts and photovoltaic power plant operation and maintenance personnel jointly determine the corresponding cleaning robot adjustment strategy (i.e., the adjustment values for the wheel height and tilt angle). The cleaning robot maintains stability and evenly distributes force across the photovoltaic panels as it passes between the two panels with a height difference. The corresponding wheel height and tilt angle adjustment strategy is recorded and used as the corresponding label for each set of data, which serves as training data.

[0125] The first machine learning model training method includes selecting a decision tree model, such as CART (Classification and Regression Tree). Decision tree models are well-suited for tasks involving classification or decision-making based on multiple features. In this scenario, a complex logical relationship exists between the input spatial posture data (the standard height difference and tilt angle of photovoltaic panels) and dual-axis balance data (the cleaning robot's front and rear wheel tracks, left and right wheel tracks, and center of gravity distribution) and the output cleaning robot adjustment strategy (wheel height and tilt angle adjustment). The decision tree constructs a tree structure, gradually dividing samples according to different feature values, intuitively demonstrating the decision-making process from input features to output strategy. This model is highly interpretable and easy to understand and analyze.

[0126] The collected training data is divided into training set, validation set and test set in the ratio of 70%, 15% and 15%. The training set is used for model parameter learning. A large amount of data can enable the model to fully learn the rules between input features and output adjustment strategies. The validation set is used to evaluate the model performance in real time during the training process. By adjusting the model's hyperparameters (such as the maximum depth of the decision tree, the minimum number of samples, etc.), the model is prevented from overfitting and ensures that the model has good generalization ability on different data. The test set is used to finally evaluate the model's performance on new data and determine whether the model can accurately output effective cleaning robot adjustment strategies, thereby verifying the practicality and reliability of the model and making full use of data resources for model training and optimization.

[0127] The decision tree model is trained using the training set. During training, the model calculates the information gain metric based on the input spatial posture data and biaxial balance data, selects the optimal splitting features and points, and constructs a decision tree. As training progresses, the decision tree gradually grows, continuously dividing sample data into different subnodes until a specific stopping condition is met (such as reaching the maximum depth or the number of subnode samples falling below the minimum number). By continuously adjusting the decision tree structure and parameters, the model accurately predicts the corresponding cleaning robot adjustment strategy based on the input features, improving the prediction accuracy of the training data.

[0128] Use accuracy on the validation set to evaluate model performance. Accuracy reflects the proportion of samples correctly predicted by the model out of the total number of predicted samples. Based on this evaluation, adjust the decision tree hyperparameters. For example, if the model overfits on the validation set, reduce the maximum depth to limit the complexity of the decision tree. If the model underfits, increase the maximum depth or reduce the minimum number of samples to allow the decision tree to better learn the features in the data. By repeatedly adjusting hyperparameters and evaluating model performance, find the model parameter configuration that performs best on the validation set.

[0129] The tuned model is finally tested on the test set, and the model's accuracy on the test set is calculated. If the test results meet expectations, the model has good generalization capabilities and can be applied to real-world scenarios, effectively adjusting strategies based on the characteristic outputs of photovoltaic panels and cleaning robots. If the test results are unsatisfactory, it is necessary to re-examine steps such as data preprocessing, model selection, and hyperparameter adjustment to identify problems and make improvements. The test set is independent of the training set and validation set. Using it to evaluate the model can truly reflect the model's performance in real-world applications, ensuring that the model can accurately output adjustment strategies even when faced with unseen data, and improving the stability of the cleaning robot when passing photovoltaic panels with drop heights.

[0130] In this embodiment, by accurately obtaining the standard height difference and the inclination angle of the photovoltaic panel, and then combining the robot's center of gravity distribution and wheelbase, an adjustment strategy is generated through the first machine learning model to adjust the wheel angle and height so that the force on each wheel of the robot is balanced when crossing the height difference. This not only prevents the wheel on one side from being suspended or overloaded, but also ensures that the cleaning components always maintain good contact with the surface of the photovoltaic panel, improving cleaning efficiency and reducing the risk of damage to the photovoltaic panel due to uneven force. In summary, this control method can significantly improve the driving stability and cleaning efficiency of the photovoltaic cleaning robot when crossing height differences and photovoltaic panels with different inclination angles through multi-angle data fusion, intelligent decision-making and adaptive adjustment, while enhancing the system's adaptability to complex environments.

[0131] Example 3:

[0132] See also Figure 4 This embodiment discloses a walking control system for a photovoltaic robot, which is used to implement the walking control method of the photovoltaic robot. The system includes:

[0133] The first data acquisition module installs a laser radar on the cleaning robot, scans the surfaces of two adjacent photovoltaic panels in the forward direction, and collects point cloud data of the two photovoltaic panels respectively;

[0134] The first data processing module fits the photovoltaic panel surface equation based on the collected point cloud data of the two photovoltaic panels, and calculates the first height difference between the two photovoltaic panels according to the fitted photovoltaic panel surface equation. ;

[0135] The second data acquisition module installs a depth camera in front of the robot to obtain the depth map of the surfaces of two adjacent photovoltaic panels in the forward direction;

[0136] The second data processing module calculates the depth gradient of the photovoltaic panel surface through the depth map, obtains the gradient matrix, finds the pixel position with the maximum gradient value, marks it as the target pixel, and obtains the three-dimensional coordinates of the target pixel through the depth camera;

[0137] The data analysis module obtains the second height difference of adjacent photovoltaic panels based on the three-dimensional coordinates of the target pixel point obtained by the depth camera , the first height difference obtained and the second height difference , perform weighted summation to obtain the standard height difference and collect the tilt angles of the two photovoltaic panels;

[0138] The walking control module obtains the spatial posture data of the photovoltaic panel and the dual-axis balance data of the cleaning robot, inputs them into the pre-built first machine learning model, and outputs the cleaning robot adjustment strategy, including controlling the height and angle adjustment of the wheels.

[0139] Example 4:

[0140] See also Figure 5-Figure 7 Based on the above embodiment, this embodiment applies the walking control system of the photovoltaic robot to a photovoltaic cleaning device for use;

[0141] The photovoltaic cleaning device includes a main frame 1, two walking components and an arc-shaped cleaning frame 2 installed on the lower surface of the main frame 1. A driving shaft 4 is provided on the two vertical ends of the arc-shaped cleaning frame 2. A bearing seat 21 is provided at the connection between the driving shaft 4 and the arc-shaped cleaning frame 2. A cleaning roller is installed between the two driving shafts 4. The cleaning roller is driven by the driving shaft 4 to rotate to clean the surface of the photovoltaic panel.

[0142] The main frame 1 is equipped with a laser radar and a depth camera 5;

[0143] The walking assembly includes a base 31 and two walking wheels 32. A rotating shaft 33 is provided on the base 31. An arc-shaped movable hole 22 is opened on the two vertical ends of the arc-shaped cleaning frame 2 above the drive shaft 4. The rotating shaft 33 protrudes and is placed in the arc-shaped movable hole 22, and one end of the rotating shaft 33 extends out of the arc-shaped movable hole 22. A limit nut 34 is provided on the extended part of the rotating shaft 33. When the walking assembly is twisted, the arc-shaped movable hole 22 moves in the arc-shaped movable hole 22 relative to the arc-shaped cleaning frame 2. The end of the arc-shaped movable hole 22 is also the limit position of the movement of the rotating shaft 33, which corresponds to the limit position of the twisting of the walking assembly.

[0144] A driving pulley 35 is rotatably provided on one of the bases 31. The driving pulley 35 and the traveling wheel 32 are arranged perpendicular to each other. There are two driving pulleys 35 in total, and a limiting guide wheel 36 is provided between the two driving pulleys 35. The limiting guide wheel 36 is bolted to the base 31.

[0145] A protection hook 37 is provided on the other base 31 , and the protection hook 37 is located between the two running wheels 32 .

[0146] Furthermore, the walking assembly is provided with a rotating shaft 33. When the device is in use, the rotating shaft 33 can rotate in the arc-shaped movable hole 22, thereby realizing a movable connection between the main frame 1 and the walking assembly. On this basis, the structural feature of the rotating shaft 33 rotating around the rotating shaft 33 as the rotation center is realized, and the angle of the walking assembly is adjusted, thereby realizing the adjustment of the angle of the walking wheel 32.

[0147] Furthermore, the walking wheel 32 is slidably set on the base 31 and can be adjusted in height. It is equipped with a hydraulic telescopic rod (not shown in the figure). By controlling the hydraulic telescopic rod, the walking wheel 32 is driven to slide on the base 31 to adjust the height of the walking wheel 32.

[0148] When the control system for the automatic walking of the photovoltaic cleaning robot is applied to the photovoltaic cleaning device, the laser radar is installed on the main frame 1, and the surfaces of two photovoltaic panels adjacent to each other in the forward direction are scanned. The point cloud data of the two photovoltaic panels are collected respectively, and the photovoltaic panel surface equation is fitted based on the collected point cloud data of the two photovoltaic panels. The first height difference between the two photovoltaic panels is calculated based on the fitted photovoltaic panel surface equation. ;

[0149] An RGB-D depth camera is installed in front of the robot to obtain a depth map of the surfaces of two adjacent photovoltaic panels in the forward direction. An RGB-D depth camera is installed in front of the photovoltaic cleaning device to obtain a depth map of the surfaces of two adjacent photovoltaic panels in the forward direction. The depth gradient of the photovoltaic panel surface is calculated using the depth map, and a gradient matrix is obtained. The pixel position with the largest gradient value is found and marked as the target pixel. The three-dimensional coordinates of the target pixel are obtained through the depth camera. The second height difference of the adjacent photovoltaic panels is obtained based on the three-dimensional coordinates of the target pixel obtained by the depth camera. , the first height difference obtained and the second height difference , perform weighted summation to obtain the standard height difference, and collect the inclination angles of the two photovoltaic panels; obtain the spatial posture data of the photovoltaic panels and the dual-axis balance data of the cleaning robot, input them into the pre-built first machine learning model, output the cleaning robot adjustment strategy, and adjust the height and angle of the walking wheel 32 based on the generated adjustment strategy.

[0150] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above-described embodiments. The above-described embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A walking control method for a photovoltaic robot, characterized in that: include: Install the LiDAR on the cleaning robot, scan the surfaces of two adjacent photovoltaic panels in the forward direction, and collect point cloud data of the two photovoltaic panels respectively; The surface equation of the photovoltaic panels is fitted based on the collected point cloud data of the two photovoltaic panels, and the first height difference between the two photovoltaic panels is calculated according to the fitted surface equation of the photovoltaic panels; A depth camera is installed in front of the robot to obtain depth images of the surfaces of two adjacent photovoltaic panels in the forward direction; Calculate the depth gradient of the photovoltaic panel surface through the depth map, obtain the gradient matrix, find the pixel with the largest gradient value, mark it as the target pixel, and obtain the 3D coordinates of the target pixel through the depth camera; The second height difference between adjacent photovoltaic panels is obtained based on the three-dimensional coordinates of the target pixel point obtained by the depth camera. The first height difference and the second height difference are weighted and summed to obtain the standard height difference, and the inclination angle of the two photovoltaic panels is collected; The spatial posture data of the photovoltaic panel and the dual-axis balance data of the cleaning robot are obtained and input into a pre-built first machine learning model to output an adjustment strategy for the cleaning robot. The adjustment strategy for the cleaning robot includes controlling the height and angle adjustment of the wheels. The height adjustment is to adjust the height of the wheels to keep the robot chassis level. The angle adjustment is to adjust the inclination angle of the wheels to adapt to the angle change of the photovoltaic panel surface.

2. The walking control method of the photovoltaic robot according to claim 1, characterized in that: Methods for fitting the surface equation of photovoltaic panels based on the collected point cloud data of two photovoltaic panels include: Centralize the point cloud data of the first photovoltaic panel, calculate the mean of the data points, and construct a centralized point matrix; Calculate the covariance matrix, perform eigenvalue decomposition on the covariance matrix, take the eigenvector corresponding to the minimum eigenvalue, and record the eigenvector as the normal vector of the fitting plane; The plane offset is generated by calculating the normal vector of the fitted plane using a formula; The fitted photovoltaic panel surface equation of the first photovoltaic panel is obtained based on the normal vector and the plane offset of the fitted plane. Similarly, the fitted photovoltaic panel surface equation of the second photovoltaic panel is obtained.

3. The walking control method of the photovoltaic robot according to claim 2, characterized in that: The method for calculating the first height difference between the two photovoltaic panels based on the fitted photovoltaic panel surface equation is: Get the absolute value of the plane offset difference between the two photovoltaic panels; The first height difference is obtained by dividing the absolute value of the obtained plane offset difference by the modulus of the normal vector of the second photovoltaic panel.

4. The walking control method of the photovoltaic robot according to claim 1, characterized in that: The spatial posture data of the photovoltaic panels include the standard height difference and tilt angle of the two photovoltaic panels; The dual-axis balance data of the cleaning robot includes the wheelbase and center of gravity distribution of the cleaning robot, wherein the wheelbase includes the front and rear wheelbase and the left and right wheelbase.

5. The walking control method of the photovoltaic robot according to claim 4, characterized in that: The method for obtaining the center of gravity distribution includes: Install pressure sensors on the four wheels of the cleaning robot to obtain the load borne by each wheel; With the center of the cleaning robot's chassis as the origin and the length of the cleaning robot as the horizontal axis, a rectangular coordinate system is established to obtain the coordinates of the four wheels respectively; Based on the load borne by the wheels and the coordinates of the four wheels, the horizontal and vertical coordinates of the center of gravity are calculated; The inclination angle of the cleaning robot is detected by an inertial measurement unit, and the vertical coordinate is calculated. The inertial measurement unit includes an accelerometer and a gyroscope, and the distribution of the center of gravity of the cleaning robot is represented by the horizontal coordinate, the vertical coordinate and the vertical coordinate.

6. The walking control method of the photovoltaic robot according to claim 5, characterized in that: The method of detecting the tilt angle of the robot by an inertial measurement unit and calculating and obtaining the vertical coordinate includes: The acceleration of the cleaning robot in the horizontal, longitudinal and vertical directions is measured by an accelerometer; The angular velocity of the cleaning robot around the horizontal, longitudinal and vertical axes is measured by a gyroscope; Calculate and obtain the inclination angles around the horizontal and vertical axes; The inclination angles around the horizontal axis and the vertical axis and the three-dimensional model data of the cleaning robot are input into the constructed second machine learning model, and the vertical coordinate of the center of gravity distribution of the robot is output.

7. The walking control method of the photovoltaic robot according to claim 6, characterized in that: The method for collecting the second machine learning model training data includes: When the cleaning robot is in different working states and postures, an inclination sensor is installed on the cleaning robot body to collect the inclination angles around the horizontal and vertical axes in real time, and collect different inclination angle combinations, including the inclination angles of the robot on flat ground, climbing, and turning. The three-dimensional modeling software or laser scanning technology is used to obtain the three-dimensional model data of the cleaning robot. The vertical coordinates of the center of gravity distribution of the cleaning robot in different working states and postures are obtained through actual measurement or calculation based on physical principles. A mapping relationship is established based on the collected inclination angles around the horizontal and vertical axes and the three-dimensional model data and the corresponding vertical coordinates of the center of gravity distribution as training data.

8. The walking control method of the photovoltaic robot according to claim 7, characterized in that: The training method of the second machine learning model includes: Use convolutional neural network model; Divide the collected data into training set, validation set and test set according to the preset ratio; Use the training set to train the model, set the optimizer and learning rate, update the model parameters through the backpropagation algorithm, minimize the loss function, and verify it through the validation set to achieve the best performance of the model on the validation set; Use the test set to evaluate the trained model, calculate the mean square error index of the model, and evaluate the performance of the model. After the performance evaluation of the model meets the standards, the trained second machine learning model will be deployed and applied.

9. The walking control method of the photovoltaic robot according to claim 5, characterized in that: The method for calculating the horizontal coordinate of the center of gravity is: multiplying the horizontal coordinate of each wheel of the cleaning robot by the load it bears, and dividing the sum by the total load, where the total load is the sum of the loads borne by each wheel; The method for calculating the ordinate of the center of gravity is as follows: multiply the ordinate of each wheel of the cleaning robot by the load it bears, and then divide the sum by the total load.

10. A walking control system for a photovoltaic robot, which is used to implement the walking control method for a photovoltaic robot according to any one of claims 1 to 9, characterized in that: The system comprises: The first data acquisition module installs a laser radar on the cleaning robot, scans the surfaces of two adjacent photovoltaic panels in the forward direction, and collects point cloud data of the two photovoltaic panels respectively; A first data processing module is configured to fit a surface equation of the two photovoltaic panels based on the collected point cloud data of the two photovoltaic panels, and calculate a first height difference between the two photovoltaic panels according to the fitted surface equation of the photovoltaic panels; The second data acquisition module installs a depth camera in front of the robot to obtain the depth map of the surfaces of two adjacent photovoltaic panels in the forward direction; The second data processing module calculates the depth gradient of the photovoltaic panel surface through the depth map, obtains the gradient matrix, finds the pixel position with the maximum gradient value, marks it as the target pixel, and obtains the three-dimensional coordinates of the target pixel through the depth camera; The data analysis module obtains the second height difference between adjacent photovoltaic panels based on the three-dimensional coordinates of the target pixel point obtained by the depth camera, performs a weighted sum of the obtained first height difference and second height difference to obtain the standard height difference, and collects the inclination angle of the two photovoltaic panels; The walking control module obtains the spatial posture data of the photovoltaic panel and the dual-axis balance data of the cleaning robot, inputs them into the pre-built first machine learning model, and outputs the cleaning robot adjustment strategy, including controlling the height and angle adjustment of the wheels.

Images