Autonomous path planning system and method based on photovoltaic cleaning robot

Abstract

The invention relates to the technical field of photovoltaic equipment cleaning, and discloses an autonomous path planning system and method based on a photovoltaic cleaning robot. The system comprises an environment sensing module which is used for collecting three-dimensional point cloud data on the surface of a photovoltaic panel and reducing noise; the path planning module is used for constructing a topological map and optimizing a global path; the obstacle detection module is used for identifying obstacles and extracting physical attributes of the obstacles; the dynamic adjustment module is used for analyzing path conflicts and dynamically re-planning; and the execution control module generates a robot motion instruction. According to the method, through cooperative work of the modules, efficient path planning and cleaning work of the photovoltaic cleaning robot in a complex environment are achieved. Compared with a traditional cleaning mode, the cleaning efficiency and quality are improved, the cost is reduced, the adaptability of equipment to a complex environment is enhanced, and powerful technical support is provided for development of the photovoltaic industry.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic device cleaning, and particularly to an autonomous path planning system and method based on a photovoltaic cleaning robot. Background Art

[0002] Under the background of the global energy transformation, as a sustainable clean energy, the application scale of photovoltaic power generation is expanding rapidly. As the core component of the photovoltaic power generation system, the cleanliness of the surface of the photovoltaic panel directly affects the power generation efficiency. However, the current cleaning work of photovoltaic panels faces many problems.

[0003] Traditional manual cleaning methods have many drawbacks. On the one hand, the cost of manual cleaning is high. As the scale of photovoltaic power stations continues to increase, the required number of manual workers also increases accordingly, which not only involves the expenditure of labor wages, but also includes a series of costs such as personnel training and safety guarantee. For example, a large photovoltaic power station may require a cleaning team of dozens or even hundreds of people, and the annual labor cost can reach several million yuan. On the other hand, the efficiency of manual cleaning is low. Manual cleaning mainly relies on manual workers holding cleaning tools to wipe, with a slow speed, and it is difficult to meet the cleaning needs of large-scale photovoltaic power stations. Moreover, manual cleaning is greatly affected by factors such as weather and terrain. Under harsh weather conditions, such as high temperature, heavy rain, strong wind, etc., cleaning operations cannot be carried out, resulting in delays in cleaning work.

[0004] Although mechanical cleaning equipment has improved the cleaning efficiency to a certain extent, it also has obvious deficiencies. Most of the existing mechanical cleaning equipment lacks adaptability to complex environments. The terrain of photovoltaic power stations is complex and diverse, and there may be different terrains such as mountains, hills, deserts, etc. Some mechanical cleaning equipment is difficult to move and operate smoothly on these complex terrains. In a mountain photovoltaic power station, the equipment may have safety problems such as slipping and tipping due to the large terrain slope. At the same time, the ability of these equipment to identify and avoid obstacles is limited. There may be various obstacles such as weeds, stones, and bird droppings around the photovoltaic panel. During the operation of mechanical cleaning equipment, if these obstacles cannot be detected and avoided in time, it will not only affect the cleaning effect, but also may cause damage to the equipment itself, increasing the maintenance cost.

[0005] In terms of path planning, the existing path planning methods of cleaning equipment are relatively simple. Some equipment adopts fixed path planning and cannot be adjusted according to the actual situation of the photovoltaic panel, resulting in incomplete cleaning and cleaning dead corners. For some equipment with path planning functions, their algorithms often cannot effectively integrate multi-source data and have an insufficient perception of the environment. For example, relying only on a single sensor to obtain information, it is difficult to accurately judge the surface condition of the photovoltaic panel, the specific position and attributes of obstacles, etc., so that the optimal cleaning path cannot be planned, reducing the cleaning efficiency and energy utilization rate.

[0006] In addition, the current cleaning equipment has a low level of intelligence and lacks the ability of autonomous decision-making and adaptive adjustment. During the cleaning process, once unexpected situations occur, such as sensor failures, equipment component damages, etc., the equipment often cannot handle them by itself and requires manual intervention, which not only affects the continuity of the cleaning work but also increases the operation and maintenance costs. With the continuous development of the photovoltaic industry, higher requirements are put forward for the intelligence, high efficiency, and adaptability of photovoltaic cleaning technology. Therefore, it is of great practical significance to develop an autonomous path planning system and method based on a photovoltaic cleaning robot that can solve the above problems. Summary of the Invention

[0007] The purpose of the present invention is to provide an autonomous path planning system and method based on a photovoltaic cleaning robot to solve the problems raised in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solutions: An autonomous path planning system based on a photovoltaic cleaning robot, the system includes:

[0009] An environmental perception module, which is used to perform three-dimensional point cloud acquisition and processing on the surface of the photovoltaic panel by using multi-source heterogeneous sensors to obtain photovoltaic surface point cloud data; and perform noise removal processing on the photovoltaic surface point cloud data based on an adaptive filtering algorithm to obtain photovoltaic surface denoised data;

[0010] A path planning module, which is used to perform dynamic map construction processing on the photovoltaic surface denoised data through an improved SLAM algorithm to obtain a photovoltaic surface topological map; and perform global path optimization processing on the photovoltaic surface topological map based on a hierarchical reinforcement learning algorithm to obtain an initial optimal path;

[0011] An obstacle detection module, which is used to perform obstacle recognition processing on the real-time collected photovoltaic panel images by using a spatio-temporal fusion convolutional neural network to obtain obstacle distribution information; and perform feature extraction processing on the physical properties of the obstacles in combination with infrared thermal imaging technology to obtain obstacle attribute data;

[0012] A dynamic adjustment module, which is used to perform conflict analysis processing on the obstacle attribute data and the initial optimal path through a fuzzy logic controller to obtain a path conflict coefficient; perform dynamic path replanning processing on the path conflict coefficient based on a quantum genetic algorithm to obtain an adjusted global path;

[0013] An execution control module, which is used to perform joint trajectory planning processing on the adjusted global path by using an inverse kinematics algorithm to obtain a robot motion trajectory instruction; perform drive signal conversion processing on the robot motion trajectory instruction through a multi-axis cooperative control program to obtain a photovoltaic cleaning robot execution instruction.

[0014] Preferably, the environmental perception module includes:

[0015] A point cloud acquisition unit, configured to perform three-dimensional point cloud acquisition on the surface of a photovoltaic panel through a technology of fusing a lidar and a depth camera;

[0016] An adaptive filtering unit, configured to perform noise separation processing on point cloud data based on a hybrid algorithm of wavelet transform and Kalman filtering;

[0017] A data fusion unit, configured to perform spatio-temporal alignment on the denoised point cloud data and the data of a photovoltaic panel tilt sensor to generate denoised data on the photovoltaic surface.

[0018] Preferably, the functional formula of the improved SLAM algorithm in the path planning module is as follows:

[0019]

[0020] In the formula, M t is the dynamic map constructed at time t, M is the set of candidate maps, z t is the sensor observation data, x t is the robot pose, α k is the confidence weight of the k-th sub-map, m k is the spatial feature description of the k-th sub-map, β is the map smoothing constraint coefficient, is the sub-map gradient, and K is the total number of sub-maps.

[0021] Preferably, the structure of the spatio-temporal fusion convolutional neural network in the obstacle detection module includes:

[0022] A spatial feature extraction layer, configured to extract spatial dimension features of a photovoltaic image through a 3D convolution kernel;

[0023] A time series correlation layer, configured to model the temporal changes of consecutive frame images by using a long short-term memory network;

[0024] An attribute classification layer, configured to perform multi-modal fusion classification by combining infrared thermal imaging data and image features and output obstacle attribute data.

[0025] Preferably, the optimization objective function of the quantum genetic algorithm in the dynamic adjustment module is:

[0026]

[0027] In the formula, P is the set of paths, d i is the length of the i-th path segment, E i is the path energy consumption, w1 and w2 are weight coefficients, the optimization objective is to maximize F(P), and N is the total number of path segments.

[0028] Preferably, the multi-axis collaborative control program in the execution control module includes:

[0029] A trajectory interpolation unit for smoothing the interpolation of the joint trajectory based on a quintic polynomial curve;

[0030] A drive conversion unit for converting the trajectory command into pulse signals of each joint motor through an inverse kinematics model;

[0031] A feedback correction unit for compensating the phase error of the pulse signal in real time by combining the encoder feedback data.

[0032] Preferably, the improved SLAM algorithm further includes:

[0033] A dynamic object elimination unit for detecting moving obstacles by an optical flow method and removing the corresponding point cloud from the map;

[0034] A map update unit for incrementally updating the map of static environmental features based on a Bayesian probability model.

[0035] Preferably, the spatio-temporal fusion convolutional neural network further includes:

[0036] An attention mechanism unit for weighted fusion of the significant regions of the infrared thermal imaging data through a channel attention module;

[0037] An anti-occlusion unit for complementing and reconstructing the partially occluded obstacle image by using a generative adversarial network.

[0038] Preferably, the quantum genetic algorithm adopts the following coding method:

[0039] Quantum chromosome coding, encoding the coordinates and steering angles of the path nodes into a superposition state of quantum bits;

[0040] Adaptive rotation gate update, dynamically adjusting the rotation angle of the quantum bits according to the path conflict coefficient to accelerate convergence.

[0041] Preferably, the present invention further includes an autonomous path planning method based on a photovoltaic cleaning robot, and the method includes the following steps:

[0042] S1: Using multi-source heterogeneous sensors to perform three-dimensional point cloud acquisition and processing on the surface of the photovoltaic panel to obtain photovoltaic surface point cloud data, and then performing noise elimination processing on the photovoltaic surface point cloud data based on an adaptive filtering algorithm to obtain photovoltaic surface denoised data;

[0043] S2: Performing dynamic map construction processing on the photovoltaic surface denoised data through the improved SLAM algorithm to obtain a photovoltaic surface topological map, and then performing global path optimization processing on the photovoltaic surface topological map based on a hierarchical reinforcement learning algorithm to obtain an initial optimal path;

[0044] S3: Use a spatio-temporal fusion convolutional neural network to perform obstacle recognition processing on the real-time collected photovoltaic panel images to obtain obstacle distribution information, and combine infrared thermal imaging technology to perform feature extraction processing on the physical attributes of the obstacles to obtain obstacle attribute data;

[0045] S4: Through a fuzzy logic controller, perform conflict analysis processing on the obstacle attribute data and the initial optimal path to obtain a path conflict coefficient, and perform dynamic path replanning processing on the path conflict coefficient based on the quantum genetic algorithm to obtain an adjusted global path;

[0046] S5: Use the inverse kinematics algorithm to perform joint trajectory planning processing on the adjusted global path to obtain a robot motion trajectory instruction, and perform drive signal conversion processing on the robot motion trajectory instruction through a multi-axis cooperative control program to obtain a photovoltaic cleaning robot execution instruction.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] The environmental perception module uses multi-source heterogeneous sensors to collect three-dimensional point clouds and combines an adaptive filtering algorithm to reduce noise. The lidar and depth camera fusion technology can comprehensively collect the surface information of the photovoltaic panel from different angles, and can accurately obtain data even under complex lighting conditions. The adaptive filtering algorithm effectively removes noise through a hybrid method of wavelet transform and Kalman filtering, provides high-precision data for subsequent path planning, makes the robot's perception of the surface condition of the photovoltaic panel more accurate, greatly improves the reliability of the cleaning operation, and reduces cleaning omissions or equipment damage caused by misjudgment. The path planning module constructs a dynamic map through an improved SLAM algorithm and optimizes the global path with the help of a hierarchical reinforcement learning algorithm. The formula of the improved SLAM algorithm comprehensively considers various factors to construct a map, and the dynamic object elimination unit and the map update unit ensure the real-time accuracy of the map, enabling the robot to quickly locate its own position and the surrounding environment. The hierarchical reinforcement learning algorithm plans the path from the macro and micro levels to find the initial optimal path. Compared with the traditional fixed path or simple planning method, it greatly improves the cleaning efficiency, reduces the cleaning time, and improves the overall power generation efficiency of the photovoltaic power station.

[0049] The obstacle detection module utilizes spatio-temporal fusion convolutional neural network and infrared thermal imaging technology, which can not only identify the distribution of obstacles, but also extract physical properties. The spatial feature extraction layer, time series correlation layer and attribute classification layer work together to accurately judge obstacles. The attention mechanism unit and anti-occlusion unit handle occlusion problems to improve detection accuracy. Combining infrared thermal imaging technology to distinguish obstacles of different materials provides detailed information for path planning, enabling the robot to avoid in advance, ensuring equipment safety and reducing maintenance costs. The dynamic adjustment module analyzes path conflicts through a fuzzy logic controller and re-plans the path using a quantum genetic algorithm. The fuzzy logic controller obtains a conflict coefficient based on the obstacle attributes and path relationship. The quantum genetic algorithm is guided by an optimization objective function, adopts quantum chromosome encoding and adaptive rotation gate update, and quickly adjusts the path in a complex environment to ensure that the robot can successfully avoid obstacles and continue to complete the cleaning task, improving the adaptability and operation ability of the robot in a complex environment.

[0050] The execution control module converts the planned path into precise motion instructions for the robot through an inverse kinematics algorithm and a multi-axis collaborative control program. The trajectory interpolation unit performs smooth interpolation based on a fifth-order polynomial curve to make the robot move smoothly. The drive conversion unit accurately converts signals to drive the motor, and the feedback correction unit combines encoder data to compensate errors in real time, ensuring that the robot moves precisely along the planned path, ensuring uniform cleaning quality and avoiding uneven cleaning from affecting the power generation efficiency of the photovoltaic panel.

[0051] The present invention realizes the automation and intelligence of photovoltaic cleaning, reduces the dependence on manual labor and lowers the labor cost. It improves the cleaning efficiency and equipment safety, reduces equipment damage and maintenance costs. Precise cleaning and path planning improve the power generation efficiency of the photovoltaic panel, increase power generation, create more economic benefits for the photovoltaic industry and promote its sustainable development. Description of the Drawings

[0052] Figure 1 It is the working principle diagram of the autonomous path planning system for the photovoltaic cleaning robot described in the present invention;

[0053] Figure 2 It is the flowchart of the map update of the improved SLAM algorithm;

[0054] Figure 3 It is the flowchart of the obstacle processing of the spatio-temporal fusion convolutional neural network;

[0055] Figure 4 It is the flowchart of the encoding and update of the quantum genetic algorithm. Detailed Embodiments

[0056] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0057] Please refer to Figures 1-4 , the present invention relates to an autonomous path planning system based on a photovoltaic cleaning robot, and its specific implementation solution will be elaborated in detail below.

[0058] This system mainly includes an environment perception module, a path planning module, an obstacle detection module, a dynamic adjustment module, and an execution control module.

[0059] The environment perception module is responsible for data collection and processing of the photovoltaic panel surface. Using multi-source heterogeneous sensors, three-dimensional point cloud collection and processing of the photovoltaic panel surface are carried out through the fusion technology of lidar and depth camera to obtain the point cloud data of the photovoltaic surface. Then, based on the adaptive filtering algorithm, which adopts a hybrid algorithm of wavelet transform and Kalman filter, noise removal processing is carried out on the collected point cloud data of the photovoltaic surface to remove interference data. Then, the denoised point cloud data is spatially and temporally aligned with the data of the photovoltaic panel inclination sensor, and finally the denoised data of the photovoltaic surface is obtained.

[0060] The path planning module works based on the denoised data of the photovoltaic surface obtained by the environment perception module. Through an improved SLAM algorithm, dynamic map construction processing is carried out on these data to obtain a topological map of the photovoltaic surface. Then, based on the hierarchical reinforcement learning algorithm, global path optimization processing is carried out on the generated topological map of the photovoltaic surface to generate an initial optimal path, and a preliminary cleaning route is planned for the robot.

[0061] The obstacle detection module processes the real-time collected images of the photovoltaic panel using a spatio-temporal fusion convolutional neural network. This network realizes the recognition and processing of obstacles through a spatial feature extraction layer, a time series correlation layer, and an attribute classification layer to obtain the obstacle distribution information. At the same time, infrared thermal imaging technology is combined to perform feature extraction processing on the physical attributes of the obstacles to obtain obstacle attribute data for subsequent judgment of the impact of obstacles on path planning.

[0062] The dynamic adjustment module receives the obstacle attribute data obtained by the obstacle detection module and the initial optimal path generated by the path planning module. Through a fuzzy logic controller, conflict analysis processing is carried out on the two to obtain a path conflict coefficient. Based on the quantum genetic algorithm, with the maximization of the objective function as the optimization direction, dynamic path replanning processing is carried out on the path conflict coefficient to obtain an adjusted global path, ensuring that the robot can adjust the path in time when encountering obstacles.

[0063] The execution control module uses the inverse kinematics algorithm to perform joint trajectory planning on the adjusted global path, obtaining the robot motion trajectory instruction. Through the multi-axis collaborative control program, which includes a trajectory interpolation unit, a drive conversion unit, and a feedback correction unit, the drive signal conversion process is carried out on the robot motion trajectory instruction, and finally the execution instruction of the photovoltaic cleaning robot is obtained to control the robot to perform cleaning operations according to the planned path.

[0064] The following elaborates the specific implementation manner of the present invention in the form of embodiments.

[0065] Embodiment 1:

[0066] This embodiment is used to describe the environmental perception module. In actual application scenarios, photovoltaic panels are usually in outdoor complex environments, and there may be interference factors such as dust and rain. The fusion technology of lidar and depth camera plays a key role. The lidar obtains distance information by emitting laser beams and measuring the time of the reflected light, while the depth camera can directly capture the depth data of objects. The combination of the two can obtain the three-dimensional point cloud data of the photovoltaic panel surface from different angles, comprehensively and accurately reflecting the shape and characteristics of the panel surface.

[0067] An adaptive filtering unit using a hybrid algorithm based on wavelet transform and Kalman filter is adopted. The wavelet transform can perform multi-scale decomposition on the signal, separating the noise in the point cloud data from the useful signal, and its multi-resolution analysis characteristics can effectively capture the noise in different frequency bands. The Kalman filter, based on the estimated value of the previous moment and the observed value of the current moment, recursively performs the optimal estimation of the state, continuously correcting the point cloud data in the presence of noise to remove the noise interference.

[0068] The data fusion unit performs spatio-temporal alignment on the denoised point cloud data and the photovoltaic panel inclination sensor data. Assuming that the inclination angle measured by the photovoltaic panel inclination sensor is θ, at time t, the timestamps of the denoised point cloud data are matched with the timestamps of the inclination sensor data, and the spatial positions of the point cloud data at different positions are adjusted according to the inclination angle θ, so that the point cloud data can more accurately reflect the actual state of the photovoltaic panel, generating the noise-reduced data of the photovoltaic surface.

[0069] Embodiment 2: The path planning module is one of the core parts of this system. When the improved SLAM algorithm constructs a dynamic map, its function formula is:

[0070]

[0071] Where M tDenotes the dynamic map constructed at time t, which is constructed based on the current sensor observation data and the robot pose information. M is the set of candidate maps, which contains possible map construction schemes, and the system needs to select the optimal one from these candidate schemes to construct the dynamic map at the current moment. z t Is the sensor observation data, including the data collected by sensors such as lidar and depth cameras, and these data are the basis for map construction. x t Represents the robot pose, that is, the position and orientation information of the robot at time t. Only by clarifying the position of the robot in the map can the map be accurately constructed. α k Is the confidence weight of the k-th sub-map. Different sub-maps have different importance levels in the overall map construction, and the confidence weight is used to measure this difference. m k Is the spatial feature description of the k-th sub-map, which is used to characterize the spatial features such as the shape and position of the sub-map. β is the map smoothing constraint coefficient, which is used to control the smoothness of the map and avoid excessive fluctuations or discontinuities in the map. Is the sub-map gradient, which reflects the change of the spatial features of the sub-map. K is the total number of sub-maps. Dividing the entire map into multiple sub-maps for processing helps to improve the efficiency and accuracy of map construction.

[0072] During the process of map construction, the system continuously calculates this formula according to the collected sensor data to find the optimal dynamic map construction scheme. Based on the obtained photovoltaic surface topology map, the hierarchical reinforcement learning algorithm further optimizes the global path. Considering factors such as the shape, layout of the photovoltaic panels and the requirements of the cleaning task, an initial optimal path is generated. For example, in a large-scale photovoltaic power station where the photovoltaic panels are distributed in an array, the hierarchical reinforcement learning algorithm will plan an initial optimal path that can efficiently clean the panels and avoid frequent turning and collisions of the robot according to the panel distribution in different areas.

[0073] Embodiment 3:

[0074] In the obstacle detection module, the spatio-temporal fusion convolutional neural network is the key technology for obstacle recognition. The spatial feature extraction layer extracts the spatial dimension features of the photovoltaic image through a 3D convolutional kernel. The 3D convolutional kernel performs convolutional operations on the length, width, and time dimensions of the image. Assume that the size of the 3D convolutional kernel is n×m×p, where n and m represent the convolutional kernel size on the image plane, and p represents the convolutional kernel size on the time dimension. Through such convolutional operations, the spatial position, shape and other features of the objects in the image can be extracted, such as the contour and size of the obstacles.

[0075] The time - series correlation layer uses a long short - term memory network (LSTM) to model the temporal changes in consecutive frame images. LSTM can effectively handle the long - term dependence problem in time - series data. In photovoltaic panel images, by analyzing consecutive multiple - frame images, it can capture the movement trajectories and change trends of obstacles. For example, when a bird or other moving object passes by the photovoltaic panel, LSTM can identify its motion state and determine whether it is an obstacle.

[0076] The attribute classification layer combines infrared thermal imaging data and image features for multi - modal fusion classification and outputs obstacle attribute data. Infrared thermal imaging technology can detect the temperature differences of objects. Different objects have different temperature characteristics. By fusing infrared thermal imaging data and image features and using classification algorithms, physical attributes such as the material of the obstacle and whether it is charged can be determined. Suppose the infrared thermal imaging data is represented as I ir , and the image features are represented as F img , and the classification algorithm can be represented as C(I ir , F img ). Through this algorithm, obstacle attribute data is obtained, providing more detailed information for subsequent path planning.

[0077] Example 4:

[0078] The dynamic adjustment module plays a key role in the entire system in dealing with emergencies and ensuring the smooth operation of the robot. The optimization objective function of the quantum genetic algorithm is:

[0079]

[0080] In this formula, P is the set of paths, which contains all possible path schemes that the robot can choose during the cleaning process. d i represents the length of the i - th path segment. The path length is an important indicator to measure the path efficiency. A shorter path can reduce the running time and energy consumption of the robot. E i is the path energy consumption. Due to factors such as terrain and the robot's movement mode, the energy consumption of different paths will vary. w1 and w2 are weight coefficients, which are used to adjust the importance of path length and path energy consumption in the optimization objective. According to the actual application scenarios and requirements, these two weight coefficients can be flexibly set. N is the total number of path segments. In actual path planning, dividing the entire path into multiple path segments for analysis and optimization helps to plan the path more precisely.

[0081] The fuzzy logic controller performs conflict analysis and processing based on the obstacle attribute data and the initial optimal path to obtain the path conflict coefficient. Suppose the path conflict coefficient is C conflict, which reflects the degree of interference of obstacles to the initial optimal path. The quantum genetic algorithm dynamically adjusts the path according to the path conflict coefficient, and continuously iterates to optimize the objective function F(P) to find the path scheme that maximizes F(P), that is, the adjusted global path. In practical applications, when encountering large obstacles, the path conflict coefficient will increase, and the quantum genetic algorithm will be more inclined to select a path with lower energy consumption and bypass the obstacles to ensure that the robot can complete the cleaning task safely and efficiently.

[0082] Embodiment 5:

[0083] The execution control module is responsible for converting the planned path into the actual movement of the robot. The trajectory interpolation unit performs smooth interpolation on the joint trajectory based on the quintic polynomial curve. The expression of the quintic polynomial curve is q(t) = a0 + a1t + a2t 2 + a3t 3 + a4t 4 + a5t 5 , where q(t) represents the position of the joint at time t, and a0 to a5 are polynomial coefficients. By giving the boundary conditions such as the positions, velocities, and accelerations of the starting point and the ending point, these coefficients can be determined, so as to achieve a smooth transition of the joint trajectory and avoid jitter and impact during the movement of the robot.

[0084] The drive conversion unit converts the trajectory instruction into the pulse signal of each joint motor through the inverse kinematic model. Assume that the kinematic model of the robot is f(q), where q is the joint angle vector, and the inverse kinematic model is f -1 (x), where x is the position and attitude of the end effector of the robot. Through the inverse kinematic model, the angles that each joint needs to rotate can be calculated according to the expected position and attitude of the end effector of the robot, and then converted into the pulse signals of each joint motor to control the rotation of the motor.

[0085] The feedback correction unit combines the encoder feedback data to compensate the phase error of the pulse signal in real time. The encoder can measure the actual rotation angle of the joint in real time. Assume that the measured angle by the encoder is θ enc , and the expected joint angle is θ des , and the phase error is Δθ = θ enc - θ des . According to the phase error, the feedback correction unit adjusts the pulse signal to compensate for the rotation of the motor, ensuring that the actual movement of the robot is consistent with the planned path and improving the accuracy and stability of the robot movement.

[0086] Embodiment 6:

[0087] In addition to the core formula for constructing the map, the improved SLAM algorithm also includes a dynamic object removal unit and a map update unit. The dynamic object removal unit detects moving obstacles through the optical flow method and removes their corresponding point clouds from the map. The optical flow method calculates the motion information of an object based on the brightness change of pixel points in an image. Assuming that in two adjacent frames of images I1(x,y) and I2(x,y), the optical flow vector of the pixel point (x,y) is (u,v), according to the optical flow constraint equation I x u + I y v + I t = 0, where I x 、I y are the gradients of the image in the x and y directions respectively, and I t is the change rate of the image over time. By solving this equation, the motion speed of the pixel point can be obtained, thereby detecting moving obstacles. The point clouds corresponding to the detected moving obstacles are removed from the map to avoid interference with map construction and path planning.

[0088] The map update unit performs incremental map updates on the static environment features based on the Bayesian probability model. The basic formula of the Bayesian probability model is In map updates, A represents the state of the map (such as the features of a certain area in the map), and B represents the new sensor observation data. According to the new observation data, the probability distribution of the map state is updated using Bayes' formula to achieve incremental updates of the map. For example, when the robot moves to a new area and obtains new sensor data, the map update unit updates the map according to the Bayesian probability model, combining the previous map information, so that the map can more accurately reflect environmental changes.

[0089] Example 7:

[0090] In addition to the basic structure, the spatio-temporal fusion convolutional neural network also includes an attention mechanism unit and an anti-occlusion unit. The attention mechanism unit performs weighted fusion on the salient regions of the infrared thermal imaging data through a channel attention module. Assuming that the number of channels of the infrared thermal imaging data is C, the channel attention module first performs global average pooling and global max pooling on the data of each channel to obtain two feature vectors F avg and F max , then processes these two feature vectors through a multi-layer perceptron (MLP) to obtain the channel attention weight vector W, and finally performs weighted fusion on the infrared thermal imaging data and the weight vector W to highlight the features of the salient regions and enhance the ability to identify obstacles.

[0091] The anti-occlusion unit uses a generative adversarial network (GAN) to complete and reconstruct the partially occluded obstacle image. The generative adversarial network consists of a generator G and a discriminator D. The goal of the generator is to generate a completed image similar to the real image, and the discriminator is used to judge whether the generated image is real. During the training process, the generator and the discriminator continuously confront and optimize. Assuming the input of the generator is the partially occluded image I occluded , the generated completed image is I generated , the output of the discriminator is D(I generated ), and the generator optimizes itself by minimizing 1 - D(I generated ), and the discriminator optimizes itself by maximizing D(I generated ). In this way, the anti-occlusion unit can complete and reconstruct the partially occluded obstacle image, improving the accuracy of obstacle detection.

[0092] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0093] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An autonomous path planning system based on a photovoltaic cleaning robot, characterized in that: include: The environmental perception module is used to collect and process the three-dimensional point cloud of the photovoltaic panel surface using multi-source heterogeneous sensors to obtain photovoltaic surface point cloud data; And based on the adaptive filtering algorithm, the photovoltaic surface point cloud data is processed to remove noise and obtain the photovoltaic surface noise reduction data; The path planning module is used to dynamically construct a map of the photovoltaic surface noise reduction data using an improved SLAM algorithm to obtain a photovoltaic surface topology map; and to perform global path optimization on the photovoltaic surface topology map based on a hierarchical reinforcement learning algorithm to obtain an initial optimal path; The obstacle detection module is used to use the spatiotemporal fusion convolutional neural network to perform obstacle recognition processing on the real-time collected photovoltaic panel images to obtain obstacle distribution information; it combines infrared thermal imaging technology to perform feature extraction processing on the physical properties of obstacles to obtain obstacle attribute data; The dynamic adjustment module is used to perform conflict analysis between the obstacle attribute data and the initial optimal path through a fuzzy logic controller to obtain a path conflict coefficient; Based on quantum genetic algorithm, dynamic path replanning is performed on the path conflict coefficient to obtain the adjusted global path. The execution control module is used to use the kinematic inverse solution algorithm to perform joint trajectory planning processing on the adjusted global path to obtain the robot motion trajectory instruction; the robot motion trajectory instruction is converted into a drive signal through a multi-axis collaborative control program to obtain the photovoltaic cleaning robot execution instruction.

2. The autonomous path planning system based on a photovoltaic cleaning robot according to claim 1, characterized in that: The environment perception module comprises: Point cloud acquisition unit, used to collect three-dimensional point cloud of the photovoltaic panel surface through the fusion technology of laser radar and depth camera; An adaptive filtering unit is used to perform noise separation processing on point cloud data based on a hybrid algorithm of wavelet transform and Kalman filter; The data fusion unit is used to align the de-noised point cloud data with the photovoltaic panel tilt sensor data in time and space to generate photovoltaic surface de-noised data.

3. The autonomous path planning system based on a photovoltaic cleaning robot according to claim 1, characterized in that: The function formula of the improved SLAM algorithm in the path planning module is as follows: Where M t is the dynamic map constructed at time t, M is the candidate map set, z t is the sensor observation data, x t is the robot pose, α k is the confidence weight of the kth submap, m k is the spatial feature description of the kth sub-map, β is the map smoothing constraint coefficient, is the submap gradient, and K is the total number of submaps.

4. The autonomous path planning system based on a photovoltaic cleaning robot according to claim 1, characterized in that: The structure of the spatiotemporal fusion convolutional neural network in the obstacle detection module includes: The spatial feature extraction layer is used to extract the spatial dimension features of the photovoltaic image through a 3D convolution kernel; The time series association layer is used to model the temporal changes of consecutive frame images using a long short-term memory network; The attribute classification layer is used to combine infrared thermal imaging data with image features for multimodal fusion classification and output obstacle attribute data.

5. The autonomous path planning system based on a photovoltaic cleaning robot according to claim 1, characterized in that: The optimization objective function of the quantum genetic algorithm in the dynamic adjustment module is: Where P is the path set, d i is the length of the i-th path, E i is the path energy consumption, w1 and w2 are weight coefficients, the optimization goal is to maximize F(P), and N is the total number of path segments.

6. The autonomous path planning system based on a photovoltaic cleaning robot according to claim 1, characterized in that: The multi-axis collaborative control program in the execution control module includes: A trajectory interpolation unit, used for smoothly interpolating the joint trajectory based on a quintic polynomial curve; A drive conversion unit, used for converting trajectory instructions into pulse signals of motors of each joint through an inverse kinematics model; The feedback correction unit is used to compensate the phase error of the pulse signal in real time in combination with the encoder feedback data.

7. The autonomous path planning system based on a photovoltaic cleaning robot according to claim 3, characterized in that: The improved SLAM algorithm also includes: Dynamic object removal unit, which detects moving obstacles using optical flow and removes their corresponding point clouds from the map; A map update unit is used to perform incremental map updates of static environment features based on a Bayesian probability model.

8. The autonomous path planning system based on a photovoltaic cleaning robot according to claim 4, characterized in that: The spatiotemporal fusion convolutional neural network also includes: The attention mechanism unit is used to perform weighted fusion of the salient areas of infrared thermal imaging data through the channel attention module; The anti-occlusion unit is used to complete and reconstruct partially occluded obstacle images using a generative adversarial network.

9. The autonomous path planning system based on a photovoltaic cleaning robot according to claim 5, characterized in that: The quantum genetic algorithm adopts the following encoding method: Quantum chromosome encoding, which encodes the coordinates of the path nodes and the steering angles into the superposition state of quantum bits; Adaptive revolving gate update dynamically adjusts the rotation angle of quantum bits according to the path conflict coefficient to accelerate convergence.

10. An autonomous path planning method based on a photovoltaic cleaning robot, characterized in that: The following steps are involved: S1: Use multi-source heterogeneous sensors to collect and process the three-dimensional point cloud of the photovoltaic panel surface to obtain photovoltaic surface point cloud data, and then use the adaptive filtering algorithm to remove noise from the photovoltaic surface point cloud data to obtain photovoltaic surface noise reduction data; S2: Dynamically map the photovoltaic surface noise reduction data using the improved SLAM algorithm to obtain the photovoltaic surface topology map, and then perform global path optimization on the photovoltaic surface topology map based on the hierarchical reinforcement learning algorithm to obtain the initial optimal path; S3: Use the spatiotemporal fusion convolutional neural network to perform obstacle recognition processing on the real-time collected photovoltaic panel images to obtain obstacle distribution information, and combine the infrared thermal imaging technology to perform feature extraction processing on the physical properties of the obstacles to obtain obstacle attribute data; S4: Conflict analysis is performed on the obstacle attribute data and the initial optimal path through the fuzzy logic controller to obtain the path conflict coefficient, and the path conflict coefficient is dynamically replanned based on the quantum genetic algorithm to obtain the adjusted global path; S5: Use the kinematic inverse solution algorithm to perform joint trajectory planning on the adjusted global path to obtain the robot motion trajectory instruction, and use the multi-axis collaborative control program to convert the robot motion trajectory instruction into a drive signal to obtain the photovoltaic cleaning robot execution instruction.

Images