A cleaning robot intelligent control method and system for a photovoltaic power station

Abstract

The application discloses a kind of cleaning robot intelligent control method and system for photovoltaic power station, it is related to cleaning robot intelligent control technical field, including the image data of photovoltaic power station is collected, using self-organizing evolution segmentation technology constructs semantic model, generates static semantic map based on semantic model, and the comprehensive pollution state of each area is analyzed;Using improved ant colony algorithm to calculate the optimal path of each robot, in the process of executing cleaning task, using DWA algorithm to sense obstacle and intelligently avoid obstacle, after obstacle avoidance, dynamically update path and continue to execute cleaning task, using optical texture feature detection method to analyze the pollution characteristics of pollution area, dynamically formulate cleaning strategy according to analysis result.The precise identification of pollution area, intelligent optimization of path planning, dynamic obstacle avoidance and adaptive cleaning strategy function are realized, and the intelligentization, refinement and operation efficiency of photovoltaic power station cleaning operation are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for cleaning robots, and in particular to an intelligent control method and system for cleaning robots used in photovoltaic power plants. Background Technology

[0002] With the continuous development of new energy technologies, photovoltaic power generation, as a green and clean energy form, has been widely used globally. Photovoltaic modules deployed in photovoltaic power plants are exposed to the outdoor environment for extended periods, making them susceptible to various pollutants such as dust, bird droppings, and fallen leaves. These pollutants significantly reduce the light intensity on the module surface, thus affecting overall power generation efficiency. To maintain the stable operation and high-efficiency production of photovoltaic power plants, regular cleaning operations have become an essential maintenance method. With the development of intelligent manufacturing and automated control technologies, more and more research and products are exploring automated cleaning solutions for photovoltaic modules based on robotics, aiming to reduce costs and increase efficiency, alleviate manual labor burdens, and improve cleaning quality. Against this backdrop, photovoltaic cleaning robot systems based on technologies such as image recognition, path planning, and intelligent control have become a hot research and industrial focus.

[0003] However, existing photovoltaic cleaning robot systems generally suffer from several technical bottlenecks. First, most systems rely solely on traditional image segmentation or manually defined navigation paths, lacking a deep understanding of the complex semantic information within the photovoltaic environment. This makes it difficult to accurately identify the spatial distribution and category characteristics of contaminated areas, resulting in simplistic cleaning strategies and low efficiency. Second, in multi-robot collaborative operation scenarios, most path planning algorithms employ traditional ant colony algorithms or static path methods, failing to effectively combine factors such as contamination levels, component distribution, and task priorities for dynamic optimization. This leads to path redundancy and frequent task conflicts. Furthermore, existing systems typically rely on simple rule-based judgments for obstacle avoidance, lacking real-time responses to dynamic changes in obstacles in complex environments, easily causing risks such as jamming and collisions. At the cleaning decision-making level, most systems have not yet established a linkage mechanism between contamination type and cleaning method, making it difficult to dynamically adjust cleaning strategies based on actual contamination characteristics, resulting in a low level of intelligence. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an intelligent control method and system for cleaning robots in photovoltaic power plants, which solves the problems of low accuracy of photovoltaic module contamination identification, poor path planning efficiency of cleaning robots, and lack of dynamic adaptation of cleaning strategies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent control method for a cleaning robot used in a photovoltaic power plant, comprising,

[0008] Image data of photovoltaic power plants are collected, a semantic model is constructed using self-organizing evolutionary segmentation technology, a static semantic map is generated based on the semantic model, and the overall pollution status of each area is analyzed.

[0009] An improved ant colony algorithm is used to calculate the optimal path for each robot. During the cleaning task, the DWA algorithm is used to perceive obstacles and perform intelligent obstacle avoidance. After obstacle avoidance, the path is dynamically updated to continue the cleaning task. The optical texture feature detection method is used to analyze the pollution characteristics of the polluted area, and the cleaning strategy is dynamically formulated based on the analysis results.

[0010] Collect operational data and robot adjustment records during the cleaning process, and upload them to the database for storage.

[0011] As a preferred embodiment of the intelligent control method for cleaning robots used in photovoltaic power plants according to the present invention, the following steps are described: Collecting image data from the photovoltaic power plant and constructing a semantic model using self-organizing evolutionary segmentation technology refers to collecting image data from the power plant, generating an image data set V, extracting multi-dimensional feature vectors from the images, and obtaining the feature point cloud set F corresponding to the i-th image. i ;

[0012] Clustering algorithm was used to analyze F. i Classification is performed to obtain cluster segmentation regions C. k This includes both normal and contaminated areas;

[0013] Cluster segmentation region C k Each feature point f in j Define the segmentation contribution function S(f) j );

[0014] Set a feature point contribution threshold τ, if S(f j If )≥τ, retain feature points f j Otherwise, it will not be retained;

[0015] Reconstructed and filtered set of clustered segmentation regions C ′ ;

[0016] Cluster the region set C ′ Compared with the original image V i Point cloud F i Combined, generate the pixel-level supervision label matrix L of the i-th image. i Image V i With the supervised label matrix L i Pairing to generate training samples (V) i ,L i );

[0017] Define a semantic model M, and select DeepLabv3+ as the architecture for the semantic segmentation network. The network parameters of the model are optimized through training and fitted, and the image data V is used. i Input the semantic model M, and the output will be the predicted label matrix L corresponding to the image. i,pred ;

[0018] The cross-entropy loss method is used to calculate the value based on the predicted label matrix L. i,pred With the supervised label matrix L i Calculate the loss function

[0019] Update the network parameters μ of the model using gradient descent, and then update the model network parameters μ. (t+1) Used for the next training session;

[0020] When the number of training rounds reaches the set limit, the final calculated μ is used. ＊ The semantic model M is obtained by using the network parameters of the model.

[0021] As a preferred embodiment of the intelligent control method for cleaning robots used in photovoltaic power plants according to the present invention, the step of generating a static semantic map based on a semantic model and analyzing the comprehensive pollution status of each area refers to inputting all image data into the semantic model M to obtain the prediction label matrix L of the corresponding image. i,pred ;

[0022] By integrating all the predicted label matrices in image spatial order, a complete static semantic map S is obtained;

[0023] The static semantic map is divided into O fixed-size grid regions, and the percentage ρ of contaminated pixels in each grid is calculated. o ;

[0024] The proportion of pollution ρ o As a pollution level for the area, the higher the pollution level, the higher the cleaning priority. A minimum pollution threshold ρ is set. min If ρ o >ρ min If the condition is met, a regional pollution cleaning task will be generated; otherwise, it will not be generated.

[0025] As a preferred embodiment of the intelligent control method for cleaning robots used in photovoltaic power plants according to the present invention, wherein: the step of using the improved ant colony algorithm to calculate the optimal path for each robot refers to constructing a pollution weighted graph based on a static semantic map S, and recording each grid as a node R;

[0026] For any two nodes R x and R y The x-th node R x Starting from the y-th node Ry If the endpoint is adjacent to and traversable between two points, then an undirected edge e is constructed. xy Calculate the guiding value g for each edge. xy ;

[0027] For each edge e xy Set initial pheromone values

[0028] Constructing the path selection probability of the cleaning robot based on guiding values ​​and pheromone values.

[0029] Initialize the robot path Road, causing robot z to start from the initial node c0, and set the path to Road. z :

[0030] according to Randomly select the next target node R y And select node R y Add a path, denoted as c1, until the run reaches the set maximum path length or cannot proceed further, then output the complete path Road. z ;

[0031] Calculate the amount of contamination P removed by robot z along its path. z With the total path length Q z ;

[0032] Based on pollution value P z With total path length Q z Evaluate the quality of the path and compute its opposite edge e. xy Pheromones contribute Δτ xy ;

[0033] Update the pheromone values ​​of all edges based on the results of this round.

[0034] Repeatedly update pheromone values ​​and generate new paths until the maximum number of iterations is reached, then output the final set of optimal paths (Road). * This serves as the optimal path for each robot.

[0035] As a preferred embodiment of the intelligent control method for cleaning robots used in photovoltaic power plants according to the present invention, wherein: during the execution of the cleaning task, the DWA algorithm is used to perceive obstacles and perform intelligent obstacle avoidance, and the path is dynamically updated after obstacle avoidance to continue the cleaning task, the static semantic map is embedded into a two-dimensional Cartesian coordinate system to initialize the robot's motion state Robot0.

[0036] Constructing a feasible speed space for robots (Speed) acc Speed bounand Speed safe ;

[0037] The Speed acc Speed ​​is the range of speeds a robot can achieve in the next cycle based on its current speed, under acceleration constraints. boun Speed ​​is the maximum and minimum speed supported by the robot's basic hardware. safe The set of velocities that will not collide with obstacles in the predicted trajectory is used to ultimately obtain the feasible velocity space Speed. fea ;

[0038] Select local navigation target point Based on the feasible speed space Speed fea With navigation target point Perform trajectory prediction;

[0039] By integrating the robot's pose at each moment, we obtain the pose sequence Seq of the robot's obstacle avoidance trajectory;

[0040] With c ＊ As the target point, the trajectory score G(v,ω) is calculated based on the robot's pose;

[0041] Choose the speed combination with the highest score (v) * ω * () as an obstacle avoidance command.

[0042] As a preferred embodiment of the intelligent control method for cleaning robots used in photovoltaic power plants according to the present invention, wherein: the method of using optical texture feature detection to analyze the pollution characteristics of the polluted area, and dynamically formulating a cleaning strategy based on the analysis results, refers to using path planning (Road). * and obstacle avoidance commands (v * ω * Upon reaching the task point, the acquired image data is converted to grayscale to obtain V. gray This allows us to obtain the pixel grayscale value (Gray) and LBP value for each image.

[0043] For the grayscale image V gray Local contrast enhancement is performed to obtain the preprocessed image V. norm ;

[0044] The preprocessed image V norm Divide into num image blocks W s ;

[0045] Evaluate the image texture sharpness (Contrast), image texture coarsness (LBPVar), and image edge density (EdgeDensity);

[0046] Set feature threshold Contrast thr ,LBPVar thr and EdgeDensity thr Based on the characteristics of the pollution, determine the type of pollution and formulate corresponding cleaning strategies;

[0047] After the cleaning strategy is completed, the pollution status of the area is reassessed. If pollutants still exist, a corresponding cleaning strategy is generated again; otherwise, no strategy is generated. If the number of times a cleaning strategy is initiated in a single cleaning task reaches the set limit, the cleaning is abandoned and an alarm is triggered.

[0048] As a preferred embodiment of the intelligent control method for cleaning robots used in photovoltaic power plants according to the present invention, the step of collecting operating data and robot adjustment records during the cleaning process and uploading them to the database for storage refers to uploading the pollution detection results, path optimization process, selected cleaning strategies, and actual trajectory adjustment data recorded by the robot during the cleaning process to the database.

[0049] Secondly, the present invention provides an intelligent control system for a cleaning robot in a photovoltaic power plant, comprising,

[0050] The image acquisition module is used to collect raw image data of the photovoltaic power station, providing input data for subsequent semantic modeling and pollution analysis;

[0051] The semantic modeling module is used to segment images using self-organizing evolutionary segmentation technology and construct a semantic model of a photovoltaic power station.

[0052] The pollution status analysis module is used to generate static semantic maps using semantic models, analyze the pollution level of each area in the map, and use this as a basis for task allocation.

[0053] The path planning module is used to plan the optimal cleaning path for each robot using an improved ant colony algorithm.

[0054] The dynamic obstacle avoidance module is used to perceive obstacles and dynamically avoid them using the DWA algorithm during the robot's cleaning task. After avoiding obstacles, the path is updated in real time to continue the task.

[0055] The pollution feature analysis and cleaning strategy customization module is used to analyze the specific characteristics of the polluted area using optical texture feature detection methods and dynamically formulate cleaning strategies.

[0056] The data upload module is used to collect the robot's operating data and adjustment records during the cleaning process and upload them to the database for centralized storage.

[0057] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the intelligent control method for a cleaning robot for a photovoltaic power station as described in the first aspect of the present invention.

[0058] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent control method for a cleaning robot for a photovoltaic power plant as described in the first aspect of the present invention.

[0059] The beneficial effects of this invention are as follows: By using a high-precision semantic modeling mechanism to identify contaminated areas, the pixel-level recognition capability of contaminated areas of photovoltaic modules is significantly improved, solving the problem of inaccurate judgment of contaminated boundaries in traditional image processing methods; by integrating the image prediction matrix in spatial order to generate a static semantic map, a structured expression and spatial reconstruction of contamination information is realized, providing reliable data support for subsequent task scheduling and path planning; when optimizing the path, a contamination weight map is constructed and an improved ant colony algorithm is used to enhance the robot's path selection tendency to cover highly contaminated areas, and the global optimality and stability of the search path are enhanced through a pheromone adjustment mechanism, optimizing the task allocation and execution efficiency of the multi-robot system; combined with the DWA algorithm for local obstacle avoidance control, the robot's ability to cope with dynamic or unknown obstacles in the actual working environment is improved, ensuring the continuity and safety of the path execution process; in the cleaning stage, the type of contamination is automatically classified by optical texture feature detection, and differentiated cleaning strategies are matched for liquid, solid, or viscous contamination to achieve precise and differentiated cleaning operations; after cleaning, the cleaning robot can dynamically adjust the operation strategy or trigger an alarm according to the cleaning effect, further enhancing the closed-loop control capability of the system. Attached Figure Description

[0060] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a flowchart of an intelligent control method for a cleaning robot used in a photovoltaic power station, as shown in Example 1.

[0062] Figure 2 This is a structural diagram of an intelligent control system for a cleaning robot used in a photovoltaic power station, as shown in Example 1.

[0063] Figure 3This is a flowchart of an intelligent control system for a cleaning robot used in a photovoltaic power station, as shown in Example 1. Detailed Implementation

[0064] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0065] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0066] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0067] Example 1, referring to Figures 1-3 This is the first embodiment of the present invention, which provides an intelligent control method for a cleaning robot in a photovoltaic power station, comprising the following steps:

[0068] S1: Collect image data of photovoltaic power plants, construct semantic models using self-organizing evolutionary segmentation technology, generate static semantic maps based on semantic models, and analyze the comprehensive pollution status of each region.

[0069] Specifically, power plant image data is collected, generating an image data set V. Multidimensional feature vectors are extracted from the images to obtain the feature point cloud set F corresponding to the i-th image. i :

[0070] F i ={f1,f2,…,f m},

[0071] Among them, F i Let f be the feature set corresponding to the i-th image. m The feature set contains a total of m feature points;

[0072] Clustering algorithm was used to analyze F. i Classification is performed to obtain cluster segmentation regions C. k This includes both normal and contaminated areas;

[0073] Cluster segmentation region C k Each feature point f in j Define the segmentation contribution function S(f) j ):

[0074] S(f j )=α·Intra(f j C k )-β·Inter(f j C other ),

[0075] Wherein, S(f j ) is the feature point f j For the current cluster segmentation region C k The segmentation quality contribution score, Intra(f) j C k ) is the feature point f j Its cluster segmentation region C k The similarity of other feature points within the region, Inter(f) j C other ) is the feature point f j Region C, segmented from other clusters other The similarity of feature points is defined by α, which is the weight of consistency within the cluster segmentation region, and β, which is the weight of difference between the cluster segmentation regions.

[0076] The Intra(f) j C k The specific expression for ) is:

[0077]

[0078] in, For cluster segmentation region C k The total number of feature points in f h For the h-th feature point, sim(f j ,f h ) is the feature point f j With feature point f h The similarity is specifically expressed as:

[0079]

[0080] Among them, ||f j -f h || represents the Euclidean distance between two feature points;

[0081] The Inter(f) j C other The specific expression for ) is:

[0082]

[0083] Among them, |C other| represents the total number of feature points in other cluster segmentation regions;

[0084] Set a feature point contribution threshold τ, if S(f j If )≥τ, retain the feature point f j Otherwise, it will not be retained:

[0085] C ′ K ={f j ∈C k |S(f j )≥τ},

[0086] Among them, C ′ K For the selected cluster segmentation region C k ;

[0087] Reconstructed and filtered set of clustered segmentation regions C ′ :

[0088] C ′ ={C ′ 1,C ′ 2,…,C ′ K},

[0089] Cluster the region set C ′ Compared with the original image V i Point cloud F i Combined, generate the pixel-level supervision label matrix L of the i-th image. i Image V i With the supervised label matrix L i Pairing to generate training samples (V) i ,L i );

[0090] Define a semantic model M, and select DeepLabv3+ as the architecture for the semantic segmentation network. The network parameters of the model are optimized through training and fitted, and the image data V is used. i Input the semantic model M, and the output will be the predicted label matrix L corresponding to the image. i,pred ;

[0091] The cross-entropy loss method is used to calculate the value based on the predicted label matrix L. i,pred With the supervised label matrix L i Calculate the loss function

[0092]

[0093] Where, N imag Let be the total number of pixels in the i-th image, and pixels be the pixel index in the image. Let L be the indicator function, and let L be the predicted label matrix. i,pred The pixels in the supervision label matrix L i If the number of pixels in the array is different, the value is 1; otherwise, the value is 0.

[0094] Update the network parameters μ of the model using gradient descent:

[0095]

[0096] Where, μ (t) μ represents the network parameters currently used by the model network. (t+1) For the updated model network parameters, loss function The gradient with respect to the model parameter μ, where η is the learning rate obtained using the adaptive optimizer;

[0097] The updated model network parameters μ (t+1) Used for the next training session;

[0098] When the training epochs reach the set upper limit, use the final calculated μ. * The semantic model M is obtained by using the network parameters of the model.

[0099] By constructing an image feature point cloud set, the multi-dimensional information such as optical texture, color gradient, and edge morphology of the photovoltaic module surface can be comprehensively reflected. Clustering the feature point cloud set effectively distinguishes the structural differences between normal and contaminated areas in the image. The clustering results are filtered and reconstructed using a feature point segmentation contribution function and a set contribution threshold mechanism, effectively eliminating feature points with blurred edges and enhancing the annotation accuracy of semantic labels. The semantic segmentation model built on the DeepLabv3+ architecture achieves fine segmentation of contaminated areas in complex scenes. The final model network parameters are calculated by setting the number of training rounds, ensuring the convergence stability and prediction consistency of the semantic model.

[0100] Furthermore, all image data are input into the semantic model M to obtain the predicted label matrix L for the corresponding image. i,pred :

[0101] L i,pred =M(V) i ),

[0102] By integrating all the predicted label matrices in image spatial order, a complete static semantic map S is obtained:

[0103] S={L 1,pred ,L 2,pred ,…,L n,pred},

[0104] Among them, Ln,pred There are a total of n predicted label matrices;

[0105] The static semantic map is divided into O fixed-size grid regions, and the percentage ρ of contaminated pixels in each grid is calculated. o :

[0106]

[0107] Where, ρ o N represents the pollution percentage of the o-th region. o Let be the total number of pixels in the o-th region. This represents the total number of pixels in the o-th region that fall under the pollution category;

[0108] The proportion of pollution ρ o The higher the pollution level of the area, the higher the cleaning priority. A minimum pollution threshold ρ is set. min If ρ o >ρ min If the condition is met, a regional pollution cleaning task will be generated; otherwise, it will not be generated.

[0109] By inputting all image data into a semantic model to obtain a predicted label matrix, high-precision semantic parsing of the image content of photovoltaic power plants can be performed, enabling automatic labeling and classification of polluted areas in each image. Integrating all predicted label matrices according to the spatial arrangement of the images generates a complete static semantic map, effectively restoring the layout information of photovoltaic power plant components in physical space and constructing a global pollution distribution map with spatial correlation. Dividing the static semantic map into grid regions and calculating the pixel proportion of polluted areas within each grid enables a quantitative assessment of pollution distribution, improving the readability of the pollution information structure. Using the pixel proportion of polluted areas as the regional pollution value and determining whether to generate a cleaning task based on the pollution value enables dynamic evaluation of cleaning task priority, avoiding ineffective cleaning of slightly polluted areas and optimizing resource allocation and cleaning efficiency.

[0110] S2: The improved ant colony algorithm is used to calculate the optimal path for each robot. During the cleaning task, the DWA algorithm is used to perceive obstacles and perform intelligent obstacle avoidance. After obstacle avoidance, the path is dynamically updated to continue the cleaning task. The optical texture feature detection method is used to analyze the pollution characteristics of the polluted area, and the cleaning strategy is dynamically formulated based on the analysis results.

[0111] Specifically, a pollution-weighted graph is constructed based on the static semantic map S, and each grid is recorded as a node:

[0112] R = {R1, R2, ..., R} O},

[0113] Among them, R O There are a total of O nodes;

[0114] For any two nodes R x and R y The x-th node R x Starting from the y-th node R y If the endpoint is adjacent to and traversable between two points, then an undirected edge e is constructed. xy Calculate the guiding value g for each edge. xy :

[0115] g xy =γ·d xy +δ·(1+ρ y ),

[0116] Where γ and δ are coefficients empirically set to adjust the proportion of physical distance and pollution value in the guide value, and γ + δ = 1, d xy For node R x R y The Euclidean distance between them, ρ y Let be the pollution value of the y-th node;

[0117] For each edge e xy Set initial pheromone values

[0118]

[0119] in, For node R x With node R y The edge between them has the pheromone value at the 0th iteration, i.e., the initial value, and Z is the number of robots participating in the cleaning task. The average distance between any two adjacent passable nodes;

[0120] Constructing the path selection probability of the cleaning robot based on guiding values ​​and pheromone values.

[0121]

[0122] in, For the z-th robot at node R x Select to go to node R y The probability, For edge e xy The pheromone value E at the r-th iteration x For the current node R x Let l be the set of all reachable neighboring nodes, and let l be a node belonging to set E. x Nodes l and g xl For edge e xl The guiding value, ∈ are control parameters obtained through multiple sets of simulation experiments, used to regulate the degree of influence between pheromone value and guidance value;

[0123] Initialize the robot path Road, causing robot z to start from the initial node c0, and set the path to Road. z :

[0124] Road z =[c0],

[0125] according to Randomly select the next target node R y And select node R y Add a path, denoted as c1, until the run reaches the set maximum path length or cannot proceed further, then output the complete path Road. z :

[0126] Road z =[c0,c1,...,c q ],

[0127] Among them, c q Robot z progressed a total of q nodes.

[0128] Calculate the amount of contamination P removed by robot z along its path. z :

[0129] P z =∑y∈Road z ρ y ,

[0130] Calculate the total path length Q of robot z. z :

[0131]

[0132] in, For node R x To node R y The connected edge e xy The distance;

[0133] Based on pollution value P z With total path length Q z Evaluate the quality of the path and compute its opposite edge e. xy Pheromones contribute Δτ xy :

[0134]

[0135] Update the pheromone values ​​of all edges based on the results of this round.

[0136]

[0137] in, For the update of e xy The pheromone value, where θ is the set pheromone volatility coefficient;

[0138] Repeatedly update pheromone values ​​and generate new paths until the maximum number of iterations is reached, then output the final set of optimal paths (Road). * This serves as the optimal path for each robot.

[0139] By constructing a grid-like pollution-weighted graph using a static semantic map, the pollution level of different areas can be effectively quantified, and the spatial layout can be transformed into a graph structure, providing a mathematical modeling foundation for path optimization. Establishing undirected edges between nodes and calculating their guidance and pheromone values ​​allows the path search process to consider both spatial accessibility and pollution removal efficiency, improving the algorithm's search tendency for highly polluted areas. Calculating the robot's path selection probability, and comprehensively considering guidance and pheromone values, enhances the diversity and convergence of path planning. Updating the pheromone value based on the result of each path selection allows for dynamic adjustment of the pheromone guidance direction during path search, improving the reinforcement effect of efficient paths, and ultimately outputting a stable, efficient, and comprehensive set of optimal paths.

[0140] Furthermore, the static semantic map is embedded into a two-dimensional Cartesian coordinate system to initialize the robot's motion state, Robot0:

[0141] Robot0=[X0,Y0,θ0,v0,ω0],

[0142] Where X0 and Y0 are the robot's initial positions in the coordinate system, θ0 is the counterclockwise orientation angle relative to the x-axis, and v0 and ω0 are the robot's initial linear velocity and angular velocity.

[0143] Constructing a feasible speed space for robots (Speed) acc Speed boun and Speed safe :

[0144] The Speed acc The range of speeds the robot can reach in the next cycle based on its current speed, under acceleration constraints, is specifically expressed as:

[0145] Speed acc =

[0146] {(v,ω)∣v∈[v t -a v ·Δt,v t +a v ·Δt],ω∈[ω t-a ω ·Δt,ω t +a ω ·Δt]},

[0147] Where v is the optional linear velocity for the next step, and ω is the optional angular velocity for the next step. t Let a be the linear velocity of the robot at the current moment. v ω represents the maximum acceleration of the robot at its online speed, Δt is the time step of the control cycle, and ω is the maximum acceleration of the robot at its online speed. t Let a be the angular velocity of the robot at the current moment. ω This represents the robot's maximum acceleration in terms of angular velocity.

[0148] The Speed boun The maximum and minimum speed boundaries supported by the robot's basic hardware are specifically represented as follows:

[0149] Speed boun ={(v,ω)|v∈[v min ,v max ],ω∈[ω min ,ω max ]},

[0150] Among them, v min v max ω represents the minimum and maximum speeds supported by the robot hardware. min ω max The minimum and maximum angular velocities supported by the robot hardware;

[0151] The Speed safe Let be the set of velocities that will not collide with obstacles in the predicted trajectory, specifically represented as:

[0152]

[0153] in, d represents the shortest braking distance at the current speed. min This represents the distance between the robot and the nearest obstacle on the current trajectory.

[0154] The final feasible speed space is Speed fea :

[0155] Speed fea =Speed acc ∩Speed boun ∩Speed safe ,

[0156] Select local navigation target point

[0157]

[0158] Where, goal is the final target point of the path, c * Road is the path closest to the robot. * The point in the middle contains coordinates (X... * ,Y * ), location t The current position of the robot, including coordinates (X, Y, Z). t ,Y t ), dis obs (location t ) represents the distance between the current robot position and the nearest obstacle, dis. safe dis is the safe distance threshold between the robot and obstacles. thr For the distance threshold obtained through Euclidean algorithm, dis(c) * ,location t ) represents the robot's current position and path point c. * The Euclidean distance is specifically expressed as:

[0159]

[0160] Among them, X * and Y * For path point c * x and y coordinates, X t and Y t These are the x and y coordinates of the robot's current position;

[0161] Based on the feasible speed space Speed fea With navigation target point Perform trajectory prediction:

[0162]

[0163] Among them, X t+1 Let Y be the horizontal coordinate of the robot at the next moment. t+1 Let θ be the robot's ordinate at the next moment. t+1 The robot's orientation angle at the next moment;

[0164] By integrating the robot's pose at each moment, we obtain the pose sequence Seq of the robot's obstacle avoidance trajectory:

[0165] Seq={(X0,Y0,θ0),(X1,Y1,θ1),…,(X T ,Y T ,θ T )},

[0166] Among them, (X)T ,Y T ,θ T () represents the robot's final pose at the end of simulation cycle T;

[0167] With c * As the target point, the trajectory score G(v,ω) is calculated based on the robot's pose:

[0168]

[0169] in, The angular score for the robot at the end of the trajectory is 1, with the score closer to the target. * ,location T ) represents the robot's position at the end of the trajectory and the path point c. * Euclidean distance, dis max The maximum acceptable Euclidean distance for obstacle avoidance;

[0170] Choose the speed combination with the highest score (v) * ω * ):

[0171]

[0172] Finally, the velocity combination (v) * ω * () as an obstacle avoidance command.

[0173] By constructing a feasible velocity space for the robot, and combining velocity variation constraints, hardware performance boundaries, and a set of velocities that do not collide with obstacles, a comprehensive screening of the robot's dynamic feasible solution space is achieved. This avoids generating uncontrollable or high-risk velocity commands, improving the safety and robustness of obstacle avoidance. Selecting a local navigation target point and predicting the trajectory within the feasible velocity space can effectively guide the robot towards the target area while maintaining a sensitive response to surrounding obstacles. Integrating the robot's pose at continuous moments to form a complete obstacle avoidance trajectory pose sequence can intuitively reflect the continuity and dynamic feasibility of the obstacle avoidance path in space. Based on the trajectory scoring function, different velocity combinations are scored, and the velocity command with the best score is selected, improving the robot's safety, path smoothness, and target proximity during obstacle avoidance.

[0174] Going a step further, using route planning (Road) * and obstacle avoidance commands (v * ω * Upon reaching the task point, the acquired image data is converted to grayscale to obtain V. gray :

[0175] V gray =Grayscale(V),

[0176] Among them, Grayscale is an existing image grayscale conversion function in the image processing library, which can obtain the pixel grayscale value Gray and LBP value of each image;

[0177] For the grayscale image V gray Local contrast enhancement is performed to obtain the preprocessed image V. norm :

[0178] V norm =CLAHE(V gray ),

[0179] Wherein, CLAHE is an existing function in the image processing library for enhancing local image contrast;

[0180] The preprocessed image V norm Divide into num image blocks W s :

[0181] W s ={W1,W2,…,W num},

[0182] Among them, W s For the s-th image patch, W num There are a total of num image patches in the image patch set;

[0183] Evaluate the sharpness of image textures (Contrast):

[0184]

[0185] in, Let be the total number of pixels in the s-th image patch, and let Gray... pixel Let be the grayscale value of the 11th pixel. It is the average of the grayscale values ​​of all pixels;

[0186] LBPVar is used to evaluate the coarseness of image texture.

[0187]

[0188] Among them, LBP pixel Let LBP feature value be the value of the eleventh pixel. The average value of the LBP feature values ​​of all pixels;

[0189] Evaluate the edge density of the image:

[0190]

[0191] in, The number of pixels identified as edge pixels within image block s;

[0192] Set feature threshold Contrast thr ,LBPVar thr and EdgeDensity thr Based on the characteristics of the pollution, determine the type of pollution and use the corresponding cleaning strategy;

[0193] If the image texture clarity is greater than or equal to the threshold, and the image texture roughness and image edge density are less than the threshold, then the contamination type is liquid contamination. Use water spray to rinse, wait a moment for the flow to settle, and then wipe with a wet brush.

[0194] If the image texture clarity and image edge density are greater than or equal to the threshold, and the image texture roughness is less than the threshold, then the pollution type is solid pollution. Use a negative pressure vacuum cleaner to remove the particles and then wipe with a wet brush.

[0195] If the image texture clarity, image texture roughness, and image edge density are all greater than or equal to the threshold, the contamination type is sticky contamination. Use water spray to rinse multiple times, use a vacuum cleaner to remove the cleaning material, and then wipe with a wet brush.

[0196] After the cleaning strategy is completed, the pollution status of the area is reassessed. If pollutants still exist, a corresponding cleaning strategy is generated again; otherwise, no strategy is generated. If the number of times a cleaning strategy is initiated in a single cleaning task reaches the set limit, the cleaning is abandoned and an alarm is triggered.

[0197] By converting image data to grayscale, pixel grayscale values ​​and LBP values ​​are extracted, enhancing the image's optical texture feature representation capability and facilitating numerical analysis of the surface state of contaminated areas. Local contrast enhancement is performed on the grayscale image to improve the sensitivity of subsequent feature extraction by enhancing edge contours and texture details in the contaminated image. This improves the separability of local visual characteristics of contaminated areas and effectively overcomes misjudgments of contamination caused by uneven lighting or image blur. The image texture clarity, roughness, and edge density are evaluated sequentially to help accurately determine the nature of the contamination. Corresponding cleaning strategies are matched to different contamination types, and cleaning strategies can be dynamically formulated based on the optical characteristics of the actual contaminated area. After cleaning, the contamination status of the area is reassessed. If the contamination is not completely removed, a corresponding cleaning strategy is generated again. If the cleaning failure exceeds the maximum number of attempts, an alarm mechanism is triggered, ensuring that the system has a feedback response mechanism under extreme contamination or equipment malfunction conditions, improving system safety and task closed-loop control capabilities.

[0198] S3: Collect operational data and robot adjustment records during the cleaning process, and upload them to the database for storage.

[0199] Specifically, the data recorded by the robot during the cleaning process, including pollution detection results, path optimization, selected cleaning strategies, and actual trajectory adjustments, are uploaded to the database to form a closed-loop feedback loop.

[0200] By uploading the pollution detection results, path optimization process, selected cleaning strategies, and actual trajectory adjustment data recorded by the robot during the cleaning process to the database, unified data storage and management are achieved, forming a complete and traceable operation record. This effectively solves the problem of data breakpoints and loss, and provides a reliable data foundation for subsequent data analysis, state trend prediction, and adjustment strategy optimization.

[0201] This embodiment also provides an intelligent control system for a cleaning robot in a photovoltaic power plant, including:

[0202] The image acquisition module is used to collect raw image data of the photovoltaic power station, providing input data for subsequent semantic modeling and pollution analysis;

[0203] The semantic modeling module is used to segment images using self-organizing evolutionary segmentation technology and construct a semantic model of a photovoltaic power station.

[0204] The pollution status analysis module is used to generate static semantic maps using semantic models, analyze the pollution level of each area in the map, and use this as a basis for task allocation.

[0205] The path planning module is used to plan the optimal cleaning path for each robot using an improved ant colony algorithm.

[0206] The dynamic obstacle avoidance module is used to perceive obstacles and dynamically avoid them using the DWA algorithm during the robot's cleaning task. After avoiding obstacles, the path is updated in real time to continue the task.

[0207] The pollution feature analysis and cleaning strategy customization module is used to analyze the specific characteristics of the polluted area using optical texture feature detection methods and dynamically formulate cleaning strategies.

[0208] The data upload module is used to collect the robot's operating data and adjustment records during the cleaning process and upload them to the database for centralized storage.

[0209] This embodiment also provides a computer device applicable to an intelligent control method for a cleaning robot in a photovoltaic power plant, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent control method for a cleaning robot in a photovoltaic power plant as proposed in the above embodiment.

[0210] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0211] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the intelligent control method for a cleaning robot used in a photovoltaic power station as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0212] In summary, this invention significantly improves the pixel-level recognition capability of contaminated areas in photovoltaic modules by using a high-precision semantic modeling mechanism to identify contaminated areas, solving the problem of inaccurate contamination boundary judgment in traditional image processing methods. It integrates image prediction matrices in spatial order to generate a static semantic map, achieving structured expression and spatial reconstruction of contamination information, providing reliable data support for subsequent task scheduling and path planning. When optimizing paths, a contamination weight map is constructed and an improved ant colony algorithm is used to enhance the robot's path selection tendency towards highly contaminated areas. Furthermore, a pheromone adjustment mechanism enhances the global optimality and stability of the search path, optimizing task allocation and execution efficiency in multi-robot systems. Combining the DWA algorithm for local obstacle avoidance control improves the robot's ability to cope with dynamic or unknown obstacles in actual working environments, ensuring the continuity and safety of path execution. During the cleaning phase, optical texture feature detection automatically classifies contamination types, matching differentiated cleaning strategies for liquid, solid, or viscous contamination to achieve precise and differentiated cleaning operations. After cleaning, the cleaning robot can dynamically adjust its operating strategy or trigger alarms based on the cleaning effect, further enhancing the system's closed-loop control capability.

[0213] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An intelligent control method for a cleaning robot used in photovoltaic power plants, characterized in that: include, Image data of photovoltaic power plants are collected, a semantic model is constructed using self-organizing evolutionary segmentation technology, a static semantic map is generated based on the semantic model, and the overall pollution status of each area is analyzed. An improved ant colony algorithm is used to calculate the optimal path for each robot. During the cleaning task, the DWA algorithm is used to perceive obstacles and perform intelligent obstacle avoidance. After obstacle avoidance, the path is dynamically updated to continue the cleaning task. The optical texture feature detection method is used to analyze the pollution characteristics of the polluted area, and the cleaning strategy is dynamically formulated based on the analysis results. Collect operational data and robot adjustment records during the cleaning process, and upload them to the database for storage; The image data collected from the photovoltaic power station is used to construct a semantic model using self-organizing evolutionary segmentation technology, generating an image data set. Extract the multidimensional feature vectors of the image to obtain the feature point cloud set corresponding to the i-th image. ; Clustering algorithms were used to... Classification is performed to obtain clustered segmentation regions. This includes both normal and contaminated areas; Cluster segmentation regions Each feature point in Define the segmentation contribution function : ， in, For feature points For the current cluster segmentation region The contribution score of segmentation quality. For feature points Its cluster segmentation region The similarity of other feature points within the same area. For feature points Other clustering segmentation regions The degree of similarity of feature points in the middle The weights for the consistency within the cluster segmentation regions are set. Weights are set to account for the differences between cluster segmentation regions. Set feature point contribution threshold ,like Preserve feature points Otherwise, it will not be retained; Reconstructed and filtered cluster segmentation region set ; Clustering and segmenting regions Compared with the original image Point clouds Combined, a pixel-level supervised label matrix for the i-th image is generated. , to image With supervised label matrix Pairing to generate training samples ; A semantic model M is defined, and DeepLabv3+ is selected as the architecture for the semantic segmentation network. The network parameters of the model are optimized through training by fitting the image data. Input the semantic model M, and the output will be the predicted label matrix corresponding to the image. ; The cross-entropy loss method is used to calculate the value based on the predicted label matrix. With supervised label matrix Calculate the loss function ; Update the network parameters of the model using gradient descent. And update the model network parameters Used for the next training session; When the training rounds reach the set limit, use the final calculated value. As parameters of the model network, the semantic model M is obtained; The method of using an improved ant colony algorithm to calculate the optimal path for each robot refers to constructing a pollution-weighted graph based on a static semantic map S, and recording each grid as a node R. For any two nodes and The xth node Starting from the y-th node If the endpoint is adjacent to and traversable between two points, then an undirected edge is constructed. Calculate the guiding value for each edge. : ， in, and This is a coefficient, set empirically, used to adjust the proportion of physical distance and pollution value in the guide value, and , For nodes , The Euclidean distance between them Let be the pollution value of the y-th node; For each edge Set initial pheromone values ; Constructing the path selection probability of the cleaning robot based on guiding values ​​and pheromone values. : ， in, For the z-th robot at node Select to go to node The probability, For the edge The pheromone value at the r-th iteration, For the current node The set of all reachable neighboring nodes, where l is the set to which they belong. node l, For the edge The guiding value, and These are control parameters obtained through multiple sets of simulation experiments, used to regulate the degree of influence between pheromone values ​​and guidance values. Initialize the robot path Road, and let robot z start from the starting node. Departure, set the path as : according to Randomly select the next target node and the selected node Add to the path, denoted as The program will continue running until it reaches the set maximum path length or can no longer proceed, at which point it will output the complete path. ; Calculate the amount of contamination removed by robot z along its path. Total running path length ; Based on pollution value With total path length Evaluate the quality of the path and compute its opposite edges. Pheromones contribution : ， Update the pheromone values ​​of all edges based on the results of this round. ; Repeatedly update pheromone values ​​and generate new paths until the maximum number of iterations is reached, then output the final set of optimal paths. This serves as the optimal path for each robot.

2. The intelligent control method for a cleaning robot used in a photovoltaic power plant as described in claim 1, characterized in that: The process involves generating a static semantic map based on a semantic model and analyzing the overall pollution status of each region. This involves inputting all image data into the semantic model M to obtain the predicted label matrix for the corresponding image. ; By integrating all the predicted label matrices in image spatial order, a complete static semantic map S is obtained; The static semantic map is divided into O fixed-size grid regions, and the percentage of contaminated pixels in each grid is calculated. ; Pollution percentage As a measure of regional pollution levels, higher pollution levels necessitate higher cleaning priorities. A minimum pollution threshold is set. ,like If the condition is met, a regional pollution cleaning task will be generated; otherwise, it will not be generated.

3. The intelligent control method for a cleaning robot used in a photovoltaic power station as described in claim 2, characterized in that: During the cleaning task, the DWA algorithm is used to perceive obstacles and perform intelligent obstacle avoidance. After obstacle avoidance, the path is dynamically updated to continue the cleaning task. This involves embedding a static semantic map into a two-dimensional Cartesian coordinate system and initializing the robot's motion state. ; Constructing a feasible velocity space for robots , and ; The This refers to the range of speeds the robot can achieve in the next cycle, based on its current speed and under acceleration constraints. The maximum and minimum speeds supported by the robot's basic hardware. The set of velocities that will not collide with obstacles in the predicted trajectory is used to ultimately obtain the feasible velocity space. ; Select local navigation target point Based on feasible velocity space With navigation target point Perform trajectory prediction; By integrating the robot's pose at each moment, a pose sequence of the robot's obstacle avoidance trajectory is obtained. ; by As the target point, the trajectory score is calculated based on the robot's pose. : ， in, The angular score is given to the robot at the end of the trajectory; the closer it is to the target, the closer it is to 1. For the robot's position at the end of the trajectory and the waypoint Euclidean distance, The maximum acceptable Euclidean distance for obstacle avoidance. The maximum speed supported by the robot hardware, v is the optional next linear speed; Choose the speed combination with the highest score. As an obstacle avoidance command.

4. The intelligent control method for a cleaning robot used in a photovoltaic power station as described in claim 3, characterized in that: The method of using optical texture feature detection to analyze the pollution characteristics of the contaminated area and dynamically formulating cleaning strategies based on the analysis results refers to the use of path planning. and obstacle avoidance commands Upon reaching the task point, the acquired image data is converted to grayscale. This allows us to obtain the pixel grayscale value (Gray) and LBP value for each image. For grayscale images Local contrast enhancement is performed to obtain the preprocessed image. ; Preprocessed image Divide into num image blocks ; Evaluate the sharpness of image textures coarseness of image texture and image edge density ; Set feature threshold , and Based on the characteristics of the pollution, determine the type of pollution and formulate corresponding cleaning strategies; After the cleaning strategy is completed, the pollution status of the area is reassessed. If pollutants still exist, a corresponding cleaning strategy is generated again; otherwise, no strategy is generated. If the number of times a cleaning strategy is initiated in a single cleaning task reaches the set limit, the cleaning is abandoned and an alarm is triggered.

5. The intelligent control method for a cleaning robot used in a photovoltaic power station as described in claim 4, characterized in that: The process of collecting operational data and robot adjustment records during the cleaning process and uploading them to the database for storage refers to uploading the pollution detection results, path optimization process, selected cleaning strategies, and actual trajectory adjustment data recorded by the robot during the cleaning process to the database.

6. An intelligent control system for a cleaning robot in a photovoltaic power plant, based on the intelligent control method for a cleaning robot in a photovoltaic power plant according to any one of claims 1 to 5, characterized in that: include, The image acquisition module is used to collect raw image data of the photovoltaic power station, providing input data for subsequent semantic modeling and pollution analysis; The semantic modeling module is used to segment images using self-organizing evolutionary segmentation technology and construct a semantic model of a photovoltaic power station. The pollution status analysis module is used to generate static semantic maps using semantic models, analyze the pollution level of each area in the map, and use this as a basis for task allocation. The path planning module is used to plan the optimal cleaning path for each robot using an improved ant colony algorithm. The dynamic obstacle avoidance module is used to perceive obstacles and dynamically avoid them using the DWA algorithm during the robot's cleaning task. After avoiding obstacles, the path is updated in real time to continue the task. The pollution feature analysis and cleaning strategy customization module is used to analyze the specific characteristics of the polluted area using optical texture feature detection methods and dynamically formulate cleaning strategies. The data upload module is used to collect the robot's operating data and adjustment records during the cleaning process and upload them to the database for centralized storage.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent control method for a cleaning robot for a photovoltaic power plant as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent control method for a cleaning robot for a photovoltaic power plant as described in any one of claims 1 to 5.

Images