Multi-robot grain surface inspection node deployment method and path planning method based on improved grey wolf algorithm
By improving the population initialization and nonlinear convergence factor of the Grey Wolf algorithm, and combining the Boolean perception model and the K-Means algorithm, the problems of boundary omission and coverage overlap in multi-robot grain surface inspection were solved, achieving efficient grain warehouse coverage and sampling inspection.
Patent Information
- Application Number
- CN202311192988.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-15
- Publication Date
- 2026-02-03
AI Technical Summary
In multi-robot grain surface inspection, there are problems such as missed detection in boundary areas and excessive overlap of the coverage area of multiple sensors, which existing algorithms cannot effectively solve.
An improved gray wolf algorithm is adopted, which initializes the population through Latin hypercube sampling to increase population diversity. A nonlinear convergence factor is introduced to balance global search and local exploitation capabilities. The node deployment position is optimized by combining Boolean perception model. The K-Means algorithm is used to classify and inspect the inspection area and solve the traveling salesman problem to plan the path.
It achieves a large coverage rate with fewer inspection nodes, meets the task requirements of coverage inspection and sampling in grain warehouses, improves inspection efficiency and coverage uniformity, and avoids the algorithm getting stuck in local optima.
Smart Images
Figure CN121453076A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field of intelligent grain warehouse inspection, and specifically relates to a multi-robot grain surface inspection node deployment method and path planning method based on an improved gray wolf algorithm. Background Technology
[0002] Grain surface inspection, capable of detecting temperature and humidity information in the storage environment as well as abnormal grain conditions such as grain insects, impurities, and condensation, has become a research hotspot in recent years. Compared to traditional manual warehouse cleaning and inspection methods, grain surface inspection robots possess autonomous perception and movement capabilities, enabling them to operate in complex storage environments such as darkness, dust, and smoke. Path planning is crucial for achieving autonomous robot movement, and determining the optimal grain surface inspection path is an urgent problem to be solved.
[0003] Domestic and international research on path planning mainly focuses on heuristic algorithms, especially swarm intelligence optimization algorithms, such as ant colony optimization, genetic algorithms, and particle swarm optimization. However, these methods also have their own problems. For example, genetic algorithms have poor adaptability to environmental changes and low search efficiency; ant colony optimization has a slow convergence speed and sometimes causes robot stagnation. Grey Wolf Optimization (GWO) is a population-based intelligent algorithm proposed by Mirja Lili et al. Compared with other swarm intelligence algorithms, it has the characteristics of simple structure, strong solution ability, and higher global search ability. In recent years, it has been widely used in scheduling, model parameter optimization, and mobile robot path planning. However, the traditional Grey Wolf algorithm also has many inherent defects. For example, using the GWO algorithm to solve the low-carbon workshop scheduling problem, it was verified that the optimization result of the GWO algorithm is the best compared with other intelligent algorithms, but it did not improve the problem of imbalance between global exploration and local exploitation. By introducing a cosine convergence factor to balance the global search and local exploitation capabilities of the Grey Wolf algorithm, it did not improve the problem of uneven initial population distribution and the algorithm's tendency to get trapped in local optima.
[0004] Unlike traditional robot path planning, when inspecting an entire large grain field or focusing on specific areas, using a single robot is inefficient and has poor anti-interference capabilities. Multi-robot collaborative operation, however, enables more efficient inspections and more reliable data transmission. Existing technologies use a linear scanning coverage algorithm, distributing multiple robots in rows to maximize coverage of the scanned area. However, this algorithm cannot sample and inspect parts of the area, resulting in poor flexibility. Another approach is multi-robot collaborative coverage search, deploying multiple sensors within a region and using clustering algorithms to decompose the entire region into multiple sub-regions, then planning paths for each sub-region. However, this algorithm does not consider sensor deployment at boundaries, leading to missed detections in these areas. Furthermore, robots performing tasks near boundaries are prone to touching them, causing safety issues. Existing combined hierarchical optimization strategies based on sensor configuration and path framework partitioning can solve multi-robot collaborative coverage problems in various irregularly shaped areas and obstacle environments. However, deploying more sensors to increase coverage results in excessive overlap between the coverage areas of different sensors. Summary of the Invention
[0005] To address the issues of missed inspections in boundary areas and excessive overlap in the coverage areas of multiple sensor deployments during multi-robot collaborative inspections, this invention provides a multi-robot grain surface inspection node deployment method and path planning method based on an improved Grey Wolf algorithm.
[0006] The solution adopted by this invention to solve its technical problem is: a multi-robot grain surface inspection node deployment method based on an improved gray wolf algorithm, which includes the following steps:
[0007] Step 1: The improved gray wolf algorithm uses Latin hypercube sampling to initialize the gray wolf population, increasing population diversity. The gray wolf algorithm is an optimization algorithm that simulates the group behavior of gray wolves. In the gray wolf algorithm, gray wolves are divided into four levels, from high to low: α wolf, β wolf, δ wolf, and ω wolf. α wolf represents the optimal solution, β and δ wolves represent suboptimal solutions, and ω wolf represents a candidate solution.
[0008] The mathematical model for gray wolves hunting their prey is as follows:
[0009]
[0010]
[0011] in, Let be the position vector of the prey in the t-th iteration. Let be the position vector of the prey. Let t represent the distance between the wolf and its prey, and t represent the current iteration number. and This is the coefficient vector.
[0012] Where vector and The expression can be:
[0013]
[0014]
[0015] r1 = r2 = rand(0,1) (5)
[0016] in, and It is a random number in the range [0,1], and a is the convergence factor.
[0017] The convergence factor 'a' decreases linearly from 2 to 0 as the number of algorithm iterations increases, i.e.
[0018]
[0019] Where t is the current iteration number and T is the maximum iteration number.
[0020] The objective function value is iterated by the gray wolf optimization algorithm to obtain the optimal solution α wolf, the suboptimal solution β wolf, and δ wolf. The positions of the other gray wolves are jointly determined by the positions of α wolf, β wolf, and δ wolf.
[0021] The process of a gray wolf chasing its prey is as follows
[0022]
[0023]
[0024]
[0025] in, These represent the distances of wolf ω from wolf α, wolf β, and wolf δ, respectively. It is a vector of random coefficients; These represent the positions of α wolf, β wolf, and δ wolf, respectively. That is the current target's location. Let ω be the step size and distance that the gray wolf moves towards the α, β, and δ wolves, and finally take... The average of the three factors determines the final location where the gray wolf will move.
[0026] The steps for improving the Grey Wolf algorithm using Latin hypercube sampling are as follows:
[0027] S1: Determine the sampling size N, i.e., the size of the gray wolf population, and set the dimension of the hypercubic variable to D, and the variable x j∈[lb,ub],j=1,2,...,D;
[0028] S2: Variable x j The domain [lb, ub] is divided into N equal intervals, i.e.
[0029]
[0030] The original hypercube was divided into N D A small hypercube;
[0031] S3: Generate a matrix A of size N×D, where each column of the matrix is a permutation of N equal intervals;
[0032] S4: Each row of matrix A corresponds to a selected small hypercube. A sample is randomly generated within each selected small hypercube, thus N samples can be extracted.
[0033] To measure the uniformity of sampling points, given a set of N points... definition Where γ i γ represents the distance from each sampling point to its nearest sampling point, and ratio represents the minimum γ among all sampling points. i With the maximum γ i The ratio, where cov represents the standard deviation of the minimum distance, and ratio and cov are expressed as...
[0034]
[0035]
[0036]
[0037] A larger ratio and a smaller cov indicate better uniformity in the distribution of sampling points.
[0038] Step 2: A nonlinear convergence factor is introduced to balance the ability of global search and local exploitation, thus preventing the improved Grey Wolf algorithm from getting trapped in local optima.
[0039] Step 3: In the scenario of grain surface inspection in a large flat warehouse, a deployment strategy for grain surface boundary inspection nodes is given, and the improved Grey Wolf algorithm is used to optimize the node deployment position with the objective function of maximizing coverage.
[0040] In step three, the target area grain surface is set as a two-dimensional plane, simplifying the node coverage problem into an optimization problem on the two-dimensional plane. A Boolean sensing model is used to establish the objective function. Suppose there is a set of inspection points Node={n1,n2,n3,...,n in the X×Y detection area. N}, any node n being detected within this regioni The two-dimensional coordinates are represented as (x i ,y i The detected area is set to (x) i ,y i A circle with center R and radius R divides the entire inspection area into x×y equal grids with an area of 1. The center of each grid represents the location of the inspection point. The set of inspection points within the inspection area can be denoted as M = {m1, m2, m3, ..., m...} j}, j = 1, 2, ..., x × y, and any point m within the region. j The two-dimensional coordinates are represented as (x j ,y j ).
[0041] n i and m j The Euclidean distance between two points is described as follows:
[0042]
[0043] node n i Point m was detected j The detection probability is expressed as
[0044]
[0045] Among them, C cov (n i ,m j ) = 1 indicates point m j It can be used by node n i Detected.
[0046] Since there is overlap between the detection areas, point m j The probability of being jointly sensed by at least one or more nodes is expressed as:
[0047]
[0048] The coverage Cov of the entire detection area is defined as the ratio of the number of covered grid cells to the total number of grid cells in the area.
[0049]
[0050] Cov is used as the fitness function for regional coverage. When the function reaches its maximum value, it indicates that the coverage of the region is at its maximum.
[0051] Step two employs nonlinear control parameters.
[0052]
[0053] Among them, ai =0, a f =2, t is the current iteration number, T max denoted as the maximum number of iterations, and 'n' as the nonlinear adjustment coefficient. The convergence factor 'a' is updated using nonlinear control parameters.
[0054] Furthermore, considering the algorithm complexity, the nonlinear adjustment coefficient n is set to 2.
[0055] A path planning method for grain surface inspection robots based on an improved gray wolf algorithm is proposed. The K-Means algorithm is used to classify the inspection areas of the deployed inspection nodes according to the number of robots. Each robot is responsible for one inspection area, and the optimal inspection path for each robot is obtained by solving the traveling salesman problem in each area.
[0056] The specific process of the K-means algorithm is as follows:
[0057] S4-1: Determine the number of groups K;
[0058] S4-2: Randomly select k points as initial cluster centers m i ;
[0059] S4-3: Calculate x for each sample i Euclidean distance to cluster center
[0060]
[0061] S4-4: Based on the distance J, assign each sample to the nearest cluster;
[0062] S4-5: Update cluster centers
[0063]
[0064] Where N i It is the number of samples belonging to the i-th cluster.
[0065] The beneficial effects of this invention are as follows: The improved gray wolf algorithm provided by this invention can provide a large coverage rate with fewer inspection nodes, and can meet the task requirements of coverage inspection and sampling detection in grain warehouses, providing a new idea for the optimal inspection path design of multiple robots in grain warehouses.
[0066] Latin hypercube sampling divides the parameter space into equally divided intervals and ensures that there is only one sample point in each interval. This ensures that the initial position of the gray wolf is more evenly distributed and avoids the problem of sampling points clustering in certain areas.
[0067] The nonlinear convergence factor provided by this invention decays slowly in the early stages of algorithm iteration, which can improve global search capability and prevent the algorithm from getting trapped in local optima; in the later stages of iteration, the nonlinear convergence factor decays rapidly, which can improve solution efficiency. Attached Figure Description
[0068] Figure 1 This is a schematic diagram of the Boolean perception model.
[0069] Figure 2 This is a diagram of the Grey Wolf algorithm structure.
[0070] Figure 3 This is a schematic diagram for initializing the population distribution.
[0071] Figure 4 This is a comparison chart of convergence factors.
[0072] Figure 5 To improve the LH-GWO process flowchart of the Grey Wolf algorithm.
[0073] Figure 6 A diagram illustrating the deployment strategy for the boundary.
[0074] Figure 7 Convergence curves for different algorithms based on the test function.
[0075] Figure 8 The diagram shows the deployment results of different algorithms.
[0076] Figure 9 This is a comparison chart of coverage rates for different algorithms.
[0077] Figure 10 A comparison chart of coverage as the number of nodes increases for different algorithms.
[0078] Figure 11 A flowchart illustrating the path planning method for a grain inspection robot.
[0079] Figure 12 A diagram showing the deployment of inspection nodes within the grain depot.
[0080] Figure 13 Path planning diagrams for inspection robots using different detection methods. Detailed Implementation
[0081] Example: This invention provides an improved Gray Wolf Algorithm (LH-GWO). This algorithm uses Latin Hypercube Sampling (LHS) to initialize the Gray Wolf population, increasing population diversity, and introduces a nonlinear convergence factor to balance global search and local exploitation capabilities, preventing the algorithm from getting trapped in local optima. Then, in a large flat warehouse grain surface inspection scenario, a grain surface boundary inspection node deployment strategy is given, and the proposed algorithm is used to optimize node deployment positions with maximum coverage as the objective function. Finally, according to specific task requirements, the K-Means algorithm is used to classify inspection nodes into inspection areas based on the number of robots. Each robot is responsible for one inspection area, and within each area, the optimal inspection path for each robot is obtained by solving the Traveling Salesman Problem (TSP).
[0082] The algorithm described in this invention can provide a large coverage rate with fewer inspection nodes and can meet the task requirements of coverage inspection and sampling detection in grain warehouses, providing a new approach for the optimal inspection path design of multiple robots in grain warehouses.
[0083] In this invention, Latin hypercube sampling is used to initialize the population of the gray wolf algorithm, increasing population diversity. The traditional gray wolf algorithm is an optimization algorithm that simulates the group behavior of gray wolves, who mostly prefer group living and have a strict social hierarchy. Figure 2 As shown, the gray wolves are divided into four levels, from highest to lowest: α wolf, β wolf, δ wolf, and ω wolf. Among them, α wolf is the optimal solution, β wolf and δ wolf are suboptimal solutions, and ω wolf is a candidate solution.
[0084] The mathematical model for gray wolves hunting their prey is as follows:
[0085]
[0086]
[0087] in, Let be the position vector of the prey in the t-th iteration. Let be the position vector of the prey. Let t represent the distance between the wolf and its prey, and t represent the current iteration number. and This is a coefficient vector. and The expression can be:
[0088]
[0089]
[0090] r1 = r2 = rand(0,1) (5)
[0091] in, and It is a random number in the range [0,1], as shown in equation (5). a is the convergence factor, which decreases linearly from 2 to 0 as the number of algorithm iterations increases, i.e.
[0092]
[0093] Where t is the current iteration number and T is the maximum iteration number.
[0094] The objective function value is iteratively calculated using the gray wolf optimization algorithm to obtain the optimal solution α wolf, the second-best solution β wolf, and δ wolf. The positions of the other gray wolves are jointly determined by the positions of α wolf, β wolf, and δ wolf. The process of gray wolves chasing prey is as follows:
[0095]
[0096]
[0097]
[0098] in, These represent the distances of wolf ω from wolf α, wolf β, and wolf δ, respectively. This is a vector of random coefficients. These represent the positions of α wolf, β wolf, and δ wolf, respectively. This is the current target position. Equation (8) defines the step size and distance that the gray wolf ω moves towards the α wolf, β wolf, and δ wolf. Finally, the average of the three values is taken to determine the final position of the gray wolf, as shown in Equation (9).
[0099] In swarm intelligence algorithms, the initialization of the population determines the starting position of the algorithm in the search space, which has a direct impact on the algorithm's optimization efficiency and accuracy.
[0100] The initial population of the traditional gray wolf algorithm is generated based on a random function within the feasible region, such as... Figure 3 As shown in (a), this initialization method easily leads to uneven spatial distribution of gray wolf individuals, affecting the optimization efficiency of the algorithm. Therefore, this invention introduces the Latin Hypercube Sampling (LHS) method to initialize the population, such as... Figure 3 As shown in (b). Figure 3 In the example, the population size is N=40, the dimension is dim=2, and the coordinate range is [0,50].
[0101] The key to Latin hypercube sampling is dividing the parameter space into equally divided intervals and ensuring that each interval contains only one sample point. This guarantees a more uniform distribution of the initial positions of the gray wolves and avoids the problem of sample points clustering in certain areas. The steps of Latin hypercube sampling are as follows:
[0102] Step 1: Determine the sampling size N, i.e., the gray wolf population size, and set the dimension of the hypercubic variable to D, and the variable x j ∈[lb,ub],j=1,2,...,D;
[0103] Step 2: Set variable x j The domain [lb, ub] is divided into N equal intervals, i.e.
[0104]
[0105] Therefore, the original hypercube is divided into N. D A small hypercube;
[0106] Step 3: Generate a matrix A of size N×D, where each column of the matrix is a permutation of N equal intervals;
[0107] Step 4: Each row of matrix A corresponds to a selected small hypercube. A sample is randomly generated within each selected small hypercube, thus N samples can be extracted.
[0108] To measure the uniformity of sampling points, this invention adopts a uniformity judgment criterion, which is defined as follows.
[0109] Given a set of N points definition Where γ i γ represents the distance from each sampling point to its nearest sampling point, and raito represents the minimum γ among all sampling points. i With the maximum γ i The ratio of , where cov represents the standard deviation of the minimum distance, and raito and cov can be expressed as
[0110]
[0111]
[0112]
[0113] A larger ratio and a smaller cov indicate better uniformity in the distribution of sampling points. This invention measures the uniformity of the initial population by calculating the ratio and cov for different methods, as shown in Table 1.
[0114] Table 1 Comparison of uniformity
[0115] name Ratio Cov LHS 0.2825 1.8208 Rand 0.1639 2.5704
[0116] As can be seen from Table 1, the population initialization based on LHS sampling has the largest raito value and the smallest cov value. That is, the population initialization using LHS is more uniform in location distribution than the population initialization using random sampling, which is more conducive to algorithm optimization.
[0117] Furthermore, this invention assumes the target region is a two-dimensional plane, thus simplifying the node coverage problem into an optimization problem on that plane. This invention uses a Boolean sensing model to establish the objective function. For example... Figure 1 As shown, assume there is a set of inspection points Node = {n1, n2, n3, ..., n} within the region X×Y. N}, any node n being detected within this region i The two-dimensional coordinates are represented as (x i ,y i A detected region can be considered as (x) i ,y i Let R be a circle with center R and radius R. For ease of calculation, the entire region is divided into x×y equal grids with an area of 1. The center of each grid represents the location of the detection point. Then, the set of detection target points within the inspection area can be denoted as M={m1,m2,m3,...,m...} j}, j = 1, 2, ..., x × y, and any point m within the region. j The two-dimensional coordinates are represented as (x j ,y j ).
[0118] n i and m j The Euclidean distance between two points is described as follows:
[0119]
[0120] node n i Point m was detected j The detection probability can be expressed as
[0121]
[0122] Among them, C cov (n i ,m j ) = 1 indicates point m j It can be used by node n i Detected. Since there may be overlap between detection areas, point m... j The probability of being jointly sensed by at least one or more nodes can be expressed as:
[0123]
[0124] The coverage rate Cov of the entire area is defined as the ratio of the number of covered grid cells to the total number of grid cells in the area, i.e.
[0125]
[0126] In this invention, equation (16) is used as the fitness function for regional coverage. When the function reaches its maximum value, it indicates that the regional coverage is at its maximum.
[0127] Global search capability and local exploitation capability are two common attributes used to measure the optimization performance of intelligent optimization algorithms. Typically, a global search strategy is used in the early stages of iteration, while a local exploitation strategy is used in the later stages. The traditional GWO algorithm mainly uses parameter A to balance the algorithm's global exploration and local exploitation capabilities. When |A|>1, the gray wolf population will expand its search range and conduct global exploration. When |A|<1, the gray wolf population will narrow its search range and conduct local exploitation. According to equations (3) and (6), during the iteration process, the value of A decreases linearly within [0,2] as the control parameter a changes. However, the iteration process of the GWO algorithm is non-linear, and the linear change of parameter a greatly limits the global optimization capability of the GWO algorithm.
[0128] This invention provides a new nonlinear control parameter
[0129]
[0130] Among them, a i =0, a f =2, t is the current iteration number, T max Let n be the maximum number of iterations and n be the nonlinear adjustment coefficient. Considering the algorithm complexity, this invention sets n to 2.
[0131] Figure 4 The nonlinear convergence factor and the linear convergence factor provided in this invention were compared and analyzed. The linear convergence factor changes at the same rate throughout the entire iteration process, which cannot effectively balance the global search and local search processes of the algorithm. In contrast, the proposed nonlinear convergence factor decays more slowly in the early stages of the algorithm iteration, which can improve the global search capability and avoid the algorithm getting trapped in local optima; in the later stages of the iteration, the nonlinear convergence factor decays more rapidly, which can improve the solution efficiency.
[0132] Therefore, the LH-GWO algorithm flow provided by this invention is as follows: Figure 5 As shown.
[0133] When robots inspect grain surfaces, the distance between the robot's position and the boundary becomes a crucial factor. If the robot is too close to the boundary, it may collide with it, causing safety issues. To achieve complete boundary coverage within the grain silo, the optimal method is to deploy inspection nodes along the perimeter of the entire area. Let r be the radius of the inspection node, then for a rectangular area of length L and width W, the number of nodes n required to cover its perimeter is...
[0134]
[0135] However, the inspection node center of this strategy is on the boundary, and there will be overlap at the vertices of the rectangles. Therefore, this invention proposes a new boundary deployment strategy, such as... Figure 6 As shown. Let r be the radius of the inspection node. Based on the minimum overlap deployment, the distance d between any two adjacent nodes in the same row or column is... For a rectangular region of length L and width W, the number of nodes n′ required to be deployed along the boundary of this rectangle is:
[0136]
[0137] Where l1 = 2r cosθ; l2 = 2r sinθ, are the overlap lengths of the bottom edge deployment points on the two sides.
[0138] As shown in Table 2, this invention selects four commonly used benchmark functions to test the performance of the proposed Gray Wolf Optimization Algorithm (LH-GWO).
[0139] Table 2 Function Information
[0140]
[0141] In the table, Sphere(F1) and Schwefel 2.22(F2) are unimodal functions; Rastrigin(F3) and Griewank(F4) are multimodal functions. In addition, the performance of the LH-GWO algorithm proposed in this invention was compared and analyzed with that of the PSO algorithm, GWO algorithm, and GWO-CS algorithm.
[0142] In the test, the population size was set to 30, the dimension of the test function was 30, and the maximum number of iterations was 500. To avoid the randomness of the algorithm's optimization results, each test function was simulated independently 30 times. The mean and standard deviation were selected as performance indicators to evaluate the algorithm. The mean reflects the stability of the algorithm, and the standard deviation reflects the robustness of the algorithm. The simulation platform was MATLAB 2022b.
[0143] Table 3 Function Test Results
[0144]
[0145] As shown in Table 3, the average performance of the LH-GWO algorithm is superior to the other three algorithms, especially in the unimodal function F1 and multimodal function F3. This demonstrates that, under the same dimensionality and number of iterations, the LH-GWO algorithm exhibits better convergence accuracy and stability. This is because, in the early stages of the algorithm, initializing the population with the LHS sequence increases population diversity, thereby improving the algorithm's convergence speed. During the search process, the algorithm introduces a nonlinear convergence factor to balance the ability to explore globally and exploit locally, while other algorithms have poor nonlinear convergence factors, making them prone to getting trapped in local optima during the search process.
[0146] To more intuitively reflect the optimization effect of the algorithm, such as Figure 7 The figure shows a comparison of the convergence curves of the four algorithms for the test function.
[0147] As can be seen from the figure, the LH-GWO algorithm proposed in this invention can more accurately lock the global optimal solution compared to the other three algorithms.
[0148] Further testing was conducted on the regional coverage of the present invention. Specifically, the optimization effects of four algorithms in solving the node coverage problem were tested. The test area was 50m×50m, the radius of the inspected node was r=5m, the population size of the four algorithms was set to 50, and the maximum number of iterations was set to 200. Figure 8 The simulation results are shown in the figure when the number of inspection nodes is 45 and the number of iterations is 200. Figure 8 (a) shows the coverage results of deploying nodes in random mode. It can be seen that the nodes are unevenly distributed in the figure, and some nodes even overlap. Figure 3 (b) shows the coverage result after optimizing the node deployment using the LH-GWO algorithm. The node positions in the figure are evenly distributed and can cover the entire area well.
[0149] Figure 9 This section compares the coverage rates of four algorithms when there are 45 nodes and 200 iterations. It shows that the LH-GWO algorithm has a significantly higher coverage rate than the other three algorithms. Table 4 lists the coverage rates of the four algorithms when the number of inspected nodes ranges from 35 to 60 (increasing by 5 nodes per group). Figure 10 This is a comparison chart of coverage rates for four algorithms as the number of nodes increases. Figure 10 It allows for a clearer observation of the changing trends in relevant data. Figure 10 It can be seen that when the area size is the same, the coverage of all four algorithms increases with the increase of the number of nodes, but the coverage of the LH-GWO algorithm is better than the other three algorithms, and it can achieve a higher coverage effect with fewer nodes.
[0150] Table 4 Coverage Results Test (%)
[0151]
[0152] Finally, simulation tests were conducted on the grain surface inspection robot. Multi-robot grain surface inspection mainly consists of two stages: inspection node deployment and path planning. Figure 11 As shown.
[0153] Regarding the deployment diagram of grain surface inspection nodes, this invention takes a large flat warehouse with a length of 60m and a width of 21m as an example. Using the LH-GWO algorithm combined with a grain surface boundary node deployment strategy, the deployment of grain surface inspection nodes is as follows: Figure 12 As shown.
[0154] Based on the obtained grain inspection node deployment scheme, the K-Means algorithm is used to classify the inspection areas of the inspection nodes according to the number of robots. Each robot is responsible for one inspection area, and the optimal inspection path of each robot is obtained by solving the traveling salesman problem in each area.
[0155] The K-means algorithm is an unsupervised learning method and a classic partition-based clustering analysis method. The K-means algorithm divides N data points into K clusters based on the distance between the data points. The final clustering goal is to achieve high similarity within clusters and low similarity between clusters.
[0156] The specific process of the K-means algorithm is as follows:
[0157] Step 1: Determine the number of groups K;
[0158] Step 2: Randomly select k points as the initial cluster centers m i ;
[0159] Step 3: Calculate x for each sample according to formula (21) i Euclidean distance to cluster center
[0160]
[0161] Step 4: Based on the distance J, assign each sample to the nearest cluster;
[0162] Step 5: Update cluster centers, m i As shown in equation (19)
[0163]
[0164] Where N i It is the number of samples belonging to the i-th cluster.
[0165] Suppose there are 3 inspection robots, and the robots are required to start from the same location. Solving the Traveling Salesman Problem (TSP) yields the optimal path planning for each robot, as follows: Figure 13 As shown. Figure 13 (a) and 13(b) are sampling inspection path planning, and 13(c) is coverage inspection path planning.
[0166] The Traveling Salesman Problem (TSP) is an optimization problem in computer science and mathematics. It involves finding the shortest path a salesman can take to visit a set of cities and return to the starting point, visiting each city only once. TSP is classified as NP-hard, meaning that finding the optimal solution is computationally difficult for large-scale instances. Various algorithms and heuristics have been developed to solve the TSP, including brute-force search, dynamic programming, and approximation algorithms. TSP has many real-world applications, and this invention uses it to derive optimal path planning for various inspection robots.
[0167] This invention addresses the problem of low node deployment coverage caused by uneven distribution of inspection nodes in grain surface inspection robots within large flat warehouses. It introduces an LHS sequence and a nonlinear convergence factor into the traditional Grey Wolf algorithm to initialize the population and balance global search with local utilization capabilities. Combined with a grain surface boundary inspection node deployment strategy, an optimal inspection node deployment map is obtained. The K-means algorithm is used to divide the inspection nodes into multiple regions, and the optimal inspection path for each robot in that region is obtained by solving the Time-of-Sight (TSS) problem. Two sets of comparative experiments were conducted on the LH-GWO, PSO, GWO, and GWO-CS algorithms: standard function test and regional coverage test. In the standard function test, the LH-GWO algorithm outperformed the other three algorithms in terms of mean, standard deviation, and convergence performance. In the regional coverage test, the coverage rate was improved by 2.28%, 3.6%, and 5.59%, respectively. The proposed path planning algorithm for grain surface inspection robots is applicable to the tasks of grain surface coverage inspection and sampling detection. Future work will further improve this algorithm by incorporating real-time obstacle avoidance capabilities.
[0168] The specific embodiments described above are merely illustrative or explanatory of the principles of the present invention and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for deploying multi-robot grain surface inspection nodes based on an improved gray wolf algorithm, characterized in that, Includes the following steps, Step 1: The improved gray wolf algorithm uses Latin hypercube sampling to initialize the gray wolf population, increasing population diversity. The gray wolf algorithm is an optimization algorithm that simulates the group behavior of gray wolves. In the gray wolf algorithm, gray wolves are divided into four levels, from highest to lowest: α wolf, β wolf, δ wolf, and ω wolf. α wolf represents the optimal solution, β and δ wolves represent suboptimal solutions, and ω wolf represents a candidate solution. The mathematical model for gray wolves hunting their prey is as follows: in, Let be the position vector of the prey in the t-th iteration. Let be the position vector of the prey. Let t represent the distance between the wolf and its prey, and t represent the current iteration number. and A coefficient vector; a vector and The expression can be: r1 = r2 = rand(0,1) (5) in, and It is a random number in the range [0,1], and 'a' is the convergence factor, which decreases linearly from 2 to 0 as the number of algorithm iterations increases. Where t is the current iteration number and T is the maximum iteration number; The objective function value is iterated using the gray wolf optimization algorithm to obtain the optimal solution α wolf, the suboptimal solution β wolf, and the δ wolf. The positions of the other gray wolves are jointly determined by the positions of α wolf, β wolf, and δ wolf. The process of gray wolves chasing prey is as follows: in, These represent the distances of wolf ω from wolf α, wolf β, and wolf δ, respectively. It is a vector of random coefficients; These represent the positions of α wolf, β wolf, and δ wolf, respectively. That is the current target's location. Let ω be the step size and distance that the gray wolf moves towards the α, β, and δ wolves, and finally take... The average of the three factors determines the final location of the gray wolf; The Grey Wolf algorithm is improved using Latin hypercube sampling, and the steps are as follows: S1: Determine the sampling size N, i.e., the size of the gray wolf population, and set the dimension of the hypercubic variable to D, and the variable x j ∈[lb,ub],j=1,2,...,D; S2: Variable x j The domain [lb, ub] is divided into N equal intervals, i.e. The original hypercube was divided into N D A small hypercube; S3: Generate a matrix A of size N×D, where each column of the matrix is a permutation of N equal intervals; S4: Each row of matrix A corresponds to a selected small hypercube. A sample is randomly generated within each selected small hypercube, thus N samples can be extracted. To measure the uniformity of sampling points, given a set of N points... definition Where γ i γ represents the distance from each sampling point to its nearest sampling point, and ratio represents the minimum γ among all sampling points. i With the maximum γ i The ratio, where cov represents the standard deviation of the minimum distance, and ratio and cov are expressed as... The larger the ratio and the smaller the cov, the better the uniformity of the sampling point distribution. Step 2: A nonlinear convergence factor is introduced to balance the ability of global search and local exploitation, so as to avoid the improved Grey Wolf algorithm from getting trapped in local optima; Step 3: In the scenario of grain surface inspection in a large flat warehouse, a deployment strategy for grain surface boundary inspection nodes is given, and the improved Grey Wolf algorithm is used to optimize the node deployment position with the objective function of maximizing coverage.
2. The multi-robot grain surface inspection node deployment method based on the improved gray wolf algorithm according to claim 1, characterized in that, In step three, the target area grain surface is set as a two-dimensional plane, simplifying the node coverage problem into an optimization problem on the two-dimensional plane. A Boolean sensing model is used to establish the objective function. Suppose there is a set of inspection points Node={n1,n2,n3,...,n...} within the X×Y detection area. N }, any node n being detected within this region i The two-dimensional coordinates are represented as (x i ,y i The detected area is set to (x) i ,y i A circle with center R and radius R divides the entire inspection area into x×y equal grids with an area of 1. The center of each grid represents the location of the inspection point. The set of inspection points within the inspection area can be denoted as M = {m1, m2, m3, ..., m...} j }, j = 1, 2, ..., x × y, and any point m within the region. j The two-dimensional coordinates are represented as (x j ,y j ), n i and m j The Euclidean distance between two points is described as follows: node n i Point m was detected j The detection probability is expressed as Among them, C cov (n i ,m j ) = 1 indicates point m j It can be used by node n i Detected; Since there is overlap between the detection areas, point m j The probability of being jointly sensed by at least one or more nodes is expressed as: The coverage Cov of the entire detection area is defined as the ratio of the number of covered grid cells to the total number of grid cells in the area. Cov is used as the fitness function for regional coverage. When the function reaches its maximum value, it indicates that the coverage of the region is at its maximum.
3. The multi-robot grain surface inspection node deployment method based on the improved gray wolf algorithm according to claim 1, characterized in that, Step 2 uses nonlinear control parameters Among them, a i =0, a f =2, t is the current iteration number, T max denoted as the maximum number of iterations, and 'n' as the nonlinear adjustment coefficient. The convergence factor 'a' is updated using nonlinear control parameters.
4. The multi-robot grain surface inspection node deployment method based on the improved gray wolf algorithm according to claim 3, characterized in that, The nonlinear adjustment coefficient n is set to 2.
5. A path planning method for a grain surface inspection robot based on an improved gray wolf algorithm, characterized in that, The K-Means algorithm is used to classify the inspection areas of the deployed inspection nodes according to the number of robots. Each robot is responsible for one inspection area, and the optimal inspection path for each robot is obtained by solving the traveling salesman problem within each area. The K-means algorithm includes the following steps: S4-1: Determine the number of groups K; S4-2: Randomly select k points as initial cluster centers m i ; S4-3: Calculate x for each sample i Euclidean distance to cluster center S4-4: Based on the distance J, assign each sample to the nearest cluster; S4-5: Update cluster centers Where N i It is the number of samples belonging to the i-th cluster.