Unmanned aerial vehicle inspection path planning method considering risk level

Abstract

The invention discloses an unmanned aerial vehicle routing inspection path planning method considering risk levels, which comprises the following steps: 1) simulating an experimental map, and establishing a related mathematical model of a path planning algorithm; 2) using an improved BI-RRT * adaptive step length path planning algorithm to carry out primary measurement recording on the path distance between the inspection points of the unmanned aerial vehicle; 3) designing a grey wolf algorithm evaluation function, enabling the evaluation target to be comprehensive optimum of the inspection task point risk and the unmanned aerial vehicle inspection path cost, and enhancing the local optimum separation capability of the algorithm; and 4) carrying out secondary planning on the path according to the sequence to ensure that the path is optimal. According to the method, the routing inspection cost of the unmanned aerial vehicle is considered in the routing inspection process, and routing inspection points with high risk degrees are preferentially inspected; an IRRT *-Connect path planning algorithm based on environmental complexity is designed, and it is ensured that the unmanned aerial vehicle plans an optimal path with high efficiency when facing a complex and narrow area.

Description

Technical Field

[0001] The present invention relates to the design and application fields of inspection robots, and in particular to a method for planning an inspection path of an unmanned aerial vehicle (UAV) taking risk levels into consideration. Background Art

[0002] The UAV intelligent inspection system is composed of UAVs, UAV hangars, flight brains, and UAV control platforms. It attracts all walks of life with its advantages of high efficiency, safety, and intelligence. The traditional power inspection method is inefficient due to factors such as complex terrain and long lines. The UAV intelligent inspection system can achieve high-altitude aerial photography, night monitoring, precise positioning, and real-time data transmission through autonomous navigation, high-definition video, infrared thermal imaging, and other technologies, effectively solving various limitations of human power inspections and greatly improving the efficiency and accuracy of power inspections. The UAV intelligent inspection system has been widely used in the fields of UAV power inspection, UAV pipeline inspection, UAV river inspection, and traffic patrol, and has achieved remarkable results. During the working process of the inspection system, the system needs to plan the working path according to the working environment to avoid incomplete inspection content due to deviations in the inspection route, or redundant routes that cause the UAV to run out of power and fall at a certain location. Therefore, the path planning problem in the UAV intelligent inspection system has important theoretical significance and good practical application prospects.

[0003] The Rapidly Exploring Random Trees (RRT) algorithm was first proposed by Professor Steven M. LaValle of Iowa State University in 1998 as a classic path planning algorithm. Due to its strong randomness and fast planning efficiency, it has been widely used and developed in recent years. However, the Rapidly Exploring Random Trees (RRT) algorithm has a slow convergence speed and low search efficiency. To address this problem, some scholars have proposed a simple and effective randomization algorithm, namely RRT-connect. The working principle of this method is to gradually establish two rapidly exploring random trees rooted in the start and target configurations. Each tree explores the surrounding space and approaches each other to speed up the search efficiency by using a simple greedy heuristic algorithm; other scholars have proposed the RRT* algorithm, which adds random geometric graphs and pruning optimization theory on the basis of RRT node expansion to ensure that the nodes of the random tree can converge to the current optimal value. RRT* can converge to the global optimal solution. However, when the number of nodes is too large, the memory consumption and computational complexity of the algorithm will increase exponentially, reducing the algorithm's running speed and is not suitable for environments with strong real-time requirements. Based on the above two improved methods, a single-query bidirectional planning method, namely Informed-RRT*-Connect, is proposed. The single-query bidirectional planning method Informed-RRT*-Connect is the same as RRT*-Connect until the first path is found, after which the proposed method only takes samples from a subset of states that may improve the solution. Like other asymptotically optimal versions of RRT, the single-query bidirectional planning methods Informed-RRT*-Connect and RRT*-Connect continue to explore the state space after finding the first solution to return a near-optimal solution. However, restricting the state space is a function of the shortest path obtained, so that the planner cannot restrict the state space to a subset before finding a solution, and the algorithm converges poorly.

[0004] In addition, some current intelligent inspection drones can only complete inspection tasks in specified areas, and do not reflect the inspection priority of risky operation points when traversing inspection tasks at multiple target points. In addition to setting priorities based on risks, it is also necessary to re-inspect task points where faults are frequent.

[0005] Most of the existing planning algorithms only plan paths for all task points in sequence, without taking the risk level and failure frequency of the task points as reference. This may result in problems with the previously inspected location points after inspection and not be re-inspected in time, or failures at high-risk task points but delay in inspection, resulting in a series of production problems or even production accidents. Therefore, before studying the path planning algorithm, it is of great theoretical significance and potential application value to re-sort the inspection sequence with the task priority as a reference item through a multi-objective optimization algorithm.

[0006] In summary, there are two main limitations in the path planning problem of UAV intelligent inspection system: 1) The traditional path planning algorithm has slow convergence speed, low node utilization in the later stage, and unstable path; 2) The traditional sorting of target points during multi-target point traversal only considers the path distance factor, while ignoring factors such as the risk coefficient of the task point itself. Therefore, the study of the UAV inspection path planning method based on the risk level of the inspection point has major research significance and application value. Summary of the invention

[0007] The technical problem to be solved by the present invention is that the path planning algorithm has a slow convergence speed. The present invention provides an inspection path planning method, comprising the following steps:

[0008] Step 1: Simulate the experimental map and establish the mathematical model of the path planning algorithm;

[0009] The mathematical model includes a map model and a path planning model;

[0010] Mark inspection task points in the map model, and set a risk level for each task point. The risk level includes two types: high risk and normal risk.

[0011] Step 2, construct the BI-RRT* adaptive step-length path planning algorithm;

[0012] Use the BI-RRT* adaptive step-size path planning algorithm to initially measure and record the path distance between any two inspection task points required by the drone;

[0013] Step 3: Use the grey wolf algorithm to obtain the inspection sequence of inspection task points.

[0014] Step 4, according to the inspection sequence, the BI-RRT* adaptive step length path planning algorithm is used to calculate the distance between two adjacent task points, and to perform path planning between two adjacent task points;

[0015] When planning the path between two adjacent task points, the number of planning iterations is set according to the distance between the two adjacent task points, so that the number of iterations is proportional to the known distance, making the planning speed fast and the obtained path optimal, and obtaining the complete optimal path for the drone to complete the inspection task at the minimum cost.

[0016] Furthermore, the map model in step 1 refers to: simulating a two-dimensional view of the real environment according to the real environment, and in the two-dimensional view, obstacles and feasible spaces are distinguished by the same color.

[0017] The path planning model described in step 1 refers to:

[0018] Let X be the state space of the planning problem, is the collision space, representing the collision state with the obstacle, and its complement X free =X\X obs is a free space, and any state in this set is a non-collision state; x start ∈X free is the initial state of the path, x goal ∈X free is the final state of the desired path, and the final planned feasible path sequence state is σ[0,1]→X free , where σ(0) = x start ,σ(1)=x goal , the set of all feasible paths is Σ;

[0019] The optimal path planning problem is to * The optimal path is in the free space X free Connect from x start to x goal The cost of c:Σ→R ≥0 The minimum path is shown in (1)

[0020]

[0021] Furthermore, in step 2, the BI-RRT* adaptive step-length path planning algorithm is used to calculate the path distance between any two task points of the inspection task points required by the drone. The steps are as follows:

[0022] Step 2.1: define any two task points as starting points x start and the end point x goal ; By starting from the starting point x start and the end point x goal Build a search tree G respectively a and search tree G b ;

[0023] Step 2.2, in the Sample() sampling function, get a random point x rand ; Get the search tree G through the Nearest() function a Middle distance x rand The nearest node x nearest ;

[0024] Step 2.3, use the Steer() function to calculate the random point x rand and node x nearest The two node information and step length determine the new node x new Location;

[0025] If x new With x nearestIf the path between them is feasible, then x new Add to the search tree G a middle;

[0026] If x new With x nearest If the path between them is not feasible, return to step 2.2 and reselect a random point x rand , until the path is feasible;

[0027] Step 2.4, use the Near() function to find the search tree G a Middle distance x new The set of all neighboring nodes whose distance is less than the set radius X near , calculate each neighboring node x near To starting point x start The path cost and x new to x near The path cost of

[0028] If there is a neighboring node x near As the parent node, from x_ start to x new The total distance cost is greater than the total distance cost of node x nearest When the total distance cost is smaller when it is the parent node, the node is taken as x new The parent node of the random tree G a Rewiring is performed;

[0029] Step 2.5, check the search tree G b Middle distance x new Whether the nearest node can be found in the search tree G a Establish a connection directly;

[0030] If the connection can be established, the path planning is completed, and the cost value with the minimum path cost among all feasible paths is taken as c best ,If the connection cannot be established, repeat the above steps 2.2-2.5 until the conditions for establishing a connection are met;

[0031] G b Enter the next iteration of the search tree, that is, the search tree G in steps 2.2-2.5 b Replace with search tree G b , until the maximum number of iterations is reached; get the path between any two inspection task points.

[0032] Furthermore, the operation process of the Sample() function in step 2.2 is as follows:

[0033] At the starting point x start and the end point x goalRandom sampling is performed between until the initial path is obtained;

[0034] According to the Informed-RRT* algorithm, the sampling area is restricted to the ellipse subset, and there is a certain probability that the original ellipse subset of the Informed RRT* algorithm is used as the sampling area, or sampling is performed in the quadratic constraint area restricted by the quadratic constraint based on the sampling area;

[0035] By retrieving the current optimal feasible path σ * The location information of , the rectangular area where its path is located, and the optimal feasible path σ * The total path cost cost;

[0036] The constraint probability p is calculated according to the total path cost, and its formula is as shown in formula (7);

[0037] p=(c_min / cost)×ζ (16)

[0038] Where ζ is the constraint coefficient, c_min is the starting point x start and the end point x goal The straight-line distance between them; the probability that a sampling point falls within the quadratic constraint region is proportional to the ζ value;

[0039] Use the random number rand. When rand is less than the probability p, sampling is performed in the quadratic constraint region subset. Otherwise, sampling is still performed in the original elliptical region subset.

[0040] Furthermore, the sampling area is restricted to the ellipse subset according to the Informed-RRT* algorithm, specifically:

[0041] In the Informed RRT* algorithm, starting point x start , end point x goal As the focus of the ellipse, the major axis length a is taken as the initial path length c best half of Pick The minor axis length At this point, all the parameters of the elliptic equation have been obtained;

[0042] First, the transformation matrix L required to convert the uniformly distributed samples in the n-dimensional unit sphere to the uniformly distributed samples in the hyperellipsoid is as shown in formula (2);

[0043]

[0044] The rotation matrix C from the ellipsoidal coordinate system to the world coordinate system is as shown in formula (3);

[0045] C=Udiag{1,...,1,det(U)det(V)}V T(18)

[0046] Where det(U) and det(V) are the determinants of matrices U and V respectively. U and V are unitary matrices obtained by singular value decomposition of matrix M, as shown in formula (4);

[0047] UΣV T ≡M(19) where M is the horizontal axis a of the super ellipsoid in the world coordinate system 1 With the identity matrix l 1 The outer product of the first column is as shown in formula (5);

[0048] M=a 1 × 1 (20)

[0049] After calculation, we finally get the sample x in the sampling domain of the world coordinate system. f ~u(X f ) as formula (6);

[0050] x f =CLx ball +x center (twenty one).

[0051] Furthermore, in step 2.2, when obtaining the search tree G a Middle distance x rand The closest node x nearest In the process, an adaptive step size strategy algorithm is adopted, as follows:

[0052] Step 2.2.1, obtain the sampling point x rand and the nearest node x nearest After the information is obtained, the sampling point x rand and the nearest node x nearest The midpoint of is the vertex, and the Divide Map() function is used to divide the two main obstacle areas into Local Maps;

[0053] Step 2.2.2, calculate the ratio of the area occupied by obstacles in the Local Map of the two main obstacle areas, denoted as the obstacle coefficient μ;

[0054] If the obstacle coefficient μ of the area is less than the set value, it is considered that there are few obstacles, and the initial defined step size prepstep is used;

[0055] On the contrary, when the obstacle coefficient μ is greater than or equal to the set value, it is considered that there are many obstacles, and the initial defined step length prep step is reduced; the implementation is based on the sampling point x rand and the nearest node x nearest The distance and obstacle coefficient μ are used to preliminarily determine the step length;

[0056] Step 2.2.3, introduce the correction factor λ after obtaining the preliminary step length;

[0057] λ = cost_goal / cost_rand;

[0058] Among them, cost_goal represents x nearest When the target point x goal Distance cost, cost_rand represents x nearest When the target point x rond Distance cost;

[0059] Step 2.2.4, design x new The calculation method is:

[0060] x new =x nearest +step×[cos(theta)sin(theta)]

[0061] Among them, step is equal to the smaller value of pre step×λ and cost_rand; theta represents x nearest and x rond The angle between two points;

[0062] Furthermore, in step 3, the gray wolf algorithm is used to obtain the inspection sequence of the inspection task points. The specific process is as follows:

[0063] In the gray wolf algorithm, each gray wolf represents a candidate solution in the population, which is the inspection sequence;

[0064] The optimal solution is regarded as α wolf, the second and third best candidate solutions are regarded as β wolf and δ wolf, and the remaining candidate solutions are regarded as ω; assuming that the population of gray wolves is N, the search space is Dim dimensional space, and the position vector of the i-th gray wolf in the Dim dimensional space can be expressed as X i Dim The mathematical model of gray wolves hunting prey is described by the formula (7) (8):

[0065] D=|C·X p (t)-X(t)| (22)

[0066] X(t+1)=X p (t)-A·D (23)

[0067] Where: t represents the current iteration number; A and C are coefficient vectors; X p is the position vector of the prey; X represents the position vector of the gray wolf;

[0068] The coefficient vectors A and C are calculated as:

[0069] A=2a·r1 -a (24)

[0070] C=2·r 2 (25)

[0071] Where: r 1 and r 2 The modulus is a random number between [0,1]; a is the convergence factor; it decreases linearly from 2 to 0 as the number of iterations increases, as shown in formula (11);

[0072]

[0073] Where: t is the current iteration number; t max is the maximum number of iterations;

[0074] The mathematical model of the gray wolf tracking the location of its prey is described as follows:

[0075]

[0076] Where: C 1 , C 2 With C 3 is a random vector; D α , D β and D δ are the distances between wolf α, wolf β, and wolf δ and other members of the wolf pack, i.e., the distances between the points in step (2); X α , X β With X δ are the positions of wolf α, wolf β and wolf δ respectively; X is the position of the current wolf.

[0077] Furthermore, each mission point is set with a risk level, and the risk level includes two types, one is high risk and the other is normal;

[0078] In step 3, when the gray wolf algorithm is used to obtain the inspection sequence of the inspection task points, the evaluation function is as follows:

[0079]

[0080] in, γ is the weight coefficient; first_length is the path distance required for the first sequence solution, length is the path distance of the current planned sequence solution, and danger_length is the path distance required from the starting point to the last inspection point with a higher risk factor.

[0081] Furthermore, after the gray wolf algorithm falls into a local optimum, the convergence factor a value is modified so that A>1 or A<-1. At this time, the expression of the convergence factor a is:

[0082]

[0083] Beneficial effects: The present invention evaluates the cost of factors that need to be considered in comprehensive inspections, assigns tasks according to the principle of minimum cost, transforms the problem of sorting the assigned task points into a traveling salesman problem to find the minimum cost solution, ensures the safety of the inspection environment, and finally uses an improved planning algorithm to traverse the task points to complete the inspection tasks, making the inspection efficiency faster and the inspection cost the lowest, optimizing the existing working method of the intelligent inspection robot, and improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The present invention is further described in detail below with reference to the accompanying drawings.

[0085] Figure 1 It is a flow chart of the method of the present invention.

[0086] Figure 2 It is the task sequence optimization and path planning framework of the method described in the present invention.

[0087] Figure 3 It is a schematic diagram of the sampling area adopted by the path planning method described in the present invention.

[0088] Figure 4 It is a schematic diagram of the adaptive step size expansion adopted by the path planning method of the present invention.

[0089] Figure 5 This is the final planning effect diagram of the embodiment of the present invention. DETAILED DESCRIPTION

[0090] The present invention proposes a patrol route planning method, such as Figure 2 As shown in the figure, this method evaluates the path cost required in the comprehensive inspection and the risk level of the inspection task point, uses the improved gray wolf optimization algorithm to reorder the task point sequence according to the minimum cost principle, so that the comprehensive inspection cost is minimized, and finally uses the BI-RRT adaptive step size path planning algorithm to complete the path planning, as shown in Figure 1 As shown, the following steps are included:

[0091] (1): Establish mathematical models related to path planning algorithms

[0092] A. Map simulation

[0093] A 500*500 two-dimensional view of the real environment is simulated based on the real environment, in which obstacles are black squares, feasible spaces are white spaces, and 20 position coordinates are randomly set as inspection task points.

[0094] B. Path Planning Model Definition

[0095] Let X be the state space of the planning problem, is the collision space, representing the collision state with the obstacle, and its complement X free =X\X obs is a free space, and any state in this set is a non-collision state; x start ∈X free is the initial state of the path, x goal ∈X free is the final state of the desired path, and the final planned feasible path sequence state is σ[0,1]→X free , where σ(0) = x start ,σ(1)=x goal , the set of all feasible paths is Σ.

[0096] The optimal path planning problem is to * The optimal path is in the free space X free Connect from x start to x goal The cost of c:Σ→R ≥0 The minimum path is shown in (1)

[0097]

[0098] The fast exploration random tree RRT path planning algorithm is used to plan from the initial state x start To the target state x goal Even though the algorithm can always find a feasible path due to its probabilistic completeness, the planned path σ is not necessarily the optimal path σ due to the random probability. * .

[0099] C. Definition of sampling area for ellipse subset of Informed-RRT* algorithm

[0100] The Informed-RRT* algorithm is an improved method based on the RRT algorithm. It samples near the initial path generated by RRT* and iteratively obtains a better path to achieve the optimization purpose. The difference from RRT* is that it restricts its secondary sampling area to ensure that the sampling points are all in the elliptical area near the optimal path, that is, x ellipse ∈X ellipse By analogy with the standard elliptic equation, Figure 3 As shown, in the Informed RRT* algorithm, starting point x start , end point x goal As the focus of the ellipse, the major axis length a is taken as the initial path length c best half of Pick Among them, c min The value of the starting point x start , end point xgoal The Euclidean distance between .

[0101] The minor axis length At this point, all parameters of the ellipse equation have been obtained.

[0102] First, the transformation matrix L required to convert samples uniformly distributed in the n-dimensional unit sphere to samples uniformly distributed in the hyperellipsoid is as shown in formula (2).

[0103]

[0104] The rotation matrix C from the ellipsoidal coordinate system to the world coordinate system is as shown in formula (3).

[0105] C=Udiag{1,...,1,det(U)det(V)}V T (31)

[0106] Where det(U) and det(V) are the determinants of matrices U and V respectively. U and V are unitary matrices obtained by singular value decomposition of matrix M, as shown in formula (4).

[0107] UΣV T ≡M(32) where M is the horizontal axis a of the super ellipsoid in the world coordinate system 1 With the identity matrix l 1 The outer product of the first column is as shown in formula (5).

[0108] M=a 1 × 1 (33)

[0109] Through calculation, the coordinate information of the sampling point in the world coordinate system is finally obtained, as shown in formula (6).

[0110] x f =CLx ball +x center (34)

[0111] Among them, x ball Represents the position coordinates obtained by random sampling on the unit circle, x center Represents the coordinates of the center of the ellipse.

[0112] (2): BI-RRT* adaptive step-length path planning algorithm based on environmental complexity

[0113] A. BI-RRT* Algorithm

[0114] In order to balance the number of nodes around the starting point and improve the search efficiency, the present invention adopts BI-RRT* as the search algorithm to complete the planning task. BI-RRT* introduces the RRT-Connect bidirectional search algorithm based on the Informed-RRT* algorithm.

[0115] First, use the Sample() sampling function to get a random point x rand , get the search tree G through the Nearest() function a Middle distance x rand The nearest node x nearest , the Steer() function is based on the node x rand and node x nearest The two node information and step length determine the new node x new Location;

[0116] If x new With x nearest If the path between them is feasible, then x new Add to the search tree G a In the example above, we use the Near() function to find the search tree G. a Middle distance x new The set of all neighboring nodes whose distance is less than the set radius X near , and calculate each neighboring point x near To starting point x goal The path cost and x new to x near The path cost, if there is x near As the parent node, from x goal to x new The total distance cost is smaller, and the node is taken as x new After the parent node is changed, in order to further reduce the path cost, the search tree G a Rewire and change x new As the parent node of makes the path cost smaller, then make corresponding changes. After completing the above steps, check the search tree G b Middle distance x new Whether the nearest node can be found in the search tree G a Directly establish a connection. If possible, the path planning is completed. The cost value of the path with the minimum cost among all feasible paths is taken as c best , with G b Enter the next iteration for the search tree until the maximum number of iterations is reached.

[0117] B.Sample() function

[0118] The present invention proposes a sampling scheme to further restrict the ellipse subset, and restrict the sampling points again with a certain probability, so that the search tree grows closer to the minimum distance between the starting point and the target point. The probability is determined by the current optimal feasible path σ *The path cost is obtained by the distance between the starting and ending points, and the probability of the sampling point falling in the area increases as the path cost decreases iteratively. The sampling diagram is shown in the attached figure. Figure 3 shown.

[0119] If the initial path planning is not completed, random sampling is performed on the entire map until the initial path is obtained; when the initial path planning is completed, the sampling area is limited to the ellipse subset according to the Informed-RRT* algorithm, and the current optimal feasible path σ is retrieved. * The location information of , the rectangular area where its path is located, and the optimal feasible path σ * The total path cost cost is used to calculate the constraint probability p, and its formula is as shown in formula (7).

[0120] p=(c_min / cost)×ζ (35)

[0121] Where ζ is the constraint coefficient, and c_min is the straight-line distance between the starting and ending points. The probability that the sampling point falls within the quadratic constraint region is proportional to the ζ value. Using the random number rand, when rand is less than the probability p, sampling is performed in the region subset within the reduced elliptical region, otherwise sampling is still performed in the original elliptical region subset. The algorithm flow of the Sample() function using the sampling region constraint scheme is shown in Table 1.

[0122] Table 1. Sample function algorithm flow

[0123]

[0124] Among them, c_max represents the optimal distance cost in the planned path. When it is not planned, its value is assigned to ∞, so that after the initial planning, the secondary sampling area can be limited according to the real-time distance value. The function of probability() is to calculate the sampling constraint probability p according to the path information; Min Elliptic_sample() function is a specific sampling function for sampling in the reduced elliptical area subset; Elliptic_sample() function is a specific sampling function for sampling in the elliptical area subset; rand_sample() function is a specific sampling function for sampling in the entire area.

[0125] C. Adaptive step size strategy algorithm

[0126] The present invention uses the Self-Adaption-Steer() function to determine the new growth node x new A method based on sampling point x is proposed rand and the nearest node x nearest The adaptive step size strategy algorithm of the interval obstacle degree is based on the acquisition of sampling points x rand and the nearest node xnearest The complexity of the obstacles between the two points determines the step length step, and the sampling point x rand Large deviation from the target point x goal The step size is modified in the case of Figure 4 shown.

[0127] The specific method of the adaptive step size adjustment strategy is as follows: rand and the nearest node x nearest After the information is obtained, the sampling point x rand and the nearest node x nearest The midpoint of is the vertex, dividing the two main obstacle areas into Local Map, with x nearest The density of obstacles in the nearby area is the first criterion. The obstacle coefficient μ is calculated. If there are few obstacles in the area, the initial definition step length prep step is used. Otherwise, the initial definition step length prep step is reduced. Then, according to the sampling point x rand and the nearest node x nearest The distance and μ are used to initially determine the step size, and to avoid the generation of sampling points x rand Large deviation from the target point x goal In order to speed up the path planning efficiency, a correction coefficient λ is introduced after obtaining the initial step length. This coefficient is calculated based on the nearest node x. nearest Respectively with the sampling point x rand With the target point x goal The Euclidean distance between the sampling points can be adjusted. If the distance to the target point is small, it means that there is a deviation. According to the correction coefficient λ, the step size can be appropriately reduced. If the distance to the sampling point x rand If it is smaller, the step size is appropriately increased according to the correction coefficient λ, so the value of the correction coefficient λ is directly proportional to the ratio of the distance cost from the target point to the distance cost from the sampling point. The algorithm flow of the Self-Adaption-Steer() function is shown in Table 2.

[0128] Table 2. Self-Adaption-Steer function algorithm flow

[0129]

[0130]

[0131] The Divide Map() function is to divide the sampling point x rand and the nearest node x nearestThe partial rectangular map Local Map as the diagonal line is extracted from the overall map; the function of the Obs_area() function is to calculate the area occupied by obstacles in the area; the function of the Size() function is to calculate the total area of ​​the Local Map area; the function of the Angle cost() function is to calculate the angle between two points.

[0132] (3) Inspection sequence optimization algorithm based on the grey wolf algorithm

[0133] A. Grey Wolf Optimization Algorithm

[0134] The Grey Wolf Optimize (GWO) algorithm is a group intelligence optimization algorithm formed by simulating the process of foraging behavior of wolves in the biological world. The wolf pack has a strict social hierarchy and organizational system. The top wolf in the wolf pack is the α wolf (the leader of the wolf pack), the β wolf is the subordinate wolf of the α wolf, and obeys the instructions of the α wolf. The δ wolf is second only to the α wolf and the β wolf, and the remaining wolves are the ω wolves, with the lowest status. The hunting of the wolf pack mainly includes three stages: searching, encircling and attacking. In the mathematical model of the Grey Wolf Optimization GWO algorithm, each grey wolf represents a candidate solution in the population, the optimal solution is regarded as the α wolf, the second and third best candidate solutions are regarded as the β wolf and the δ wolf, and the remaining candidate solutions are regarded as ω. The solution in the present invention is the inspection sequence. Assuming that the population size of the grey wolf is N (that is, the number of solutions preset by the algorithm), the search space is the Dim dimensional space (that is, the number of inspection points), and the position vector of the i-th grey wolf in the Dim dimensional space can be expressed as X i Dim , the mathematical model of gray wolves hunting prey is described as follows:

[0135] D=|C·X p (t)-X(t)| (36)

[0136] X(t+1)=X p (t)-A·D (37)

[0137] Where: t represents the current iteration number; A and C are coefficient vectors; X p is the position vector of the prey, that is, the position information of each inspection point; X represents the position vector of the gray wolf, that is, the position information preset by the current algorithm. The calculation formula of the coefficient vectors A and C is:

[0138] A=2a·r 1 -a (38)

[0139] C=2·r 2 (39)

[0140] Where: r 1 and r 2The modulus is a random number between [0, 1]; a is the convergence factor, which linearly decreases from 2 to 0 as the number of iterations increases.

[0141]

[0142] In the formula: t is the current number of iterations; t max is the maximum number of iterations.

[0143] The mathematical model for the gray wolves to track the prey's position is described as follows:

[0144]

[0145] In the formula: C 1 , C 2 and C 3 are random vectors; D α , D β and D δ are the distances between the alpha wolf, beta wolf, and delta wolf and the other members of the wolf pack respectively, that is, the distances between the points obtained in step (2); X α , X β and X δ are the positions of the alpha wolf, beta wolf, and delta wolf respectively; X is the position of the current wolf.

[0146] B. Improved gray wolf optimization algorithm

[0147] When the prey stops moving, the gray wolves complete the hunting process by attacking. To simulate approaching the prey, the value of the convergence factor a is gradually decreased, so the fluctuation range of the coefficient vector A also decreases accordingly. That is, during the iteration process, when the value of the convergence factor a linearly decreases from 2 to 0, the value of its corresponding coefficient vector A also changes within the interval [-a, a]. When the value of the coefficient vector A is within the interval [-a, a], the next position of the gray wolf can be at any position between its current position and the prey's position. When -1 < A < 1, the wolf pack attacks the prey and falls into local optimum. When A > 1 or A < -1, the gray wolves move away from the prey and explore other areas to continue searching for the global optimum solution.

[0148] To avoid the algorithm falling into local optimum, a local optimum detection algorithm is added to the algorithm. The detection algorithm obtains the number of times of getting the same solution. If the number of iterations of obtaining the same solution exceeds a certain limit, it is determined that the algorithm has fallen into local optimum. After determining that the algorithm has fallen into local optimum, the value of the convergence factor a is modified according to the current number of iterations t, such that A > 1 or A < -1. At this time, the calculation formula for the value of a is as in (14):

[0149]

[0150] At the same time, in order to avoid the sequence solution given by the algorithm not giving priority to the task points with higher risk coefficients during the execution process, the evaluation function of the optimization algorithm is modified in the present invention, and the evaluation function is (15):

[0151]

[0152] in, γ is the weight coefficient. first_length is the path distance required for the first sequence solution, length is the path distance of the current planned sequence solution, and danger_length is the path distance required to reach the last inspection point with a higher risk factor from the starting point. Figure 5 As shown, this is the path distance from the starting point 1 to the last inspection point 16 with a higher risk factor.

[0153] (4) Final inspection route planning

[0154] Through the above steps, the final inspection order can be obtained. The algorithm described in step (2) can be used again, while increasing the number of algorithm iterations to ensure that the path cost between the two points is optimal, thereby completing the entire inspection path planning.

[0155] In the experimental map, the number of inspection points is set to 20, the number of inspection points with higher risk coefficient is set to 5, and the population of the gray wolf algorithm is set to 100. The inspection path is planned, and it is proved that the path planned by this method can achieve the optimal path cost under the premise of ensuring that all inspection points with higher risk coefficient are inspected first, and there are no other redundant inspection paths. The planning results are shown in the figure. Figure 5 As shown, task points 1, 2, 8, 11, and 16 are task points with higher risk systems.

[0156] The method of the present invention comprehensively evaluates the path cost required in the inspection and the risk level of the inspection task point, uses the improved gray wolf optimization algorithm to reorder the task point sequence according to the minimum cost principle, so that the comprehensive inspection cost is minimized, and finally uses the IRRT*-Connect* adaptive step size path planning algorithm to complete the path planning.

[0157] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for planning a drone inspection path, characterized in that: The following steps are involved: Step 1, simulate the experimental map and establish the mathematical model of the path planning algorithm; The mathematical model includes a map model and a path planning model; inspection task points are marked in the map model; Step 2: Initially plan and record the path distance between any two inspection task points required by the drone; Step 3, using the grey wolf algorithm to obtain the inspection sequence of the inspection task points; Step 4: Calculate the distance between two adjacent task points according to the inspection sequence, and plan the path between the two adjacent task points.

2. The method for planning a drone inspection path according to claim 1, characterized in that: The map model mentioned in step 1 refers to: A two-dimensional view of the real environment is simulated according to the real environment, and obstacles and feasible spaces are distinguished by using different colors in the two-dimensional view.

3. The method for planning a drone inspection path according to claim 1, characterized in that: The path planning model described in step 1 refers to: Let X be the state space of the planning problem, is the collision space, representing the collision state with the obstacle, and its complement X free =X\X obs is a free space, and any state in this set is a non-collision state; x start ∈X free is the initial state of the path, x goal ∈X free is the final state of the desired path, and the final planned feasible path sequence state is σ[0,1]→X free , where σ(0) = x start ,σ(1)=x goal , the set of all feasible paths is Σ; The optimal path planning problem is to * The optimal path is in the free space X free Connect from x start to x goal The cost of c:Σ→R ≥0 The minimum path is shown in (1) 4. The method for planning a drone inspection path according to claim 1, characterized in that: In step 2, the BI-RRT* adaptive step-length path planning algorithm is used to calculate the path distance between any two task points of the inspection task points required by the drone. The steps are as follows: Step 2.1: define any two task points as starting points x start and the end point x goal ; By starting from the starting point x start and the end point x goal Build a search tree G respectively a and search tree G b ; Step 2.2, in the Sample() sampling function, get a random point x rand ; Get the search tree G through the Nearest() function a Middle distance x rand The nearest node x nearest ; Step 2.3, use the Steer() function to calculate the random point x rand and node x nearest The two node information and step length determine the new node x new Location; If x new With x nearest If the path between them is feasible, then x new Add to the search tree G a middle; If x new With x nearest If the path between them is not feasible, return to step 2.2 and reselect a random point x rand , until the path is feasible; Step 2.4, use the Near() function to find the search tree G a Middle distance x new The set of all neighboring nodes whose distance is less than the set radius X near , calculate each neighboring node x near To starting point x start The path cost and x new to x near The path cost of If there is a neighboring node x near As the parent node, from x_ start to x new The total distance cost is greater than the total distance cost of node x nearest When the total distance cost is smaller when it is the parent node, the node is taken as x new The parent node of the random tree G a Rewiring is performed; Step 2.5, check the search tree G b Middle distance x new Whether the nearest node can be found in the search tree G a Establish a connection directly; If the connection can be established, the path planning is completed, and the cost value with the minimum path cost among all feasible paths is taken as c best ,If the connection cannot be established, repeat the above steps 2.2-2.5 until the conditions for establishing a connection are met; G b Enter the next iteration for the search tree until the maximum number of iterations is reached; Get the path between any two inspection task points.

5. The method for planning a drone inspection path according to claim 4, characterized in that: The operation process of the Sample() function is as follows: At the starting point x start and the end point x goal Random sampling is performed between until the initial path is obtained; According to the Informed-RRT* algorithm, the sampling area is limited to the ellipse subset, and the current optimal feasible path σ is retrieved. * The location information of , the rectangular area where its path is located, and the optimal feasible path σ * The total path cost cost; The constraint probability p is calculated according to the total path cost, and its formula is as shown in formula (7); p=(c_min / cost)×ζ (2) Where ζ is the constraint coefficient, c_min is the starting point x start and the end point x goal The straight-line distance between them; the probability that a sampling point falls within the quadratic constraint region is proportional to the ζ value; Use the random number rand. When rand is less than the probability p, sampling is performed in the quadratic constraint region subset, otherwise sampling is still performed in the original elliptical region subset.

6. A method for planning a drone inspection path according to claim 5, characterized in that: The sampling area is restricted to the ellipse subset according to the Informed-RRT* algorithm, specifically: In the Informed RRT* algorithm, starting point x start , end point x goal As the focus of the ellipse, the major axis length a is taken as the initial path length c best half of Pick The minor axis length At this point, all the parameters of the elliptic equation have been obtained; First, the transformation matrix L required to convert the uniformly distributed samples in the n-dimensional unit sphere to the uniformly distributed samples in the hyperellipsoid is as shown in formula (2); The rotation matrix C from the ellipsoidal coordinate system to the world coordinate system is as shown in formula (3); C=Udiag{1,...,1,det(U)det(V)}V T (4) Where det(U) and det(V) are the determinants of matrices U and V respectively. U and V are unitary matrices obtained by singular value decomposition of matrix M, as shown in formula (4); UΣV T ≡M(5) where M is the outer product of the horizontal axis a1 of the super ellipsoid and the first column of the identity matrix l1 in the world coordinate system, as shown in formula (5); M=a1×l1 (6) Through calculation, we finally get the sample x in the sampling domain of the world coordinate system f ~u(X f ) as formula (6); x f =CLx ball +x center (7)。 7. The method for planning a drone inspection path according to claim 4, characterized in that: In step 2.2, we obtain the search tree G a Middle distance x rand The closest node x nearest In the process, an adaptive step size strategy algorithm is adopted, as follows: Step 2.2.1, obtain the sampling point x rand and the nearest node x nearest After the information is obtained, the sampling point x rand and the nearest node x nearest The midpoint of is the vertex, and the Divide Map() function is used to divide the two main obstacle areas into Local Maps; Step 2.2.2, calculate the ratio of the area occupied by obstacles in the Local Map of the two main obstacle areas, denoted as the obstacle coefficient μ; If the obstacle coefficient μ of the area is less than the set value, it is considered that there are few obstacles, and the initial defined step size prepstep is used; On the contrary, when the obstacle coefficient μ is greater than or equal to the set value, it is considered that there are many obstacles, and the initial defined step length prepstep is reduced; the implementation is based on the sampling point x rand and the nearest node x nearest The distance and obstacle coefficient μ are used to preliminarily determine the step length; Step 2.2.3, introduce the correction factor λ after obtaining the preliminary step length; λ = cost_goal / cost_rand; Among them, cost_goal represents x nearest When the target point x goal Distance cost, cost_rand represents x nearest When the target point x rond Distance cost; Step 2.2.4, design x new The calculation method is: x new =x nearest +step×[cos(theta)sin(theta)] Among them, step is equal to the smaller value of pre step×λ and cost_rand; theta represents x nearest and x rond The angle between two points.

8. The method for planning a drone inspection path according to claim 1, characterized in that: In step 3, the gray wolf algorithm is used to obtain the inspection sequence of the inspection task points. The specific process is as follows: In the gray wolf algorithm, each gray wolf represents a candidate solution in the population, which is the inspection sequence; The optimal solution is regarded as α wolf, the second and third best candidate solutions are regarded as β wolf and δ wolf, and the remaining candidate solutions are regarded as ω; assuming that the population of gray wolves is N, the search space is Dim dimensional space, and the position vector of the i-th gray wolf in the Dim dimensional space can be expressed as X iDim The mathematical model of gray wolves hunting prey is described by the formula (7) (8): D=|C·X p (t)-X(t)| (8) X(t+1)=X p (t)-A·D (9) Where: t represents the current iteration number; A and C are coefficient vectors; X p is the position vector of the prey; X represents the position vector of the gray wolf; The coefficient vectors A and C are calculated as: A=2a·r1-a (10) C＝2·r2 (11) Where: the modulus of r1 and r2 is a random number between [0,1]; a is the convergence factor; it decreases linearly from 2 to 0 as the number of iterations increases, as shown in formula (11); Where: t is the current iteration number; t max is the maximum number of iterations; The mathematical model of the gray wolf tracking the location of its prey is described as follows: Where: C1, C2 and C3 are random vectors; D α , D β and D δ are the distances between wolf α, wolf β, and wolf δ and other members of the wolf pack, i.e., the distances between the points in step (2); X α , X β With X δ are the positions of wolf α, wolf β and wolf δ respectively; X is the position of the current wolf.

9. The method for planning a drone inspection path according to claim 1, characterized in that: Each mission point is set with a risk level, which includes two risk levels: high risk and normal risk. In step 3, when the gray wolf algorithm is used to obtain the inspection sequence of the inspection task points, the evaluation function is as follows: in, γ is the weight coefficient; first_length is the path distance required for the first sequence solution, length is the path distance of the current planned sequence solution, and danger_length is the path distance required from the starting point to the last inspection point with a higher risk factor.

10. The method for planning a drone inspection path according to claim 1, characterized in that: After the Gray Wolf Algorithm falls into a local optimum, the convergence factor a value is modified so that A>1 or A<-1. At this time, the expression of the convergence factor a is:

Images