Multi-unmanned aerial vehicle cooperative inspection path planning method for angle tower insulator in mountainous area

Abstract

The invention belongs to the technical field of unmanned aerial vehicle inspection of a power system, and particularly relates to a multi-unmanned aerial vehicle cooperative inspection path planning method for an angle tower insulator in a mountainous area. Comprising the steps that S1, a mountainous area angle tower insulator inspection multi-constraint model is constructed, wherein the mountainous area angle tower insulator inspection multi-constraint model comprises a mountainous area three-dimensional terrain obstacle model, an angle tower insulator detection task model, an electrical safety distance model and a multi-target comprehensive cost model; s2, a WHA algorithm: initializing a population by adopting segmented chaotic mapping to improve the distribution uniformity of a solution, calculating tracking energy to judge the conversion between a global wide-area pursuit stage and a local precise pursuit stage, optimizing path details in a plurality of differentiation modes in the local precise pursuit stage, and introducing a self-adaptive random walk strategy to prevent premature convergence of the algorithm; and S3, multi-machine collaborative path optimization and dynamic adjustment. Unification of high efficiency, safety and detection integrity of the routing inspection path can be realized, and the problem of path optimization of angle tower insulators executed by multiple machines cooperatively under complex terrains is solved.

Description

Technical Field

[0001] This invention belongs to the field of power system unmanned aerial vehicle (UAV) inspection technology, specifically relating to a multi-UAV collaborative inspection path planning method for insulators of corner towers in mountainous areas. Background Technology

[0002] Corner towers of power transmission lines are critical nodes in mountainous power transmission networks. The insulators they house, as core components for insulation and mechanical support, are highly susceptible to damage such as line tripping and equipment burnout if defects like cracks, dirt accumulation, or broken skirts occur. Therefore, regular inspections are essential. However, inspecting insulators on corner towers in mountainous areas faces multiple technical challenges: 1. Complex terrain limits single-drone access and endurance. Mountainous inspection areas commonly feature steep slopes, dense forests, and deep ravines, making it easy for drones to be obstructed by terrain, increasing detour distances. Simultaneously, the low air pressure and strong airflow in mountainous areas exacerbate drone energy consumption, resulting in low single-drone endurance and frequent return-to-base recharging, leading to low inspection efficiency. 2. The unique nature of insulator inspection tasks makes locations prone to poor inspection results. Insulators on corner towers are often arranged in a V-shape or string pattern and are easily obstructed by the tower's steel frame. If the drone's shooting position is poor, the insulator caps and discs can easily block each other, making it impossible to meet the defect identification requirements and easily missing key defects such as cracks and edge contamination, making it difficult to guarantee the integrity of the inspection. 3. Traditional path planning algorithms have insufficient performance and poor adaptability to multi-machine collaboration. Existing path planning algorithms, such as the basic particle swarm optimization algorithm, wolf pack algorithm, and ant colony algorithm, all have problems such as uneven distribution of initial solutions and premature convergence in the later stages of iteration. Among them, the basic particle swarm optimization algorithm is prone to incomplete solution space coverage due to population initialization; the wolf pack algorithm's collaboration logic does not integrate insulator detection perspective constraints and electrical safety distance constraints; and the ant colony algorithm relies on the positive feedback mechanism of pheromone accumulation, which is prone to getting stuck in local optima in complex mountainous terrain. Moreover, each algorithm is difficult to balance the optimization accuracy and collaborative efficiency of multi-machine paths. At the same time, these algorithms generally lack unified modeling for mountainous terrain, insulator detection requirements, and electrical safety constraints; when multiple machines execute tasks, problems such as path conflicts and unbalanced task allocation are prone to occur.

[0003] In other words, as the core insulation and support component of the corner tower of the mountain transmission line, the performance of the insulator directly determines the safe operation of the power system. However, the mountain inspection environment has terrain obstacles such as steep slopes and dense forests, and the corner tower body can easily block the detection of the insulator. Coupled with the strict constraints of electrical safety distance, the traditional single-machine inspection mode has problems such as insufficient endurance, many blind spots, and low efficiency. At the same time, conventional path planning algorithms are prone to problems that are difficult to adapt to the needs of multi-machine collaboration.

[0004] Chinese invention patent application CN202010957742.9 discloses a method for planning the inspection path of a power plant drone, including the following steps: 1) On-site survey: Conduct an on-site survey of the route to be inspected, and record the take-off point, landing point, and path points; 2) Determine the inspection route and analyze its rationality; 3) Drone inspection: After take-off, remotely control the drone to fly along the inspection route determined in step 2); 4) Record the coordinates of the starting point in step 1), and then collect path points sequentially along the coordinates. This method for planning the inspection path of a power plant drone, by conducting an on-site survey of the route to be inspected before drone inspection, converting the surveyed route into a digital elevation map, extracting existing obstacles, and judging the rationality of the inspection route, achieves high safety, avoids excessive invalid flight during drone flight, and improves the efficiency of inspection. The core of the process in the patent application, "site survey - route determination - remote control flight - fault diagnosis", is a process-based manual-assisted planning that requires human intervention and lacks complex algorithms and design constraints such as mountainous terrain and insulator viewing angle coverage.

[0005] Chinese invention patent application CN202410210462.X provides a method, device, and electronic device for planning waypoints for unmanned aerial vehicles (UAVs). The method includes: modeling an inspection area to obtain an obstacle model; determining multiple tangent points between a straight line passing through the current position and the obstacle model based on the current position and the obstacle model; determining an initial set of waypoint positions based on the multiple tangent points and a critical distance, where the distance between each initial waypoint position and any tangent point is greater than the critical distance; determining flight constraints based on current operational information; establishing an objective function with the waypoint positions of the target UAV as the dependent variable and minimizing the total flight path length of the target UAV as the objective; solving the objective function based on the flight constraints to obtain multiple optimal waypoint positions from the initial set of waypoint positions; determining the flight trajectory based on the current position and the multiple optimal waypoint positions, and controlling the target UAV to fly according to the flight trajectory. This application solves the problems of poor flight path planning and weak adaptive capability in existing UAV technologies. While this patent application uses a basic particle swarm optimization algorithm, it still suffers from the uneven distribution and premature convergence problems of traditional algorithms, and its planning objective is only "minimizing the total flight path length."

[0006] In summary, existing single-machine inspection modes and traditional swarm intelligence algorithms cannot meet the complex requirements of inspecting insulators on angle towers in mountainous areas. There is an urgent need for a path planning method that integrates a multi-machine collaborative framework, accurate environmental modeling, and high-performance optimization algorithms to overcome the aforementioned technical bottlenecks. Summary of the Invention

[0007] This invention addresses the aforementioned problems by studying the coupling characteristics of complex terrain constraints in mountainous areas, the need for full-view insulator inspection, and electrical safety regulations. It overcomes the shortcomings of traditional single-machine inspection methods, such as insufficient endurance and numerous blind spots, as well as the uneven initial population distribution and premature convergence in later iterations of conventional swarm intelligence algorithms like Particle Swarm Optimization (PSO) and Wolf Pack algorithms. The proposed method is a multi-UAV collaborative inspection path planning approach for corner tower insulators in mountainous areas, based on the WHA (Wisdom Hunter Algorithm). This algorithm combines the population optimization logic of PSO with the collaborative hunting concept of Wolf Pack. By constructing a multi-machine collaborative division of labor and leveraging 3D point cloud virtual composite mountain inspection scenarios with obstacles, safety constraints, and inspection requirements, it achieves multi-objective optimization of the inspection path, improving inspection path safety, inspection completeness, and multi-machine collaborative efficiency, and enhancing the method's adaptability to complex mountainous inspection scenarios.

[0008] The technical solution of this invention is as follows: First, a multi-constraint model for the inspection of insulators at corner towers in mountainous areas is constructed. A three-dimensional terrain obstacle model is established through voxel discretization. A task model is designed in conjunction with the insulator inspection perspective requirements. An electrical safety distance model is constructed based on power industry standards. A multi-objective comprehensive cost model is formed by integrating terrain, safety, inspection, energy consumption, and turning costs. Then, the WHA algorithm is adopted. Population initialization is optimized through piecewise chaotic mapping. A stage transition of the tracking energy control algorithm is introduced. Four precise pursuit strategies are designed to optimize path details. An adaptive random walk strategy is used to suppress premature convergence of the algorithm. The algorithm is used to solve for the optimal path for multiple machines. Finally, multi-machine collaborative path optimization and dynamic adjustment are implemented. Inspection tasks are allocated according to the principle of "proximity allocation + load balancing". Multi-machine spacing constraints are added to avoid collisions. When encountering sudden obstacles, local replanning is triggered to ensure the continuous execution of inspection tasks.

[0009] This invention provides a multi-UAV collaborative inspection path planning method for insulators on corner towers in mountainous areas, comprising the following steps: S1: Constructing a multi-constraint model for insulator inspection on corner towers in mountainous areas: The multi-constraint model includes a three-dimensional terrain obstacle model for mountainous areas, a corner tower insulator detection task model, an electrical safety distance model, and a multi-objective comprehensive cost model; S2: Solving for the optimal multi-UAV inspection path based on the WHA algorithm: The WHA algorithm includes first initializing the population using a piecewise chaotic mapping to improve the uniformity of solution distribution, calculating the tracking energy to determine the transition between global wide-area pursuit and local precise pursuit stages, using various differentiated methods to optimize path details in the local precise pursuit stage, and introducing an adaptive random walk strategy to prevent premature convergence of the algorithm; S3: Multi-UAV collaborative path optimization and dynamic adjustment: Tasks are allocated according to the number of corner towers and the distribution of insulator detection points, multi-UAV spacing constraints are added to avoid collisions, and dynamic adjustments are made when unexpected situations occur, triggering local replanning to update path segments.

[0010] In one specific implementation, the three-dimensional terrain obstacle model of the mountainous area described in step S1 uses a "voxel discretization" method to discretize the inspection area into voxel units of fixed size and assign them access level labels, where 0 = gentle terrain, 1 = sparse forest, 2 = steep slope, and 3 = no passage. The terrain obstacle cost function is J. terrain (i) = k·grade(i); where k is the obstacle weight coefficient and grade(i) is the access level label of the i-th voxel unit; preferably, sparse forest k = 2.5, steep slope k = 3.2, and no-passage k = +∞.

[0011] In one specific implementation, in the corner tower insulator detection task model described in step S1, the corner tower coordinates T(x) are defined. T ,y T ,z T The coordinates of key detection points on the insulator string are used to constrain the angle between the UAV and the insulator normal to be less than or equal to a certain specific angle. The detection coverage cost function is J. cover =β× θ 2 β is the coverage penalty coefficient. θ is the angle deviation value.

[0012] In one specific implementation, in the corner tower insulator detection task model described in step S1, the coordinates of the key detection points of the insulator string include the string vertex coordinates I1(x1,y1,z1) and the string end coordinates I2(x2,y2,z2), and the angle between the UAV and the insulator normal is constrained to be ≤30°, and β=1.8.

[0013] In one specific implementation, the electrical safety distance function of the electrical safety distance model in step S1 is J. safety =γ·(d safe -d) 3 Where γ is the safety weighting coefficient, d is the actual distance, and d safe For a safe distance; if d ≥ d safe Then J safety =0.

[0014] In one specific implementation, the multi-objective integrated cost model in step S1 is J. total =ω1J terrain +ω2J safety +ω3J cover +ω4J power +ω5J steer ω1, ω2, ω3, ω4, and ω5 are all weighting coefficients, J terrain Let J be the terrain obstacle cost function. safety J is the electrical safety distance function. coverTo detect the coverage cost function, J power Let J be the energy consumption cost function. power =k·L, where k is the energy consumption factor, L is the path length, and J steer This is the turning cost function.

[0015] In one specific implementation, in the multi-objective comprehensive cost model described in step S1, the weight coefficients ω1=75, ω2=110, ω3=95, ω4=12, ω5=9; k=0.8, J steer The function is as follows:

[0016] Where, θ i This is the drone's turning angle, with the maximum turning angle set to 60°.

[0017] In one specific implementation, the tracking energy formula in step S2 is: E0 [-1.2, 1.2] represents the initial tracking energy. For the number of iterations, The maximum number of iterations is determined by |E| ≥ 1.1, at which point the algorithm enters the global wide-area pursuit stage; when |E| < 1.1, the algorithm enters the local precise pursuit stage.

[0018] In one specific implementation, during the local precision tracking stage in step S2, the tracking energy is E, and the random factor is s, s [0,1] Based on |E| and s, four methods are selected for local precision pursuit: flexible pursuit with 0.6≤|E|<1.1 and s≥0.55, rigid pursuit with |E|<0.6 and s≥0.55, diving flexible pursuit with 0.6≤|E|<1.1 and s<0.55, and diving rigid pursuit with |E|<0.6 and s<0.55.

[0019] In one specific implementation, the perturbation formula for the adaptive random walk strategy in step S2 is: ,in, Let be the optimal solution for the t-th iteration, δ(t) = 0.8·(1-t / T) be the dynamic step size, and s5 be the value of t. [0,1] is a random number; if no better solution is found in 12 consecutive iterations, the step size δ(t) is halved, and the iteration is terminated when δ(t) < 0.1m.

[0020] Regarding its beneficial effects, this invention is a multi-UAV collaborative path planning method that integrates swarm intelligence optimization ideas such as particle swarm optimization and wolf pack optimization, combined with virtual world modeling, and is adapted to the needs of complex mountainous terrain and corner tower insulator inspection. It is applicable to defect detection and safety inspection of corner tower insulators of 110kV~500kV mountainous transmission lines, and can achieve a balance between high efficiency, safety and completeness of inspection paths, solving the path optimization problem of multi-UAV collaborative execution of special inspection tasks in complex terrain. Attached Figure Description

[0021] Figure 1 A diagram showing the locations of corner towers and key insulator points in mountainous areas.

[0022] Figure 2 A top-down view of a 3D voxel model of a mountainous area. Where 0 = gentle terrain, 1 = sparse forest, and 2 = steep slope.

[0023] Figure 3 This is a three-dimensional view of the multi-machine collaborative inspection path.

[0024] Figure 4 for Figure 3 The path planning diagram of UAV 1.

[0025] Figure 5 for Figure 3 The path planning diagram of the Chinese drone 2.

[0026] Figure 6 for Figure 3 The path planning diagram of the Chinese UAV 3. Detailed Implementation

[0027] This invention proposes a multi-UAV collaborative inspection path planning method for insulators on corner towers in mountainous areas. First, a multi-constraint model for mountainous inspection is constructed: a three-dimensional terrain obstacle model is established using voxel discretization to quantify the impact of different terrains on the path; a task model is designed based on the insulator inspection perspective requirements to avoid blind spots caused by tower obstruction; an electrical safety distance model is constructed to ensure inspection compliance; and a multi-objective comprehensive cost model is formed by integrating terrain, safety, inspection, energy consumption, and turning costs. Second, the WHA algorithm is used: the population is initialized through piecewise chaotic mapping to improve the uniformity of the initial solution distribution; the tracking energy is calculated to control the transition between global wide-area pursuit and local precise pursuit stages, adapting to the "exploration first, refinement later" path requirement; four precise pursuit methods are designed to optimize path details, and an adaptive random walk strategy is introduced to suppress premature convergence of the algorithm. Finally, multi-UAV collaborative optimization is achieved: inspection tasks are allocated according to the principle of "proximity allocation + load balancing," and multi-UAV spacing constraints are added to avoid collisions.

[0028] The method for multi-UAV collaborative inspection path planning for insulators on corner towers in mountainous areas includes the following steps.

[0029] S1: Construct a multi-constraint model for the inspection of insulators on corner towers in mountainous areas.

[0030] 1. 3D Terrain Obstacle Model for Mountainous Areas. A voxel discretization method is used to model the mountainous terrain, accurately identifying obstacles and providing a clear optimization environment boundary for the algorithm. Specifically, this includes: 1) Discretizing the inspection area into fixed-size voxel units, assigning each unit a "passage level label," with the highest priority terrain labeled: 0 = gentle terrain (no obstacles), 1 = sparse forest (requires additional energy), 2 = steep slope (slope > 35°, restricted passage), 3 = ravine / rocky area (no passage). 2) Defining a terrain obstacle cost function to quantify the impact of different terrains on the path: J terrain (i) = k·grade(i); where grade(i) is the access level label of the i-th voxel, k is the obstacle weight coefficient, where sparse forest k = 2.5, steep slope k = 3.2, and no-passage k = +∞, and the cost of the no-passage area is set to infinity to ensure path avoidance.

[0031] 2. Corner Tower Insulator Inspection Task Model. A task constraint model is established to address the insulator inspection perspective and coverage requirements. Specifically, this includes: 1) Parameterization of the corner tower and insulator: defining the corner tower coordinates T(x... T ,y T ,z T ), Key detection points of the insulator string (string vertices I1(x1,y1,z1), string ends I2(x2,y2,z2)), clearly define the detection viewpoint constraint—the angle between the UAV and the insulator normal ≤ 30°, avoiding porcelain insulators. 2) Calculation of detection coverage cost: If the UAV path point P(x,y,z) satisfies the viewpoint constraint, the coverage cost J cover =0; otherwise, J cover =β× θ 2 ( θ is the viewing angle deviation angle, and β=1.8 is the coverage penalty coefficient. The cost sensitivity of the viewing angle deviation is enhanced by the squared term, which urges the path to meet the detection requirements.

[0032] 3. Electrical Safety Distance Model. An electrical safety constraint model for conductors and towers is constructed to address different voltage levels and basic safety distance requirements. Specifically, this includes: 1) Primary safety distance requirements, i.e., the distance between the drone and the conductor at each voltage level: 110kV line d safe1 ≥2m, 220kV line d safe1 ≥3m, 500kV line d safe1 ≥5m; 2) Secondary safety distance requirement, i.e., the distance between the drone and the tower: all voltage levels d safe2 ≥3m; 3) Safety distance cost function: if the actual distance d < d safe Then Jsafety =γ·(d safe -d) 3 Where γ=200 is the safety weight coefficient; if the distance d≥d safe Then J safety =0, the cubic term amplifies the cost of close-range violations, ensuring that the path strictly adheres to safety standards.

[0033] 4. Multi-objective integrated cost model. This model integrates five cost categories: terrain, safety, detection, energy consumption, and turning, to construct a total cost function and achieve multi-objective optimization: J total =ω1J terrain +ω2J safety+ ω3J cover+ ω4J power+ ω5J steer , where: J power =k·L; where L is the path length, k is 0.8, i.e., the energy consumption factor; the longer the path, the higher the J. power The larger; ; where θ i The turning angle of the drone is set to a maximum of 60°. When the turning angle exceeds half the threshold, a penalty of 1.5 is applied to reduce energy loss from large turns in the flight path. The weighting coefficients ω1=75, ω2=110, ω3=95, ω4=12, and ω5=9 are based on the priority of safety, with detection as the main focus, while also taking efficiency into account, to ensure the core requirements are met.

[0034] S2: WHA algorithm.

[0035] 1. Segmented Chaotic Mapping for Population Initialization. To address the problem of uneven population distribution during random initialization in traditional swarm intelligence algorithms such as Particle Swarm Optimization, Fruit Fly Algorithm, and Wolf Pack Algorithm, segmented chaotic mapping is used to generate candidate initial path solutions. Specifically, this includes: 1) Segmented Chaotic Mapping Formula: ;where x i A random number in [0,1] The result after chaotic mapping is that the control parameters are divided into two segments (4.1 and 3.9). Compared with fixed parameters, this can enhance the randomness and uniformity of the chaotic sequence and cover a wider solution space; 2) Population selection strategy: compare the fitness of "randomly initialized population" and "segmented chaotic population", that is, the comprehensive cost J total By retaining the top 45% of optimal solutions as the initial population, the overall quality of the initial solutions is improved, laying the foundation for subsequent iterations and optimizing the problem of insufficient initial population quality in the traditional wolf pack algorithm.

[0036] 2. Tracking Energy Calculation and Stage Transition. A "tracking energy" algorithm stage is defined. This mechanism references the stage cooperation concept of the wolf pack algorithm while incorporating the adaptive adjustment logic of the particle swarm algorithm, enabling dynamic switching between global exploration and local development. Specifically, it includes: 1) Tracking Energy Formula: Among them, E0 [-1.2, 1.2] represents the initial tracking energy, and E0 is a random number within this data range. For the number of iterations, To maximize the number of iterations, the square root sign ensures that the energy decays smoothly with the iteration process, which aligns with the "explore first, refine later" requirement of path optimization; 2) Stage transition rules: When |E|≥1.1, the algorithm enters the global wide-area pursuit stage, which avoids getting trapped in local optima; when |E|<1.1, it enters the local precise pursuit stage, which optimizes the route details.

[0037] 3. Global Wide-Area Pursuit Phase. Two stochastic strategies, random reference pursuit and mean-guided pursuit, are employed to generate new path candidate solutions, expanding the search scope. Similar to the population information interaction mechanism in particle swarm optimization, this covers more potential optimal paths. Specifically, this includes: 1) Random Reference Pursuit: Where t is the current iteration number, Let i be the position after the t-th iteration. For individuals randomly selected in the t-th iteration, s1, s2 [0,1] represents uniformly distributed random numbers. The search diversity is enhanced by guiding the update of solutions through random reference individuals; 2) Mean-oriented pursuit: ;in, Let be the optimal solution in the t-th iteration. s3 and s4 are uniformly distributed random numbers in [0,1], high is the upper bound of the search space, and low is the lower bound. Combining the optimal solution with the population mean, the exploration direction and search efficiency are balanced, continuing the core guiding logic of the particle swarm algorithm, while taking into account the cooperative efficiency advantage of the wolf pack algorithm; 3) Strategy selection: Generate a random number q. If q≥0.45, choose random reference pursuit; otherwise, choose mean-oriented pursuit. The two strategies complement each other through probability triggering.

[0038] 4. Localized Precision Tracking Phase. Based on the tracking energy E and the random factor s [0,1], four local pursuit methods are adopted: flexible pursuit, rigid pursuit, diving flexible pursuit, and diving rigid pursuit, to optimize path details and adapt to insulator detection and safety constraints. Specifically, it includes: 1) Flexible pursuit, i.e., when 0.6≤|E|<1.1 and s≥0.55: Among them, K [0,2.2] represents the pursuit intensity coefficient. The search range is gradually reduced, and the path is fine-tuned to meet the insulator's perspective constraint. 2) Rigid pursuit, i.e., when |E| < 0.6 and s ≥ 0.55: ; Fix the optimal solution direction and quickly converge to a high-quality path, ensuring the distance is greater than the safe distance. 3) Dive-type flexible pursuit, i.e., when 0.6≤|E|<1.1 and s<0.55: , Compare x1, x2, The total cost is calculated by taking the minimum value of x as... Where LF(D) is the Lévy flight function, Given a 1×D dimensional random vector, each dimension takes values ​​uniformly distributed in [0,1]. The optimal solution neighborhood is explored through irregular perturbations to optimize insulator detection coverage. 4) Dive-style rigid pursuit, i.e., when |E| < 0.6 and s < 0.55: , Compare x1, x2, The total cost is calculated by taking the minimum value of x as... Adjusting the position based on the average position of the population avoids cross-path collisions between multiple machines and improves coordination efficiency.

[0039] 5. Adaptive Random Walk Anti-Premature Convergence Strategy. To avoid the algorithm getting stuck in local optima in the later stages of iteration, a dynamic perturbation is applied to the current optimal solution to compensate for the key drawback of premature convergence in traditional particle swarm optimization algorithms such as Particle Swarm Optimization, Wolf Pack, and Fruit Fly algorithms. Specifically, this includes: 1) Perturbation formula: ; where δ(t) = 0.8·(1-t / T) is the dynamic step size, which decreases with the number of iterations, with a large exploration range in the early stage and fine-tuning in the later stage; s5 [0,1] is a random number; 2) Termination judgment: If there is no better solution after 12 consecutive iterations, the step size δ(t) is halved; if δ(t) < 0.1m, the iteration is terminated to ensure that the algorithm balances convergence speed and optimization accuracy.

[0040] S3: Multi-machine collaborative path optimization and dynamic adjustment.

[0041] 1. Multi-drone task allocation: Based on the number of corner towers and the distribution of insulator inspection points, the principle of "prioritizing safety, allocating nearby, and taking into account balance" is adopted to allocate inspection tasks to m drones, ensuring that the number of towers inspected by each drone is ≤3, and the flight distance of a single drone is ≤ its maximum endurance radius.

[0042] 2. Path collision avoidance constraints: Add multi-aircraft spacing constraints to the path planning, generally set to ≥12m, to avoid collisions during flight.

[0043] 3. Dynamic replanning: If sudden obstacles such as temporary construction areas or strong airflow are encountered during the inspection, WHA local replanning is triggered to reduce the amount of computation, ensure the continuous execution of the inspection task, and improve the algorithm's adaptability to dynamic environments.

[0044] Example 1

[0045] Parameter settings. Specifically, these include: 1) Inspection scenario: A 220kV transmission line section in a mountainous area, containing 4 towers, including 3 corner towers. The terrain includes flat terrain (label 0), sparse forest (label 1), and steep slopes (label 2). 2) Drone parameters: Three DJI Mavic 3T drones are used, with a maximum flight speed of 15m / s, a flight time of 45 minutes, and support for real-time image transmission. 3) Constraint parameters: Conductor distance d safe1 ≥4m, tower safety distance d safe2 ≥3m, insulator inspection viewing angle deviation θ≤30°. 4) Algorithm parameters: WHA population size 32, maximum number of iterations 160, weight coefficients ω1=75, ω2=110, ω3=95, ω4=12, ω5=9.

[0046] Experimental Results. The generated paths and UAV takeoff points were all at latitude and longitude coordinates 102.338649244096 and 29.0872447320626, with an altitude of 1458.679m. For safety reasons, waypoint 1 below refers to the point above the middle of the tower, as shown in the table below. Table 1 shows the path of UAV 1. Table 2 shows the path of UAV 2. Table 3 shows the path of UAV 3. The total flight distance of UAV 1 was 2262.5 meters, the total flight distance of UAV 2 was 1924.6 meters, and the total flight distance of UAV 3 was 3062.9 meters.

[0047] Table 1

[0048] Table 2

[0049] Table 3

[0050] Because the slope between tower #2 and tower #3 is more than 35° steep, the task allocation process prioritizes safety and assigns drone 3 to inspect towers #3 and #4.

[0051] Path compliance verification: The planned paths of all three drones successfully avoided steep slopes and sparse forests. The distance between the drone and the power line was ≥5.1m and the distance between the drone and the tower was ≥5m, meeting the electrical safety constraints. The insulator detection viewing angle deviation was ≤8°, with no blind spots in coverage.

[0052] Shooting effect verification: The shooting effect of the three drones was good. The size of the insulator target was appropriate, the background was clean and simple, and there were no complex objects such as poles, wires, trees or so as to block it. The steel caps of the insulators were clear and without overlap, which met the requirements for insulator deterioration detection.

[0053] Table 4 is a performance comparison table between the WHA algorithm provided by this invention and existing algorithms.

[0054] Table 4

[0055] Experimental results show that the method of the present invention outperforms traditional swarm intelligence algorithms in terms of convergence speed, overall cost optimization, and flight efficiency, and is more suitable for the inspection needs of insulators on corner towers in mountainous areas.

[0056] In summary, this invention addresses the core pain points of insulator inspection at angle towers in mountainous areas, including complex terrain, numerous blind spots, difficulties in multi-machine collaboration, and premature convergence of algorithms. Through multi-constraint coupled modeling and improved optimization of existing algorithms, the WHA algorithm is derived. Combined with an innovative multi-machine collaborative mechanism, this invention enhances inspection safety, detection completeness, and execution efficiency. On one hand, by using voxel discretization terrain modeling and electrical safety distance constraints, obstacles such as steep slopes and dense forests in mountainous areas are accurately avoided, reducing the risk of drone collisions and solving the problem of poor terrain adaptability in traditional methods. On the other hand, a dedicated cost model is designed for the insulator inspection perspective requirements, combined with multi-machine, multi-angle collaborative inspection, improving the drone inspection effect. Simultaneously, the WHA algorithm improves convergence speed and effectively avoids premature convergence through piecewise chaotic initialization and adaptive random walk strategies. Furthermore, multi-machine "proximity allocation + load balancing" task allocation and ≥12m collision avoidance constraints improve drone inspection efficiency. This invention fully adapts to the efficient, safe, and accurate inspection needs of insulators on high-voltage transmission lines in mountainous areas, providing reliable technical support for the stable operation of power systems.

[0057] In summary, this invention belongs to the field of unmanned aerial vehicle (UAV) inspection technology in power systems, specifically involving a multi-UAV collaborative inspection path planning method for corner tower insulators in mountainous areas. The method includes: S1: Constructing a multi-constraint model for corner tower insulator inspection in mountainous areas: including a three-dimensional terrain obstacle model, a corner tower insulator detection task model, an electrical safety distance model, and a multi-objective comprehensive cost model; S2: The WHA algorithm: including initializing the population using piecewise chaotic mapping to improve the uniformity of solution distribution, calculating tracking energy to determine the transition between global wide-area pursuit and local precise pursuit stages, using various differentiated methods to optimize path details in the local precise pursuit stage, and introducing an adaptive random walk strategy to prevent premature convergence; S3: Multi-UAV collaborative path optimization and dynamic adjustment. This invention can achieve a balance between efficiency, safety, and detection integrity in inspection paths, solving the path optimization problem for multi-UAV collaborative execution of corner tower insulators in complex terrain.

[0058] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for multi-UAV collaborative inspection path planning of insulators on corner towers in mountainous areas, characterized in that, Includes the following steps: S1: Construct a multi-constraint model for the inspection of insulators at corner towers in mountainous areas: The multi-constraint model includes a three-dimensional terrain obstacle model for mountainous areas, a task model for the inspection of insulators at corner towers, an electrical safety distance model, and a multi-objective comprehensive cost model; S2: Solving the optimal inspection path for multiple machines based on the WHA algorithm: The WHA algorithm includes first using a segmented chaotic mapping to initialize the population to improve the uniformity of solution distribution, calculating the tracking energy to determine the transition between the global wide-area pursuit and the local precise pursuit stage, using a variety of differentiated methods to optimize path details in the local precise pursuit stage, and introducing an adaptive random walk strategy to prevent premature convergence of the algorithm. S3: Multi-machine collaborative path optimization and dynamic adjustment: Tasks are allocated based on the number of corner towers and the distribution of insulator detection points. Multi-machine spacing constraints are added to avoid collisions. Dynamic adjustments are made in case of emergencies, triggering local replanning to update path segments.

2. According to the method described in claim 1, the three-dimensional terrain obstacle model in the mountainous area in step S1 adopts the "voxel discretization" method, discretizing the inspection area into voxel units of fixed size and assigning them access level labels, where 0 = gentle terrain, 1 = sparse forest, 2 = steep slope, 3 = no passage, and the terrain obstacle cost function is J. terrain (i)=k·grade(i); where, k is the obstacle weight coefficient, and grade(i) is the access level label of the i-th voxel unit; preferably, sparse forest k=2.5, steep slope k=3.2, and no-passage k=+∞.

3. The method according to claim 1, characterized in that, In the corner tower insulator detection task model described in step S1, the corner tower coordinates T(x) are defined. T ,y T ,z T The coordinates of key detection points on the insulator string are used to constrain the angle between the UAV and the insulator normal to be less than or equal to a certain specific angle. The detection coverage cost function is J. cover =β× θ 2 β is the coverage penalty coefficient. θ is the angle deviation value.

4. The method according to claim 3, characterized in that, In the corner tower insulator detection task model described in step S1, the coordinates of the key detection points of the insulator string include the string vertex coordinates I1(x1,y1,z1) and the string end coordinates I2(x2,y2,z2), and the angle between the UAV and the insulator normal is constrained to be ≤30°, and β=1.

8.

5. The method according to claim 1, characterized in that, The electrical safety distance function of the electrical safety distance model in step S1 is J safety = γ · (d safe - d 3 ) ; wherein γ is a safety weight coefficient, d is an actual distance, d safe is a safety distance; if d ≥ d safe , then J safety = 0.

6. The method according to claim 1, characterized in that, The multi-objective integrated cost model mentioned in step S1 is J total =ω1J terrain +ω2J safety +ω3J cover +ω4J power +ω5J steer ω1, ω2, ω3, ω4, and ω5 are all weighting coefficients, J terrain J is the terrain obstacle cost function. safety J is the electrical safety distance function. cover To detect the coverage cost function, J power Let J be the energy consumption cost function. power =k·L, where k is the energy consumption factor, L is the path length, and J steer This is the turning cost function.

7. The method according to claim 6, characterized in that, In the multi-objective integrated cost model described in step S1, the weight coefficients are ω1=75, ω2=110, ω3=95, ω4=12, ω5=9; k=0.8, J steer The function is as follows: Where, θ i This is the drone's turning angle, with the maximum turning angle set to 60°.

8. The method according to claim 1, characterized in that, The formula for tracking energy in step S2 is as follows: E0 [-1.2, 1.2] represents the initial tracking energy. For the number of iterations, The maximum number of iterations is determined by |E| ≥ 1.1, at which point the algorithm enters the global wide-area pursuit stage; when |E| < 1.1, the algorithm enters the local precise pursuit stage.

9. The method according to claim 8, wherein in the local precision tracking stage of step S2, the tracking energy is E, and the random factor is s, s [0,1] Based on |E| and s, four methods are selected for local precision pursuit: flexible pursuit with 0.6≤|E|<1.1 and s≥0.55, rigid pursuit with |E|<0.6 and s≥0.55, diving flexible pursuit with 0.6≤|E|<1.1 and s<0.55, and diving rigid pursuit with |E|<0.6 and s<0.

55.

10. The method according to claim 1, characterized in that, The perturbation formula for the adaptive random walk strategy in step S2 is: ,in, Let be the optimal solution for the t-th iteration, δ(t) = 0.8·(1-t / T) be the dynamic step size, and s5 be the value of t. [0,1] is a random number; if no better solution is found in 12 consecutive iterations, the step size δ(t) is halved, and the iteration is terminated when δ(t) < 0.1m.

Images