Unmanned aerial vehicle flight path planning method based on global optimization method

Abstract

The invention provides an unmanned aerial vehicle flight path planning method based on a global optimization method, and the method comprises the steps: S1, building a three-dimensional space environment model of a construction site, and setting the coordinates of a starting point and a target point of an unmanned aerial vehicle; s2, constructing a multi-dimensional physical constraint model and a track cost function; s3, obtaining an optimal global path of the unmanned aerial vehicle in the known obstacle environment through a global optimization method in combination with the three-dimensional space environment model of the construction site and the track cost function; s4, when the unmanned aerial vehicle encounters an unknown dynamic obstacle in the flight process, in combination with an RRT * algorithm and a dynamic obstacle detection method, triggering local path re-planning, generating an obstacle avoidance path in real time, fusing the obstacle avoidance path with the optimal global path, and generating a final flight path; s5, smoothing the final track by adopting a B spline curve, and eliminating sharp corners and sharp turns; according to the invention, autonomous track generation and dynamic obstacle avoidance of the unmanned aerial vehicle in a complex construction environment are realized, and the continuity of the path is ensured.

Description

Technical Field

[0001] This application relates to the field of UAV trajectory planning technology, specifically to a UAV trajectory planning method based on global optimization. Background Technology

[0002] With the acceleration of urban construction, the demand for on-site safety monitoring at construction sites is increasing, and drones are being used more and more for inspection and monitoring. The efficiency and quality of drone operations at construction sites are closely related to their flight paths, so flight path planning has become one of the key research directions.

[0003] However, construction site environments are complex, with numerous known obstacles such as building frames, pipelines, and tower cranes, as well as various dynamic unknown obstacles such as construction machinery, stockpiled materials, personnel, and vehicles. These obstacles make the trajectory planning problem extremely challenging. Using only global planning algorithms often ignores sudden obstacles, resulting in slow planning speeds or difficulty in obtaining the optimal route. On the other hand, using only local obstacle avoidance algorithms may lead to getting stuck in local optima and failing to react in time. Therefore, this application aims to solve the above-mentioned technical problems by implementing global obstacle avoidance planning in a known environment and quickly switching to local obstacle avoidance strategies when encountering unknown obstacles, thereby ensuring the flight safety of UAVs.

[0004] After a detailed search, no relevant technical solutions were found. Therefore, a detailed technical solution is required. Summary of the Invention

[0005] This application provides a UAV trajectory planning method based on a global optimization approach, comprising the following steps: S1: Establish a three-dimensional spatial environment model of the construction site and set the starting point coordinates and target point coordinates of the UAV; S2: Based on the flight characteristics of the UAV, construct a multi-dimensional physical constraint model, and at the same time, construct a trajectory cost function; S3: By introducing a global optimization method that integrates the Wild Dog optimization algorithm and the Symbiotic Biology Search algorithm, and combining it with the 3D spatial environment model of the construction site and the trajectory cost function, the optimal global path of the UAV in a known obstacle environment is obtained. S4: When the UAV encounters an unknown dynamic obstacle during flight, it combines the RRT* algorithm with the dynamic obstacle detection method to trigger local path replanning, generate an obstacle avoidance path in real time and merge it with the optimal global path to generate the final trajectory. S5: The final trajectory is smoothed using B-spline curves to eliminate sharp angles and abrupt turns.

[0006] As a preferred embodiment, the construction site environmental parameters collected in S1 are used to establish a three-dimensional spatial environment model of the construction site. The construction site environmental parameters include the spatial location and size information of known obstacles and the dynamic characteristics of unknown obstacles.

[0007] As a preferred embodiment, the multi-dimensional physical constraint model constructed in S2 includes setting the minimum turning radius, maximum pitch angle, flight altitude limit, and maximum range.

[0008] As a preferred option, the formula for the trajectory cost function is: Among them, J cost J represents the trajectory cost function. length J represents the cost of track length. height J represents the cost of track altitude. turn w1 represents the cost of the maximum turning angle, w2 represents the weight of the track length cost function, w3 represents the weight of the track altitude cost function, and w4 represents the weight of the maximum turning angle cost function. i i=1, 2, 3.

[0009] As a preferred option, the formula for calculating the track length cost is: in, x i This represents the x-axis coordinate of the i-th track node in the three-dimensional coordinate system; y i This represents the y-axis coordinate of the i-th track node in the three-dimensional coordinate system; z i This represents the z-axis coordinate of the i-th track node in the three-dimensional coordinate system; n is the total number of track nodes.

[0010] As a preferred option, the formula for calculating the altitude cost of a flight path is: , where p i Let p be the coordinates of the i-th track node.

[0011] As a preferred option, the formula for calculating the maximum turning cost is: Where δ is the maximum steering angle, γ i For a certain flight turn angle, a i Let be the vector of the i-th segment of the track.

[0012] As a preferred embodiment, the global optimization method in S3 follows this process: B1: Initialize parameters and randomly generate the initial population; B2: Fitness value calculation; B3: Determine if the random number P is less than 0.5. If so, use the group attack-mutual benefit mechanism to update the current position of the individual; otherwise, use the sweeping behavior-symbiosis mechanism to update the current position of the individual. P is a uniformly distributed random number in the interval (0, 1). B4: Determine if the individual survival rate is less than 0.3. If yes, continue to B5; otherwise, proceed to B6. B5: Employs a survival-parasitism mechanism to update individual location; B6: Update the optimal fitness value; B7: Determine whether all individuals have been traversed. If yes, execute B8; otherwise, jump to B2. B8: Determine if the iteration has ended. If so, output the optimal global path and end the process; otherwise, jump to B2.

[0013] As a preferred approach, the final trajectory is smoothed using a cubic B-spline curve.

[0014] As a preferred embodiment, the flow of the RRT* algorithm in S4 is as follows: A1: Initialize random tree T; A2: During the sampling process, the RRT* algorithm either directly selects the target point or randomly samples a point in space; A3: After sampling, the RRT* algorithm searches for the distance X during the tree search process. rand The nearest node X nearest It expands towards the sampling point direction with a fixed step size Δx, generating a new node X. new ; A4: Determine the new node X new If a collision occurs with an unknown dynamic obstacle, return to A2 if yes, otherwise proceed to A5; A5: The RRT* algorithm searches for the optimal parent node within a specific radius; where the specific radius is 1.5 to 3 times the step size. A6: Add the optimal parent node to the random tree T; A7: Determine if a path that meets the requirements has been found; otherwise, return to A2; otherwise, execute A8. A8: Determine if the algorithm's termination condition is met; if so, terminate the process; otherwise, execute A9. A9: Update the starting node of the path planning; A10: Update obstacle environment. Based on real-time information or environmental changes, correct or re-record the position, shape, and status information of obstacles in the current scene. A11: Replan and return to A2 to execute again.

[0015] As a preferred embodiment, the formula for A2 is: in: X rand P represents the coordinates of a random node obtained from sampling, where ξ is a uniformly random number. goal X represents the target bias sampling probability parameter. goal Represents the directly selected target point, used for guidance, X free Represents points randomly sampled in space, used for exploration; RandomSample ( X free This indicates the operation of randomly sampling feasible coordinate points within a free space region.

[0016] As a preferred embodiment, the new node X in A3 new The calculation formula is: Among them, X new Represents a new node, X nearest Δx represents the nearest node, and Δx represents a fixed step size.

[0017] As a preferred option, the formula for selecting the optimal parent node in A5 is: X parent = argmin {Cos t ( X near )+ d ( X near , X new )} X parent Cos represents the optimal parent node. t ( X ) represents the cumulative cost from the starting point to node X. d Represents Euclidean distance; X near Indicates that it is located at X new Candidate nearest neighbor nodes within the neighborhood are typically any node in the set of nodes that satisfy the distance constraint within the existing set of nodes in the tree. These nodes are used as candidate parent nodes, and the nodes in the candidate nearest neighbor set are traversed and compared when selecting the optimal parent node.

[0018] As a preferred option, the formula for a B-spline curve is: Where P(u) is a B-spline curve, d i (i = 0,1,2,...,n) represent control points, Ni ,k (u) represents the K-order normalized B-spline basis function defined by the following recursive formula; n represents the highest index of the control point sequence of the B-spline curve, used to characterize the range of the number of control points, i.e., the total number of control points is n+1, and the set of control points is { d 0 , d 1 ,…, d n}; Summation symbol This represents the weighted summation of all control points and their corresponding basis functions, thus obtaining the point coordinates of the B-spline curve P(u); ; Where i represents the index of the basis function, k represents the order of the B-spline, and u represents the curve parameter. i represents the i-th node in the node vector; define represents a mathematical convention used to handle indeterminate forms that may appear in recursive formulas.

[0019] This application, based on a fusion strategy of global planning and local obstacle avoidance, enables UAVs to generate autonomous flight paths and dynamically avoid obstacles in complex construction environments. It solves the problems of insufficient path optimization capabilities and lack of effective avoidance strategies when encountering unknown obstacles in existing methods in construction site environments. This invention uses B-spline curves for flight path smoothing, ensuring the continuity of the path and flight stability, enabling UAVs to perform construction site safety monitoring tasks efficiently and safely. This application has strong engineering practical value and promotion potential. Attached Figure Description

[0020] Figure 1 This is the logic block diagram for this application; Figure 2 This is a schematic diagram of the pitch angle θ constraint in this application; Figure 3 This is the B-spline curve of this application; Figure 4 A flowchart of the global optimization method; Figure 5 The convergence curve for the two-dimensional experiment; Figure 6 This is a comparison chart of fitness values ​​in two-dimensional experiments; Figure 7 The convergence curve for the three-dimensional experiment; Figure 8 A comparison chart of fitness values ​​in three-dimensional experiments; Figure 9 This is the flowchart of the RRT* algorithm. Detailed Implementation

[0021] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings; it should be noted that the specific embodiments described herein are only for illustration and explanation of the present invention and are not intended to limit the present invention. Example 1

[0022] This embodiment provides a UAV trajectory planning method based on a global optimization approach, more specifically, a UAV trajectory planning method for construction sites based on a global optimization approach. It aims to solve the problems of low efficiency in UAV path planning and poor obstacle avoidance capabilities in complex construction environments, and to achieve high-precision, highly robust safety monitoring tasks. The method includes the following steps: S1: Before the mission begins, a 3D spatial environment model of the construction site is established, and the starting point coordinates and target point coordinates of the UAV are set to provide a data foundation for subsequent trajectory planning. Specifically, environmental parameters of the construction site are collected to establish the 3D spatial environment model. These parameters include the spatial location and size information of known obstacles and the dynamic characteristics of unknown obstacles. The 3D spatial environment model of the construction site is implemented using a programming language. More specifically, a simplified modeling strategy is adopted. Known obstacles such as buildings, tower cranes, and construction equipment are represented as regular bodies, such as cylinders, through geometric simplification modeling. Unknown obstacles such as transport vehicles, construction personnel, or temporarily piled materials are modeled as spheres. In this way, a 3D spatial environment model of a complex construction site can be quickly constructed. By constructing a complete 3D spatial environment model, basic data support is provided for subsequent trajectory planning and simulation.

[0023] S2: Based on the flight characteristics of the UAV, parameters such as minimum turning radius, maximum pitch angle, flight altitude limit, and maximum range are set to construct a multi-dimensional physical constraint model. This ensures that the generated trajectory conforms to the dynamic constraints of the aircraft, guaranteeing not only mathematical feasibility but also safe execution by the UAV's flight control system. Simultaneously, based on the principle of minimizing trajectory parameters, a multi-parameter trajectory cost function is constructed, considering factors such as trajectory length cost, altitude cost, and maximum turning angle cost. This provides quantitative evaluation indicators for optimizing path planning, ensuring obstacle avoidance capabilities while also considering energy economy and mission completion efficiency. The trajectory cost function is constructed using a programming language.

[0024] S3: A global optimization method combining the Wild Dog Optimization Algorithm (DOA) and the Symbiotic Organism Search Algorithm (SOS) is introduced to perform global trajectory planning on the 3D spatial environment model, generating an initial global path that avoids known obstacles; then, by combining the environment model and the cost function, the optimal global path of the UAV in the known obstacle environment is obtained; the global trajectory planning is implemented through a programming language, and the global optimization method is used for simulation to obtain the globally optimal path.

[0025] The global optimization method in S3 combines the swarm attack, sweeping behavior, and survival rate mechanism of the Wild Dog Optimization Algorithm with the mutualistic, symbiotic, and parasitic phases of the Symbiotic Biology Search Algorithm, respectively, to achieve synergistic optimization of the two algorithms. That is, it introduces three mechanisms: swarm attack-mutualistic, sweeping-symbiotic, and survival rate-parasitic, to improve search efficiency and convergence speed and obtain the optimal global path (optimal track). Compared with the traditional single algorithm, the global optimization method can find feasible paths faster and avoid getting trapped in local optima, thereby generating an overall optimized global track in complex construction site environments.

[0026] S4: When the UAV flies along the optimal global path and encounters unknown obstacles in real time, such as temporarily entering engineering vehicles or moving construction personnel, it triggers local path replanning by combining the RRT* algorithm and dynamic obstacle detection method. It generates an obstacle avoidance path in real time to achieve real-time obstacle avoidance and merges it with the optimal global path. Through a smooth transition, it is seamlessly connected to the original global trajectory to generate the final trajectory, ensuring that the UAV can safely continue to execute the remaining flight distance. This can guarantee the safety of the UAV and the continuity of the mission, and prevent the risk of collision.

[0027] S5: The final flight path is smoothed by using cubic B-spline curves to eliminate sharp corners and abrupt turns, making the curvature of each segment of the flight path continuous and changing smoothly to meet the requirements of the UAV's turning radius and attitude changes, thus obtaining a smooth flight path suitable for the actual flight of the UAV; improving the continuity and stability of the flight path, making the final flight path more consistent with the actual flight characteristics of the UAV; enabling the UAV to perform construction site safety monitoring tasks efficiently and safely.

[0028] The formula for a B-spline curve is: Where P(u) is a B-spline curve, d i (i = 0,1,2,...,n) represent control points, N i ,k (u) represents the K-order normalized B-spline basis function defined by the following recursive formula; n represents the highest index of the control point sequence of the B-spline curve, used to characterize the range of the number of control points, i.e., the total number of control points is n+1, and the set of control points is { d 0 , d 1 ,…, d n}; Summation symbol This represents the weighted summation of all control points and their corresponding basis functions, thus obtaining the point coordinates of the B-spline curve P(u); Where i represents the index of the basis function, k represents the order of the B-spline, and u represents the curve parameter. irepresents the i-th node in the node vector; define represents a mathematical convention used to handle indeterminate forms that may appear in recursive formulas.

[0029] This step corrects the inflection points in the final trajectory, making the path curvature continuous and the turns smooth. The smoothed and optimized trajectory can effectively reduce the frequency of attitude adjustments during flight, reduce the burden on the control system, and improve the stability and energy efficiency of the flight process. Example 2

[0030] This embodiment provides a specific technical method for a trajectory cost function, specifically:

[0031] The formula for the trajectory cost function is: Among them, J cost J represents the trajectory cost function. length J represents the cost of track length. height Representing a high cost, J turn Let w1 represent the cost of the maximum turning angle, w2 represent the weight of the trajectory length cost function, and w3 represent the weight of the altitude cost function. Simultaneously, it must satisfy the following conditions: and , i=1,2,3.

[0032] Specifically, the track length cost directly affects mission efficiency. The smaller the track length cost function, the shorter the path, the lower the energy consumption, and the higher the completion rate. The distance cost (track length cost) of the flight phase is shown in the formula. The distance cost is the sum of the distances between each segment of the track nodes, where n is the total number of track nodes.

[0033] The formula for calculating the cost of track length is: Specifically, to avoid collisions and improve energy efficiency, drones need to maintain a reasonable flight altitude; in construction site flight path optimization, altitude configuration is particularly critical, and altitude costs are introduced to enable drones to generate suitable routes. in, x i This represents the x-axis coordinate of the i-th track node in the three-dimensional coordinate system; y i This represents the y-axis coordinate of the i-th track node in the three-dimensional coordinate system; z i This represents the z-axis coordinate of the i-th track node in the three-dimensional coordinate system; n is the total number of track nodes.

[0034] The formula for calculating altitude cost is: , where p i Let p be the coordinates of the i-th track node.

[0035] Specifically, to avoid sharp path turns and excessively steep bends, the formula for calculating the maximum turning cost is as follows: Where δ is the maximum steering angle, γ i For a certain flight turn angle, a i Let be the vector of the i-th segment of the track. Example 3

[0036] This embodiment describes the process of the RRT* algorithm in S4: The RRT* algorithm is a sampling-based path planning algorithm, which is an improved version of RRT (Rapid Exploratory Random Tree). Its core advantage is that it can efficiently search for near-optimal paths in complex environments. A1: Initialize random tree T; A2: During the sampling process, the RRT* algorithm can directly select the target point with a certain probability or randomly sample a point in space; the formula for the random sampling rule is as follows: in: X rand Represents the coordinates of the random nodes obtained from sampling, used for node expansion in the RRT* tree; ξ It is a uniformly distributed random number, specifically a random number that is uniformly distributed within the interval (0,1); P goal This represents the target bias sampling probability parameter; X goal This represents the directly selected target point, used for guidance. X free These represent points randomly sampled in space for exploration, i.e., unobstructed free space areas; RandomSample ( X free This indicates the operation of randomly sampling feasible coordinate points within a free space region; In this step, to improve the algorithm's convergence efficiency, a target bias sampling strategy is introduced when... ξ < P goal At that time, X rand Set as target point coordinates X goal This guides the search process towards the target direction; otherwise, it operates in unobstructed free space. X free Random sampling was performed inside to obtain X rand = RandomSample ( X free ).

[0037] A3: After sampling, the RRT* algorithm searches for the distance X during the tree search process. rand The nearest node X nearest It expands towards the sampling point direction with a fixed step size Δx, generating a new node X. new ; New node X new The calculation formula is: X new Represents a new node, X nearest Δx represents the nearest node, and Δx represents a fixed step size.

[0038] A4: Determine the new node X new If a collision occurs with an unknown dynamic obstacle, return to A2 if yes, otherwise proceed to A5; A5: Generate a new node X new Then, the RRT* algorithm searches for the optimal parent node within a specific radius; where the specific radius is 1.5 to 3 times the step size; the key to the RRT* algorithm is to find the optimal parent node for the new node X. new The formula for finding the optimal parent node is as follows: X parent = argmin {Cos t ( X near )+ d ( X near , X new )} X parent Cos represents the optimal parent node. t ( X ) represents the cumulative cost from the starting point to node X. d Represents Euclidean distance; X near Indicates that it is located at X new Candidate nearest neighbor nodes within the neighborhood are typically any node in the set of nodes that satisfy the distance constraint within the existing set of nodes in the tree. These nodes are used as candidate parent nodes, and the nodes in the candidate nearest neighbor set are traversed and compared when selecting the optimal parent node.

[0039] Compared to the RRT algorithm, the RRT* algorithm has a range-based reconnection strategy. This rewiring mechanism ensures that the path from the starting point to each node is as short as possible, thereby continuously optimizing the overall path quality.

[0040] A6: Add the optimal parent node to the random tree T; A7: Determine if a path that meets the requirements has been found; otherwise, return to A2; otherwise, execute A8. A8: Determine if the algorithm's termination conditions are met, such as the path length being sufficient or the number of iterations being enough. If yes, end the process; otherwise, execute A9. A9: Update the starting node of the path planning; A10: Update obstacle environment. Based on real-time information or environmental changes, correct or re-record the position, shape, and status information of obstacles in the current scene. A11: Replan and return to A2 to execute again. Example 4

[0041] This embodiment describes the global optimization method, which will be discussed in conjunction with the following. Figure 4 The logic of the global optimization method is discussed in detail. The core logic of this global optimization algorithm is to explore and develop in parallel in the solution space by simulating the co-evolution and competition mechanism of biological populations in order to find the optimal flight path of the UAV. Its process can be clearly divided into the following stages:

[0042] B1: Initialize parameters and randomly generate the initial population; The algorithm first sets control parameters, such as population size, maximum number of iterations, and weight coefficients of the cost function. Then, within the 3D environment model of the construction site, a set of initial "wild dogs" individuals is randomly generated, with each "wild dog" individual representing a complete trajectory from the starting point to the end point. This step aims to initialize a solution set with sufficient diversity to lay the foundation for the subsequent evolutionary process and ensure that the algorithm can extensively explore the entire solution space.

[0043] B2: Fitness value calculation; Based on the constructed cost function, the fitness value of the trajectory represented by each individual in the population is calculated. This step is a quantitative evaluation of the quality of the solution. The fitness value directly reflects the quality of the trajectory. For example, the lower the cost, the better the fitness. This is the fundamental basis for driving the entire algorithm to evolve towards the optimal solution.

[0044] B3: Determine if the random number P is less than 0.5. If so, use the group attack-mutual benefit mechanism to update the current position of the individual; otherwise, use the sweep behavior-symbiosis mechanism to update the current position of the individual. P represents the random number parameter used for strategy selection. P is a uniformly distributed random number in the interval (0, 1). Generate a random number in the range [0, 1] and select different positions to update the strategy based on its value; this is a branch decision point. This mechanism aims to dynamically balance the algorithm's exploration and development capabilities, using 0.5 as a boundary to guide the population to perform global search and local optimization with similar probabilities, avoiding premature entrapment in local optima.

[0045] If P < 0.5, the individual position is updated using a group attack-mutualism mechanism. Entering this branch, the mechanism combines the group attack behavior of the Wild Dog algorithm and the mutualism phase of the Symbiotic Biology algorithm. This step focuses on global exploration. It simulates group cooperative hunting, guides individuals to learn towards historical optimal solutions, and explores new areas through interaction with random individuals, aiming to discover more potential solution spaces.

[0046] The formula for calculating the group attack-mutual benefit mechanism is: ; ; ; V new = V r1 + rand 0,1 * (V best - Mutual_vector.BF); Among them, V new This indicates that the individual's position has been updated. and β1 represents the individual positions of two randomly generated wild dogs; Uniformly generated random numbers within the interval, β1 is a scaling factor that can change the size and direction of the wild dog's trajectory; na represents [2, SearchAgents The random number generated by the reciprocal of [2]; V best φk(t) represents the current optimal solution; φk(t) represents the subset of wild dogs that will attack, where φk(t) is the optimal solution. X, where X represents the number of randomly generated stray dogs; Represents the position of the current iteration of the search; Represents the best search proxy obtained in the previous iteration; Mutual_vector represents and The mutually beneficial relationship between them; BF represents the benefit factor, BF∈{1, 2}; rand 0,1 ∈{0,1}.

[0047] If P ≥ 0.5, the individual position is updated using a sweeping behavior-symbiosis mechanism. This mechanism combines the sweeping (random foraging) behavior of wild dogs with the symbiotic phase of the symbiotic biological algorithm. This step focuses on local development, which encourages individuals to conduct fine searches in their current location and learn from the information of other random individuals in the population. The aim is to deeply optimize existing good solutions and improve the quality of the solutions.

[0048] The calculation formula for cleaning behavior-symbiosis mechanism is: ; ; ; in, , Two candidate search vectors are generated for the cleaning action phase, where β2 represents... The generated random number, σ represents a randomly generated binary number, σ ∈ {0,1}; r 1. r 2 represents a random number from 1 to the wild dog population size. Representing the r The location of a search agent for a stray dog; Representing the r The location of the search agent for the two stray dogs; Represents the position of the current iteration of the search; .

[0049] B4: Determine if the individual survival rate is less than 0.3. If yes, continue to B5; otherwise, proceed to B6. After the above update, the survival rate of each individual is calculated. If the survival rate is lower than the threshold of 0.3, the individual is determined to be a weak individual. This step is the selection mechanism of survival of the fittest in the population. It is used to identify and mark individuals with poor fitness, in order to prepare for population renewal and diversification. B5: Employ a survival-parasitism mechanism to update individual location; specifically: adopt a parasitic strategy for individuals with low survival rates to update their location; For the marked inferior individuals, a parasitic strategy is used to replace them with a newly generated or strongly mutated new individual. This step is key to maintaining population diversity and escaping local optima. By introducing new genetic material, it effectively prevents the population from converging prematurely and increases the possibility of the algorithm exploring unknown regions.

[0050] The formula for calculating survival rate—parasitic mechanism—is as follows: ; in, This represents a search agent with a low survival rate that is about to be updated; r1 and r2 represent two different random numbers, r 1, r2∈ (1, SearchAgents), and r1≠ r2; (t) (t) represents the r1 and r2th search agents; σ represents the best search agent obtained in the previous iteration; σ represents a randomly generated binary number, σ∈ {0,1}.

[0051] B6: Update the optimal fitness value; After updating all individuals, the entire population is examined, and the optimal fitness value found in the current iteration and its corresponding trajectory are recorded. This step ensures that the algorithm does not lose the best solution found during the evolution process, thus guaranteeing convergence.

[0052] B7: Determine whether all individuals have been traversed. If yes, execute B8; otherwise, jump to B2. Check whether the above update operations from B3 to B6 have been completed for every individual in the population in the current iteration. This step is the termination check of the inner loop, ensuring that the entire population has a fair chance to evolve in a complete iteration, thereby coordinating the search process.

[0053] B8: Determine if the iteration has ended. If yes, output the globally optimal path and end the process; otherwise, jump to B2. Checking whether the maximum number of iterations has been reached, or whether the optimal solution has not improved significantly for several consecutive generations, is the termination check of the outer loop. It determines the timing of the end of the entire optimization process, achieving a balance between computational resources and solution quality.

[0054] When the algorithm terminates, it outputs the globally optimal individual recorded throughout the evolution process, i.e., the planned optimal trajectory. This step is the final output of the algorithm, which delivers a safe trajectory that has the lowest overall cost and is most suitable for UAV flight under all constraints.

[0055] This embodiment iterates until convergence, and finally outputs the global optimal solution. This algorithm successfully integrates the advantages of two metaheuristic algorithms and solves the global trajectory planning problem of UAVs in complex environments efficiently and robustly through the built-in balancing mechanism.

[0056] In summary, by adopting the above technical solutions, this invention solves the problems of insufficient path optimization capability and lack of effective avoidance strategies when encountering unknown obstacles in existing methods in construction site environments. Based on a fusion strategy of global planning and local obstacle avoidance, this application can achieve autonomous trajectory planning, dynamic obstacle avoidance, and safe and stable flight of UAVs in complex construction site environments, taking into account both global optimization and local obstacle avoidance, and ultimately generating a smooth and feasible flight path to accurately complete safety inspection tasks. Figures 5 to 8 Experimental results show that this method can significantly shorten the flight path, improve the obstacle avoidance success rate and mission completion efficiency compared with traditional planning algorithms. It provides a practical technical solution for intelligent monitoring and management of construction sites and has broad engineering practical value, engineering application prospects and promotion value.

[0057] Figure 5 — Figure 8In this context, the HIDOA-SOS algorithm represents a global optimization method that integrates the Wild Dog Optimization Algorithm and the Symbiotic Search Algorithm; the DOA algorithm represents the Wild Dog Optimization Algorithm; the SOS algorithm represents the Symbiotic Search Algorithm; and the SSA algorithm represents the Sparrow Search Algorithm.

[0058] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0059] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims. All of these forms are within the protection scope of this application.

Claims

1. A method for unmanned aerial vehicle path planning based on global optimization method, characterized in that, It comprises the following steps: S1: a three-dimensional space environment model of a construction site is established, and a starting point coordinate and a target point coordinate of the unmanned aerial vehicle are set; S2: a multi-dimensional physical constraint model is constructed according to the flight characteristics of the unmanned aerial vehicle; and a flight path cost function is constructed; S3: a global optimization method of introducing a wild dog optimization algorithm and a symbiotic organism search algorithm is combined with the three-dimensional space environment model of the construction site and the flight path cost function to obtain an optimal global path of the unmanned aerial vehicle in a known obstacle environment; The flow of the global optimization method is as follows: B1: parameters are initialized and an initial population is randomly generated; B2: fitness value calculation; B3: whether the random number P is less than 0.5 is judged, yes, the current position of the individual is updated by using the group attack-mutual benefit mechanism, otherwise the current position of the individual is updated by using the cleaning behavior-symbiotic mechanism; P is a random number uniformly distributed in the interval (0, 1); B4: whether the individual survival rate is less than 0.3 is judged, yes, B5 is continuously executed, otherwise, jump to B6; B5: the individual position is updated by using the survival rate-parasitic mechanism; B6: the optimal fitness value is updated; B7: whether all individuals are traversed is judged, yes, B8 is executed, otherwise, jump to B2; B8: whether the iteration is ended is judged, yes, the optimal global path is outputted and the flow is ended; otherwise, jump to B2; S4: when the unmanned aerial vehicle encounters an unknown dynamic obstacle in the flight process, a local path re-planning is triggered by combining the RRT* algorithm and the dynamic obstacle detection method, an obstacle avoidance path is generated in real time and fused with the optimal global path to generate a final flight path; S5: the final flight path is smoothed by using a B-spline curve to eliminate sharp corners and sharp turns.

2. The method of claim 1, wherein, In the multi-dimensional physical constraint model constructed in S2, the minimum turning radius, the maximum pitch angle, the navigation height limit and the maximum distance of the flight path are set.

3. The method of claim 1, wherein, The formula of the flight path cost function is: where J cost represents a track cost function, J length represents a track length cost, J height represents a track height cost, J turn represents a maximum turn cost, w1 represents a weight value of the track length cost function, w2 represents a weight value of the track height cost function, w3 represents a weight value of the maximum turn cost function, w i where i = 1, 2, 3.

4. The method of claim 3, wherein, The calculation formula of the flight path length cost is: ； wherein, x i represents the x-axis coordinate value of the i-th track node in the three-dimensional coordinate system; y i represents the y-axis coordinate value of the i-th track node in the three-dimensional coordinate system; z i represents the z-axis coordinate value of the i-th track node in the three-dimensional coordinate system; n is the total number of track nodes.

5. The method of claim 3, wherein the method is characterized by, The calculation formula of the flight path height cost is: , where p i is the coordinate of the ith track node p.

6. The method of claim 3, wherein the method is based on a global optimization method. The calculation formula of the maximum turning angle cost is: ； where δ is the maximum turning angle, γ i is the flight angle of a segment, a i is the vector of the i-th segment in the flight path.

7. The method of claim 1, wherein, In S5, the formula of the B-spline curve is: ； where P(u) is a B-spline curve, d i (i = 0,1,2,..., n) represent control points, N i ,k (u) represent K-order normalized B-spline basis functions defined by the following recursive formula; n represents the highest index of the control point sequence of the B-spline curve, used to represent the control point number range, i.e., the total number of control points is n+1; ； where i represents the index of the basis function, k represents the order of the B-spline, u represents the curve parameter, u i represents the i-th node in the node vector; define represents a mathematical convention used to handle indeterminate forms that can arise in recursive formulas.

8. The method of claim 1, wherein, The flow of the RRT* algorithm in S4 is as follows: A1: random tree T initialization; A2: in the sampling process, the RRT* algorithm directly selects a target point or randomly samples a point in the space; A3: After the sampling ends, the RRT* algorithm finds the distance X rand the nearest node X nearest and expands in the direction of the sampling point with a fixed step size Δx to generate a new node X new ; A4: Determine if new node X collides with unknown dynamic obstacle new If the unknown dynamic obstacle is not collided with, the process returns to A2, otherwise, A5 is executed. A5: the RRT* algorithm searches the neighborhood nodes within a certain radius to find the optimal parent node; wherein the certain radius is 1.5 times to 3 times of the step length; A6: the optimal parent node is added to the random tree T; A7: whether a path meeting the requirements has been found is judged, no, return to A2, yes, execute A8; A8: whether the termination condition of the algorithm is met is judged, yes, the flow is ended; otherwise, execute A9; A9: the starting node of the path planning is updated; A10: the obstacle environment is updated, the position, shape and state information of the obstacles in the current scene are corrected or re-recorded according to real-time information or environmental changes; A11: re-planning, return to A2 and execute again.

9. The method of claim 8, wherein, The formula of A2 is: ； wherein: X rand represents the randomly sampled node coordinates, ξ is a uniform random number, P goal represents the target bias sampling probability parameter, X goal represents the directly selected target point, X free represents the randomly sampled point in space, RandomSample ( X free ) represents the operation of randomly sampling a feasible coordinate point within the free space region; The new node X in the A3 new The calculation formula is: ; X new represents a new node, X nearest represents the nearest node, Δx represents a fixed step length; The selection formula of the optimal parent node in A5 is: X parent = argmin {Cos t ( X near )+ d ( X near , X new )} X parent represents the optimal parent node, Cos t X represents the cumulative cost from the origin to node X, d represents the Euclidean distance; X near denotes the set of candidate near neighbors located within X new neighborhood, any node in the set of nodes that already exist in the tree that satisfy the distance constraint, used as candidate parent nodes, each node in the set of candidate near neighbors is traversed and compared in the selection of the optimal parent node.​

Images