Unmanned aerial vehicle path planning method and system based on RRT and expert genetic algorithm

Abstract

The invention relates to an unmanned aerial vehicle path planning method and system based on RRT and an expert genetic algorithm. The method comprises the following steps: carrying out unmanned aerial vehicle task allocation by using a particle swarm-genetic algorithm; simplifying a task allocation problem into a traveling salesman problem, and calculating path cost by using an optimal track solved by an optimal boundary value problem; unmanned aerial vehicle task allocation is performed by using a particle swarm-genetic algorithm, and an optimal solution is found through particle update iteration; solving a planned path of the unmanned aerial vehicle by using a genetic algorithm based on RRT and an expert strategy; selecting the individual with the optimal fitness and taking the path of the individual as the optimal path; performing hierarchical planning on the paths of the multiple unmanned aerial vehicles by using multi-unmanned aerial vehicle path conflict detection; and performing smoothing processing through a multi-section trajectory optimization method to obtain a final trajectory. According to the invention, the problem that the unmanned aerial vehicle path planning method based on the traditional genetic algorithm is difficult in population initialization in a complex environment is solved, and the convergence speed and the optimal solution solving capability of the unmanned aerial vehicle path planning algorithm are improved.

Description

Technical Field

[0001] The present invention relates to a UAV path planning technology, and in particular to a UAV path planning method and system based on RRT (Rapidly-Exploring Random Trees) and expert genetic algorithm. Background Art

[0002] The rise of the low-altitude economy is accelerating the innovation of drone technology, and its application potential in various industries has been greatly released. With its outstanding maneuverability and reliability, drones have emerged in many fields such as agriculture, forestry, and environmental monitoring. In particular, with the support of swarm intelligence technology, cluster operations have become possible, overcoming the limitations of the capabilities and resources of single drones and providing solutions for a wider range of mission scenarios. However, multi-machine collaboration faces multiple challenges, including how to efficiently allocate tasks in complex environments, avoid obstacles, and resolve path conflicts between multiple drones. These difficulties have made path planning a research hotspot, which aims to ensure that drone clusters can work together and complete their missions safely and without hindrance. The maximization of the efficiency of drone clusters depends on their collaboration and the mechanism of planning and following the best path to complete the task. By optimizing the path planning algorithm, drone clusters will be able to serve various civil and commercial needs more intelligently and efficiently.

[0003] At present, the path planning methods of drones can be divided into search-based path planning algorithms, sampling-based path planning algorithms, evolutionary algorithm-based path planning algorithms, and reinforcement learning-based path planning algorithms. The search-based method can ensure that the optimal path is searched, and the search efficiency is constantly improving with the improvement. However, the search space of this type of method in high-dimensional space, more complex and large maps, or multi-drone clusters will grow exponentially with the increase in number, resulting in an explosion in search time. The sampling-based method can ensure that the solution is found quickly in more dimensions and spaces, but this type of method is probabilistically complete, so if time is limited, it cannot guarantee the optimal solution. The algorithm based on reinforcement learning is still in its infancy. Due to the characteristics of its action space and state space, once the state space is large, a lot of computing resources are required, and it cannot be implemented in edge devices with limited computing resources. Further development is needed.

[0004] Unlike the other three methods, the evolution-based algorithm can quickly cross-evolve in high-dimensional space and converge quickly to ensure the optimality of the solution. The traditional genetic algorithm is an evolution-based algorithm. The UAV path planning method based on the traditional genetic algorithm first uses population initialization to select a feasible point for each row of the rasterized map, and then uses the connection function to make all nodes connected, and finally obtains a feasible population solution. However, in extremely complex environments, this type of method is prone to fall into local optimal solutions, and it may even be difficult to complete population initialization. Its existing problems and defects can be summarized as follows:

[0005] (1) For maps in extreme cases, the UAV path planning method based on traditional genetic algorithms cannot obtain the initialized population.

[0006] (2) As the number of initial populations increases, the initialization speed of the UAV path planning method based on traditional genetic algorithm is not efficient.

[0007] (3) The randomness of initialization will affect the convergence results of the algorithm, resulting in instability in the performance of the UAV path planning method based on the traditional genetic algorithm. Summary of the invention

[0008] Aiming at the inefficiency and instability of UAV path planning method based on traditional genetic algorithm, a UAV path planning method and system based on RRT and expert genetic algorithm was proposed.

[0009] The technical solution of the present invention is:

[0010] A UAV path planning method based on RRT and expert genetic algorithm includes the following steps:

[0011] Step S1: Use particle swarm-genetic algorithm to allocate UAV tasks; first simplify the task allocation problem to the traveling salesman problem TSP, and use the optimal trajectory solved by the optimal boundary value problem OBVP to calculate the path cost; then use the particle swarm-genetic algorithm PSO-GA to allocate UAV tasks, the algorithm is initialized as a group of random particles, each particle represents the task allocation list; then use the greedy algorithm to quickly initialize the population, the particles pass (current position X i , individual optimal value pbest, global optimal value gbest) three values ​​are selected for genetic and crossover update of individuals; finally, the optimal solution is found through particle update iteration, and the output is the mission target point of each single UAV;

[0012] Step S2: Use the genetic algorithm based on the rapid exploration random tree RRT and the expert strategy to solve the planned path of the UAV; for each single UAV obtained in step S1, first use the rapid exploration random tree RRT to assist in optimization to initialize the population, increase the corresponding point edge weights to complete the initialization, and output a population with multiple path weights; then use the genetic iterative algorithm based on the expert strategy to solve the fitness of the initialized population; after the iteration is completed, select the individual with the best fitness and take the path of the individual as the best path;

[0013] Step S3: Use multi-UAV path conflict detection to perform hierarchical planning on the paths of multiple UAVs; first, perform temporal decomposition on the optimal paths of all UAVs, and find out the conflict points by comparing the temporal paths of different UAVs; if there are conflicting paths, return to step S2 for new planning; finally, obtain the non-conflicting paths of all UAVs and smooth the trajectories through the multi-segment trajectory optimization method MSG to obtain the final trajectory.

[0014] Further, the method specifically includes the following steps:

[0015] Step S1: Use particle swarm-genetic algorithm to allocate UAV tasks; first, simplify the task allocation problem to the traveling salesman problem TSP, and use the optimal trajectory solved by the optimal boundary value problem OBVP to calculate the path cost; then use the particle swarm-genetic algorithm PSO-GA to allocate UAV tasks, and the algorithm is initialized as a group of random particles, each particle represents the task allocation list; then use the greedy algorithm to quickly initialize the population, and the particles pass (current position x i , individual optimal value pbest, global optimal value gbest) three values ​​are selected for genetic and crossover update of individuals; finally, the optimal solution is found through particle update iteration, and the output is the mission target point of each single UAV;

[0016] In step S1, the particle passes (current position X i , individual optimal value pbeat, global optimal value gbest) three values ​​select genetic and crossover update individuals, and the specific update formula is expressed as:

[0017]

[0018] in is the position of the particle in the previous generation, is the updated particle position, ω is the inertia factor, c 1 is the individual cognitive factor, c 2 is the group cognitive factor, α is the cross-genetic factor, is the reverse path of the previous generation of particles, gbest is the maximum fitness value found by this particle in the historical search space. (j-1) is the value of the maximum fitness in the public total population historical search space;

[0019] Among them, ω, c 1 、c 2 The three parameters are adaptively adjusted, specifically expressed as follows:

[0020]

[0021] Where (ω max ,c 1max ,c 2max ) represents the maximum value of the weight, (ω min ,c 1min ,c 2min ) indicates the minimum value of the weight; iter indicates the current number of iterations, and ITER_MAX indicates the set upper limit of the total iterations;

[0022] Step S2: Use the genetic algorithm based on the rapid exploration random tree RRT and the expert strategy to solve the planned path of the UAV; for each mission target point obtained by each single UAV in step S1, first use the rapid exploration random tree RRT to assist in optimization to initialize the population, increase the corresponding point edge weights to complete the initialization, and output a population of multiple path weights; then use the genetic iterative algorithm based on the expert strategy to solve the fitness of the initialized population; after the iteration is completed, select the individual with the best fitness and take the path of the individual as the best path;

[0023] In step S2, a random tree RRT-assisted optimization initialization population method is rapidly explored, and the specific steps include:

[0024] Step A1: Get a rasterized map, the RRT sampling tree is Tree;

[0025] Step A2: Randomly sample the map to obtain a sampling point s rand , find the point s in the tree closest to the sampling point near ; through s near and rand The line of the sampled fixed number of steps of the new point s new =Δs near +(1-Δ)s rand , where Δ is the weight;

[0026] Step A3: near and new Perform collision detection; if no obstacle is encountered, s new Add Tree; until the distance between the new point and the end point is less than the preset value γ;

[0027] Step A4: After obtaining the RRT initialization path, increase the corresponding point edge weight to complete the initialization;

[0028] In step S2, the fitness of the initialized population is solved by a genetic iterative algorithm based on an expert strategy, and the specific steps include:

[0029] Step B1: Obtain a path from a population; from the starting point s 0 At the beginning, for all subsequent action nodes j of node i, define their selection probability temp ij , by obtaining the temp ij An optimal path can be derived under this individual situation;

[0030] Among them, temp ij It is expressed as:

[0031] temp ij =h(j)+β index(i,j) +μ×drive(j) i,j∈MAP

[0032] h(j)=cost(i,j)+OBVP(j,end) i,j∈MAP

[0033] Where MAP is the mapping set of all points after the map is rasterized, cost(i,j) represents the actual path cost from node i to node j, OBVP(j,end) represents the predicted smooth trajectory cost from node j to the end point end; h(j) is the heuristic function value of a successor node j; β index(i,j) is the selection weight of the edge from node i to node j; drive(j) is the aircraft steering drive loss if node j is selected as the next node, and is 0 if the subsequent node has the same direction of motion as the current node; μ is the weight parameter; temp is obtained by ij An optimal path path_popu can be obtained under this individual condition;

[0034] Step B2: Calculate the degree of advantage fit of each individual in the overall population, obtain the population's fit set FIT, and select the genetic parent set PARENT through a competitive bidding method;

[0035] Among them, the fitness fit is expressed as:

[0036] fit=cost(i,j)+drive(j) i,j∈path_popu

[0037] The parent set PARENT is represented as:

[0038] PARENT = {argmin i(fit(i))} i∈rand(NP)in FIT

[0039] Where NP represents the population size;

[0040] Step B3: obtain a new population through crossover inheritance and mutation of the parents, and then guide the new population with expert strategies including optimal guidance and wolf pack algorithm guidance;

[0041] Among them, in step B3, the expert strategy guidance including the best guidance and the wolf pack algorithm guidance is performed on the new population, and the specific steps include:

[0042] Step a1: For the best fitness fit best The individual with the smallest value in the fit function is rewarded, which is expressed as:

[0043] β x =β x +bouns x∈path(index(fit best ))

[0044] where β x represents the probability weight of path x direction selection, bouns represents reward guidance, and path represents the path set;

[0045] Step a2: For the worst fitness fit worst The individual with the maximum value in the fit function is given expert guidance by the RRT algorithm, which is expressed as:

[0046] β y =β y +guidance y∈path(index(fit worst ))

[0047] where β y represents the probability weight of path y direction selection, guidance represents expert guidance based on the RRT algorithm, and path represents the path set;

[0048] Step a3: Use the wolf pack algorithm to use the top three individuals in terms of fitness to fit the worst fitness worst The individuals receive expert guidance, which is expressed as:

[0049]

[0050] where β z represents the probability weight of path z direction selection, β first , β second , β thirdIt represents the probability weight of the path direction selection of the top three individuals in fitness ranking, and path represents the path set;

[0051] Step B4: Finally, a new iterative population is obtained; after the iteration is completed, the individual with the best fitness fit, that is, the individual with the minimum value in the fit function, is selected to obtain the best path;

[0052] Step S3: Use multi-UAV path conflict detection to perform hierarchical planning on the paths of multiple UAVs; first, perform temporal decomposition on the optimal paths of all UAVs, and find out the conflict points by comparing the temporal paths of different UAVs; if there are conflicting paths, return to step S2 for new planning; finally, obtain the non-conflicting paths of all UAVs and smooth the trajectories through the multi-segment trajectory optimization method MSG to obtain the final trajectory.

[0053] A UAV path planning system based on RRT and expert genetic algorithm is used to implement the UAV path planning method based on RRT and expert genetic algorithm as described above, and the system includes the following modules:

[0054] Task allocation module: Taking the environment map as input, the task allocation problem is simplified into the traveling salesman problem through TSP; the optimal trajectory is solved and the path cost is calculated through OBVP; PSO-GA is used to allocate tasks to the drone cluster and obtain the task path points of the drone cluster;

[0055] Path planning module: After obtaining the task path points of each drone from the task allocation module, use RRT to assist in optimizing the initialization population, increase the corresponding point edge weights to complete the initialization, and output a population with multiple path weights; obtain a new population through cross-inheritance and mutation, and use the expert strategy including the best guidance and wolf pack algorithm guidance to solve the fitness of the initialized population; after the iteration is completed, select the individual with the best fitness and take the path of the individual as the best path;

[0056] Multi-UAV path conflict detection module: After obtaining the best path for each UAV from the path planning module, the best path is decomposed in time series, and the conflict points are found by comparing the time series paths of different UAVs; if there is a conflicting path, it returns to the path planning module for new planning; finally, the non-conflicting paths of all UAVs are obtained, and the trajectories are smoothed by the multi-segment trajectory optimization method MSG to obtain the final smooth path.

[0057] The beneficial effects of the present invention are:

[0058] (1) The present invention uses the path trajectory generated by the UAV dynamics as the path cost of the task allocation problem in task allocation, which can be more in line with the actual flight cost and energy consumption of the UAV. It can also be better coupled with the back-end path planning to reduce the calculation difficulty of subsequent path planning.

[0059] (2) In the path planning, the present invention adds a population initialization strategy based on the rapid exploration random tree RRT to the genetic algorithm, which solves the problem of the difficulty of population initialization of the evolutionary algorithm in a complex environment. The algorithm adds an expert strategy including the best guidance and the wolf pack algorithm guidance, which can better guide the direction of evolution and make the path planning algorithm converge faster and more accurately. The adaptive change factor is added to the cross evolution, which makes it easier for the algorithm to jump out of the local optimal solution, ensuring the optimality of the algorithm's solution.

[0060] (3) The present invention proposes multi-UAV path conflict detection to perform hierarchical path planning for UAVs, so that multiple UAVs will not have collision conflicts due to timing problems during operation, ensuring the feasibility of the solution to the required task. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a flow chart of a UAV path planning method based on RRT and expert genetic algorithm of the present invention;

[0062] Figure 2 This is a diagram of a UAV path planning system based on RRT and expert genetic algorithm of the present invention;

[0063] Figure 3 This is a flow chart of the method for rapidly exploring random tree RRT-assisted optimization of initialization population in the present invention;

[0064] Figure 4 A flow chart of solving the fitness of the initialized population using the genetic iterative algorithm based on the expert strategy of the present invention;

[0065] Figure 5 A flowchart of expert strategy guidance for a new population including optimal guidance and wolf pack algorithm guidance provided by the present invention;

[0066] Figure 6 This is a comparison chart of the simulation experiment with the upper limit of the population size and the number of iterations of the present invention being (10, 50);

[0067] Figure 7 This is a comparison chart of the simulation experiment with the upper limit of the population size and the number of iterations of the present invention being (10,200);

[0068] Figure 8 This is a comparison chart of the simulation experiments of the present invention with the upper limit of the population size and the number of iterations being (50, 200). DETAILED DESCRIPTION

[0069] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0070] Figure 1 This is a flow chart of a method for UAV path planning based on RRT and expert genetic algorithm of the present invention. This embodiment provides a method for using the present invention, which includes the following steps:

[0071] Step S1: Use particle swarm-genetic algorithm to allocate UAV tasks; first simplify the task allocation problem to the traveling salesman problem TSP, and use the optimal trajectory solved by the optimal boundary value problem OBVP to calculate the path cost; then use the particle swarm-genetic algorithm PSO-GA to allocate UAV tasks, the algorithm is initialized as a group of random particles, each particle represents the task allocation list; then use the greedy algorithm to quickly initialize the population, the particles pass (current position X i , individual optimal value pbest, global optimal value gbest) three values ​​are selected for genetic and crossover update of individuals; finally, the optimal solution is found through particle update iteration, and the output is the mission target point of each single UAV;

[0072] Step S2: Use the genetic algorithm based on the rapid exploration random tree RRT and the expert strategy to solve the planned path of the UAV; for each single UAV obtained in step S1, first use the rapid exploration random tree RRT to assist in optimization to initialize the population, increase the corresponding point edge weights to complete the initialization, and output a population with multiple path weights; then use the genetic iterative algorithm based on the expert strategy to solve the fitness of the initialized population; after the iteration is completed, select the individual with the best fitness and take the path of the individual as the best path;

[0073] Step S3: Use multi-UAV path conflict detection to perform hierarchical planning on the paths of multiple UAVs; first, perform temporal decomposition on the optimal paths of all UAVs, and find out the conflict points by comparing the temporal paths of different UAVs; if there are conflicting paths, return to step S2 for new planning; finally, obtain the non-conflicting paths of all UAVs and smooth the trajectories through the multi-segment trajectory optimization method MSG to obtain the final trajectory.

[0074] Figure 2 The present invention is a UAV path planning system diagram based on RRT and expert genetic algorithm. The system implements a UAV path planning method based on RRT and expert genetic algorithm, and the system includes:

[0075] Task assignment module. Taking the environment map as input, the task assignment problem is simplified to the traveling salesman problem through TSP. The optimal trajectory is solved and the path cost is calculated through OBVP. PSO-GA is used to assign tasks to the drone cluster and obtain the task path points of the drone cluster.

[0076] Path planning module. After obtaining the task path points of each drone from the task allocation module, use RRT to assist in optimizing the initialization population. Initialization can be completed by increasing the corresponding point edge weights, and the output is a population with multiple path weights. A new population is obtained through crossover inheritance and mutation, and the fitness of the initialized population is solved using an expert strategy including the best guidance and wolf pack algorithm guidance. After the iteration is completed, the individual with the best fitness is selected and the path of the individual is taken as the best path.

[0077] Multi-UAV path conflict detection module. After obtaining the best path for each UAV from the path planning module, the best path is decomposed in time series, and the conflict points are found by comparing the time series paths of different UAVs. If there is a conflicting path, it returns to the path planning module for new planning. Finally, the non-conflicting paths of all UAVs are obtained, and the trajectories are smoothed by the multi-segment trajectory optimization method MSG to obtain the final smooth path.

[0078] Figure 3 This is a flow chart of the method for rapidly exploring random tree RRT-assisted optimization of the initialization population of the present invention. The method is implemented in the process of step S2 of embodiment 1, and the method includes the following steps:

[0079] Step A1: Get a rasterized map, the RRT sampling tree is Tree;

[0080] Step A2: Randomly sample the map to obtain a sampling point s rand , find the point s in the tree closest to the sampling point near ; through s near and rand The line of the sampled fixed number of steps of the new point s new =Δs near +(1-Δ)s rand , where Δ is the weight;

[0081] Step A3: near and new Perform collision detection; if no obstacle is encountered, s new Add Tree; until the distance between the new point and the end point is less than the preset value γ;

[0082] Step A4: After obtaining the RRT initialization path, increase the corresponding point edge weight to complete the initialization.

[0083] Figure 4The flowchart of solving the fitness of the initialized population by the genetic iterative algorithm based on the expert strategy of the present invention is as follows. The method is implemented in the process of step S2 of embodiment 1, and the method comprises the following steps:

[0084] Step B1: Obtain a path from a population; from the starting point s 0 At the beginning, for all subsequent action nodes j of node i, define their selection probability temp ij , by obtaining the temp ij An optimal path can be derived under this individual situation;

[0085] Among them, temp ij It is expressed as:

[0086] temp ij =h(j)+β index(i,j) +μ×drive(j) i,j∈MAP

[0087] h(j)=cost(i,j)+OBVP(j,end) i,j∈MAP

[0088] Where MAP is the mapping set of all points after the map is rasterized, cost(i,j) represents the actual path cost from node i to node j, OBVP(j,end) represents the predicted smooth trajectory cost from node j to the end point end; h(j) is the heuristic function value of a successor node j; β index(i,j) is the selection weight of the edge from node i to node j; drive(j) is the aircraft steering drive loss if node j is selected as the next node, and is 0 if the subsequent node has the same direction of motion as the current node; μ is the weight parameter; temp is obtained by ij An optimal path path_popu can be obtained under this individual condition;

[0089] Step B2: Calculate the degree of advantage fit of each individual in the overall population, obtain the population's fit set FIT, and select the genetic parent set PARENT through a competitive bidding method;

[0090] Among them, the fitness fit is expressed as:

[0091] fit=cost(i,j)+drive(j) i,j∈path_popu

[0092] The parent set PARENT is represented as:

[0093] PARENT = {argmin i (fit(i))} i∈rand(NP)in FIT

[0094] Where NP represents the population size;

[0095] Step B3: obtain a new population through crossover inheritance and mutation of the parents, and then guide the new population with expert strategies including optimal guidance and wolf pack algorithm guidance;

[0096] Among them, in step B3, the expert strategy guidance including the best guidance and the wolf pack algorithm guidance is performed on the new population, and the specific steps include:

[0097] Step a1: For the best fitness fit best The individual with the smallest value in the fit function is rewarded, which is expressed as:

[0098] β x =β x +bouns x∈path(index(fit best ))

[0099] where β x represents the probability weight of path x direction selection, bouns represents reward guidance, and path represents the path set;

[0100] Step a2: For the worst fitness fit worst The individual with the maximum value in the fit function is given expert guidance by the RRT algorithm, which is expressed as:

[0101] β y =β y +guidance y∈path(index(fit worst ))

[0102] where β y represents the probability weight of path y direction selection, guidance represents expert guidance based on the RRT algorithm, and path represents the path set;

[0103] Step a3: Use the wolf pack algorithm to use the top three individuals in terms of fitness to fit the worst fitness worst The individuals receive expert guidance, which is expressed as:

[0104]

[0105] where β z represents the probability weight of path z direction selection, β first , β second , β third It represents the probability weight of the path direction selection of the top three individuals in fitness ranking, and path represents the path set;

[0106] Step B4: Finally, a new iterative population is obtained; after the iteration is completed, the individual with the best fitness fit, that is, the individual with the minimum value in the fit function, is selected to obtain the best path;

[0107] Step S3: Use multi-UAV path conflict detection to perform hierarchical planning on the paths of multiple UAVs; first, perform temporal decomposition on the optimal paths of all UAVs, and find out the conflict points by comparing the temporal paths of different UAVs; if there are conflicting paths, return to step S2 for new planning; finally, obtain the non-conflicting paths of all UAVs and smooth the trajectories through the multi-segment trajectory optimization method MSG to obtain the final trajectory.

[0108] Simulation experiment:

[0109] In the same simulation scenario, different algorithms are used for comparative experiments with different numbers of iterations and different population sizes. The upper limits of population size and number of iterations are (10,50), (10,200) and (50,200) respectively. The algorithms compared are genetic algorithm GA, particle swarm optimization algorithm PSO, immune genetic algorithm IGA, genetic algorithm GA-RRT based on RRT initialization, immune genetic algorithm IGA-RRT based on RRT initialization, and genetic algorithm Expert-GA based on expert strategy. The method proposed in this paper is called ERRT-GA.

[0110] The upper limit of population size and iteration number is (10,50)

[0111]

[0112]

[0113] The population size and number of iterations are capped at (10,200)

[0114]

[0115] The upper limit of population size and iteration number is (50,200)

[0116]

[0117] The comparison results are shown in the table above and Figure 6 , Figure 7 and Figure 8 As shown in the figure, it can be seen that the overall running time of the ERRT-GA task proposed by the present invention is greatly reduced, and when most of the comparison algorithms have converged, it can jump out of the local optimal solution and better find the global optimal solution. The ERRT-GA method proposed by the present invention has solved the global optimal solution in different population sizes and iteration times.

[0118] The above-mentioned embodiment only expresses one implementation mode of the present invention, and its description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A UAV path planning method based on RRT and expert genetic algorithm, It is characterized in that The following steps are involved: Step S1: Use particle swarm-genetic algorithm to allocate UAV tasks; first simplify the task allocation problem to the traveling salesman problem TSP, and use the optimal trajectory solved by the optimal boundary value problem OBVP to calculate the path cost; then use the particle swarm-genetic algorithm PSO-GA to allocate UAV tasks, the algorithm is initialized as a group of random particles, each particle represents the task allocation list; then use the greedy algorithm to quickly initialize the population, the particles pass (current position X i , individual optimal value pbest, global optimal value gbest) three values ​​are selected for genetic and crossover update of individuals; finally, the optimal solution is found through particle update iteration, and the output is the mission target point of each single UAV; Step S2: Use the genetic algorithm based on the rapid exploration random tree RRT and the expert strategy to solve the planned path of the UAV; for each single UAV obtained in step S1, first use the rapid exploration random tree RRT to assist in optimization to initialize the population, increase the corresponding point edge weights to complete the initialization, and output a population with multiple path weights; then use the genetic iterative algorithm based on the expert strategy to solve the fitness of the initialized population; after the iteration is completed, select the individual with the best fitness and take the path of the individual as the best path; Step S3: Use multi-UAV path conflict detection to perform hierarchical planning on the paths of multiple UAVs; first, perform temporal decomposition on the optimal paths of all UAVs, and find out the conflict points by comparing the temporal paths of different UAVs; If there is a conflicting path, return to step S2 for new planning; finally, the non-conflicting paths of all drones are obtained and the trajectories are smoothed by the multi-segment trajectory optimization method MSG to obtain the final trajectory.

2. A UAV path planning method based on RRT and expert genetic algorithm according to claim 1, It is characterized in that The specific steps include: Step S1: Use particle swarm-genetic algorithm to allocate UAV tasks; first, simplify the task allocation problem to the traveling salesman problem TSP, and use the optimal trajectory solved by the optimal boundary value problem OBVP to calculate the path cost; then use the particle swarm-genetic algorithm PSO-GA to allocate UAV tasks, and the algorithm is initialized as a group of random particles, each particle represents the task allocation list; then use the greedy algorithm to quickly initialize the population, and the particles pass (current position x i , individual optimal value pbest, global optimal value gbest) three values ​​are selected for genetic and crossover update of individuals; finally, the optimal solution is found through particle update iteration, and the output is the mission target point of each single UAV; In step S1, the particle passes (current position X i , individual optimal value pbest, global optimal value gbest) three values ​​select genetic and crossover update individuals, and the specific update formula is expressed as: in is the position of the particle in the previous generation, is the updated particle position, ω is the inertia factor, c 1 is the individual cognitive factor, c 2 is the group cognitive factor, α is the cross-genetic factor, is the reverse path of the previous generation of particles, gbest is the maximum fitness value found by this particle in the historical search space. (j-1) is the value of the maximum fitness in the public total population historical search space; Among them, ω, c 1 、c 2 The three parameters are adaptively adjusted, specifically expressed as follows: Where (ω max ,c 1max ,c 2max ) represents the maximum value of the weight, (ω min ,c 1min ,c 2min ) indicates the minimum value of the weight; iter indicates the current number of iterations, and ITER_MAX indicates the set upper limit of the total iterations; Step S2: Use the genetic algorithm based on the rapid exploration random tree RRT and the expert strategy to solve the planned path of the UAV; for each mission target point obtained by each single UAV in step S1, first use the rapid exploration random tree RRT to assist in optimization to initialize the population, increase the corresponding point edge weights to complete the initialization, and output a population of multiple path weights; then use the genetic iterative algorithm based on the expert strategy to solve the fitness of the initialized population; after the iteration is completed, select the individual with the best fitness and take the path of the individual as the best path; In step S2, a random tree RRT-assisted optimization initialization population method is rapidly explored, and the specific steps include: Step A1: Get a rasterized map, the RRT sampling tree is Tree; Step A2: Randomly sample the map to obtain a sampling point s rand , find the point s in the tree closest to the sampling point near ; through s near and rand The line samples a new point s with a fixed number of steps new =Δs near +(1-Δ)s rand , where Δ is the weight; Step A3: near and new Perform collision detection; if no obstacle is encountered, s new Add Tree; until the distance between the new point and the end point is less than the preset value γ; Step A4: After obtaining the RRT initialization path, increase the corresponding point edge weight to complete the initialization; In step S2, the fitness of the initialized population is solved by a genetic iterative algorithm based on an expert strategy, and the specific steps include: Step B1: Obtain a path from a population; from the starting point s 0 At the beginning, for all subsequent action nodes j of node i, define their selection probability tempi j , by obtaining the tempi j An optimal path can be derived under this individual situation; Among them, tempi j It is expressed as: temp ij =h(j)+β index(i，j) +μ×drive(j) i,j∈MAP h(j)=cost(i,j)+OBVP(j,end) i,j∈MAP Where MAP is the mapping set of all points after the map is rasterized, cost(i, j) represents the actual path cost from node i to node j, OBVP(j, end) represents the predicted smooth trajectory cost from node j to the end point end; h(j) is the heuristic function value of a successor node j; β index(i，j) is the selection weight of the edge from node i to node j; drive(j) is the aircraft steering drive loss if node j is selected as the next node, and is 0 if the subsequent node has the same direction of motion as the current node; μ is the weight parameter; temp is obtained by ij An optimal path path_popu can be obtained under this individual condition; Step B2: Calculate the degree of advantage fit of each individual in the overall population, obtain the population's fit set FIT, and select the genetic parent set PARENT through a bidding competition method; Among them, the fitness fit is expressed as: fit=cost(i,j)+drive(j) i,j∈path_popu The parent set PARENT is represented as: PARENT={argmin i (fit(i))} i∈rand(NP)in FIT Where NP represents the population size; Step B3: obtain a new population through crossover inheritance and mutation of the parents, and then guide the new population with expert strategies including optimal guidance and wolf pack algorithm guidance; Among them, in step B3, the expert strategy guidance including the best guidance and the wolf pack algorithm guidance is performed on the new population, and the specific steps include: Step a1: For the best fitness fit best The individual with the smallest value in the fit function is rewarded, which is expressed as: β x =β x +bouns x∈path(index(fit best )) where β x represents the probability weight of path x direction selection, bouns represents reward guidance, and path represents the path set; Step a2: For the worst fitness fit worst The individual with the maximum value in the fit function is given expert guidance by the RRT algorithm, which is expressed as: b y =b y +guidance y∈path(index(fit worst )) where β y represents the probability weight of path y direction selection, guidance represents expert guidance based on the RRT algorithm, and path represents the path set; Step a3: Use the wolf pack algorithm to use the top three individuals in terms of fitness to fit the worst fitness worst The individuals receive expert guidance, which is expressed as: where β z represents the probability weight of path z direction selection, β first , β second , β third It represents the probability weight of the path direction selection of the top three individuals in fitness ranking, and path represents the path set; Step B4: Finally, a new iterative population is obtained; after the iteration is completed, the individual with the best fitness fit, that is, the individual with the minimum value in the fit function, is selected to obtain the best path; Step S3: Use multi-UAV path conflict detection to perform hierarchical planning on the paths of multiple UAVs; first, perform temporal decomposition on the optimal paths of all UAVs, and find out the conflict points by comparing the temporal paths of different UAVs; if there are conflicting paths, return to step S2 for new planning; finally, obtain the non-conflicting paths of all UAVs and smooth the trajectories through the multi-segment trajectory optimization method MSG to obtain the final trajectory.

3. A UAV path planning system based on RRT and expert genetic algorithm, It is characterized in that Used to implement a UAV path planning method based on RRT and expert genetic algorithm as described in any one of claims 1 or 2, the system includes the following modules: Task allocation module: Taking the environment map as input, the task allocation problem is simplified into the traveling salesman problem through TSP; the optimal trajectory is solved and the path cost is calculated through OBVP; PSO-GA is used to allocate tasks to the drone cluster and obtain the task path points of the drone cluster; Path planning module: After obtaining the task path points of each drone from the task allocation module, use RRT to assist in optimizing the initialization population, increase the corresponding point edge weights to complete the initialization, and output a population with multiple path weights; obtain a new population through cross-inheritance and mutation, and use the expert strategy including the best guidance and wolf pack algorithm guidance to solve the fitness of the initialized population; after the iteration is completed, select the individual with the best fitness and take the path of the individual as the best path; Multi-UAV path conflict detection module: After obtaining the best path for each UAV from the path planning module, the best path is decomposed in time series, and the conflict points are found by comparing the time series paths of different UAVs; If there is a conflicting path, it will return to the path planning module for new planning; finally, the non-conflicting paths of all drones will be obtained, and the trajectory will be smoothed by the multi-segment trajectory optimization method MSG to obtain the final smooth path.

Images