A multi-unmanned aerial vehicle path planning method based on proxy model optimization

Abstract

The application discloses a multi-unmanned aerial vehicle path planning method based on agent model optimization, acquires a multi-unmanned aerial vehicle planning task; a path planning model optimization target function is established, the target function is a total length of all unmanned aerial vehicle paths; Monte Carlo sampling is carried out based on the target function, a group of decision vector values are randomly generated, and a target function value corresponding to each decision vector is calculated as a sample data point; the sample data point is fitted by using a Gaussian process to generate an agent model; the agent model meeting the requirement is used as an auxiliary target; the target function and the target function of the auxiliary target are iteratively evolved by using a genetic algorithm; and the unmanned aerial vehicle path planning is output. The application establishes a path planning model, and solves the model by using an efficient algorithm to acquire an effective path planning scheme.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of unmanned aerial vehicles, and particularly relates to a multi-unmanned aerial vehicle path planning method based on agent model optimization. BACKGROUND

[0002] Unmanned aerial vehicle cluster cooperative combat as a new combat mode will play an important role in future joint operations.

[0003] Generally, the unmanned aerial vehicle formation is composed of multiple types of unmanned aerial vehicles, which carry different loads and weapon equipment systems, and the platform performance is also different, so each unmanned aerial vehicle has different task execution capabilities when performing different types of tasks.

[0004] There are many threats in the process of task execution of the unmanned aerial vehicle, such as enemy radars in the battlefield environment, various anti-aircraft weapons, etc., once the unmanned aerial vehicle is discovered by the enemy or is destroyed, it will cause great loss, and due to the limitation of the terrain of task execution, the unmanned aerial vehicle needs to avoid mountains, bad weather, etc. when flying. SUMMARY

[0005] Therefore, the application provides a multi-unmanned aerial vehicle path planning method based on agent model optimization, which specifically comprises the following steps:

[0006] Obtaining a multi-unmanned aerial vehicle planning task;

[0007] Establishing a path planning model optimization objective function, the objective function is to minimize the total length of all unmanned aerial vehicle paths;

[0008] Based on the objective function, a group of decision vector values are randomly generated by Monte Carlo sampling, and the objective function value corresponding to each decision vector is calculated as a sample data point; 80% of the sample data points are randomly selected as a training set, a Gaussian process is used for fitting to generate an initial agent model; the remaining 20% of the sample data points are used as a verification set to test the initial agent model, if the agent model does not meet the requirements, the parameters of the agent model are modified or the agent model is regenerated until the requirements are met; the agent model meeting the requirements is used as an auxiliary target;

[0009] The objective function of the target function and the auxiliary target function is evolved by genetic algorithm. In the evolution process, in order to make the algorithm converge to the optimal region quickly in the evolution process, the mutation strategy adopts the mutation operator in the differential evolution algorithm, then the offspring is generated by selective mating, and the offspring population is evaluated by selective imitation. After merging the parent population and the offspring population, the factor cost of the merged population is sorted, the factor level is re-determined, the scalar fitness and skill factor of each individual in the merged population are updated, finally the individuals in the population are selected by the elite strategy to enter the next iteration until the evolution stopping condition is met.

[0010] Outputting the path planning of the unmanned aerial vehicle.

[0011] Further, the objective function of the path planning model optimization is as follows:

[0012]

[0013] There are m unmanned aerial vehicles to perform tasks, for unmanned aerial vehicle I i , the feasible path R i is represented as a set of k coordinate points R i ={p i，1 , p i，2 ,..., p i，k}, the distance between adjacent two coordinate points p i，a and p i，b is represented as D(p i，a , p i，b ), and p i，s represents the starting point of unmanned aerial vehicle I i .

[0014] Further, the radar, air defense equipment and several peaks are regarded as spherical no-fly zones, the spherical center is set as C(x c , y c ), and the radius is r, then the no-fly zone can be represented as: (x-x c ) 2 +(y-y c ) 2 ≤r 2 , the constraints include threat constraints and unmanned aerial vehicle performance constraints;

[0015] The threat constraints are as follows:

[0016] For a line segment whose endpoints are A(x a , y a ) and B(x b , y b ), the points on the line segment are represented as: X=(x a , y a )-ε(x a -xb , y a -y b ), 0≤ε≤1;

[0017] Point X0= (x0, y0)-ε0(x a -x b , y a -y b ) is the foot of the perpendicular from the center of the circle to the line on which the segment AB lies, and the segment AB is perpendicular to the segment X0C,

[0018]

[0019] If ε0≤0, it means that the foot of the perpendicular is on the extension of the segment BA, then the point closest to the center of the sphere is A; if ε0>1, it means that the foot of the perpendicular is on the extension of the segment AB, then the point closest to the center of the sphere is B; if 0<ε0<1, it means that the point closest to the center of the sphere is X0;

[0020] The performance constraints of the unmanned aerial vehicles are as follows:

[0021] First, the distance between each two path points must be greater than the minimum step length, which is expressed as follows:

[0022] D(p i，s , p i，1 )≥maxStep i

[0023] D(p i，s , p i，k )≥maxStep i

[0024] D(p i，j , p i，j-1 )≥maxStep i，j =2,..., k

[0025] Second, due to the limitation of the fuel on board, the unmanned aerial vehicles need to complete the task within the limited maximum flight distance, and the total distance of each unmanned aerial vehicle should not exceed the maximum flight distance, that is:

[0026]

[0027] Further, the chromosome coding and decoding method is as follows:

[0028] The coding mode of the single unmanned aerial vehicle path planning is a set of k path coordinate points, and the m unmanned aerial vehicles are a real number matrix R composed of m×k path coordinate points m×k Each row is a path of an unmanned aerial vehicle, and each element r ij in the matrix represents the path of the unmanned aerial vehicle I icoordinates of the jth point passed through by the UAV I

[0029] Since each UAV has multiple tasks to perform, taskList i denotes the UAV I i The task list to be performed, assuming that the number of tasks performed by a UAV is at most b, so a three-dimensional matrix Q b×m×k denotes a complete chromosome, each element q tij denotes the UAV I i coordinates of the jth point passed through by the UAV I

[0030] In order to improve the quality of the initial solution, each UAV will generate a set of coordinates in the same direction and monotonous according to the vector direction from the starting point to the ending point of the tth task when performing the tth task.

[0031] Further, for the crossover operator, first construct a set of path point arrays that ignore threat direct target points as a conditional optimal solution denoted as C best When a certain probability is met, the conditional optimal solution is selected as one of the parents for high-quality gene segment transmission, and the other parent is randomly selected from the population; otherwise, two individuals are randomly selected from the population as parents; the two parents are denoted as C 1 and C 2 In the crossover process, the high-quality genes of each parent are selected for inheritance to generate offspring C off .

[0032] Further, the mutation operator adopts the mutation strategy of differential evolution DE / best / 1 / and the smoothing strategy, and the mutation strategy of differential evolution DE / best / 1 / is as follows:

[0033] Two different individuals C 1 and C 2 are randomly selected from the population, and the conditional optimal solution C best constructed in the crossover process, then the generated offspring q off is as follows:

[0034] C off =C best +γ·(C 1 -C 2 )

[0035] Where γ is a scaling factor, defined between [0, 2];

[0036] The smoothing strategy is as follows:

[0037] Randomly select a mutation point, when the mutation point is only one, it is a single-point mutation mode, that is, a single-point mutation is performed on the information of a bit in 2 to k-1 bits of an individual for a task, and the specific process is as follows: a mutation point muLoc is randomly selected, 2 < muLoc < k-1, the mutation point is adjusted to the center point of the front and rear two points, the first variable and the last variable are not involved in mutation, and the smoothing formula is represented as:

[0038]

[0039] In order to accelerate the mutation process, the above muLoc is a group of numbers under a certain probability, which means that the smoothing formula is executed in a loop for each point involved in muLoc.

[0040] The beneficial effects of the present application are as follows:

[0041] The present application establishes a path planning model and obtains an effective path planning scheme by using an efficient algorithm. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The path planning method flowchart of the present application;

[0043] Figure 2 The path planning scheme diagram of the present application. DETAILED DESCRIPTION

[0044] The present application will be further described below in conjunction with the drawings, but the present application is not limited in any way by the drawings, and any transformation or replacement based on the teaching of the present application belongs to the protection scope of the present application.

[0045] The unmanned aerial vehicle distribution system architecture in the present application is built in a layered mode, which is also called a multi-layer architecture mode, and the layered mode meets the idea of "high cohesion and low coupling".

[0046] Based on the system design idea, the unmanned aerial vehicle distribution system is divided into three layers: a data layer, a business layer and a presentation layer, wherein the data layer includes data collection and storage, mainly obtains battlefield environment, individual positioning and demand information through sensors, GPS, individual communication modules and the like, and stores the collected data in a database after preprocessing; the business layer is an unmanned aerial vehicle task planning system, which includes a task planning model and mainly includes two sub-modules: a task allocation module and a path planning module, and the core solving algorithm is a multi-factor evolution algorithm based on auxiliary targets, wherein the task allocation module generates a task allocation scheme according to the demand obtained by the data layer, and the path planning module quickly plans the path of the unmanned aerial vehicle group distribution according to the task allocation scheme; the presentation layer is an interactive interface, which is operated by a decision maker with management authority, and can view relevant information and issue scheduling instructions.

[0047] As described above, UAV mission planning in the operational layer is a crucial component of the UAV delivery system. It primarily refers to the rational allocation of targets of different locations, values, and threat levels to UAVs of varying types, values, and combat capabilities throughout the entire operational mission, while meeting tactical and technical specifications, operational requirements, platform and weapon performance constraints, and UAV tactical usage conditions. This involves employing efficient mission planning algorithms to determine mission execution plans for the UAVs, maximizing overall operational effectiveness while minimizing costs. This invention studies multi-UAV mission planning and solves the problem based on auxiliary target technology.

[0048] A multi-UAV mission planning scenario is described using a triple {I, T, C}, with a map size of 100km * 100km. Where I = {I1, I2, ..., I...} m Let} be a set of drones, representing the number of drones performing missions on the battlefield. Each element I in the set... i (i = 1, 2, ..., m) includes drone I i Flight speed V i ,Location Longest range S i and drone type P i In this scenario, there are 3 types of drones, denoted as P = {1, 2, 3}. i ∈P, P i =1 represents drone I i As a type of unmanned aerial vehicle, P i =2 represents drone I i For type II unmanned aerial vehicles, P i =3 represents drone I i There are three types of UAVs, each with different payloads and start-up costs, as detailed in Table 1; T = {T1, T2, ... T} n Let T be the set of tasks, representing the number of tasks that need to be performed on the battlefield. Each element T in the set... j (j = 1, 2, ..., n) includes the position of each task. And the number of battery cells required for the task points Q j C represents the constraints in the battlefield.

[0049] Table 1 Characteristics of Heterogeneous UAVs

[0050]

[0051] Multi-UAV Path Planning Service Analysis

[0052] There are many threats in the process of unmanned aerial vehicle executing tasks, such as enemy radar in battlefield environment, various air weapons, etc., once the unmanned aerial vehicle is found by the enemy or is destroyed, it will cause huge loss, and due to the limitation of the terrain of task execution, the unmanned aerial vehicle needs to avoid mountains, bad weather, etc. Forbidden flight area when flying. The present application firstly models the threats in the flight scene, then establishes a multi-unmanned aerial vehicle path planning model, then designs auxiliary targets according to the model, and finally solves the model by genetic algorithm to solve the path planning of multi-unmanned aerial vehicle in the task allocation scheme.

[0053] Objective function setting

[0054] There are m unmanned aerial vehicles to execute tasks, for unmanned aerial vehicle I i , its feasible path R i is expressed as a set of k coordinate points R i ={p i，1 , p i，2 ,..., p i，k}, the distance between adjacent two coordinate points p i，a and p i，b is expressed as D(p i，a , p i，b ), and p i，s represents the starting point of unmanned aerial vehicle I i .

[0055] The path planning requires that the total path is the shortest under the condition of meeting various constraints. When multiple unmanned aerial vehicles execute tasks, the path of each unmanned aerial vehicle needs to be planned to ensure that all unmanned aerial vehicles reach their respective task points with the optimal path of each unmanned aerial vehicle. However, due to the constraint limitation, it is impossible for each path to be the shortest path, therefore, the optimization target of the present path planning model is to minimize the total length of all unmanned aerial vehicle paths.

[0056]

[0057] Constraint condition

[0058] The threats in the battlefield, such as radar, air weapons and some mountains, are regarded as spherical forbidden flight area, the sphere center is set as C(x c , y c ), and the radius is r, then the forbidden flight area can be expressed as: (x-x c ) 2 +(y-y c ) 2 ≤r 2 , under the combat environment, the constraint is expressed as follows.

[0059] (1) Threat constraint

[0060] To determine whether a point is in the forbidden region, it is only necessary to determine whether the distance from the point to the center of the sphere is greater than the radius of the sphere. For a point (x i , y i ), it is necessary to satisfy: (x i -x c ) 2 +(y i -y c ) 2 >r 2 .

[0061] To determine whether a line is in the sphere, it is only necessary to determine whether the distance from the line to the center of the sphere is greater than the radius of the sphere. The threat constraint is as follows:

[0062] For a line segment with endpoints A (x a , y a ) and B (x b , y b ), the points on the line segment are represented as: X = (x a , y a )-ε(x a -x b , y a -y b ), 0≤ε≤1;

[0063] Point X0 = (x a , y a )-ε0(x a -x b , y a -y b ) is the foot of the perpendicular from the center of the sphere to the line on which the line segment AB lies, and the line segment AB is perpendicular to the line segment X0C,

[0064]

[0065] If ε0≤0, it means that the foot of the perpendicular is on the extension of the line segment BA, and the closest point of the line segment to the center of the sphere is A; if ε0>1, it means that the foot of the perpendicular is on the extension of the line segment AB, and the closest point of the line segment to the center of the sphere is B; if 0<ε0<1, it means that the closest point of the line segment to the center of the sphere is X0.

[0066] It is determined whether the line segment intersects the forbidden region by determining whether the distance from the closest point of the line segment to the center of the sphere is greater than the radius of the sphere.

[0067] (2) Unmanned aerial vehicle performance constraints

[0068] Firstly, the UAV needs a distance to stabilize the influence of flight action in the process of flight, so as to enter the next action more safely, and this distance is the minimum step length of the UAV flight. Therefore, we need to ensure that the distance between each two path points is greater than the minimum step length when planning the path, so as to ensure that the UAV can fly safely according to the planned flight path. The formula is as follows:

[0069] D(p i，s , p i，1 ) ≥ maxStep i

[0070] D(p i，s , p i，k ) ≥ maxStep i

[0071] D(p i，j , p i，j-1 ) ≥ maxStep i，j = 2,..., k

[0072] Secondly, due to the limitation of on-board fuel, the UAV needs to complete the task within the limited maximum flight distance, and the total distance of each UAV flight should not exceed the maximum flight distance. That is:

[0073]

[0074] Usually, when multiple UAVs perform tasks, it is necessary to ensure that the UAVs do not collide with each other during flight. However, the environment established by the present application is a two-dimensional space, and the UAVs can adjust in height to avoid collision between UAVs, so the model does not consider the collision factor between UAVs.

[0075] When establishing the auxiliary objective of the continuity optimization problem, first, the optimization model is established according to the problem, the optimization objective of the problem is determined, and the original optimization objective function is constructed. Then, the proxy model related to the original optimization objective function is established as the second optimization problem. Specifically, first, Monte Carlo sampling is performed based on the original optimization objective function, a group of decision vector values are randomly generated, and the target function value corresponding to each decision vector is calculated as a sample data point; 80% of the samples are randomly selected from the sample data points as the training set, and the Gaussian process is used for fitting to generate the initial proxy model; the remaining 20% of the samples are used as the verification set to test the initial proxy model, if the proxy model does not meet the requirements, the parameters of the proxy model are modified or the proxy model is re-generated until the requirements are met; the proxy model that meets the requirements is used as the auxiliary objective.

[0076] The commonly used proxy model for optimization includes Gaussian process, polynomial regression, radial basis function, etc. Among them, Gaussian process is often selected to establish the proxy model due to its globality and smoothness. The reasons why the application selects Gaussian process to assist the design of the target are as follows:

[0077] (1) Gaussian process has been proved to be effective and theoretically reasonable, and is simpler and has fewer hyperparameters compared with other proxy models;

[0078] (2) The calculation complexity of Gaussian process is O (N it S 3d ), wherein N it is the iteration number, S is the sampling point number, and d is the variable number;

[0079] (3) In the optimization process, we only focus on the position of the global optimal solution, and Gaussian process can smooth the shape of the original function and ignore the position of the local optimal solution.

[0080] In mathematics, the Gaussian function is defined as follows:

[0081]

[0082] Wherein, d is the number of decision variables, a and b are parameters to be estimated, x i represents the i th variable, μ i and σ i are obtained by minimizing the error between the estimated value of the Gaussian function and the actual value of the sample. In order to fit the function y = f (x) by Gaussian process, x ∈ R d , the original function is sampled to obtain n sampling points x = (x 1 , x 2 ,..., x n ) ∈ R d and its corresponding observation value y = (y 1 , y 1 ,..., y n ), for any two sampling points x, x' ∈ R d , the correlation c between them is defined as:

[0083]

[0084] Wherein, 1 ≤ p i ≤ 2, which is used to measure the smoothness of the fitted function, θ i ≥ 0 represents the importance of x i to the fitted function f (x), and the hyperparameter value of the Gaussian function is obtained by minimizing the error function, and the error function is defined as follows:

[0085]

[0086] where C is an n x n matrix composed of c(x, x'), and I is an n x 1 unit vector.

[0087] The present application simplifies the original objective function based on the idea of a proxy model, obtains sample points through Monte Carlo simulation sampling, and constructs a function similar to the original function but simpler through Gaussian fitting, and uses the function as an auxiliary target.

[0088] The genetic algorithm is used to simultaneously iteratively optimize the original optimization objective function and the objective function of the proxy model, and specifically includes:

[0089] Initialize the population, calculate the factor cost and factor level, calculate the scalar fitness of each individual according to the factor level, and determine the skill factor of the individual.

[0090] In the evolution process, in order to make the algorithm quickly converge to the optimal region in the evolution process, the mutation strategy in the algorithm uses the mutation operator in the differential evolution algorithm, then generates offspring through selective mating, and then evaluates the offspring population through selective imitation. After merging the parent population and the offspring population, the factor cost of the merged population is sorted, the factor level is re-determined, the scalar fitness and skill factor of each individual in the merged population are updated, and finally the individuals in the population are selected through the elite strategy to enter the next iteration until the evolution stopping condition is met.

[0091] Chromosome encoding and decoding method

[0092] The encoding method of single unmanned aerial vehicle path planning is usually a set of k path coordinate points (including the starting point and the ending point), and m unmanned aerial vehicles are a real number matrix R composed of path coordinate points with a size of m x k m×k Each row represents the path of an unmanned aerial vehicle, and each element r ij in the matrix represents the coordinates of the jth point passed by unmanned aerial vehicle I i .

[0093] Since each unmanned aerial vehicle has multiple tasks to perform, taskList i represents the task list to be performed by unmanned aerial vehicle I i . Assuming that the number of tasks performed by a certain unmanned aerial vehicle is at most b, we use a three-dimensional matrix Q b×m×k to represent a complete chromosome, and each element q tij in the matrix Q represents the coordinates of the jth point passed by unmanned aerial vehicle I i when performing the i-th task. It is worth noting that the number of tasks performed by some unmanned aerial vehicles is less than b, so the path point set of these unmanned aerial vehicles in the task dimension that exceeds the number of their own tasks is a set of zero vectors.

[0094] To improve the quality of initial solution, each UAV will randomly generate a set of coordinate points with consistent and monotonic direction according to the vector direction from the start point to the end point of the task when performing the tth task.

[0095] Genetic operator

[0096] By sharing genetic information between different tasks and learning from high-quality solutions, population diversity can be increased, making it more likely for the solution to escape from local optimum. For the crossover operator, a set of path points array ignoring the threat direct target point is first constructed as the conditional optimal solution denoted as C best When a certain probability is met, the conditional optimal solution is selected as one of the parents for high-quality gene segment transmission, and the other parent is randomly selected from the population. Otherwise, two individuals are randomly selected from the population as parents. The two parents are denoted as C 1 and C 2 In the crossover process, the high-quality genes of each parent are selected for inheritance to generate offspring C off . The detailed genetic operator algorithm is shown in Table 2.

[0097] Table 2 Crossover operator algorithm

[0098]

[0099] For the mutation operator, the invention designs two mutation operators. One adopts the mutation strategy of differential evolution DE / best / 1 / 1. First, two different individuals C 1 and C 2 are randomly selected from the population, and the conditional optimal solution C best constructed in the crossover process, then the generated offspring q off is as follows:

[0100] C off = C best + γ·(C 1 -C 2 )

[0101] where γ is a scaling factor, generally defined between [0, 2], usually taken as 0.5, and also set to 0.5 in this experiment.

[0102] The other is a smoothing strategy that randomly selects a mutation point. When there is only one mutation point, it is a single-point mutation method. Specifically, the single-point mutation is performed on the coding information of a certain bit from 2 to k-1 of a certain individual for a task, and the specific process is as follows. A mutation point muLoc(2 < muLoc < k-1) is randomly selected, and the mutation point is adjusted to the center point of the two points before and after it. The first variable (start point) and the last variable (end point) are not involved in the mutation. The smoothing formula can be expressed as:

[0103]

[0104] In order to accelerate the mutation process, the muLoc is a set of numbers at a certain probability, which means that the smoothing formula is executed in a loop for each point involved in muLoc.

[0105] Path planning scheme

[0106] Suppose there are 10 UAVs in this scenario to perform tasks. The algorithm population size is set to 50, the iteration number is 100, the crossover probability is 0.9, and the mutation probability is 0.5. The path planning solution result is as follows Figure 2 .

[0107] Under this scheme, the total length of all UAV path flights is 14055.52km, and the flight distance of each UAV is [946.447432248167, 1697.05627484771, 1413.97457576965, 1587.88782947176, 1616.09034240499, 1296.81514524244, 1118.91857522694, 1275.61749752816, 1597.18343236666, 1505.52799856383]. It can be seen from Figure 2 that each UAV avoids obstacles and quickly completes the task within the flight range.

[0108] The beneficial effects of the present application are as follows:

[0109] The present application establishes a path planning model and obtains an effective path planning scheme by solving it through an efficient algorithm.

[0110] The word "preferred" is used herein as a term of art to denote use as an example, instance, or illustration. Any aspect or design described herein as "preferred" is not necessarily to be construed as being advantageous over other aspects or designs. Rather, the use of the word "preferred" is intended to present a concept in a particular manner. The term "or" as used in this application is intended to mean an inclusive "or" rather than an exclusive "or". That is, unless specified otherwise, or as is clear from the context, the phrase "X employs A or B" is intended to mean any of the natural inclusive permutations. That is, the phrase "X employs A or B" is satisfied by any of the following instances: X employs A; X employs B; or X employs both A and B.

[0111] Moreover, although the present disclosure has been illustrated and described with respect to one or more implementations, equivalent alterations and modifications will occur to others skilled in the art based on the foregoing description and accompanying drawings. The present disclosure includes all such modifications and alterations and is limited only by the scope of the following claims. In particular regard to the various functions performed by the above described components (e.g., elements, engines, modules, etc.), the terms used to describe such components are intended to correspond, where appropriate, to any component which performs the specified function (e.g., that is functionally equivalent), even though not structurally equivalent to the disclosed structure which performs the function in the herein illustrated exemplary implementations of the present disclosure. In addition, while a particular feature of the disclosure can have been disclosed with respect to only one of several implementations, such feature can be combined with one or other features of the other implementations as can be desired and advantageous for any given or particular application. Furthermore, to the extent that the terms "including", "includes", "having", "has", "contain", "contains", or variants thereof to be afforded similar meanings in the context of describing common ownership in the background portion of the detailed description are used, such terms are intended to be inclusive in a manner similar to the term "comprising" as comparable terms under the doctrine of equivalents in the field of patent law.

[0112] The various functional units in the embodiments of the present application can be integrated in one processing module, or each unit can exist physically, or a plurality of or more units can be integrated in one module. The integrated module can be realized in the form of hardware, or in the form of a software functional module. When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium. The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. The above-mentioned devices or systems can execute the storage method in the corresponding method embodiments.

[0113] In summary, the above embodiments are one embodiment of the present application, but the embodiments of the present application are not limited by the above embodiments, and any changes, modifications, substitutions, combinations and simplifications made without departing from the spirit and principle of the present application are equivalent replacement methods and are included in the protection scope of the present application.

Claims

1. A multi-unmanned aerial vehicle path planning method based on proxy model optimization, characterized in that, The method comprises the following steps: acquiring a multi-unmanned aerial vehicle planning task; establishing a path planning model optimization objective function, the objective function being to minimize the total length of all unmanned aerial vehicle paths; based on the objective function, performing Monte Carlo sampling to randomly generate a set of decision vector values, and calculating the objective function value corresponding to each decision vector as a sample data point; randomly selecting 80% of the sample data points as a training set, fitting using a Gaussian process to generate an initial surrogate model; the remaining 20% of the sample data points are used as a verification set to test the initial surrogate model; if the surrogate model does not meet the requirements, the parameters of the surrogate model are modified or the surrogate model is re-generated until the requirements are met; and the surrogate model that meets the requirements is used as an auxiliary objective; the objective function and the objective function of the auxiliary objective are iteratively evolved through a genetic algorithm; in the evolution process, in order to enable the algorithm to quickly converge to an optimal region in the evolution process, a mutation operator in a differential evolution algorithm is used as a mutation strategy, then offspring are generated through selective mating, the offspring population is evaluated through selective imitation, the parent population and the offspring population are combined, the factor rank is re-determined according to the factor cost ranking of the combined population, the scalar fitness and skill factor of each individual in the combined population are updated, and finally the individuals in the population are selected through an elite strategy to enter the next iteration until an evolution stop condition is met; outputting an unmanned aerial vehicle path plan.

2. The multi-UAV path planning method based on proxy model optimization according to claim 1, characterized in that, The path planning model optimization objective function is as follows: m drones perform a task, for drone I i , its feasible path R i , is represented as a set of k coordinate points R i = {p i，1 , p i，2 ,..., p i，k}, p i，s represents the starting point of drone I i , D(p i，s , p i，1 ) represents the distance between coordinate points p i，s and p i，1 , D(p i，j , p i，j-1 ) represents the distance between coordinate points p i，j and p i，j-1 , D(p i，s , p i，k ) represents the distance between coordinate points p i，s and p i，k .

3. The method of claim 2, wherein, The radar, air defense equipment and several mountains are regarded as a spherical no-fly zone, the sphere center is set as C(x c , y c ), and the radius is r, x is the UAV coordinate, and then the no-fly zone can be expressed as: (x-x c ) 2 +(y-y c ) 2 ≤r 2 , the constraints include threat constraints and UAV performance constraints; The threat constraint is as follows: For a line segment whose endpoints are A(x a , y a ) and B(x b , y b ), a point on the line segment is represented as: X = (x a , y a ) - ε(x a - x b , y a - y b ), 0 ≤ ε ≤ 1; Point X0= (x a , y a )-e0(x a -x b , y a -y b ) is the foot of the perpendicular from the center to the line on which the segment AB lies, the segment AB being perpendicular to the segment X0C, If ε0≤0, it indicates that the perpendicular point is on the extension line of the line segment BA, and the point closest to the sphere center is A; if ε0≥1, it indicates that the perpendicular point is on the extension line of the line segment AB, and the point closest to the sphere center is B; if 0 The unmanned aerial vehicle performance constraint is as follows: First, the distance between each two path points must be greater than the minimum step length, which is expressed as follows: D(p i，s , p i，1 ) ≥ maxStep i D(p i，s , p i，k ) ≥ maxStep i D(p i，j , p i，j-1 ) ≥ maxStep i , j = 2,..., k maxStep i is the minimum step size; Second, due to the limitation of the on-board fuel, the unmanned aerial vehicle needs to complete the task within a limited maximum flight distance, and the total distance of each unmanned aerial vehicle flight should not exceed the maximum flight distance, that is:

4. The method of claim 3, wherein, The chromosome encoding and decoding method is as follows: The encoding mode of single-UAV path planning is a set of k path coordinate points, and the m UAVs are a real matrix R composed of m x k path coordinate points m×k Each row is a path of a UAV, and each element r ij in the matrix represents the j i th coordinate of the point passed by the i th UAV Since each drone has multiple tasks to perform, taskList i represents drone I i the list of tasks to be performed, assuming that the number of tasks performed by a drone is at most b, so a three-dimensional matrix Q b×m×k represents a complete chromosome, each element q tij in the matrix Q represents the coordinates of the jth point passed through by drone I i when performing the tth task; In order to improve the quality of the initial solution, a set of coordinate points with consistent and monotonic directions are randomly generated for each unmanned aerial vehicle according to the vector direction from the starting point to the ending point of the tth task.

5. The method of claim 4, wherein, For the crossover operator, a set of path points which ignore the threat to the direct target point is constructed as the condition optimal solution, denoted as C best When a certain probability is satisfied, the condition optimal solution is selected as one of the parents for the transmission of high-quality gene segments, and the other parent is randomly selected from the population; otherwise, two individuals are randomly selected from the population as parents; the two parents are denoted as C 1 and C 2 In the crossover process, the high-quality genes of each parent are selected for inheritance to generate offspring C off .

6. The method of claim 5, wherein, The mutation operator adopts the mutation strategy and smoothing strategy of differential evolution DE / best / 1 / 1, the mutation strategy of differential evolution DE / best / 1 / 1 being as follows: C is randomly selected from the population 1 and C 2 and the conditionally optimal solution C constructed during the crossover process best The resulting offspring q off is generated as follows: C off = C best + γ · (C 1 - C 2 ) Wherein, γ is a scaling factor, which is defined between 0 and 2; The smoothing strategy is as follows: A mutation point is randomly selected, and when there is only one mutation point, it is a single-point mutation method, that is, the single-point mutation is performed on the coding information of a certain bit in positions 2 to k-1 of a certain individual of a certain task, and the specific process is as follows: a mutation point muLoc is randomly selected, 2 To accelerate the mutation process, the above muLoc is a set of numbers at a certain probability, which means that the smoothing formula is executed in a loop for each point involved in muLoc.

Images