Unmanned aerial vehicle formation online route planning method and device based on rolling optimization strategy

Abstract

The invention provides an unmanned aerial vehicle formation online route planning method and device based on a rolling optimization strategy. The method comprises the following steps: 1, establishing and initializing an environment model and an unmanned aerial vehicle performance model; 2, constructing a cost function f of local route planning optimization of route performance indexes; 3, determining a global planning starting point, a final target point and a global optimization route; 4, determining a local planning starting point according to the global planning starting point; 5, updating the environment threat situation; 6, using a local target point selection strategy to determine a local planning target point; 7, taking the cost function f as an optimized target function, and performing local route planning by applying a variation particle swarm optimization algorithm to obtain a local route; 8, taking a point where the aircraft flies for only one step distance Ds along the local route as a local planning step length arrival point (x3, y3); and 9, judging whether the local planning step length arrival point (x3, y3) reaches the planned final target point or not, and if so, ending the unmanned aerial vehicle formation online route planning.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle (UAV) route planning, and in particular relates to an online route planning method and device for an unmanned aerial vehicle (UAV) formation using a rolling optimization strategy. Background Art

[0002] As one of the key technologies in the field of UAVs, online route planning for UAV formations mainly includes two parts: offline planning and online planning.

[0003] Offline planning does not need to consider unknown sudden threats. It is mainly based on pre-set tasks and known environmental information. It comprehensively considers the safety and coordination constraints when the UAV formation flies to perform global route planning and achieve the best or better specified performance indicators. Its focus is on the optimization of the desired route. Online planning is based on real-time sensor information. It is carried out in real time during the flight of the UAV according to the offline planned route. The purpose is to respond to the continuously incoming obstacle information, perform local route re-planning and make corresponding reasonable obstacle avoidance behaviors. Its focus is to quickly and reliably generate feasible routes.

[0004] There are often many uncertainties in the process of UAV formation operations and mission execution, which are manifested in the suddenness of threats / obstacles, uncertainty in location and quantity, temporary changes in missions, etc. To achieve online route planning for UAV formations in complex environments, it is necessary not only to consider the optimal solution, but also the planning speed. It is necessary to balance the two indicators of route cost and route planning time in the process of avoiding sudden threats in order to ensure flight safety to the greatest extent.

[0005] However, the existing online route planning of UAV formations has problems such as difficulty in ensuring real-time performance, low solution efficiency, and poor flexibility. Therefore, it is necessary to adopt an online route planning method for UAV formations with a rolling optimization strategy to quickly adjust the route plan according to changes in battlefield situation and mission objectives, so as to truly adapt to the needs of complex combat missions. Summary of the invention

[0006] The purpose of the present invention is to provide a method and device for online route planning of a UAV formation with a rolling optimization strategy, which can solve the problems of difficulty in ensuring real-time performance, low solution efficiency and poor flexibility in existing online route planning technologies for UAV formations.

[0007] First invention, the present application provides a method for online route planning of a UAV formation with a rolling optimization strategy, the method comprising:

[0008] Step 1: Establish and initialize the environment model and the UAV performance model. The parameter information of the environment model includes: environment boundary information, threat area, obstacle area and no-fly zone; the formation-related parameters of the UAV performance model include: UAV speed, UAV formation formation and UAV formation formation width;

[0009] Step 2: Based on the parameter information of the environmental model and the formation-related parameters of the UAV performance model, a cost function f for optimizing the local route planning based on the route performance index is constructed;

[0010] Step 3: Determine the global planning starting point, final target point and global optimization route;

[0011] Step 4: Determine the local planning starting point based on the global planning starting point;

[0012] Step 5: Update the environmental threat situation based on the environmental model, the local planning starting point, and the situation measurement values ​​within the single-step planning window period;

[0013] Step 6: Determine the local planning target point using the local target point selection strategy;

[0014] Step 7: Based on the local planning starting point and the local planning target point, the cost function f is used as the optimization objective function, and the mutation particle swarm optimization algorithm is used to perform local route planning to obtain the local route;

[0015] Step 8: The aircraft only flies a step distance D along the local route. s The point reached is used as the local planning step to reach the point (x3, y3);

[0016] Step 9: According to the preset judgment conditions, determine whether the local planning step arrival point (x3, y3) reaches the planned final target point. If it reaches the planned final target point, the online route planning of the UAV formation ends, otherwise return to step 4.

[0017] Furthermore, the cost function in step 2 is: f = L(1 + β1f obs +β2f θ ),

[0018] Where L is the route Total length, expanded waypoints Generated by the i-th particle in the particle swarm, containing m extended waypoints,

[0019] Provide routes for all threat areas, obstacle areas, and no-fly zones in the environment The total threat caused by the environment; where κ is the sum of the number of threat areas, obstacle areas, and no-fly zones in the environment, For the The distance from the center of the circumscribed circle of the obstacle to the jth route segment D of the ith route i (j) distance, is the radius of the circumscribed circle, r s is the obstacle avoidance safety distance, r d is the formation width;

[0020] For route The sum of the angle constraint cost values ​​of all adjacent segments, θ j For route The jth sub-segment D i (j) and the j+1th sub-segment D i (j) angle, θ ref It is the reference value of the minimum angle between adjacent segments of UAV formation flight;

[0021] β1 is the obstacle avoidance constraint weight, and β2 is the angle constraint weight between adjacent flight segments.

[0022] Furthermore, step 4 includes:

[0023] The first local planning starting point is determined according to the global planning starting point (x1, y1) of the aircraft, and the subsequent local planning starting points are the step arrival points of the previous round of planning.

[0024] Furthermore, the local target point selection strategy in step 6 includes a heuristic target point selection strategy and an optimal route target point selection strategy.

[0025] Furthermore, the heuristic target point selection strategy is:

[0026] After each rolling window information update, the heuristic function f(P)=g(P)+h(P) is used to select the local target point, where P is the point on the rolling window, g(P) is the cost of the drone flying from the current position to the target position, and h(P) is the cost of flying from P to the end point.

[0027] Furthermore,

[0028] Furthermore, the optimal route target point selection strategy is to use the intersection of each step rolling window radius and the global optimization route as the local target point.

[0029] Further, step 7 includes the following steps:

[0030] 701, set a particle swarm consisting of n particles, initialize the position and velocity of each particle in the particle swarm, take the i-th particle as an example (1≤i≤n), its position can be expressed as X i , whose speed can be expressed as V i, initialize the individual optimal cost function value of each particle in the particle swarm. Taking the i-th particle as an example (1≤i≤n), its individual optimal cost function value is expressed as Initialize the global optimal cost function value f of the entire particle swarm g = inf, pre-set obstacle avoidance constraint weight β1 and adjacent segment angle constraint weight β2;

[0031] 702, according to the position of the ith particle, use the cubic spline interpolation method to obtain the coordinates of the m expanded waypoints generated by the ith particle

[0032] 703, based on the expanded waypoints, calculate the Euclidean distances between two adjacent waypoints and sum them up to obtain the route The total length L;

[0033] 704, based on the expanded waypoints, using the formula Calculate the route The cost of each two adjacent waypoints and each threat is summed to get the total threat f obs ;

[0034] 705, based on the expanded waypoints, using the formula Calculate the route The sum of the angle constraint cost values ​​of all adjacent segments f θ ;

[0035] 706, according to the route The total length L, f obs , f θ , obstacle avoidance constraint weight β1 and adjacent segment angle constraint weight β2, using the cost function f = L(1 + β1f obs +β2f θ ) Calculate the cost function value f of the i-th particle i ;

[0036] 707, for the i-th particle, compare the cost function value f calculated in this round i and the individual optimal cost function value if Then update the individual optimal cost information and location information

[0037] 708, compare the individual optimal cost function value of each particle in the particle swarm and the global optimal cost function value f g ,if Then update the global optimal cost information and location information

[0038] 709, according to the particle swarm algorithm, the position and velocity of each particle are updated respectively. Taking the i-th particle as an example (1≤i≤n), the updating process of the position and velocity can be expressed as follows;

[0039]

[0040] w=w max -t((w max -w min ) / t max )

[0041] Among them, c1 is the individual learning factor, c2 is the social learning factor, t is the current iteration number, and t max is the maximum number of iterations, V i t represents the velocity of the ith particle at the tth iteration, represents the position of the i-th particle at the t-th iteration, w is the inertia weight, and w max represents the upper limit of the inertia weight, w min Indicates the lower bound of the inertia weight.

[0042] 710, each time the position and velocity of each particle are updated, premature judgment is performed. The specific process is: the global optimal cost f obtained by the particle swarm algorithm is g Monitor, if it is the same in the process of K consecutive iterations, but not all constraints are met, it is considered that there is a premature trend of falling into a local extreme point. At this time, a mutation strategy needs to be adopted, the process is: the current optimal solution P g Any maintenance of is changed to a random value within the environmental boundary to disrupt the current search direction of the particle and make it jump out of the local optimum. The specific performance is:

[0043] X g =[x1,x2,…,x u ,…,x n ]

[0044] Among them, X g is the position coordinate of the global optimal particle currently found, x u is a random perturbation term for replacement, whose value is randomly generated within the search range of the particle, and the sequence number u is also randomly generated.

[0045] 711, determine whether the termination condition is met. If the termination condition is met, the iteration and optimization are terminated, otherwise, return to step 702.

[0046] Furthermore, the preset judgment condition in step 9 is:

[0047] When the UAV formation is planning its route online, the final target point of the plan appears in a circle with the local planning starting point as the center and the rolling window distance as the radius.

[0048] Second invention, the present application provides an online route planning device for a UAV formation with a rolling optimization strategy, a modeling unit, a cost function construction unit, a determination unit, an update unit, a local route acquisition unit, and a judgment unit, wherein:

[0049] A modeling unit is used to establish and initialize an environment model and a UAV performance model. The parameter information of the environment model includes: environment boundary information, threat area, obstacle area and no-fly zone; the formation-related parameters of the UAV performance model include: UAV speed, UAV formation formation and UAV formation formation width;

[0050] The cost function construction unit is used to construct a cost function f for optimizing local route planning based on the quality of route performance indicators according to the parameter information of the environment model and the formation-related parameters of the UAV performance model;

[0051] The determination unit determines the global planning starting point, which is used for the final target point and the global optimization route; determines the local planning starting point according to the global planning starting point; and determines the local planning target point by using the local target point selection strategy;

[0052] An updating unit, used for updating the environmental threat situation according to the environmental model, the local planning starting point, the current planning starting point position and the situation measurement value within the single-step planning window period;

[0053] The local route acquisition unit is used to perform local route planning based on the local planning starting point and the local planning target point, take the cost function f as the optimization objective function, and use the mutation particle swarm optimization algorithm to obtain the local route;

[0054] The judgment unit is used to determine whether the aircraft only flies a step distance D along the local route. s The point reached is taken as the local planning step arrival point (x3, y3); according to the preset judgment conditions, it is judged whether the local planning step arrival point (x3, y3) reaches the planned final target point. If it reaches the planned final target point, the online route planning of the UAV formation ends.

[0055] Advantages of the present invention: The present invention proposes a method and device for online route planning of a UAV formation with a rolling optimization strategy, which draws on the principle of predictive control rolling optimization to dynamically plan a better flyable route for the entire UAV formation. The method is simple to implement, has strong optimization capability, fast convergence speed, convenient parameter setting, and strong practicality and versatility. The specific technical effects include the following:

[0056] (1) This method can make full use of the local environmental information detected in real time to perform online route planning in a rolling manner. By predicting and analyzing the environmental situation, multi-step planning, and single-step execution, it can better adapt to the needs of online route planning in complex dynamic flight environments.

[0057] (2) Based on the traditional particle swarm algorithm, this method dynamically adjusts the inertia weight linearly according to the iterative process, focusing on the global and local optimization of the balance algorithm at different stages, solving the defects of the traditional particle swarm algorithm that it is easy to fall into the local optimum and converge too slowly in the later stage, and has a strong ability to solve complex problems with multiple constraints;

[0058] (3) This method introduces premature detection and mutation strategies based on the traditional particle swarm algorithm, which helps to solve the premature stagnation problem that is prone to occur in the early stage of optimization of the traditional particle swarm algorithm, can effectively reduce the number of invalid iterations, improve the optimization efficiency and ability of the algorithm, and help meet the real-time requirements of online route planning;

[0059] (4) This method is highly flexible. Depending on the optimization focus of the problem to be solved, the optimal algorithm or heuristic algorithm can be flexibly selected under the framework of rolling optimization, and parameters such as cost function and weight coefficient can be set;

[0060] (5) This method is highly versatile and is applicable to both fixed-wing and rotary-wing UAVs. For different models of UAVs, one only needs to change the parameters related to the UAV performance in the algorithm to use it directly. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Execute a flow chart for a route planning method based on rolling optimization;

[0062] Figure 2 Schematic diagram of the dynamic forward process for rolling optimization;

[0063] Figure 3 Flowchart for particle swarm optimization algorithm execution. DETAILED DESCRIPTION

[0064] The invention belongs to the technical field of unmanned aerial vehicle route planning, and specifically relates to an online route planning method for an unmanned aerial vehicle formation based on an optimization strategy. Firstly, an environmental model and an unmanned aerial vehicle performance model are established; secondly, a local route planning optimization cost function f for evaluating the quality of route performance indicators is designed; then, a local planning starting point is determined, and the environmental threat situation is updated according to a current planning starting point position and a plurality of measurement values ​​within a single-step planning window period; then, a local target point selection strategy is used to determine a local planning target point; then, the cost function f is used as an optimized target function, and a variant particle swarm optimization algorithm is used to perform online local route planning; an aircraft only flies a step length along the planned local route, and the point is a local planning step length arrival point; finally, it is determined whether the planned final target point is reached, and if reached, the planning ends, otherwise, the above-mentioned local route online planning steps are executed in a rolling manner.

[0065] like Figure 1 As shown, the present application provides an online route planning method and device for a UAV formation with a rolling optimization strategy, which treats the route planning problem of a formation of M UAVs as a long-aircraft route planning problem that takes into account the formation width and the formation turning radius, combines the rolling optimization method with the variant particle swarm optimization algorithm, and repeatedly uses the variant particle swarm optimization algorithm to update the search for local routes in the rolling window according to the locally perceived situation information, and continuously moves forward with the flight process. Through multi-step planning and single-step execution, the applicability and optimality of the local window route search are effectively improved. Considering that when the flight range of the UAV is large, the UAV will not frequently change altitude during flight, the three-dimensional planning space can be simplified to a two-dimensional plane planning problem at a constant altitude.

[0066] like Figure 1 As shown, the present application provides an online route planning method for a UAV formation with a rolling optimization strategy, the method comprising:

[0067] Step 1: Establish and initialize the environment model and the UAV performance model. The parameter information of the environment model includes: environment boundary information, threat area, obstacle area and no-fly zone; the formation-related parameters of the UAV performance model include: UAV speed, UAV formation formation and UAV formation formation width;

[0068] Step 2: Based on the parameter information of the environmental model and the formation-related parameters of the UAV performance model, a cost function f for optimizing the local route planning based on the route performance index is constructed;

[0069] Optionally, the cost function in step 2 is: f = L(1 + β1f obs +β2f θ ),

[0070] Where L is the route Total length, expanded waypoints Generated by the i-th particle in the particle swarm, containing m extended waypoints,

[0071] Provide routes for all threat areas, obstacle areas, and no-fly zones in the environment The total threat caused by the environment; where κ is the sum of the number of threat areas, obstacle areas, and no-fly zones in the environment, For the The distance from the center of the circumscribed circle of the obstacle to the jth route segment D of the ith route i (j) distance, is the radius of the circumscribed circle, r s is the obstacle avoidance safety distance, r d is the width of the formation.

[0072] For route The sum of the angle constraint cost values ​​of all adjacent segments, θ j For route The jth sub-segment D i (j) and the j+1th sub-segment D i (j) angle, θ ref It is the reference value of the minimum angle between adjacent segments of UAV formation flight.

[0073] β1 is the obstacle avoidance constraint weight, and β2 is the angle constraint weight between adjacent flight segments.

[0074] Step 3: Determine the global planning starting point, final target point and global optimization route;

[0075] Step 4: Determine the local planning starting point based on the global planning starting point;

[0076] Optionally, step 4 includes: determining the first local planning starting point according to the global planning starting point (x1, y1) of the aircraft, and the subsequent local planning starting points are the step arrival points of the previous round of planning.

[0077] Step 5: Update the environmental threat situation based on the environmental model, the local planning starting point, the current planning starting point position, and the situation measurement value within the single-step planning window period;

[0078] Step 6: Determine the local planning target point using the local target point selection strategy;

[0079] Optionally, the local target point selection strategy in step 6 includes a heuristic target point selection strategy and an optimal route target point selection strategy.

[0080] Furthermore, the heuristic target point selection strategy is: after each rolling window information update, the heuristic function f(P)=g(P)+h(P) is used to select the local target point, where P is the point on the rolling window, g(P) is the cost of the drone flying from the current position to the target position, and h(P) is the cost of flying from P to the end point.

[0081] It should be noted that g(P) can be estimated based on the position of P and the environmental information in the current window.

[0082] It should be explained that since the information outside the boundary cannot be obtained, h(P) is estimated using the Euclidean distance from P to the end point, and finally the window boundary point P with the minimum cost is used as the local planning target point P sub , when the obtained P sub If there is more than one, you can choose any one, namely:

[0083] minf(P)=g(P)+h(P)

[0084] In order to simplify the calculation, the method reduces the optimization requirements and focuses on finding a feasible algorithm. Therefore, the value of g(P) is only determined by whether P belongs to the feasible domain, that is:

[0085]

[0086] The definition of the feasible domain is: the generated local target point is not located in the threat area, obstacle area and no-fly zone.

[0087] It should be noted that the heuristic target point selection strategy is a compromise between the requirements of global optimization and the constraints of local limited information.

[0088] Furthermore, the optimal route target point selection strategy is to use the intersection of each step rolling window radius and the global optimization route as the local target point.

[0089] It should be noted that the optimal route target point selection strategy can ensure that the global optimization index is taken into account when performing local planning online in real time. The premise of its use is that there is a globally optimized route. At this time, the local target point is selected as the intersection of the rolling window radius of each step and the global optimized route.

[0090] It should be noted that if the local target point generated by this method cannot be reached due to local obstacle avoidance requirements, that is, the offline route where the local target point is located conflicts with the newly emerged threat environment, it needs to be generated according to the heuristic target point selection strategy.

[0091] Step 7: Based on the local planning starting point and the local planning target point, the cost function f is used as the optimization objective function, and the mutation particle swarm optimization algorithm is used to perform local route planning to obtain the local route;

[0092] Optionally, step 7 includes the following steps:

[0093] 701, set a particle swarm consisting of n particles, initialize the position and velocity of each particle in the particle swarm, take the i-th particle as an example (1≤i≤n), its position can be expressed as X i , whose speed can be expressed as V i , initialize the individual optimal cost function value of each particle in the particle swarm. Taking the i-th particle as an example (1≤i≤n), its individual optimal cost function value is expressed as Initialize the global optimal cost function value f of the entire particle swarm g = inf, pre-set obstacle avoidance constraint weight β1 and adjacent segment angle constraint weight β2;

[0094] 702, according to the position of the ith particle, use the cubic spline interpolation method to obtain the coordinates of the m expanded waypoints generated by the ith particle

[0095] 703, based on the expanded waypoints, calculate the Euclidean distances between two adjacent waypoints and sum them up to obtain the route The total length L;

[0096] 704, based on the expanded waypoints, using the formula Calculate the route The cost of each two adjacent waypoints and each threat is summed to get the total threat f obs ;

[0097] 705, based on the expanded waypoints, using the formula Calculate the route The sum of the angle constraint cost values ​​of all adjacent segments f θ ;

[0098] 706, according to the route The total length L, f obs , f θ , obstacle avoidance constraint weight β1 and adjacent segment angle constraint weight β2, using the cost function f = L(1 + β1f obs +β2f θ ) Calculate the cost function value f of the i-th particle i ;

[0099] 707, for the i-th particle, compare the cost function value f calculated in this round i and the individual optimal cost function value if Then update the individual optimal cost information and location information P bi =X i ;

[0100] 708, compare the individual optimal cost function value of each particle in the particle swarm and the global optimal cost function value f g ,if Then update the global optimal cost information and location information

[0101] 709, according to the particle swarm algorithm, the position and velocity of each particle are updated respectively. Taking the i-th particle as an example (1≤i≤n), the updating process of the position and velocity can be expressed as follows;

[0102]

[0103] w=w max -t((w max -w min ) / t max )

[0104] Among them, c1 is the individual learning factor, c2 is the social learning factor, t is the current iteration number, and t max is the maximum number of iterations, V i t represents the velocity of the ith particle at the tth iteration, represents the position of the i-th particle at the t-th iteration, w is the inertia weight, and w max represents the upper limit of the inertia weight, w min Indicates the lower bound of the inertia weight.

[0105] 710, each time the position and velocity of each particle are updated, premature judgment is performed. The specific process is: the global optimal cost f obtained by the particle swarm algorithm is g Monitor, if it is the same in the process of K consecutive iterations, but not all constraints are met, it is considered that there is a premature trend of falling into a local extreme point. At this time, a mutation strategy needs to be adopted, the process is: the current optimal solution P g Any maintenance of is changed to a random value within the environmental boundary to disrupt the current search direction of the particle and make it jump out of the local optimum. The specific performance is:

[0106] X g =[x1,x2,…,x u ,…,x n ]

[0107] Among them, X g is the position coordinate of the global optimal particle currently found, x u is a random perturbation term for replacement, whose value is randomly generated within the search range of the particle, and the sequence number u is also randomly generated.

[0108] 711, determine whether the termination condition is met. If the termination condition is met, the iteration and optimization are terminated, otherwise, return to step 702.

[0109] Step 8: The aircraft only flies a step distance D along the local route. s The point reached is used as the local planning step to reach the point (x3, y3);

[0110] Among them, the distance between the local planning step arrival point (x3, y3) and the local planning starting point (x1, y1) is the step distance D s .

[0111] Step 9: According to the preset judgment conditions, determine whether the local planning step arrival point (x3, y3) reaches the planned final target point. If it reaches the planned final target point, the online route planning of the UAV formation ends, otherwise return to step 4.

[0112] Optionally, the preset judgment condition in step 9 is: during planning, the final target point of planning appears in a circle with the local planning starting point as the center and the rolling window distance as the radius.

[0113] In summary, the present invention provides an online route planning method and device for a UAV formation with a rolling optimization strategy, which draws on the principle of predictive control rolling optimization, makes full use of the local environmental information measured by the UAV in real time, and performs online planning in a rolling manner in a series of mobile space windows. By predicting and analyzing the environmental situation, multi-step planning, and single-step execution, adjustments are made to the changing environmental situation in advance, so as to dynamically plan a better flyable route for the overall UAV formation.

[0114] Example 1

[0115] This embodiment provides a method for online planning of a dual-aircraft formation route based on a rolling optimization strategy. In this embodiment, the formation is composed of two drones, and its initial information is set as follows:

[0116] Set the first UAV as the leader in the formation, with an initial speed of V1 = 22 m / s, an initial track inclination of θ1 = -5 degrees, an initial track deviation of ψ1 = 10 degrees, a planned starting point position of (200 m, 200 m), and a target point position of (5000 m, 4550 m); the rolling window size is D w =1000m, the execution step distance is D s =500m, the forward spacing of the formation is 20m, the lateral spacing is 15m, the altitude is consistent, the two drones form an "oblique one" formation, the cruising altitude is 500m, and there is no pre-planned optimal route initially.

[0117] In this embodiment, combined Figure 1 , Figure 2 , Figure 3 As shown, the specific steps include:

[0118] Step 1: Set the threat zone, obstacle zone, no-fly zone and other environmental information at a cruising altitude of 500m;

[0119] Specifically, the x-direction range of the three-dimensional space is x min =0m,x max =6000m, the range in y direction is y min =0m,y max =6000m; the threat zone, obstacle zone and no-fly zone information are as follows:

[0120]

[0121] Step 2: Design a local route planning optimization cost function f to evaluate the performance indicators of the route;

[0122] Specifically, considering the route length constraints, obstacle avoidance constraints, and minimum angle constraints of adjacent segments of UAV formation flight, the optimization cost function f is designed, which is specifically expressed as follows:

[0123] f=L(1+1000f obs +100f θ )

[0124] Further, is the total length of the route;

[0125] Further, It is the sum of the threats posed to the route by all threat areas, obstacle areas, and no-fly zones in the environment;

[0126] Further, It is the sum of the angle constraint cost values ​​of all adjacent segments on the route;

[0127] Among them, the minimum turning radius of the drone is r = 100m, and the obstacle avoidance safety distance of the drone formation is r s =10m, the width of the UAV formation r d =15m, the reference value of the minimum angle between adjacent segments of UAV formation flight θ ref =60deg.

[0128] Step 3: Determine the starting point of local planning;

[0129] Specifically, the starting point of the first planning is x1=200m, y1=200m, and the starting point of each subsequent round of planning is the step length arrival point of the previous round of planning;

[0130] Step 4: Update the environmental threat situation based on the current planning starting point position and multiple measurement values ​​within the single-step planning window period;

[0131] Specifically, with the planning starting point as the center, the rolling window D w =1000m is the radius, and it is determined whether there is a threat situation that needs to be considered in this planning within this range.

[0132] Step 5: Determine the local planning target point using the local target point selection strategy;

[0133] Specifically, since there is no initial optimal route, a heuristic target point selection strategy is needed to select a local target point using a heuristic function f(P)=g(P)+h(P), where P is a point on the rolling window.

[0134] Furthermore, g(P) is the cost of the UAV flying from the current planned starting point to the planned final target point. The value of g(P) is determined by whether P belongs to the feasible domain, that is:

[0135]

[0136] Furthermore, since the information outside the boundary cannot be obtained, h(P) is estimated using the Euclidean distance from P to the end point, and finally the window boundary point P with the minimum cost is used as the local planning target point P sub ,Right now:

[0137] minf(P)=g(P)+h(P)

[0138] Furthermore, when the obtained P sub If there is more than one, you can choose any one.

[0139] Step 6: Use the cost function f designed in step 2 as the optimization objective function and use the mutation particle swarm optimization algorithm to perform local route planning;

[0140] Specifically, the optimization calculation process of the mutation particle swarm algorithm is:

[0141]

[0142] w=w max -t((w max -w min ) / t max )

[0143] Among them, c1=2.05 is the individual learning factor, c2=2.05 is the social learning factor, t is the current iteration number, t max =500 is the total number of iterations, V i t and They represent the velocity and position of the ith particle at the tth iteration, respectively. The number of particles npop = 60, w is the inertia weight, and wmax =0.98 and w min =0.4 represent the upper and lower bounds of the inertia weight respectively.

[0144] Furthermore, during the optimization iteration, the optimal solution is judged to be premature. If it is considered that there is a premature trend of falling into a local extreme point, the current optimal solution is mutated.

[0145] Specifically, the premature judgment mechanism is: monitor the optimal cost obtained by the particle swarm algorithm. If it is the same in K consecutive iterations but has not yet satisfied all constraints, it is considered that there is a premature trend of falling into a local extreme point;

[0146] Furthermore, the mutation operation is to change any maintenance of the current optimal solution to a random value within the environmental boundary, which is specifically manifested as:

[0147] x g =[x1,x2,…,x r ,…,x n ]

[0148] Among them, the precocious judgment threshold K = 10, x g is the global optimal particle obtained by optimization, x u =rand(x min ,x max ) is a random perturbation term for replacement, and the serial number u=rand(1,n) is also randomly generated.

[0149] Step 7: The aircraft flies only one step distance along the route planned in step 6, and this point is the local planning step arrival point;

[0150] Specifically, the distance between the local planning step arrival point and the local planning starting point should be the preset execution step distance D s =500m.

[0151] Step 8: Determine whether the planned final target point has been reached. If so, the planning ends; otherwise, return to step 3.

[0152] Specifically, the judgment condition for reaching the final target point of the plan is: in a certain planning, the final target point of the plan appears in a circle with the local planning starting point as the center and the rolling window distance as the radius.

[0153] In this embodiment, combined Figure 2 The rolling optimization execution process specifically includes the following steps:

[0154] Step 1: Based on the local information detected by the aircraft, starting from the current position of the aircraft (x1, y1), the particle swarm optimization algorithm is used to plan the trajectory within the rolling window to (x2, y2);

[0155] Step 2, the aircraft only flies to (x3, y3) along the planned flight path, where x3 < x2, and the distance between (x3, y3) and (x1, y1) is the single-step execution distance of the aircraft;

[0156] Step 3, during the process of the aircraft flying from (x1, y1) to (x3, y3), re-initialize the planning starting point, take (x3, y3) as the new current position and plan forward to (x4, y4), where the distance between (x4, y4) and (x3, y3) is the rolling window distance;

[0157] Step 4, roll and implement the above process.

[0158] In this embodiment, combined with Figure 3 , the optimization of the particle swarm optimization algorithm specifically includes the following steps:

[0159] Step 1, initialize the positions and velocities of the particle swarm;

[0160] Step 2, calculate the particle cost function value f according to the designed cost function i ;

[0161] Step 3, for all particles, compare the particle cost function value f i and the individual optimal cost function value f bi . If f i < f bi , then update the individual optimum, that is, f bi = f i , P bi = X i ;

[0162] Step 4, compare the individual optimal cost function values f of all particles bi with the global optimal cost function value f g . If f bi < f g , then update the global optimum, that is, f g = f bi , P g = P bi ;

[0163] Step 5, update the positions and velocities of each particle according to the particle swarm algorithm respectively;

[0164] Step 6, determine whether the termination condition is reached. If it is reached, the iteration terminates; otherwise, return to Step 2.

[0165] Example 2

[0166] This embodiment provides a method for online planning of a dual-aircraft formation route based on a rolling optimization strategy. In this embodiment, the formation is composed of three drones, and its initial information is set as follows:

[0167] Set the first UAV as the leader in the formation, with an initial speed of V1 = 24m / s, an initial track inclination of θ1 = 10deg, an initial track deviation of ψ1 = -5deg, an initial position of (350m, 350m), and a target position of (5550m, 5000m); the rolling window size is D w =800m, the execution step distance is D s =300m, the forward spacing of the formation is 20m, the lateral spacing is 12m, the altitude is consistent, the three drones form a "positive wedge" formation, the lead drone is in the first row of the formation, and the other two drones are in the second row of the formation. The cruising altitude is 300m, and there is no pre-planned optimal route initially.

[0168] In this embodiment, combined Figure 1 , Figure 2 , Figure 3 As shown, the following steps are included:

[0169] Step 1: Set the threat zone, obstacle zone, no-fly zone and other environmental information at a cruising altitude of 300m;

[0170] Specifically, the x-direction range of the three-dimensional space is x min =0m,x max =7000m, the range in y direction is y min =0m,y max =7000m; the threat zone, obstacle zone and no-fly zone information are as follows:

[0171]

[0172]

[0173] Step 2: Design a local route planning optimization cost function f to evaluate the performance indicators of the route;

[0174] Specifically, considering the route length constraints, obstacle avoidance constraints, and minimum angle constraints of adjacent segments of UAV formation flight, the optimization cost function f is designed, which is specifically expressed as follows:

[0175] f=L(1+1000f obs +100f θ )

[0176] Further, is the total length of the route;

[0177] Further, It is the sum of the threats posed to the route by all threat areas, obstacle areas, and no-fly zones in the environment;

[0178] Further, It is the sum of the angle constraint cost values ​​of all adjacent segments on the route;

[0179] Among them, the minimum turning radius of the drone is r = 105m, and the obstacle avoidance safety distance of the drone formation is r s =15m, UAV formation width r d =24m, the reference value of the minimum angle between adjacent segments of UAV formation flight θ ref =70deg.

[0180] Step 3: Determine the starting point of local planning;

[0181] Specifically, the starting point of the first planning is x1=350m, y1=350m, and the starting point of each subsequent round of planning is the step length arrival point of the previous round of planning.

[0182] Step 4: Update the environmental threat situation based on the current planning starting point position and multiple measurement values ​​within the single-step planning window period;

[0183] Specifically, with the planning starting point as the center, the rolling window D w =800m is the radius, and it is determined whether there is a threat situation that needs to be considered in this planning within this range.

[0184] Step 5: Determine the local planning target point using the local target point selection strategy;

[0185] Specifically, since there is no initial optimal route, a heuristic target point selection strategy is needed to select a local target point using a heuristic function f(P)=g(P)+h(P), where P is a point on the rolling window.

[0186] Furthermore, g(P) is the cost of the UAV flying from the current planned starting point to the planned final target point. The value of g(P) is determined by whether P belongs to the feasible domain, that is:

[0187]

[0188] Furthermore, since the information outside the boundary cannot be obtained, h(P) is estimated using the Euclidean distance from P to the end point, and finally the window boundary point P with the minimum cost is used as the local planning target point P sub ,Right now:

[0189] minf(P)=g(P)+h(P)

[0190] Furthermore, when the obtained P sub If there is more than one, you can choose any one.

[0191] Step 6: Use the cost function f designed in step 2 as the optimization objective function and use the mutation particle swarm optimization algorithm to perform local route planning;

[0192] Specifically, the optimization calculation process of the mutation particle swarm algorithm is:

[0193]

[0194] w=w max -t((w max -w min ) / t max )

[0195] Among them, c1=2.1 is the individual learning factor, c2=2.2 is the social learning factor, t is the current iteration number, t max =800 is the total number of iterations, V i t and They represent the velocity and position of the ith particle at the tth iteration, respectively. The number of particles npop = 50, w is the inertia weight, and w max =0.99 and w min =0.41 represent the upper and lower bounds of the inertia weight respectively.

[0196] Furthermore, during the optimization iteration, the optimal solution is judged to be premature. If it is considered that there is a premature trend of falling into a local extreme point, the current optimal solution is mutated.

[0197] Specifically, the premature judgment mechanism is: monitor the optimal cost obtained by the particle swarm algorithm. If it is the same in K consecutive iterations but has not yet satisfied all constraints, it is considered that there is a premature trend of falling into a local extreme point;

[0198] Furthermore, the mutation operation is to change any maintenance of the current optimal solution to a random value within the environmental boundary, which is specifically manifested as:

[0199] x g =[x1,x2,…,x u ,…,x n ]

[0200] Among them, the precocious judgment threshold K = 15, x g is the global optimal particle obtained by optimization, x u =rand(x min ,x max ) is a random perturbation term for replacement, and the serial number u=rand(1,n) is also randomly generated.

[0201] Step 7: The aircraft flies only one step length along the route planned in Step 6, and this point is the local planning step length arrival point;

[0202] Specifically, the distance between the local planning step length arrival point and the local planning starting point should be the pre-set execution step length distance D s = 300 m.

[0203] Step 8: Determine whether the planned final target point has been reached. If it has been reached, the planning ends; otherwise, return to Step 3.

[0204] Specifically, the judgment condition for reaching the planned final target point is that during a certain planning, the planned final target point appears within the circle with the local planning starting point as the center and the rolling window distance as the radius.

[0205] In this embodiment, combined with Figure 2 , the rolling optimization execution process specifically includes the following steps:

[0206] Step 1: Starting from the current position (x1, y1) of the aircraft, according to the local information detected by the aircraft, use the particle swarm optimization algorithm to plan the flight path within the rolling window to (x2, y2);

[0207] Step 2: The aircraft flies only to (x3, y3) along the above-planned flight path, where x3 < x2, and the distance between (x3, y3) and (x1, y1) is the single-step execution distance of the aircraft;

[0208] Step 3: During the process of the aircraft flying from (x1, y1) to (x3, y3), re-initialize the planning starting point, and use (x3, y3) as the new current position to plan forward to (x4, y4), where the distance between (x4, y4) and (x3, y3) is the rolling window distance;

[0209] Step 4: Roll and implement the above process.

[0210] In this embodiment, combined with Figure 3 , the particle swarm optimization algorithm optimization specifically includes the following steps:

[0211] Step 1: Initialize the position and velocity of the particle swarm;

[0212] Step 2: Calculate the particle cost function value f according to the designed cost function i ;

[0213] Step 3: For all particles, compare the particle cost function value f i and the individual optimal cost function value f bi , if f i < f bi , then update the individual optimum, that is, f bi = fi , P bi =X i ;

[0214] Step 4: Compare the individual optimal cost function values ​​f of all particles bi and the global optimal cost function value f g , if f bi <f g , then update the global optimum, that is, f g =f bi , P g =P bi ;

[0215] Step 5, update the position and velocity of each particle according to the particle swarm algorithm;

[0216] Step 6: Determine whether the termination condition is met. If so, the iteration terminates; otherwise, return to step 2.

[0217] The present invention proposes a method and device for online route planning of a UAV formation with a rolling optimization strategy, which draws on the principle of predictive control rolling optimization to dynamically plan a better flyable route for the UAV formation as a whole. The method is simple to implement, has strong optimization capability, fast convergence speed, convenient parameter setting, and strong practicality and versatility. The specific technical effects include the following:

[0218] (1) This method can make full use of the local environmental information detected in real time to perform online route planning in a rolling manner. By predicting and analyzing the environmental situation, multi-step planning, and single-step execution, it can better adapt to the needs of online route planning in complex dynamic flight environments.

[0219] (2) Based on the traditional particle swarm algorithm, this method dynamically adjusts the inertia weight linearly according to the iterative process, focusing on the global and local optimization of the balance algorithm at different stages, solving the defects of the traditional particle swarm algorithm that it is easy to fall into the local optimum and converge too slowly in the later stage, and has a strong ability to solve complex problems with multiple constraints;

[0220] (3) This method introduces premature detection and mutation strategies based on the traditional particle swarm algorithm, which helps to solve the premature stagnation problem that is prone to occur in the early stage of optimization of the traditional particle swarm algorithm, can effectively reduce the number of invalid iterations, improve the optimization efficiency and ability of the algorithm, and help meet the real-time requirements of online route planning;

[0221] (4) This method is highly flexible. Depending on the optimization focus of the problem to be solved, the optimal algorithm or heuristic algorithm can be flexibly selected under the framework of rolling optimization, and parameters such as cost function and weight coefficient can be set;

[0222] (5) This method is highly versatile and is applicable to both fixed-wing and rotary-wing UAVs. For different models of UAVs, one only needs to change the parameters related to the UAV performance in the algorithm to use it directly.

Claims

1. A method for online route planning of a UAV formation based on a rolling optimization strategy, characterized in that: The method comprises: Step 1: Establish and initialize the environment model and the UAV performance model. The parameter information of the environment model includes: environment boundary information, threat area, obstacle area and no-fly zone; the formation-related parameters of the UAV performance model include: UAV speed, UAV formation formation and UAV formation formation width; Step 2: Based on the parameter information of the environmental model and the formation-related parameters of the UAV performance model, a cost function f for optimizing the local route planning based on the route performance index is constructed; Step 3: Determine the global planning starting point, final target point and global optimization route; Step 4: Determine the local planning starting point based on the global planning starting point; Step 5: Update the environmental threat situation based on the environmental model, the local planning starting point, the current planning starting point position, and the situation measurement value within the single-step planning window period; Step 6: Determine the local planning target point using the local target point selection strategy; Step 7: Based on the local planning starting point and the local planning target point, the cost function f is used as the optimization objective function, and the mutation particle swarm optimization algorithm is used to perform local route planning to obtain the local route; Step 8: The aircraft only flies a step distance D along the local route. s The point reached is used as the local planning step to reach the point (x3, y3); Step 9: According to the preset judgment conditions, determine whether the local planning step arrival point (x3, y3) reaches the planned final target point. If it reaches the planned final target point, the online route planning of the UAV formation ends, otherwise return to step 4.

2. The method for online route planning of a UAV formation according to claim 1, characterized in that: The cost function in step 2 is: f = L(1 + β1f obs +β2f θ ), Where L is the route Total length, expanded waypoints Generated by the i-th particle in the particle swarm, containing m extended waypoints, Provide routes for all threat areas, obstacle areas, and no-fly zones in the environment The total threat caused by the environment; where κ is the sum of the number of threat areas, obstacle areas, and no-fly zones in the environment, For the The distance from the center of the circumscribed circle of the obstacle to the jth route segment D of the ith route i (j) distance, is the radius of the circumscribed circle, r s is the obstacle avoidance safety distance, r d is the formation width; For route The sum of the angle constraint cost values ​​of all adjacent segments, θ j For route The jth sub-segment D i (j) and the j+1th sub-segment D i (j) angle, θ ref It is the reference value of the minimum angle between adjacent segments of UAV formation flight; β1 is the obstacle avoidance constraint weight, and β2 is the angle constraint weight between adjacent flight segments.

3. The method for online route planning of a UAV formation according to claim 1, characterized in that: Step 4 includes: The first local planning starting point is determined according to the global planning starting point (x1, y1) of the aircraft, and the subsequent local planning starting points are the step arrival points of the previous round of planning.

4. The method for online route planning of a UAV formation according to claim 1, characterized in that: The local target point selection strategy in step 6 includes a heuristic target point selection strategy and an optimal route target point selection strategy.

5. The method for online route planning of a UAV formation according to claim 4, characterized in that: The heuristic target point selection strategy is: After each rolling window information update, the heuristic function f(P)=g(P)+h(P) is used to select the local target point, where P is the point on the rolling window, g(P) is the cost of the drone flying from the current position to the target position, and h(P) is the cost of flying from P to the end point.

6. The method for online route planning of a UAV formation according to claim 5, characterized in that:

7. The method for online route planning of a UAV formation according to claim 4, characterized in that: The optimal route target point selection strategy is to use the intersection of each step rolling window radius and the global optimization route as the local target point.

8. The method for online route planning of a UAV formation according to claim 1, characterized in that: Step 7 includes the following steps:

701. Set up a particle swarm consisting of n particles, initialize the position and velocity of each particle in the particle swarm, and express the position of the i-th particle as X i , the speed is expressed as V i , initialize the individual optimal cost function value of each particle in the particle swarm, and its individual optimal cost function value is expressed as Initialize the global optimal cost function value f of the entire particle swarm g = inf, pre-set obstacle avoidance constraint weight β1 and adjacent segment angle constraint weight β2; 702, according to the position of the ith particle, use the cubic spline interpolation method to obtain the coordinates of the m expanded waypoints generated by the ith particle 703, based on the expanded waypoints, calculate the Euclidean distances between two adjacent waypoints and sum them up to obtain the route The total length L; 704, based on the expanded waypoints, using the formula Calculate the route The cost of each two adjacent waypoints and each threat is summed to get the total threat f obs ; 705, based on the expanded waypoints, using the formula Calculate the route The sum of the angle constraint cost values ​​of all adjacent segments f θ ; 706, according to the route The total length L, f obs , f θ , obstacle avoidance constraint weight β1 and adjacent segment angle constraint weight β2, using the cost function f = L(1 + β1f obs +β2f θ ) Calculate the cost function value f of the i-th particle i ; 707, for the i-th particle, compare the cost function value f calculated in this round i and the individual optimal cost function value if Update the individual optimal cost information and location information 708, compare the individual optimal cost function value of each particle in the particle swarm and the global optimal cost function value f g ,if Then update the global optimal cost information and location information 709, according to the particle swarm algorithm, the position and velocity of each particle are updated respectively. Taking the i-th particle as an example (1≤i≤n), the updating process of the position and velocity is expressed as follows; w=w max -t((w max -w min ) / t max ) Among them, c1 is the individual learning factor, c2 is the social learning factor, t is the current iteration number, and t max is the maximum number of iterations, V i t represents the velocity of the ith particle at the tth iteration, represents the position of the i-th particle at the t-th iteration, w is the inertia weight, and w max represents the upper limit of the inertia weight, w min Indicates the lower bound of the inertia weight; 710, each time the position and velocity of each particle are updated, premature judgment is performed. The specific process is: the global optimal cost f obtained by the particle swarm algorithm is g Monitor, if it is the same in the process of K consecutive iterations, but not all constraints are met, then adopt the mutation strategy, the process is: the current optimal solution P g Any maintenance of is changed to a random value within the environmental boundary to disrupt the current search direction of the particle and make it jump out of the local optimum. The specific performance is: X g =[x1,x2,…,x u ,…,x n ] Among them, X g is the position coordinate of the global optimal particle currently found, x u is a random perturbation term for replacement, whose value is randomly generated within the search range of the particle, and the sequence number u is also randomly generated; 711, determine whether the termination condition is met; if the termination condition is met, the iterative optimization is terminated, otherwise return to step 702.

9. The method for online route planning of a UAV formation according to claim 1, characterized in that: The preset judgment conditions in step 9 are: When the UAV formation is planning its route online, the final target point of the plan appears in a circle with the local planning starting point as the center and the rolling window distance as the radius.

10. An online route planning device for a UAV formation with a rolling optimization strategy, characterized in that: Modeling unit, cost function building unit, determination unit, update unit, local route acquisition unit, judgment unit, among which: A modeling unit is used to establish and initialize an environment model and a UAV performance model. The parameter information of the environment model includes: environment boundary information, threat area, obstacle area and no-fly zone; the formation-related parameters of the UAV performance model include: UAV speed, UAV formation formation and UAV formation formation width; The cost function construction unit is used to construct a cost function f for optimizing local route planning based on the quality of route performance indicators according to the parameter information of the environment model and the formation-related parameters of the UAV performance model; The determination unit determines the global planning starting point, which is used for the final target point and the global optimization route; determines the local planning starting point according to the global planning starting point; and determines the local planning target point by using the local target point selection strategy; An updating unit, used for updating the environmental threat situation according to the environmental model, the local planning starting point, the current planning starting point position and the situation measurement value within the single-step planning window period; The local route acquisition unit is used to perform local route planning based on the local planning starting point and the local planning target point, take the cost function f as the optimization objective function, and use the mutation particle swarm optimization algorithm to obtain the local route; The judgment unit is used to determine whether the aircraft only flies a step distance D along the local route. s The point reached is taken as the local planning step arrival point (x3, y3); according to the preset judgment conditions, it is judged whether the local planning step arrival point (x3, y3) reaches the planned final target point. If it reaches the planned final target point, the online route planning of the UAV formation ends.

Images