A Method, Device and Medium for Task Allocation of Multi-UAV Forest Fire Patrol

Abstract

The present invention belongs to the technical field of multi-UAV forest fire inspection task allocation, and discloses a multi-UAV forest fire inspection task allocation method, device, and medium. Based on the geographical data and fire information of the forest fire area, the present invention constructs a multi-UAV collaborative inspection task allocation model, with the dual optimization objectives of minimizing the total inspection distance and optimizing the UAV load balance, and additionally introduces a map level cost and a UAV detour cost to improve the rationality of task allocation; selects the iterative strategy of the gray wolf population between the individual internal discrete operation and the group cooperation crossover operation according to the dynamic switching factor; in the iterative optimization, updates the elite archive population based on non-dominated sorting and crowding distance, and enhances the convergence and distribution of the solution through the "removal-insertion" strategy. The present invention solves the defect that the traditional continuous multi-objective gray wolf algorithm cannot be applied to discrete problems through unique discretization operations, and can quickly generate inspection paths that take into account timeliness and energy consumption balance.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-UAV forest fire inspection task allocation, and particularly relates to a multi-UAV forest fire inspection task allocation method, device and medium based on a discrete multi-objective grey wolf optimization algorithm. Background Art

[0002] With the development of UAV technology, multi-UAV collaborative inspection shows great application potential in forest fire monitoring. Compared with a single-UAV system, a multi-UAV system can improve the inspection efficiency and coverage through task allocation and collaborative operation. Especially in complex terrains and emergency situations, it can detect and monitor fire situations more quickly.

[0003] However, the multi-UAV task allocation problem usually involves multiple conflicting optimization objectives, such as minimizing the flight distance and balancing the task load. Minimizing the flight distance tends to preferentially select the UAV closest to the task point to execute the task. This local optimal strategy may cause some UAVs to undertake too many tasks due to geographical location advantages. On the contrary, balancing the task load requires a global perspective of task scheduling, which may force the tasks in the same area to be assigned to UAVs in different positions. Although it can avoid overloading a single UAV, it will lead to an increase in the flight path of the UAV and the total energy consumption.

[0004] Existing task allocation methods mostly adopt single-objective optimization models and merge multiple optimization objectives into one objective for solution by the method of weight summation. Although this method has a low computational complexity, it is difficult to effectively balance multiple conflicting objectives. In addition, most inspection task allocation methods are only applicable to two-dimensional environments, ignoring the influence of obstacles or terrain on the allocation results and lacking refined modeling of terrain constraints, resulting in the disconnection between path planning and actual needs.

[0005] As a swarm intelligence optimization algorithm that simulates the hunting behavior of grey wolves, the grey wolf optimization algorithm has received wide attention due to its strong search ability and convergence performance in continuous optimization problems. However, its core operations (such as surrounding and attacking prey) rely on continuous variable update mechanisms and cannot be directly applied to discrete sequence optimization problems such as multi-UAV inspection task allocation, and it is difficult to balance the diversity and convergence of the solution set in multi-objective scenarios. Summary of the Invention

[0006] In view of the above technical problems, the present invention proposes a multi-UAV forest fire inspection task allocation method. This method first constructs a dual-objective multi-UAV forest fire inspection task allocation model, and the task allocation model takes the minimization of the total inspection distance and the minimization of the UAV load difference as dual-objective functions; and designs a discrete multi-objective grey wolf algorithm for solving this task allocation model, significantly improving the inspection efficiency and task allocation effect of the multi-UAV system.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A multi-UAV forest fire inspection task allocation method includes the following steps:

[0009] Step 1. Based on the geographical data and fire information of the forest fire area, construct a multi-UAV collaborative inspection task allocation model, where the model takes the minimization of the total inspection distance and the minimization of the UAV load difference as the dual objective functions;

[0010] Step 2. Conduct gray wolf individual coding. For N t inspection tasks, encode the inspection paths of N u UAVs into a task sequence, and generate N u -1 random integers between [1, N t as the breakpoints of the task sequence;

[0011] Step 3. For the gray wolf individual coding, randomly generate multiple groups of different gray wolf individuals; by calculating the dual objective function values of each gray wolf individual, select the current better path solutions using non-dominated sorting and store them in the initial elite archive;

[0012] Set the dynamic switching factor a ∈ [0, 1];

[0013] Step 4. In the elite archive, find the three wolves with the lowest crowding degree as the leader wolves according to the crowding degree of the gray wolf individuals; select the gray wolf population iteration strategy according to the current value of a: for each gray wolf individual, first generate a random number r ∈ [0, 1]; when r ≤ a, perform the discretization operation based on the individual interior; when r > a, perform the crossover operation of group collaboration;

[0014] Step 5. Perform a "removal-insertion" operation on each gray wolf individual, calculate the distance between adjacent task points using the distance matrix, remove the task point with the largest distance and insert it near the task point closer to it;

[0015] Step 6. Update the elite archive; determine whether the maximum number of iterations is reached; if not, return to Step 4, otherwise output the optimal solution set in the elite archive as the final multi-UAV inspection task allocation plan.

[0016] In addition, based on the above multi-UAV forest fire inspection task allocation method, the present invention also proposes a computer device, which includes a memory and one or more processors. An executable code is stored in the memory. When the processor executes the executable code, it is used to implement the steps of the above multi-UAV forest fire inspection task allocation method.

[0017] In addition, based on the above multi-UAV forest fire inspection task allocation method, the present invention also proposes a computer-readable storage medium with a program stored thereon. When the program is executed by a processor, it is used to implement the steps of the above multi-UAV forest fire inspection task allocation method.

[0018] The present invention has the following advantages:

[0019] 1. Considering the complexity of the task scenario and the conflict of optimization objectives, the method of the present invention designs a dual-objective multi-UAV forest fire inspection task allocation model, including two optimization objectives of minimizing the total flight distance and minimizing the task load difference between UAVs and three task constraints. Different from the traditional optimization model that transforms multiple optimization objectives into a single objective through weighted summation, this model can effectively handle the trade-off between different conflicting objectives and provide a more comprehensive and balanced set of solutions. The present invention can take into account both inspection efficiency and task balance, ensure that the UAV group can efficiently complete the inspection task while avoiding the situation of individual UAV overload or uneven inspection task allocation, and improve the overall robustness and task execution effect of the system.

[0020] 2. In the setting of the objective function of the dual-objective optimization model, the method of the present invention additionally combines the map level cost function and the distance cost function of the UAV flying around the terrain, so that the inspection task planning can fully consider the influence of the complex forest terrain. Among them, the map level cost function optimizes the UAV task allocation efficiency by evaluating the level score of the UAV inspection sequence, making the UAV task allocation sequence more in line with the actual requirements; the distance cost function of the UAV flying around the terrain can quantify the influence of the terrain between task points on the flying around distance, thereby improving the rationality and execution efficiency of the inspection path.

[0021] 3. Aiming at the discrete characteristics of the inspection task allocation, the present invention conducts a unique discretization design on the traditional continuous multi-objective grey wolf algorithm to make it applicable to the discrete task allocation problem. First, a special encoding is performed on the discrete sequence of tasks to represent grey wolf individuals; then, aiming at the special grey wolf individual encoding method, a unique discretization operation (intra-individual discrete operation and population crossover operation) is designed to update the task encoding of grey wolf individuals and adjust the two discretization operations by introducing a dynamic switching factor a, so that the algorithm can fully explore the solution space in the early stage and quickly converge to the optimal solution in the later stage, effectively balancing the exploration stage and the development stage of the algorithm; in addition, the present invention also designs a "removal-insertion" operation to adjust the updated grey wolf individuals to avoid the algorithm falling into a local optimal state, ensure that the algorithm can effectively explore the solution space during the optimization process, improve the feasibility and allocation efficiency of the solution, and thus better meet the actual needs of the multi-UAV collaborative inspection task. Description of the Drawings

[0022] Figure 1 This is a flowchart of a multi - UAV forest fire inspection task allocation method in an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram for calculating the map - level cost introduced in the task allocation model of an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram for calculating the distance cost of UAVs flying around terrain introduced in the task allocation model of an embodiment of the present invention;

[0025] Figure 4 This is the discrete coding method of gray wolf individuals in the discrete multi - objective gray wolf optimization algorithm of an embodiment of the present invention;

[0026] Figure 5 This is a schematic diagram for calculating the crowding degree of the gray wolf population in the discrete multi - objective gray wolf optimization algorithm of an embodiment of the present invention;

[0027] Figure 6 This is a schematic diagram of the discretization operation inside a gray wolf individual in the discrete multi - objective gray wolf optimization algorithm of an embodiment of the present invention;

[0028] Figure 7 This is a schematic diagram of the crossover operation of the gray wolf group in the discrete multi - objective gray wolf optimization algorithm of an embodiment of the present invention;

[0029] Figure 8 This is a schematic diagram of the "removal - insertion" operation designed in the discrete multi - objective gray wolf optimization algorithm of an embodiment of the present invention. Detailed implementation manners

[0030] The present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners:

[0031] Embodiment 1

[0032] In view of the problems of low task allocation efficiency, insufficient multi - objective trade - off, and difficulty in handling discrete task points in the existing methods, Embodiment 1 of the present invention proposes a multi - UAV forest fire inspection task allocation method. Based on the discrete multi - objective gray wolf optimization algorithm, by finely constructing a task allocation model in a three - dimensional environment and designing an optimization mechanism and a multi - objective solution strategy adapted to the discrete environment, it can significantly improve the inspection efficiency and task allocation effect of the multi - UAV system on the basis of meeting multiple objectives such as minimizing the flight distance, balancing the task load, and the task inspection sequence constraint.

[0033] As Figure 1 shown, the multi - UAV forest fire inspection task allocation method in this embodiment includes the following steps:

[0034] Step 1. Construct a task allocation model.

[0035] Based on the geographical data and fire information of the forest fire area, a multi - UAV collaborative inspection task allocation model is constructed. The model takes the minimization of the total inspection distance and the minimization of the UAV load difference as the dual - objective functions. At the same time, a map - level cost function and a distance cost function for the UAV to fly around the terrain are introduced in the geographical constraint conditions, aiming to improve the inspection efficiency and consider the impact of the three - dimensional terrain on the inspection task allocation, so that the allocation result is closer to the real task requirements.

[0036] Among them, the specific process of constructing the multi - UAV collaborative inspection task allocation model is as follows:

[0037] First, digitize the forest map, including terrain height, inspection task point locations, and UAV nest locations.

[0038] Then, establish a dual - objective mathematical model for multi - UAV task allocation, including two objective functions and three constraint conditions.

[0039] The objective functions are:

[0040] (1)

[0041] (2)

[0042] (3)

[0043] Equation (1) is to optimize the total flight distance of the UAVs; Equation (2) is to minimize the task load difference between multiple UAVs and ensure the balance of the UAV task load quantity; Equation (3) gives the calculation method of C ij of.

[0044] In the formula, represents the total inspection distance, represents the UAV load difference, X ij is a binary variable, X ij = 1 means the UAV will fly from task point T i to task point T j , otherwise, X ij = 0; C ij represents the distance cost between task points T i and T j ; T s is the level evaluation score of the UAV task sequence, i.e., the map - level cost; m k is the task load quantity of the k - th UAV; N t / N u represents the ideal average task load quantity of each UAV; T c represents the distance cost for the UAV to fly around the terrain; d ij represents task point T i and Tj The Euclidean distance between i , y i , z i ) and (x j , y j , z j ) are task points T i and T j The three-dimensional coordinates of .

[0045] The constraints are:

[0046] (4)

[0047] (5)

[0048] (6)

[0049] Formula (4) represents each task point T i The out-degree is 1, which means that each task point can only connect to one subsequent task point; Formula (5) means that each task point T i The in-degree is 1, which means that each task point is only allowed to be connected by one predecessor task point; Formula (6) indicates that the in-degree and out-degree of the starting point T0 are both N u , which means that each drone should start from the starting point and eventually return to the starting point.

[0050] In order to make the UAV task allocation sequence more in line with actual requirements, the map level cost T is introduced into the objective function calculation of formula (1) and formula (2): s The cost of flying around the drone is T c Evaluate UAV mission sequences.

[0051] Map level cost T s Evaluate the UAV mission sequence and calculate its score. s The calculation method is described as follows:

[0052] like Figure 2 As shown, for the convenience of explanation, it is assumed that the machine nest is located in the center of the map (represented by the black triangle in the figure), and the square inspection area is divided into the same interval S t Divided into multiple concentric square grids (similar to the shape of a Chinese character "回"), the map level increases by 1 with each outward expansion (for example, the center is level 1, and the outwards are 2 and 3 respectively). t The calculation method is:

[0053] ;

[0054] ;

[0055] As can be seen from the formula, according to the number of task points N t and the number of inspection drones N u the map level N can be calculated l ; represents rounding down the variable ; according to the obtained map level N l and the side length of the square map Map size the division spacing S is calculated t .

[0056] It is required that the inspection sequence of the drones presents a "low - high - low" pattern in terms of map level (for example, the map levels corresponding to the inspection sequence should be 1 - 2 - 3 - 2 - 1); if the path level jump does not conform to this "low - high - low" pattern (such as 1 - 3 - 1 - 2 - 1), a penalty cost is added, as shown in formula (7):

[0057] (7)

[0058] In the formula, T rank is the actual map level sequence corresponding to the task sequence of each drone; I rank is the ideal map level sequence of each drone. The ideal map levels corresponding to the even - numbered task sequences are: 1 - 2 - 3 - 3 - 2 - 1, and the ideal map levels corresponding to the odd - numbered task sequences are: 1 - 2 - 3 - 2 - 1; sum|T rank −I rank | represents the sum of the difference sequences between the actual map level and the ideal map level.

[0059] Meanwhile, when calculating the distance between task points, if there is a mountain peak blocking between two task points, the drone has an additional detour distance; as Figure 3 shown, assume that task points T1 and T2 are on both sides of a mountain peak. Uniformly distributed p n path points I pi are inserted by linear interpolation on the line connecting T1 and T2, i = 1, 2,..., n; calculate the difference ΔH i between the height of each path point and the terrain height, and the formula is as follows:

[0060] ;

[0061] In the formula, (x i , y i , z i ) represents the three - dimensional coordinates of path point I pi ; terrain(x i , y i ) represents the terrain height value corresponding to the coordinates (x i , y i ).

[0062] Then the UAV's detour cost T c is expressed by formula (8):

[0063] (8)

[0064] Through this method, the mountain height between task points is converted into the additional distance cost for the UAV to fly around the mountain body.

[0065] Step 2. Encoding of gray wolf individuals.

[0066] For N t inspection tasks, the inspection paths of N u UAVs are encoded into a task sequence, and each value in the sequence corresponds to an inspection task point. Additionally, N u −1 random integers between [1, N t are generated as breakpoints of the task sequence for dividing the task sequence; N u −1 breakpoints divide the task sequence into N u sequence segments, and each segment corresponds to the inspection sequence of one UAV; the dimension of the gray wolf individual in this encoding method is 1×N t +N u −1.

[0067] Figure 4 Fig. shows the encoding method of gray wolf individuals in the context of three UAVs inspecting ten inspection task points.

[0068] The inspection paths of 3 UAVs are encoded into a task sequence, and each value in the sequence corresponds to an inspection task point. Additionally, 2 random integers (breakpoints) between [1, 10] are generated to divide the task sequence into the inspection subsequences of 3 UAVs. Figure 4 In, breakpoints 3 and 7 divide the task sequence into 3 subsequences; among them, the task sequence of UAV 1 is 7 - 4 - 3; the task sequence of UAV 2 is 1 - 2 - 9 - 8; the task sequence of UAV 3 is 6 - 5 - 10.

[0069] Step 3. Initializing the gray wolf population and the elite archive.

[0070] For the individual encoding method proposed in Step 2, multiple groups of different gray wolf individuals are randomly generated, called the "gray wolf population". By calculating the bi - objective function values of each gray wolf individual, the current relatively optimal path solutions are selected by non - dominated sorting and stored in the initial elite archive Archive; and a dynamic switching factor a∈[0,1] is set to control the update strategy of gray wolf individuals.

[0071] The specific description of the update strategy of the non - dominated sorting method is as follows:

[0072] In the multi-objective optimization problem, the dominance relationship between different solutions is described as follows: for any two solutions x a and x b in the population, if the values of x a in all objective functions are not worse than those of x b , that is , f i (x a ) ≤ f i (x b );and x a is strictly better than x b in at least one objective, that is , f j (x a ) < f j (x b );then it is said that x a dominates x b .

[0073] All solutions in the population are divided into multiple front layers according to the dominance relationship. The first layer is the set of non-dominated solutions that are not dominated by any solution, which is called the Pareto front; solutions in the first layer are preferentially selected to be added to the elite archive. If the number of solutions in the first layer exceeds the upper limit of the archive capacity, further screening is performed based on the crowding distance of the solutions, and solutions that are sparsely distributed in the objective space are retained to maintain diversity.

[0074] The dynamic switching factor a is used to adaptively adjust the search strategy of the gray wolf population, for balancing the exploration and exploitation of the algorithm. The calculation method of the dynamic switching factor a is shown in formula (9):

[0075] (9)

[0076] In the formula, iter represents the current iteration number of the algorithm; iter max represents the maximum value of the iteration number of the algorithm.

[0077] The value of the dynamic switching factor a will non-linearly decrease from 1 to 0 with the iteration number. At the beginning of each iteration, a random number r is first generated. When r ≤ a, the discretization operation within the individual is performed; when r > a, the crossover operation of group collaboration is performed; from the changing trend of the dynamic switching factor a, it can be seen that in the early stage of the algorithm, the value of a is larger, and there is a greater probability of performing the discretization operation within the individual, which is beneficial to the gray wolf population to explore the solution space; while in the later stage of the algorithm, the value of a decreases, and at this time there is a greater probability of performing the crossover operation of group collaboration, and the gray wolf population converges to the vicinity of the leader wolf through crossover with the leader wolf.

[0078] Step 4. Selection of the discrete optimization strategy for the task sequence.

[0079] In the elite archive, find the three wolves with the lowest crowding degree among the gray wolf individuals according to the crowding degree of the gray wolf individuals as the leader wolves (α wolf, β wolf, δ wolf); select the gray wolf population iteration strategy according to the current value of a.

[0080] In this embodiment, a grid mechanism is used to calculate the crowding degree of the gray wolf population in the elite archive; as Figure 5 shown, first find the maximum value max and the minimum value min of the non-dominated solutions in each objective function; then divide the interval [min, max] into N equal parts on each objective function dimension, and divide all gray wolf individuals into different grids; for each grid, calculate the number of solutions in the grid and use this number as the crowding degree of each solution in the grid; the more solutions there are in each grid, the greater the crowding degree of the corresponding solution; in the elite archive, the solutions with a larger crowding degree (such as Figure 5 the red area in) will be preferentially removed.

[0081] To solve the defect that the traditional gray wolf algorithm cannot solve discrete problems, the present invention designs a unique discretization operation (that is, the following processing process in step 4) for the discrete coding method of the gray wolf individuals proposed in step 2 to replace the update strategy of the traditional gray wolf algorithm; at the beginning of each iteration, first generate a random number r, r ∈ [0, 1].

[0082] When r ≤ a, perform the discretization operation based on the individual internal, including swap, shift, and flip operations.

[0083] The individual discrete operation is as Figure 6 shown, the swap operation will randomly select two positions in the gray wolf individual for swapping; in the shift operation, first select a sequence segment with a random length, and then randomly select a sequence position in the remaining segment to insert the selected segment; in the flip operation, also select a sequence segment with a random length and reverse the segment order.

[0084] In this embodiment, a roulette wheel-based method is used to select the specific discrete operation. Among them, the swap operation probability P1 = 0.35, the shift operation probability P2 = 0.35, and the flip operation probability P3 = 0.3.

[0085] When the gray wolf individual is updated, the individual discretization operation will cause the number of inspection subsequence tasks of each UAV to change; therefore, the UAV breakpoint position needs to be updated correspondingly, that is, first determine the breakpoint positions at both ends of the longest inspection subsequence task of the UAV; then move the left breakpoint to the right or the right breakpoint to the left to balance the inspection tasks of each UAV.

[0086] Among them, whether to move the left breakpoint to the right or the right breakpoint to the left depends on the position of the longest subsequence in the gray wolf individual.

[0087] When r > a, the crossover operation of group collaboration is performed, and the crossover operation is carried out between the gray wolf individuals and the randomly selected leader wolf. The selection of the leader wolf is based on the preset probability parameter P α , P β , P δ .

[0088] The crossover operation of group collaboration can be further described as follows: First, divide the task sequence of the gray wolf individual according to the breakpoint of the leader wolf to ensure that the gray wolf individual and the leader wolf have task subsequences of the same length; then set the crossover probability c p , and generate N u random numbers c ri between 0 and 1; if the c ri corresponding to the UAV i is greater than the crossover probability c p , then the UAV i subsequence in the new gray wolf individual will inherit the corresponding task sequence of the leader wolf; otherwise, if the c ri corresponding to the UAV i is less than the crossover probability c p , then the UAV subsequence of the gray wolf individual will be retained in the new individual.

[0089] In this embodiment, the selection of the leader wolf is based on the preset probability parameters P α = 0.45, P β = 0.3, P δ = 0.25; the group crossover operation is as Figure 7 shown. First, divide the task sequence of the gray wolf individual according to the breakpoint of the leader wolf to ensure that the gray wolf individual and the leader wolf have task subsequences of the same length. Then set the crossover probability c p = 0.4, and generate N u random numbers c ri . Figure 7 Among them, the c r1 corresponding to the UAV 1 is 0.67, which is greater than the crossover probability, so the UAV 1 subsequence of the new gray wolf individual will inherit the corresponding task sequence of the leader wolf; otherwise, for the c r3 corresponding to the UAV 3 is 0.32, which is less than the crossover probability, so the UAV subsequence of the gray wolf individual will be retained in the new individual.

[0090] After the crossover operation of group collaboration, the task segments of the gray wolf individuals will be exchanged with those of the leader wolf, resulting in possible repetition of task points in the updated gray wolf individuals; therefore, corresponding individual repair operations need to be performed:

[0091] First, determine the inspection task numbers that are repeated with the remaining segments in the segments of the gray wolf individual that have not been exchanged ( Figure 7 repeated task points in it: 2, 6, 3), and find out the missing inspection task numbers (Figure 7 Missing task points: 9, 10, 11); then randomly replace the missing task numbers with duplicate task numbers.

[0092] This step 4 designs a unique discretization operation (intra-individual discretization operation and population crossover operation) for the coding method in step 2, which solves the defect that the traditional continuous grey wolf algorithm cannot handle discrete problems.

[0093] Step 5. Perform the "removal-insertion" operation. After all grey wolf individuals are updated, perform the "removal-insertion" operation on each grey wolf individual; calculate the distance between adjacent task points using the distance matrix Dist, remove the task point with the largest distance, and insert it near the task point closer to it, further optimizing the quality of grey wolf individuals and avoiding the algorithm falling into local optima.

[0094] Such as Figure 8 shown, the "removal-insertion" operation performed on grey wolf individuals is described as follows:

[0095] First, construct a distance matrix Dist with matrix dimension N t ×N t , which stores the distances between task points; then, calculate the distances between adjacent task points of the current grey wolf individual, select the task point A with the largest adjacent distance (such as Figure 8 task point 10 in) and remove it; find the task point B with the smallest distance from this task point in the distance matrix Dist (such as task point 11), insert task point 10 before and after task point 11 respectively, and update the breakpoint positions accordingly; calculate the objective function values of the individual after the two insertions respectively, and select the individual with the smallest objective function value as the new individual to complete the update.

[0096] Step 6. Update the elite archive Archive.

[0097] After the above update operation is completed, re-evaluate the bi-objective function values of all paths and update the elite archive Archive; if the archive is full, eliminate the dominated or overly crowded individuals through non-dominated sorting and crowding degree calculation operations.

[0098] When the elite archive reaches the storage limit, the newly added non-dominated solutions will be compared with the solutions in the archive.

[0099] If a new solution is dominated by at least one solution in the archive, then this new solution cannot enter the archive; if a new solution dominates one or more solutions in the archive, then the dominated solutions must be eliminated and the new solution can enter the archive; if the new solution and the solutions in the archive do not dominate each other, then run the grid mechanism to delete the crowded solutions and store the new solution in the archive.

[0100] Set the maximum number of iterations T, and determine whether the current iteration has reached the maximum number of iterations T; if not, return to step 4, otherwise output the optimal solution set in the elite archive as the final multi-UAV inspection task allocation plan.

[0101] The method of the present invention solves the defect that the traditional continuous multi-objective grey wolf algorithm cannot be applied to discrete problems by designing a unique discretization operation, can quickly generate inspection paths that take into account timeliness and energy consumption balance, can adapt to the characteristics of discrete task allocation problems while ensuring the multi-objective optimization effect, effectively improve the fire monitoring efficiency, reduce the risk of missed inspections and resource waste, and provide efficient decision support for forest fire emergency response, so it has important application prospects.

[0102] Embodiment 2

[0103] This Embodiment 2 describes a computer device. The computer device includes a memory and one or more processors. Executable code is stored in the memory. When the processor executes the executable code, it is used to implement the steps of the multi-UAV forest fire inspection task allocation method in Embodiment 1 above.

[0104] In this embodiment, the computer device is any device or apparatus with data processing capabilities, which will not be elaborated here.

[0105] Embodiment 3

[0106] This Embodiment 3 describes a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, it is used to implement the steps of the multi-UAV forest fire inspection task allocation method in Embodiment 1 above.

[0107] The computer-readable storage medium can be an internal storage unit of any device or apparatus with data processing capabilities, such as a hard disk or memory, or an external storage device of any device with data processing capabilities, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device.

[0108] Of course, the above description is only the preferred embodiments of the present invention. The present invention is not limited to listing the above embodiments. It should be noted that all equivalent substitutions and obvious deformation forms made by any person skilled in the art under the teaching of this specification fall within the substantial scope of this specification and should be protected by the present invention.

Claims

1. A method for task allocation of multi-UAV forest fire inspection, characterized in that Including the following steps: Step 1. Based on the geographical data and fire information of the forest fire area, construct a multi-UAV collaborative inspection task allocation model. The model takes the minimization of the total inspection distance and the minimization of the UAV load difference as the dual objective functions; introduce the map level cost function and the distance cost function of the UAV flying around the terrain into the multi-UAV collaborative inspection task allocation model; The objective function is: Where D represents the total inspection distance, L represents the difference in drone load, and X ij is a binary variable, X ij =1 means the drone will start from the mission point T i Fly to mission point T j Otherwise, X ij =0;C ij Represents the task point T i and T j The distance cost between s is the level evaluation score of the UAV mission sequence, i.e., the map level cost; m k is the number of mission loads of the kth UAV; N t / N u represents the ideal average mission load of each UAV; T c represents the distance cost of the UAV flying around the terrain; d ij Represents the task point T i and T j The Euclidean distance between i ,y i ,z i ) and (x j ,y j ,z j ) are task points T i and T j The three-dimensional coordinates of The constraint conditions are: Equation (4) indicates that the out-degree of each task point T i is 1, representing that each task point can only be connected to one subsequent task point; Equation (5) indicates that the in-degree of each task point T i is 1, representing that each task point is only allowed to be connected by one preceding task point; Equation (6) indicates that the in-degree and out-degree of the starting point T0 are both N u , representing that each drone should start from the starting point and finally return to the starting point; Step 2. Conduct individual coding for grey wolves. For N t patrol tasks, encode the patrol paths of N u UAVs into a task sequence, and generate N u -1 random integers between [1, N t as breakpoints of the task sequence; Step 4. For the gray wolf individual coding, randomly generate multiple groups of different gray wolf individuals; By calculating the dual objective function values of each gray wolf individual, use non-dominated sorting to select the current better path scheme and store it in the initial elite archive; And set the dynamic switching factor a ∈ [0, 1] to control the update strategy of the gray wolf individuals; Step 7. In the elite archive, find the three wolves with the lowest crowding degree as the leader wolves according to the crowding degree of the gray wolf individuals; select the gray wolf population iteration strategy according to the current value of a: for each gray wolf individual, first generate a random number r ∈ [0, 1]; when r ≤ a, perform the discretization operation based on the individual; when r > a, perform the crossover operation of group collaboration; Step 8. Perform the "removal-insertion" operation on each gray wolf individual, calculate the distance between adjacent task points using the distance matrix, remove the task point with the largest distance and insert it near the task point closer to it; Step 9. Update the elite archive; determine whether the maximum number of iterations is reached; if not, return to Step 4, otherwise output the optimal solution set in the elite archive as the final multi-UAV inspection task allocation scheme.

2. The multi-UAV forest fire inspection task allocation method according to claim 1, characterized in that: In the said Step 1, the construction process of the multi-UAV collaborative inspection task allocation model includes: First, digitize the forest map, including terrain height, inspection task point positions, and UAV nest positions; then establish a dual-objective mathematical model for multi-UAV task allocation, including objective functions and constraint conditions.

3. The multi-UAV forest fire inspection task allocation method according to claim 1, characterized in that: In the above step 1, the map level cost T s is calculated as follows: Assume that the nest position is at the center of the map, divide the inspection area into multiple concentric square grids at a preset interval, and the map level increases from the inside to the outside; it is required that the inspection sequence of the UAVs presents a "low-high-low" pattern on the map level; if the path level jump does not conform to this "low-high-low" pattern, a penalty cost is added, as shown in formula (7): where, T rank is the actual map level sequence corresponding to the mission sequence of each UAV; I rank is the ideal map level sequence of each UAV; sum|T rank - I rank | represents the sum of the difference sequences between the actual map level and the ideal map level; Meanwhile, when calculating the distance between computing task points, if there is a mountain blocking between two task points, the UAV makes a detour for an additional distance. Assume that task points T1 and T2 are on both sides of a mountain. Uniformly distributed p n waypoints I are inserted through linear interpolation on the line connecting T1 and T2 pi , where i = 1, 2,..., n; calculate the difference ΔH between the height of each waypoint and the terrain height i , and the formula is as follows: ΔH i = terrain(x i , y i ) - z i ; where (x i , y i , z i ) represents the three-dimensional coordinates of path point I pi ; terrain(x i , y i ) represents the terrain height value corresponding to the coordinates (x i , y i ). Then the UAV's detour cost T c is expressed by formula (8): Convert the mountain height between task points into the additional distance cost of the UAV flying around the mountain.

4. The multi-UAV forest fire inspection task allocation method according to claim 1, characterized in that: In the said Step 3, the specific description of the update strategy of the non-dominated sorting method is: In the multi-objective optimization problem, the dominance relationship between different solutions is described as: For any two solutions x in the population a and x b , if the values of x a in all objective functions are not worse than those of x b , that is f i (x a ) ≤ f i (x b ) ; and x a is strictly better than x b in at least one objective, that is f j (x a ) < f j (x b ) ; then x a is said to dominate x b ; All solutions in the population are divided into multiple front layers according to the dominance relationship. The first layer is a set of non-dominated solutions that are not dominated by any solution, which is called the Pareto front. The solutions in the first layer are preferentially selected to be added to the elite archive. If the number of solutions in the first layer exceeds the upper limit of the archive capacity, further screening is performed based on the crowding distance of the solutions, and the solutions that are sparsely distributed in the objective space are retained to maintain diversity; The calculation method of the dynamic switching factor a is shown in formula (9): a = 1 - (iter / iter max ) 2 (9) where iter represents the current algorithm iteration count; iter max represents the maximum value of the algorithm iteration count.

5. The multi-UAV forest fire inspection task allocation method according to claim 1, characterized in that: In the step 4, the grid mechanism is used to calculate the crowding degree of the gray wolf population in the elite archive; First, find the maximum value max and the minimum value min of each objective function of the non-dominated solutions; then divide the interval [min, max] into N equal parts in each objective function dimension, and divide all gray wolf individuals into different grids; for each grid, calculate the number of solutions in the grid, and use this number as the crowding degree of each solution in the grid; the more solutions there are in each grid, the greater the crowding degree of the corresponding solution.

6. The multi-UAV forest fire inspection task allocation method according to claim 1, characterized in that: In the step 4, when r ≤ a, the discretization operation based on the individual is performed, including swapping, shifting, and flipping; the specific discretization operation is selected based on the roulette wheel method; When the update of the gray wolf individual is completed, the discretization operation of the individual will cause the change in the number of inspection sub-sequence tasks of each UAV; therefore, the breakpoint position of the UAV needs to be updated accordingly, that is, first determine the breakpoint positions at both ends of the longest inspection sub-sequence task of the UAV; then move the left breakpoint to the right or the right breakpoint to the left to balance the inspection tasks of each UAV; When r > a, perform the crossover operation of group collaboration, and perform the crossover operation between the gray wolf individual and the randomly selected leader wolf; the selection of the leader wolf is based on the preset probability parameters P α , P β and P δ ; In the crossover operation of group cooperation, it will cause the repeated task points to appear in the updated gray wolf individuals; the corresponding individual repair operation needs to be performed. The individual repair operation is as follows: first, determine the repeated inspection task numbers in the gray wolf individual, and find the missing inspection task numbers; then randomly replace the missing task numbers with the repeated task numbers.

7. The multi-UAV forest fire inspection task allocation method according to claim 1, characterized in that: The "removal-insertion" operation performed on the gray wolf individual in the step 5 is described as follows: First, construct a distance matrix Dist with dimensions N t ×N t , which stores the distances between task points; Then calculate the distance between adjacent task points of the current gray wolf individual, select the task point A with the largest adjacent distance and remove it; find the task point B with the smallest distance from this task point in the distance matrix Dist, insert the task point A before and after the task point B respectively, and update the breakpoint position accordingly; Calculate the objective function values of the gray wolf individual after the two insertions respectively, and select the gray wolf individual with the smallest objective function value as the new gray wolf individual to complete the update.

8. The multi-UAV forest fire inspection task allocation method according to claim 1, characterized in that: In the step 6, after the update operation is completed, re-evaluate the bi-objective function values of all paths and update the elite archive; if the archive is full, remove the dominated or overly crowded individuals through the non-dominated sorting and crowding degree calculation operations; When the elite archive reaches the storage limit, the newly added non-dominated solutions are compared with the solutions in the archive; if a new solution is dominated by at least one solution in the archive, then this new solution cannot enter the archive; if a new solution dominates one or more solutions in the archive, then the dominated solutions must be removed, and the new solution can enter the archive; if the new solution and the solutions in the archive do not dominate each other, the grid mechanism is run to delete the crowded solutions, and the new solution is stored in the archive.

9. A computer device includes a memory and one or more processors; executable code is stored in the memory; characterized in that, When the processor executes the executable code, it is used to implement the steps of the multi-UAV forest fire inspection task allocation method described in any one of claims 1 to 8 above.

10. A computer-readable storage medium having a program stored thereon; characterized in that, When the program is executed by the processor, it is used to implement the steps of the multi-UAV forest fire inspection task allocation method described in any one of claims 1 to 8 above.

Images