Post-disaster unmanned aerial vehicle task allocation method based on confidence degree multi-target PSO solution

Abstract

The invention discloses a method for solving post-disaster unmanned aerial vehicle rescue task allocation through a confidence-based multi-target particle swarm optimization algorithm, and belongs to the technical field of rescue, and the method comprises the steps: minimizing the task failure number and unmanned aerial vehicle energy consumption under a constraint condition, and constructing an unmanned aerial vehicle rescue task allocation model; a new particle decoding method is designed based on the energy consumption matrix, a particle variation strategy is proposed accordingly, and premature local falling is avoided; a confidence degree based on an emotional infection mechanism is introduced to guide particle speed updating, a particle confidence degree value is calculated according to factors such as task completion conditions, and the state and fitness change, dynamic balance exploration and development ability of adjacent particles are affected. And designing a global extreme value selection mechanism according to a solution set distribution evaluation index, and further improving the distribution of the Pareto optimal solution set obtained by the algorithm. According to the method, the task failure frequency and the energy consumption of the unmanned aerial vehicle can be effectively reduced in post-disaster practical application, and the rescue working efficiency is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of rescue technology, and particularly relates to a post-disaster unmanned aerial vehicle task allocation method based on confidence degree and multi-target PSO solving. BACKGROUND

[0002] Natural disasters and man-made disasters are often unavoidable, especially disasters such as earthquakes, fires and explosions, which have the characteristics of suddenness, randomness and destructiveness, and are difficult to prevent in advance. After the disaster, due to the collapse of buildings and the damage of traffic roads, the rescue work is seriously affected. In addition, the post-disaster scene is complex and may be accompanied by secondary disasters, which will threaten the safety of rescue personnel. How to timely protect the life safety of the disaster-stricken people and quickly provide rescue materials has become a key problem to be solved.

[0003] The traditional emergency rescue method has the disadvantages of slow response and limited rescue range; while the unmanned aerial vehicle can break through the terrain and traffic restrictions, quickly enter the disaster center, and transmit disaster information in real time, helping command decision-making, and its efficient, flexible and safe characteristics effectively make up for the shortcomings of traditional emergency rescue. With the significant improvement of the flight and precise landing capabilities of unmanned aerial vehicles, various types of equipment can be carried to perform reconnaissance, search and rescue, and perform material delivery and other types of tasks, thereby improving the efficiency of emergency rescue.

[0004] The use of unmanned aerial vehicles to transport rescue materials can greatly reduce rescue time and cost. However, with the increasing diversity and complexity of tasks, a single unmanned aerial vehicle is limited in resources and execution capabilities, and a multi-unmanned aerial vehicle system can more effectively cope with various complex tasks due to its higher flexibility and stronger fault tolerance. However, if there is a lack of effective cooperation mechanism, simple combination of unmanned aerial vehicles can easily lead to task conflicts and resource waste. Multi-unmanned aerial vehicle cooperation needs to consider multiple constraints such as environmental complexity, task demand and unmanned aerial vehicle resources, consider the performance of the unmanned aerial vehicle and the matching degree between the unmanned aerial vehicle and the task, and perform global planning and reasonable task allocation to improve the efficiency of task completion.

[0005] In post-disaster rescue, the earlier the rescue is, the higher the survival rate of survivors is. Therefore, when using unmanned aerial vehicles to perform the above task allocation, time constraints, unmanned aerial vehicle range and load constraints need to be considered, so as to improve the efficiency of unmanned aerial vehicle task execution and optimize the allocation result. Existing methods mostly focus on minimizing task completion time or maximizing the success rate of unmanned aerial vehicle task execution, or simply normalize the target, lacking comprehensive consideration of the problem. Since the battery capacity of the unmanned aerial vehicle is limited, if the number of task successes is excessively pursued, the power may be depleted, the unmanned aerial vehicle may not be able to return or even crash, thereby affecting the rescue effect. Therefore, it is necessary to reduce the number of task failures while comprehensively considering the energy consumption of the unmanned aerial vehicle. SUMMARY

[0006] To solve the above technical problems, the present application provides a post-disaster UAV task allocation method based on confidence degree and multi-objective PSO solution, which promotes the quick and effective completion of rescue tasks to at least solve some of the technical problems mentioned in the background art.

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

[0008] The present application provides a post-disaster UAV task allocation method based on confidence degree and multi-objective particle swarm optimization algorithm (CD-MOPSO) solution, which comprises the following steps:

[0009] Based on the deadline of the rescue task and the time required for the UAV to reach the task, the minimum task failure number objective function is determined;

[0010] Based on the weight of the material that the UAV can carry and the operating power of the UAV, the minimum UAV energy consumption objective function is determined;

[0011] According to the minimum task failure number objective function and the minimum UAV energy consumption objective function, a post-disaster UAV task allocation mathematical model is constructed;

[0012] Further, the minimum task failure number objective function is expressed as:

[0013]

[0014] Wherein, F1 is the number of failed rescue tasks, H(·) is the Heaviside step function, which is 1 when and 0 when That is, when the UAV arrival time exceeds the deadline, the number of failed tasks is increased by 1. Δt j is the deadline of task j (target point), that is, the remaining survival time of trapped person j, is the total flight time of UAV k to reach task j.

[0015] Further, the flight time of each UAV to reach the rescue task corresponds to the acquisition step, which comprises:

[0016]

[0017] Wherein, is the decision variable, representing the UAV k from the i-th task point to the j-th task point; is the time used by the k-th UAV from the target point i to the target point j;

[0018] Based on the UAV speed and the distance between the two tasks, the time of the UAV to reach the rescue task is acquired, which is expressed as:

[0019]

[0020] wherein, denotes the Euclidean distance between target point i and j, and v is the flight speed of the UAV.

[0021] Further, the minimum energy consumption objective function of the UAV is represented as:

[0022]

[0023] wherein, F2 is the energy consumption sum of all UAVs, is a decision variable, representing the UAV k from the i-th target point to the j-th target point; is the time used by the k-th UAV from the target point i to the target point j, is the running power of the UAV k after weight update between the target point i and the target point j.

[0024] Based on the UAV weight and the UAV rotor and rotor rotating area, the running power of the UAV from one target point to the rescue task required is obtained, represented as:

[0025]

[0026] wherein, is the weight of the UAV k before reaching the target j, g is the gravitational constant, p is the air density, ζ is the UAV rotor rotating area, and n0 is the number of UAV rotors.

[0027] Based on the weight of the materials required by the target point, the weight change of the UAV between two target points is obtained, represented as:

[0028]

[0029] wherein, w i is the weight of the materials required by the target point i.

[0030] Further, the decoding strategy of the confidence-based multi-target particle swarm optimization algorithm specifically includes:

[0031] For any particle X y =[x y1 ,x y2 ,...,x yn ], perform decoding operation to obtain task allocation matrix, sort the elements in the particle from large to small, and return the corresponding original position index (i.e. task number), to obtain a matrix X' y =[x' y1 ,x' y2 ,…,x' yn ] with interval [1, n] size, calculate the energy consumption E between any two tasksij :

[0032]

[0033] where E ij represents the energy consumption from base to the target point, i≠j ij represents the energy consumption between target points i, j. The energy consumption matrix {[E 11 ,E 12 ,…,E 1n ],…,[E n1 ,E n2 ,…,E nn ]} T .

[0034] Each m elements in X' y are divided into a group, then the following can be obtained The first group is assigned to the m unmanned aerial vehicles in turn, and the subsequent groups are matched according to the remaining time of the target point and the unmanned aerial vehicle. The matching adopts a greedy algorithm, and the target points with the smallest energy consumption difference and meeting the material demand in the previous group task are preferentially selected by referring to the energy consumption matrix. The assignment of all tasks is completed in turn. The task assignment matrix R is obtained.

[0035] Further, the multi-objective particle swarm optimization algorithm based on confidence degree is used to solve the unmanned aerial vehicle post-disaster rescue task allocation model, and specifically includes:

[0036] The velocity of the particle is updated based on the confidence degree to obtain a new solution;

[0037] The mutation operation is performed on different particles, including crossover, intra-group task exchange and single-point mutation operation;

[0038] Based on the crowding distance and diversity between particles, the external reserve set and global extreme value of the particle are updated.

[0039] Further, the post-disaster unmanned aerial vehicle task allocation method solved by the multi-objective particle swarm optimization algorithm based on confidence degree obtains a particle update formula of confidence degree emotional contagion, which is expressed as:

[0040] v y (t+1)=ω(cd y )v y (t)+r1c1(cd y )(p best (t)-x y (t))+r2c2(cd y )(p gbest (t)-x y (t))

[0041] x y(t+1) = x y (t) + v y (t+1)

[0042] where ω (cd y ), c1 (cd y ) and c2 (cd y ) are automatically selected according to Table 1.

[0043] Table 1 Particle switching search strategy considering confidence

[0044]

[0045] Based on the fitness gain, confidence inertia and emotional contagion coefficient, the confidence calculation method is represented as:

[0046]

[0047] where λ ∈ (0, 1) represents the confidence retention coefficient, that is, the inertia of confidence; η is proportional to the fitness gain Δf y (t), increasing information when the fitness is improved, and decreasing when the fitness is reduced, Δf y (t) = f (pbest y (t)) - f (pbest y (t-1)), where f is a multi-objective aggregation index; μ is an emotional contagion coefficient; represents the adjacent set of particles.

[0048] Based on the individual optimal value of the particle, the multi-objective aggregation index is obtained, which is represented as:

[0049] Δf y (t) = F1 (pbest y (t)) - f (pbest y (t-1)) + F2 (pbest y (t)) - f (pbest y (t-1))

[0050] where Δf y (t) represents the fitness gain, and pbest y represents the optimal extreme value of the particle y.

[0051] Based on the path similarity of the rescue task allocation, the adjacent set of selected particles is obtained, and the similarity between particles y and q is calculated using the Jaccard index, which is represented as:

[0052]

[0053] where s yqSimilarity of particle y and particle q, P y = {P y1 , P y2 , …, P yn} and P q = {P q1 , P q2 , …, P qn} respectively represent the node order of the rescue task allocation path of particles y and q.

[0054] Set a similarity threshold s min , only select particles s yq ≥ s min to form an adjacent set to ensure the effectiveness of the adjacent relationship. Generally, in order to balance the information exchange of the population and the calculation efficiency, the number of sets is set to 3-5, if the current population diversity is low, the number is reduced to 3; if the diversity is high, the number is increased to 5.

[0055] Based on the size relationship of the confidence degree, the emotion contagion coefficient is set, that is, the neighbor with high confidence degree is allowed to infect the emotion to the particle with low confidence degree, which is represented as:

[0056]

[0057] Wherein, μ0∈[0.3, 0.4]. is the set of particles adjacent to particle y, and the emotion influence is realized through the adjacent relationship in .

[0058] Further, the mutation operation performed specifically includes:

[0059] Cross operation: divided into cross between different particles and single particle u-opt operation: the cross between different particles is to select two particles and cross part of the segment. The cross of a single particle is to set the value of u according to the size of the problem, u∈[2, 5], select the particle to perform u-opt operation, the main steps are to randomly divide the particle into u segments, and recombine these segments. By comparing the objective function value of the combined particle with the dominance relationship of the original particle, it is determined whether to update the particle.

[0060] Group task exchange: based on the decoding strategy proposed in the present application, the task allocation result is mainly affected by the group where the task is located. By adjusting the group where the task is located to change the task allocation result, the task is re-allocated, so as to change the task allocation result.

[0061] Single-point mutation: randomly select a dimension e in particle x y , apply Gaussian disturbance in the value interval range of the particle, and the update formula is as follows:

[0062]

[0063] where x max and x min are the maximum and minimum of the particle value interval, respectively. This operation is mainly aimed at the dominated solution with poor fitness value to ensure the convergence of particles while increasing the diversity of solutions.

[0064] Further, the updating of the particle extreme value and the external reserve set based on the crowding distance and diversity between particles comprises:

[0065] Extreme value updating: based on the crowding distance and diversity index, the updating mode of individual extreme value and global extreme value is obtained, including: selecting the non-dominated solution of the particle in the evolution process as the individual extreme value; setting an adjustment probability, and using a selection method combining crowding degree and random selection to obtain the global extreme value, and the adjustment probability calculation mode is represented as:

[0066]

[0067] where P adj is the adjustment probability, α = β = 0.5, Δ max = 1, it is the current iteration number, and IT is the total iteration number. P adj gradually decreases with the increase of the iteration number. When P adj is less than a random number uniformly distributed on [0, 1], the particle with the maximum crowding degree is selected as the global extreme value; otherwise, a particle is randomly selected from the Pareto solution set as the global extreme value. This strategy adjusts the global extreme value selection mode according to the probability, which helps to make the solution set of the non-dominated front more uniform.

[0068] Based on the objective function value of the particle, the calculation mode of the particle crowding degree is obtained, which is represented as:

[0069]

[0070] where F r (y-1) and F r (y+1) are the values of the two individuals adjacent to individual y on the rth objective function, and F rmax and F rmin represent the maximum and minimum values on the rth objective, respectively.

[0071] Based on the Euclidean distance between non-dominated solutions, the calculation mode of the diversity evaluation index Δ is obtained, which is represented as:

[0072]

[0073] where N represents the number of current non-dominated solutions, and dh denotes the Euclidean distance between two neighboring individuals in the non-dominated solution set, d f and d l are the Euclidean distances between the leftmost solution and the boundary solution, and the rightmost solution and the boundary solution in the non-dominated solution set, respectively, denotes the average distance, and the smaller the value of Δ, the better the distribution and diversity of the non-dominated solution.

[0074] External reserve set update: when the number of non-dominated solutions exceeds the maximum capacity of the reserve set, the difference between the newly generated non-dominated particle and the particle in the external reserve set needs to be compared. When the objective function values are different, the non-dominated solution with the smallest crowding degree is selected for replacement; if the objective function values are equal, the difference Dis(y, q) of the particle X y and X q after grouping is compared for screening.

[0075] Based on the task grouping information, the difference between the two particles is obtained, which is denoted as:

[0076]

[0077] According to the technical solutions described above, compared with the prior art, the present application discloses a post-disaster unmanned aerial vehicle task allocation method based on confidence degree multi-objective PSO solution, which has the following beneficial effects:

[0078] The present application takes the minimization of task failure number and unmanned aerial vehicle energy consumption as evaluation indexes, establishes a multi-unmanned aerial vehicle rescue task allocation model in a post-disaster environment, which can effectively solve the problem of multi-unmanned aerial vehicle multi-task allocation, so as to improve the efficiency of post-disaster rescue work.

[0079] The confidence degree multi-objective particle swarm algorithm based on the emotional contagion mechanism provided by the present application proposes a decoding method that takes into account the number of trapped personnel and the energy consumption of unmanned aerial vehicles, updates the particle speed guided by the confidence degree, gives the updating method of the particle extreme value and the external reserve set, and mutates the particles. The algorithm not only can obtain a better solution in a complex post-disaster scene, but also can dynamically balance the exploration and exploitation capabilities of the algorithm. BRIEF DESCRIPTION OF DRAWINGS

[0080] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of the present application illustrated in the drawings and their descriptions are used to explain the present application and do not limit the present application. In the drawings:

[0081] Figure 1 is a schematic diagram of unmanned aerial vehicle task allocation;

[0082] Figure 2 is a flowchart of a multi-objective particle swarm algorithm;

[0083] Figure 3 schematic diagram for decoding particles;

[0084] Figure 4 schematic diagram for particle mutation strategy instance particles;

[0085] Figure 5 schematic diagram for difference comparison instance;

[0086] Figure 6 schematic diagram for distribution of pareto solution set obtained by contrast algorithm;

[0087] Figure 7 schematic diagram for specific task allocation scheme of unmanned aerial vehicle obtained by improved algorithm. DETAILED DESCRIPTION

[0088] The technical solutions in the embodiments of the present application will be apparently and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0089] Referring to Figure 1 The embodiment of the present application discloses a confidence-based multi-objective particle swarm optimization algorithm for unmanned aerial vehicle rescue task allocation problem, comprising the following steps:

[0090] Based on the deadline of the rescue task and the time required for the unmanned aerial vehicle to reach the task, a target function of minimizing the number of task failures is determined;

[0091] Based on the weight of the material that can be carried by the unmanned aerial vehicle and the operating power of the unmanned aerial vehicle, a target function of minimizing the energy consumption of the unmanned aerial vehicle is determined;

[0092] According to the target function of minimizing the number of task failures and the target function of minimizing the energy consumption of the unmanned aerial vehicle, a mathematical model of post-disaster task allocation of the unmanned aerial vehicle is constructed.

[0093] Next, the above content will be described in detail.

[0094] I. Problem and environment description

[0095] In order to effectively solve the above problem, without loss of generality, in order to simplify the construction of the model, the following assumptions are given:

[0096] 1. The flight speed of the unmanned aerial vehicle remains constant;

[0097] 2. The position, survival time and resource demand information of the personnel to be rescued are known;

[0098] 3. The unmanned aerial vehicles are of the same model and have fixed capacity.

[0099] Based on the above assumptions, the problem to be solved by the present application is as follows: in a post-disaster scenario, the available resources of the unmanned aerial vehicle are limited, m unmanned aerial vehicles (numbered 1 to m) all start from the same location, and need to complete the rescue tasks of n targets to be rescued. The personnel to be rescued are randomly distributed in the disaster area, and their life conditions are different. Different types of material rescue are needed due to different degrees of injury. In addition, the unmanned aerial vehicle is limited by energy consumption and endurance during flight, and needs to be based on the actual needs of the rescue personnel to minimize the number of failures with the least energy consumption.

[0100] As shown in Figure 1 , there are three unmanned aerial vehicles and eight target points to be rescued in the post-disaster environment. The three unmanned aerial vehicles start from the same starting point (rescue center) and return to the starting point after completing the rescue. The rescue sequence of the unmanned aerial vehicles in the figure is as follows: UAV1 performs T1→T6→T4, UAV2 performs T2→T7→T5, and UAV3 performs T8→T3.

[0101] Table 1 shows the specific symbols involved in the model and the corresponding explanations.

[0102] Table 1 conforms to the definition

[0103]

[0104]

[0105] II. Model establishment

[0106] 2.1 Objective function

[0107] The initial position of the unmanned aerial vehicle is H0(X0, Y0), and the flight speed is a constant v. The rescue task allocation matrix R is established m×W , which represents the rescue sequence of m unmanned aerial vehicles, where each column of the matrix corresponds to the length of the rescue task sequence When the number of tasks cannot be evenly distributed, the empty positions are filled with 0 to ensure the integrity of the matrix. The matrix form is as follows:

[0108]

[0109] where r ki represents the i-th rescue task performed by the k-th unmanned aerial vehicle, 1≤k≤m, 1≤i≤W.

[0110] It is easy to know that if the time of the UAV reaching the trapped person is less than the survival time of the trapped person, the rescue is successful, otherwise the rescue fails. And in the process of executing the task, the flight cost and the flight energy consumption of the UAV are closely related. With the completion of the task, the load of the UAV changes constantly, and the UAV needs to maintain a constant speed all the time, so the energy consumption of the UAV is also changing constantly. Therefore, the flight energy consumption of the UAV cannot be simply determined by the length of the flight path, and the energy consumption of the UAV needs to be calculated according to the change of the load.

[0111] A mathematical model of UAV cooperative task allocation is constructed with the number of failed rescue tasks and the energy consumption of the UAV as the optimization objective function:

[0112]

[0113] Wherein, F1 is the number of failed rescue tasks, H(·) is the Heaviside step function, which is 1 when , and 0 when . That is, when the arrival time of the UAV exceeds the deadline, the number of failed tasks is increased by 1. Δt j is the deadline of task j (target point), that is, the remaining survival time of the trapped person j, is the total flight time of the UAV k when reaching the target point j. F2 is the sum of the energy consumption of all UAVs, E k is the energy consumption of the kth UAV.

[0114] The total flight time of the UAV k when reaching the target point j in F1 is The calculation formula is as follows:

[0115]

[0116] Wherein, is the decision variable, which represents the UAV k from the i th target point to the j th target point; is the time used by the kth UAV from the target point i to the target point j;

[0117] Based on the speed of the UAV and the distance between the two tasks, the time of the UAV from one target point to the rescue task is obtained, which is represented as:

[0118]

[0119] Wherein, represents the Euclidean distance from the target point i to j, and v is the flight speed of the UAV. The energy consumption E k of the kth UAV in F2 is calculated as:

[0120]

[0121] Wherein, The running power of the UAV k after weight update between the target point i and the target point j is calculated as follows:

[0122]

[0123] wherein, is the weight of the UAV k after unloading the supplies at the target point i, g is the gravitational constant, p is the air density, z is the rotor rotation area of the UAV, and n0 is the number of rotors of the UAV.

[0124] is the weight change of the UAV k between the i-th target point and the j-th target point, and is calculated as follows:

[0125]

[0126] wherein, w i is the weight of the supplies required at the target point i.

[0127] 2.2 Constraint conditions

[0128] In order to consider the actual rescue, the UAV load and the rescue distance have certain limitations. In order to make the model more realistic and maximize the use of UAV resources, the following constraints are given:

[0129] 1) Each target point can only be completed by one UAV:

[0130]

[0131] 2) The energy consumption E consumed by the UAV k during flight k is less than the maximum available energy E0 of the UAV:

[0132] E k <E0 (8)

[0133] 3) The total weight M of the UAV k performing the task k does not exceed the maximum load weight M UAV of the UAV:

[0134] M k ≤M UAV (9)

[0135] 4) Each UAV k starts from the rescue center:

[0136]

[0137] 5) Each UAV k returns to the rescue center after completing all tasks:

[0138]

[0139] III. Algorithm solution

[0140] The improved multi-objective particle swarm optimization algorithm is used for solving the present application, and the solving process of the algorithm is as shown in Figure 2 , specifically:

[0141] Step 1: initialize particle position, velocity, individual optimum and global optimum, and initialize particle confidence and adjacency relationship at the same time;

[0142] Step 2: decode and calculate fitness, update individual optimum, global optimum, mutation probability and Pareto optimum;

[0143] Step 3: for each particle, the following operations are performed;

[0144] Step 3.1: calculate the confidence according to the fitness change and the neighbor, and determine the inertia weight and the learning factor according to the confidence;

[0145] Step 3.2: update the position and velocity of the particle;

[0146] Step 3.3: adaptive mutation of the particle, roulette selection of mutation strategy;

[0147] Step 4: calculate the fitness value and sort the particles according to non-dominance and compare the crowding degree;

[0148] Step 5: update the individual optimum, global optimum and external storage set;

[0149] Step 6: screen the solutions meeting the constraints;

[0150] Step 7: judge whether the algorithm meets the ending condition, if yes, the algorithm ends and the external storage set is output; otherwise, return to step 3.

[0151] 3.1 Particle encoding and decoding

[0152] The present application adopts real number encoding method, and each particle represents a problem solution. When decoding, the rescue time and the unmanned aerial vehicle energy consumption are comprehensively considered: the earlier the arrival time, the greater the probability of successful rescue; the smaller the energy consumption, the more target points the unmanned aerial vehicle can execute. Assuming that any particle is represented as X y =[x y1 ,x y2 ,...,x yn ], the decoding method is as follows:

[0153] ·Sort the particle elements from large to small, and return the corresponding original position index (i.e. task number) to obtain a matrix X' y =[x' y1 ,x' y2 ,…,x' ynEijdenotes the energy consumption between any two tasks ij :

[0154]

[0155] where E ij denotes the energy consumption from base to the target point, i≠j ij denotes the energy consumption between target points i, j. The energy consumption matrix {[E 11 ,E 12 ,…,E 1n ],…,[E n1 ,E n2 ,…,E nn ]} T .

[0156] · Each m (number of UAVs) elements in X y are grouped into a set, and the matrix is obtained:

[0157]

[0158] · The first group is assigned to the m UAVs in order, and the subsequent groups are matched according to the remaining time of the target point and the UAV. The matching uses a greedy algorithm, which refers to the energy consumption matrix and preferentially selects the target point with the smallest energy consumption difference in the previous group of tasks and meets the material demand. Among them,

[0159] The energy consumption matrix E ij is based on the energy consumption difference between each target point, which is calculated as shown in equation (12):

[0160] To illustrate the decoding process in detail, assume that there are 3 UAVs (m = 3) and 7 target points (T1, T2, …, T7). Assume that the algorithm solves the particle X y is X y = [19, 22, 5, 61, 38, 10, 47], and after sorting and updating the particle, X y = [4, 7, 5, 2, 1, 6, 3], the elements in X y are grouped to obtain

[0161] The first group is assigned to the three UAVs in turn: UAV1 is assigned task T4, UAV2 is assigned task T7, and UAV3 is assigned task T5. The second group is matched according to the urgency of the remaining time of the task, starting from the target point T2 with the shortest time, and combined with the energy consumption matrix, the target point with the smallest energy consumption difference in the first group of tasks is selected for matching. Assume that E 42 <E 52 <E 72Then task 2 is assigned to UAV2, and then target point T6 is matched in the remaining UAV1 and UAV3. All task assignments are completed in turn to obtain a task assignment matrix:

[0162]

[0163] Finally, the task execution order is: UAV1: T4→T1, UAV2: T7→T2, UAV3: T5→T6→T3. The particle decoding schematic diagram is as shown in Figure 3 .

[0164] 3.2 Particle mutation operation

[0165] In the process of exploring the solution space, in order to avoid falling into a local optimal solution, three particle local search strategies are designed in combination with the constructed mathematical model and decoding mode.

[0166] (1) Cross operation: including cross between different particles and u-opt operation of a single particle.

[0167] • Cross between different particles: randomly select two particles, exchange part of the segment, realize particle recombination, as shown in Figure 4 (a).

[0168] • U-opt cross operation of a single particle: randomly divide the particle into u segments, u∈[2, 5], recombine these segments. By comparing the target function value of the combined particle with the dominance relation of the original particle, it is determined whether to update the particle, as shown in Figure 4 (b).

[0169] (2) Group task exchange

[0170] Based on the decoding strategy proposed in the application, the task assignment result is mainly affected by the group where the task is located. By adjusting the group where the task is located to change the task assignment result, the task is re-assigned, as shown in Figure 4 (c).

[0171] (3) Single-point mutation

[0172] Randomly select a dimension e in the particle x y , and apply Gaussian disturbance in the value interval range of the particle, and its update is shown in formula (13):

[0173]

[0174] Where, x max and x min are the maximum and minimum values of the particle value interval. This operation is mainly aimed at the dominated solution with poor fitness value, so as to ensure the convergence of the particle and increase the diversity of the solution.

[0175] The initial selection probability of the four variation modes is set as s1=s2=s3=s4=0.25, the selection probability is adaptively adjusted according to the number of successful variations, i.e. whether the solution after variation is superior, and selection is performed through roulette gambling, and the higher the probability, the greater the selection probability.

[0176] 3.3 Reserve set update

[0177] The selection of global optimal particles is a key link in multi-objective optimization, and MOPSO highly depends on the storage of excellent particles. Generally, an external reserve set is used for storage, and is updated by increasing or deleting to keep a fixed size. In the present application, the crowding distance calculation is used to calculate the crowding degree D of each particle in the non-dominated front:

[0178]

[0179] wherein F r (y-1), F r (y+1) are the values of the rth objective function of the two individuals adjacent to the individual y, F rmax and F rmin represent the maximum value and the minimum value on the rth objective, respectively.

[0180] The diversity evaluation index Δ is used to measure the uniformity of the distribution of non-dominated solutions, and the calculation method is shown in formula (15):

[0181]

[0182] wherein N represents the number of current non-dominated solutions, d h represents the Euclidean distance between two adjacent solutions in the non-dominated solution set, d f and d l are the Euclidean distances between the leftmost solution and the boundary solution and the rightmost solution and the boundary solution in the non-dominated solution set, respectively, and d is the average distance. If Δ=0, it indicates that all solutions in the Pareto optimal solution set are distributed at equal intervals. Generally, the smaller the value of Δ, the better the distribution and diversity of non-dominated solutions.

[0183] In order to enhance the diversity of solutions, when the number of non-dominated solutions exceeds the maximum capacity of the reserve set, the difference between the newly generated non-dominated particle and the particles in the external reserve set needs to be compared. If the objective function values are different, the non-dominated solution with the smallest crowding degree is selected for replacement; if the objective function values are equal, the difference degree Dis(y,q) of the particles X y and X q after grouping is compared according to formula (16) for screening.

[0184]

[0185] For example,Figure 5 Non-dominated solution X' y = {[1, 4, 6], [2, 7, 5], [3, 8]}, X' q = {[1, 4, 7], [2, 5, 6], [3, 8]}, the objective function value is the same, the difference degree Dis(y, q) = 2, the task allocation result is different, and it is replaced.

[0186] 3.4 Global extreme value update

[0187] Based on the diversity evaluation index in section 3.3, the global extreme value is selected by using the congestion degree and random selection switching mode. First, the adjustment probability P adj of the global extreme value selection mode is generated according to the diversity index, which is specifically shown in formula (17):

[0188]

[0189] Wherein, α = β = 0.5, Δ max = 1, it is the current iteration number, and IT is the total iteration number. P adj decreases gradually with the increase of iteration number. When P adj is less than a random number uniformly distributed on [0, 1], the particle with the largest congestion degree is selected as the global extreme value; otherwise, a particle is randomly selected from the Pareto solution set as the global extreme value. This strategy adjusts the global extreme value selection mode according to the probability, which helps to make the solution set of the non-dominated front more uniform.

[0190] 3.5 "confidence" driven particle update formula

[0191] Referring to social psychology and multi-agent system theory, the application considers the behavior difference and collaborative optimization of particles under different "confidence", and the particle update is affected by the emotional feedback of "confidence", forming the heterogeneous search behavior in the group. The specific content is as follows:

[0192] Each particle carries a "confidence" variable, which is dynamically adjusted by factors such as task completion, environmental complexity and historical success rate; particles with high confidence tend to conduct local fine search, and particles with low confidence tend to conduct large-scale exploration. The confidence is periodically updated based on particle task completion feedback and group performance. At the same time, the "emotional contagion" mechanism is introduced, and adjacent particles can affect each other's confidence, thereby enhancing the group synergy effect. Under different confidence states, the inertia weight and learning factor of the particle will be adjusted respectively, and the emotional threshold is designed to realize automatic switching of behavior mode, which is specifically manifested as fine approximation of high-confidence particles to good solutions and increase of random disturbance of low-confidence particles to expand the search range.

[0193] To realize the above mechanism, the application expands the traditional particle swarm structure by introducing the concept of "confidence degree", increases the confidence degree attribute, and dynamically adjusts the confidence degree according to the search effect of particles and the group performance. The specific calculation of the confidence degree is shown in formula (18):

[0194]

[0195] Wherein, λ ∈ (0, 1) represents a confidence retention coefficient, that is, the inertia of confidence; η and the fitness gain Δf y (t) are proportional, the confidence degree is increased when the fitness is improved, and is reduced when the fitness is reduced, Δf y (t) = f (pbest y (t)) - f (pbest y (t-1)), wherein f is a multi-objective aggregation index; μ is an emotional contagion coefficient, which is set based on the confidence degree, that is, the neighbors with high confidence degree are allowed to infect the particles with low confidence degree, and the specific formula (19) is as follows:

[0196]

[0197] Wherein, μ0 ∈ [0.3, 0.4]. The set of particles adjacent to the particle y is realized through the adjacent relationship in According to the path similarity of the rescue task allocation, the adjacent set of particles is selected, and it is assumed that the node order of the rescue task allocation path of particles y and q is represented as P y ={P y1 ,P y2 ,…,P yn} and P q ={P q1 ,P q2 ,…,P qn}, the similarity of particles y and q is calculated by using Jaccard index, and the specific formula (20) is as follows:

[0198]

[0199] For each particle y, the similarity s yq of the particle y with all other particles is calculated, the similarity threshold s min is set, and only the particles with s yq ≥ s min are selected to form the adjacent set to ensure the effectiveness of the adjacent relationship. Generally, in order to balance the population information exchange and the calculation efficiency, the number of the set is set to 3-5, if the current population diversity is low, the number is reduced to 3; if the diversity is high, the number is increased to 5.

[0200] According to the current confidence level, the particle switching search strategy is shown in Table 2:

[0201] Table 2 Particle switching search strategy considering confidence

[0202]

[0203] The particle velocity and position update formula designed based on this is shown in (21):

[0204]

[0205] Among them, ω(cd y ), c1(cd y ) and c2(cd y ) is automatically selected according to the confidence level in Table 2.

[0206] 4. Simulation Experiment

[0207] Consider the following scenario: Target point locations and mission durations are generated within a 3.5 km x 3.5 km area, including n target points (n = 11, 18, 25, 35). There are m drones participating in the rescue (m = 2, 3, 4, 6). The drones are required to depart from the origin (rescue center), complete all tasks in sequence, and then return to the origin. The target point's material requirements are set to three levels based on the duration: 1 kg, 2 kg, and 3 kg. The maximum payload of the drone is 15 kg. See Table 3 for details of the initial parameters.

[0208] Table 3 Initial values ​​of parameters

[0209]

[0210] The Pareto solution obtained through simulation experiments is as follows Figure 6 As shown in the figure, the specific plan for the UAV to perform the mission is as follows Figure 7 The figure shows key information such as the specific tasks and execution order of each drone. These detailed task allocation results will further verify the effectiveness and practicality of the improved CD-MOPSO algorithm in practical applications.

[0211] Table 4 HV values ​​of the compared algorithms in different scenarios

[0212]

[0213]

[0214] It can be seen from the comparison results of the HV values of different algorithms that the improved algorithm has the best result. Compared with the traditional MOPSO algorithm, the improved CD-MOPSO algorithm combines particle confidence and different local search strategies, and the particles with high confidence can be fully explored to obtain a better PF. The traditional MOPSO uses a roulette method to select the global best particle and maintains an external archive set. The traditional algorithm mainly targets the objective value of the particle, and in fact, the task allocation results may differ for the same objective value. Finding different task allocation results can help the particle explore more solutions. Similarly, the comparison algorithm MOPGA mainly proposes different mutation strategies. This method mutates particles during the evolution process, but may not generate new task allocation schemes. Therefore, the solution set obtained by the improved algorithm is closer to the real PF. Compared with the NSGA-II algorithm, the decoding and mutation strategies of the particles are the same during the experiment, but all particles participate in crossover and mutation at each iteration of the NSGA-II algorithm. However, the crossover and mutation are all new solutions generated randomly. Compared with the improved algorithm, the efficiency of the NSGA-II algorithm is lower because the probability of using different strategies is adjusted according to the different effects of the strategies. The MOPSO algorithm continuously learns from the global best particle during the evolution process, which can quickly generate solutions and produce more possibilities. Therefore, as can be seen from Table 4, the average HV value of the improved algorithm is the best in different scenarios.

[0215] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other.

[0216] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined in the present application can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown in the present application, but will conform to the widest scope consistent with the principles and novel features disclosed in the present application.

Claims

1. A post-disaster UAV task allocation method based on a multi-objective particle swarm optimization algorithm based on confidence is characterized by: The steps include: Based on the rescue mission deadline and the time required for the drone to reach the mission, determine the objective function to minimize the number of mission failures; Based on the weight of the materials that the drone can carry and the operating power of the drone, determine the objective function to minimize the energy consumption of the drone; Constructing a mathematical model for post-disaster UAV task allocation based on the objective function of minimizing the number of mission failures and the objective function of minimizing UAV energy consumption; A multi-objective particle swarm optimization algorithm based on confidence is used to solve the mathematical model of the UAV post-disaster rescue task allocation, and the UAV rescue tasks are allocated based on the solution results.

2. The post-disaster UAV task allocation method based on the confidence-based multi-objective particle swarm optimization algorithm according to claim 1 is characterized in that: The objective function of minimizing the number of task failures is expressed as: Where F1 is the number of rescue mission failures, H(·) is the Heaviside step function, and when When 1, When the UAV arrival time is greater than the deadline, the mission failure number increases by 1. j is the deadline of task j (target point), that is, the remaining survival time of trapped person j, is the total flight time for UAV k to reach task j.

3. The post-disaster UAV task allocation method based on the confidence-based multi-objective particle swarm optimization algorithm according to claim 1 is characterized in that: The total time taken by the drone to complete all tasks is obtained by: Based on the UAV flight speed, the flight time of each UAV to reach the rescue mission is obtained; it is expressed as: in, is the decision variable, indicating that UAV k moves from the i-th mission point to the j-th mission point; is the time taken by the kth UAV to travel from target point i to target point j; Based on the speed of the drone and the distance between the two tasks, the time it takes for the drone to arrive at the rescue task is obtained, which is expressed as: in, represents the Euclidean distance from target point i to j, and v is the flight speed of the drone.

4. The post-disaster UAV task allocation method based on the confidence-based multi-objective particle swarm optimization algorithm according to claim 1 is characterized in that: The objective function of minimizing the energy consumption of the UAV is expressed as: Among them, F2 is the energy consumption of all drones, is the decision variable, indicating that UAV k moves from the i-th target point to the j-th target point; is the time taken by the kth UAV to travel from target point i to target point j, is the operating power of UAV k after weight update from target point i to target point j. Based on the weight of the UAV and the rotating area of ​​the UAV rotor and rotor blades, the operating power of the UAV after moving from one target point to the next target point is obtained, which is expressed as: in, is the weight of UAV k before it reaches the target point j, g is the gravity constant, ρ is the air density, ζ is the rotating area of ​​the UAV rotor, and n0 is the number of UAV rotors. Based on the weight of the materials required at the target point, the weight change of the drone between the two target points is obtained, which is expressed as: Among them, w i is the weight of materials required for target point i.

5. The post-disaster UAV task allocation method based on the confidence-based multi-objective particle swarm optimization algorithm according to claim 1 is characterized in that: The encoding and decoding strategies of the multi-objective particle swarm algorithm specifically include: For any particle X y =[x y1 ,x y2 ,...,x yn ]Perform the decoding operation to obtain the task allocation matrix: Step 5.1: Sort the elements in the particle from large to small and return the corresponding original position index (i.e. task number) to obtain a matrix X' with a size interval of [1,n] y =[x' y1 ,x' y2 ,…,x' yn ], calculate the energy consumption E between any two tasks ij : Among them, when i=j, E ij Indicates the energy consumption from the base to the target point, when i≠j, E ij Represents the energy consumption between target points i and j. The energy consumption matrix {[E 11 ,E 12 ,…,E 1n ],…,[E n1 ,E n2 ,…,E nn ]} T . Step 5.2: Place X' y If every m elements in the _{\mathbf { ... Step 5.3: Assign the first group to the m drones in order. Subsequent groups are matched to drones based on the remaining time to their target points. This matching algorithm uses a greedy algorithm, referencing the energy consumption matrix, to prioritize the target points that minimize energy consumption and meet the material requirements of the previous group of tasks. This completes the task assignment process, yielding the task assignment matrix R.

6. The post-disaster UAV task allocation method based on the confidence-based multi-objective particle swarm optimization algorithm according to claim 1 is characterized in that: The confidence-based multi-objective particle swarm optimization algorithm solves the UAV post-disaster rescue task allocation model, specifically including: Update the particle velocity based on the confidence level to obtain a new solution; Perform mutation operations on different particles, including crossover, intra-group task exchange, and single-point mutation operations; Based on the crowding distance and diversity between particles, the external reserve set and global extreme value of particles are updated.

7. The post-disaster UAV task allocation method based on the confidence-based multi-objective particle swarm optimization algorithm according to claim 1 is characterized in that: The particle update formula influenced by confidence sentiment is expressed as: v y (t+1)=ω(cd y )v y (t)+r1c1(cd y )(p best (t)-x y (t))+r2c2(cd y )(p gbest (t)-x y (t)) x y (t+1)=x y (t)+v y (t+1) Among them, ω(cd y ), c1(cd y ) and c2(cd y ) is automatically selected according to the confidence level in Table 1. Table 1 Particle switching search strategy considering confidence Based on fitness gain, confidence inertia, and emotional contagion coefficient, the "confidence" calculation method is expressed as: Among them, λ∈(0,1) represents the confidence retention coefficient, that is, the inertia of confidence; η is related to the fitness gain Δf y (t), confidence increases when fitness improves and decreases when fitness decreases, Δf y (t) = f(pbest y (t))-f(pbest y (t-1)), where f is the multi-target aggregation index; μ is the emotional contagion coefficient; represents the neighboring set of particle y. Based on the individual optimal value of the particle, its multi-objective aggregation index is obtained, which is expressed as: Δf y (t)=F1(pbest y (t))-f(pbest y (t-1))+F2(pbest y (t))-f(pbest y (t-1)) Where Δf y (t) represents the fitness gain, pbest y Represents the optimal extreme value of particle y. Based on the path similarity of the rescue mission assignment, the adjacent set of the selected particles is obtained. The similarity between particles y and q is calculated using the Jaccard index, which is expressed as: Among them, s yq Indicates the similarity between each particle y and particle q, P y ={P y1 ,P y2 ,…,P yn } and P q ={P q1 ,P q2 ,…,P qn } represent the node order of the rescue mission assignment path of particles y and q respectively. Set the similarity threshold s min , select only s yq ≥s min The particles constitute the adjacent set To ensure the validity of the adjacency relationship. Generally, in order to take into account both population information exchange and computational efficiency, The number of sets is set to 3 to 5. If the current population diversity is low, the number is reduced to 3; if the diversity is high, the number is increased to 5. Based on the relationship between the confidence levels, the emotion contagion coefficient is set, that is, neighbors with high confidence are allowed to spread emotions to particles with lower confidence levels, which can be expressed as: Where μ0∈[0.3,0.4]. is the set of particles adjacent to particle y, through The adjacency relationship in realizes emotional influence.

8. The post-disaster UAV task allocation method based on the confidence-based multi-objective particle swarm optimization algorithm according to claim 1 is characterized in that: The mutation steps performed include: Crossover operations are divided into crossover between different particles and u-opt operations for individual particles. Crossover between different particles selects two particles and creates a partial crossover. For single-particle crossover, the u value is set based on the problem size, and particles are selected for u-opt operations. The main steps are to randomly divide the particles into u segments and then recombine these segments. The decision to update the particle is made by comparing the objective function value of the recombined particles with the dominance relationship of the original particles. Inter-group task exchange: Based on the decoding strategy proposed by the present invention, the task allocation result is mainly affected by the group in which the task is located. By adjusting and changing the group in which the task is located, the task allocation result can be changed, thereby achieving task redistribution and thus changing the task allocation result. Single point mutation: randomly select particle x y A certain dimension e in , applies Gaussian perturbation within the value range of the particle, and its update formula is: Among them, x max and x min are the maximum and minimum values ​​of the particle value range, respectively. This operation is mainly aimed at the dominant solution with poor fitness value to ensure particle convergence while increasing the diversity of solutions.

9. The post-disaster UAV task allocation method based on the confidence-based multi-objective particle swarm optimization algorithm according to claim 1 is characterized in that: Based on the crowding distance and diversity between particles, the extreme value and external reserve set of particles are updated. Extremum update: Based on the crowding distance and diversity index, the update method of obtaining individual extremum and global extremum includes: selecting the non-dominated solution of particles in the evolution process as the individual extremum; setting the adjustment probability, using a combination of crowding and random selection to obtain the global extremum, and adjusting the probability calculation method, which is expressed as: Among them, P adj represents the adjustment probability, α=β=0.5, Δ max =1, it is the current iteration number, IT is the total iteration number. adj As the number of iterations increases, it gradually decreases. adj When the value is less than a random number uniformly distributed on [0,1], the particle with the largest crowding degree is selected as the global extreme value; otherwise, a particle is randomly selected from the Pareto solution set as the global extreme value. This strategy adjusts the global extreme value selection method according to the probability, which helps to make the solution set distribution of the non-dominated frontier more uniform. Based on the objective function value of the particle, the calculation method for obtaining the particle crowding degree is expressed as: Among them, F r (y-1), F r (y+1) are the values ​​of the rth objective function of the two individuals adjacent to individual y, F rmax and F rmin Represent the maximum and minimum values ​​on the r-th target respectively. Based on the Euclidean distance between non-dominated solutions, the calculation method of the diversity evaluation index Δ is obtained, which is expressed as: Where N represents the number of current non-dominated solutions, d h represents the Euclidean distance between two neighbor individuals in the non-dominated solution set, d f and d l are the Euclidean distances between the leftmost solution and the boundary solution, and between the rightmost solution and the boundary solution in the non-dominated solution set, Represents the average distance, and the smaller the Δ value, the better the distribution and diversity of non-dominated solutions. External reserve set update: When the number of non-dominated solutions exceeds the maximum capacity of the reserve set, the difference between the newly generated non-dominated particles and the particles in the external reserve set needs to be compared. If the objective function values ​​are different, the non-dominated solution with the smallest congestion is selected for replacement; if the objective function values ​​are equal, the particle X is compared. y and X q The difference degree Dis(y,q) after grouping is used for screening. Based on the task grouping information, the difference between the two particles is obtained, which is expressed as:

Images