CN112230675B

CN112230675B - Unmanned aerial vehicle task allocation method considering operation environment and performance in collaborative search and rescue

Info

Publication number: CN112230675B
Application number: CN202011117615.4A
Authority: CN
Inventors: 张明; 李松锐; 李伯权
Original assignee: Nanjing University of Aeronautics and Astronautics
Current assignee: Nanjing University of Aeronautics and Astronautics
Priority date: 2020-10-19
Filing date: 2020-10-19
Publication date: 2022-06-24
Anticipated expiration: 2040-10-19
Also published as: CN112230675A

Abstract

The invention relates to an unmanned aerial vehicle task allocation method considering operation environment and performance in collaborative search and rescue, and belongs to the field of task allocation methods of navigation and unmanned aerial vehicle collaborative search and rescue. The invention considers the disaster level of the search and rescue points to endow each search and rescue point with a level attribute, considers the influence of the detection range of the unmanned aerial vehicle on the disaster level, corrects the level, and converts the level result into the hovering time of the unmanned aerial vehicle. Selecting a certain number of unmanned aerial vehicle release positions according to the minimum total search and rescue cost, allocating a proper search and rescue point to each position, considering the influence of unmanned aerial vehicle performance, terrain and low-altitude wind on unmanned aerial vehicle battery energy consumption and other factors, establishing a helicopter release unmanned aerial vehicle position location optimization model, solving by using an improved binary bat algorithm, and comparing and analyzing with a result before improvement. The invention greatly improves the practicability and accuracy of the task allocation research of the unmanned aerial vehicle.

Description

Unmanned aerial vehicle task allocation method considering operation environment and performance in collaborative search and rescue

Technical Field

The invention relates to an unmanned aerial vehicle task allocation method considering operation environment and performance in collaborative search and rescue, and belongs to the field of task allocation methods of navigation and unmanned aerial vehicle collaborative search and rescue.

Background

Aiming at the problem of cooperative allocation, transportation and search and rescue work are implemented through unmanned aerial vehicle formation, and the cooperative task allocation is more and more favored, is one of key technologies for formation flight, and aims to reasonably allocate tasks to different unmanned aerial vehicles according to a set target under the condition of meeting constraints in various aspects and fully exert the overall efficiency of formation. The main approaches are centralized allocation, distributed allocation, and hierarchical distributed allocation.

The centralized allocation means that communication, control and the like among unmanned aerial vehicles in a formation are carried out by a unique control center, the problems of multiple traveling salesmen and vehicle paths are mainly solved, the task allocation of the unmanned aerial vehicle battery energy saving is considered, the adopted optimization methods comprise mixed integer linear programming, dynamic network flow optimization and a multi-agent reinforcement learning method, the task allocation mode has the advantages that the optimal solution is obtained or a more satisfactory solution is obtained within an acceptable time range, and the time is consumed when the general problem scale is large.

The distributed distribution mode has higher requirement on the unmanned aerial vehicle, the unmanned aerial vehicle is required to have the capability of independent calculation, analysis and decision-making, the unmanned aerial vehicle can communicate in a formation through a random planning model, an improved ant colony algorithm, a random incentive mechanism, a contract net auction model, a balanced clustering algorithm and the like, the flexibility is high, but the task distribution has high requirement on the unmanned aerial vehicle, and the independent calculation and decision-making are required.

The hierarchical distributed distribution mode integrates the advantages of a centralized distribution mode and a distributed distribution mode, and is a mixed distribution mode. The unmanned aerial vehicle task allocation mode is used for layering and classifying all unmanned aerial vehicles according to a certain rule. The unmanned aerial vehicles are grouped according to categories, the large classes of the unmanned aerial vehicles at the same level and the unmanned aerial vehicles in the same group select a centralized distribution mode, and different combined control centers select a distributed control mode. The main models and algorithms are: a hierarchical distributed task allocation model, a VAMU (vehicle assisted multiple unmanned aerial vehicle) algorithm, a Wolf Pack search algorithm, an improved coverage search algorithm, a genetic and clustering combination, a locust algorithm, an algorithm for a coverage path planning problem with limited battery energy consumption, and the like.

The method comprises the steps that a UGV (unmanned vehicle) -UAV (unmanned aerial vehicle) collaborative exploration path planning method is adopted, the UGV is used as a route of the UAV, the UGV is used as an energy gas station of the UAV to supplement the energy consumption of a battery of the UAV, the built model is solved by taking the minimum total distance of the UAV, the minimum total distance of the UGV and the minimum sum of the UAV and the energy consumption of the battery of the UAV as main constraints, and the collaborative task planning result of the UAV and the UAV is obtained. However, the problem scale is small, the influence of terrain, environmental factors and the like on the planning result cannot be considered, modeling is carried out in a two-dimensional plane, and the problem is simplified excessively. Pariksit Maini proposed and studied the use of mobile gasoline stations to increase the operating range, saving costs in space and time. This document considers the effect of terrain on mobile station siting, but there is a certain deficiency in the assignment of missions taking into account the actual circumstances.

More logistics unmanned aerial vehicle task allocation models are single targets, such as the shortest total path or the smaller transportation cost, but in the rescue situation, it is very important to consider the search and rescue time, and in addition, the number limit of unmanned aerial vehicles should be considered.

Some researches usually show that the flight path of the unmanned aerial vehicle is relatively ideal, and influence of the unmanned aerial vehicle on task allocation of the unmanned aerial vehicle, such as factors of avoiding terrain obstacles and weather, and the like, the performance of the unmanned aerial vehicle, the release and recovery of the unmanned aerial vehicle, the energy consumption difference of the actual flight of the unmanned aerial vehicle at different stages and the like are not considered; some researches do not relate to the constraint of battery energy consumption of the unmanned aerial vehicle; in the search and rescue hovering process of the rescue point, the severity of the disaster area in the area is often related, the classification of the grade attributes of the rescue point belongs to the problem of disaster classification, and the relevance of the disaster classification and the hovering time of unmanned aerial vehicle search and rescue needs to be further considered in combination with the rescue practice.

The above problems make the search process not very suitable for the reality of emergency rescue, and specifically have the following disadvantages:

(1) the influence of the detection range of the unmanned aerial vehicle on the classification of the grade attributes of the target points is less considered, and the results of the unmanned aerial vehicle position address selection and the search and rescue task allocation are further influenced.

(2) The influence of the actual flight environment of the unmanned aerial vehicle (mainly including the influence of low-altitude wind on the energy consumption of the battery of the unmanned aerial vehicle and other factors) on the task allocation result is rarely considered.

(3) The influence of the performance of the unmanned aerial vehicle on the position address selection and task allocation result release of the unmanned aerial vehicle is considered.

Disclosure of Invention

The invention provides an unmanned aerial vehicle task allocation method considering the operation environment and performance in collaborative search and rescue. And giving a grade attribute to each search and rescue point by considering the disaster grade of the search and rescue point, correcting and grading by considering the influence of the detection range of the unmanned aerial vehicle on the disaster grade, and converting the grading result into the hovering time of the unmanned aerial vehicle. Selecting a certain number of unmanned aerial vehicle release positions according to the minimum total search and rescue cost, allocating a proper search and rescue point to each position, considering the influence of unmanned aerial vehicle performance, terrain and low-altitude wind on unmanned aerial vehicle battery energy consumption and other factors, establishing a helicopter release unmanned aerial vehicle position location optimization model, solving by using an improved binary bat algorithm, and comparing and analyzing with a result before improvement.

The invention adopts the following technical scheme for solving the technical problems:

an unmanned aerial vehicle task allocation method considering operation environment and performance in collaborative search and rescue comprises the following steps:

(1) acquiring unmanned aerial vehicle performance data, including maximum endurance time, maximum hover time, flight endurance time under the influence of 3-level low-altitude wind on unmanned aerial vehicle battery energy consumption, a rise limit, price, maximum charging times, terrain data of a researched area and main eight disaster indicators;

(2) preliminarily dividing disaster grades of administrative units according to disaster indexes, simulating search and rescue points according to population density distribution and area proportion of each county, and giving the administrative units where the simulation points are located the same disaster grade attributes; solving the detection range of the unmanned aerial vehicle at each simulation point according to a blind area calculation model and an algorithm, bringing the detection range into a grading index, and obtaining a modified grading result by using a clustering method;

(3) the influence of flight endurance performance, disaster level, wind and terrain of the unmanned aerial vehicle on the endurance time of the unmanned aerial vehicle is considered, the minimum total search and rescue cost is taken as a target, and the problem of determining the position of the unmanned aerial vehicle released in navigation is modeled; improving a binary bat algorithm, introducing a differential evolution mechanism, solving the model, determining the position of the unmanned aerial vehicle released by navigation, and comparing and analyzing the position with a result obtained by the unmodified binary bat algorithm;

(4) after the unmanned aerial vehicle position is released, carrying out unmanned aerial vehicle task allocation on the target search and rescue points allocated to each position point: at each position for releasing the unmanned aerial vehicles, an allocation model is established for a target function by considering the search and rescue cost, the using number of the unmanned aerial vehicles and the balance of the unmanned aerial vehicle allocation tasks; and solving the multi-target model by using an NSGA-II algorithm to obtain a distribution result, and comparing and analyzing the distribution result with a result obtained by the single-target model.

The invention has the following beneficial effects:

firstly: the task allocation model for searching and rescuing the multiple unmanned aerial vehicles considering the operating environment and performance of the unmanned aerial vehicles is provided. According to various disaster indexes of a disaster area, population distribution is used for simulating disaster-affected search and rescue points, and disaster grades and corresponding hovering search and rescue time of target search points are determined by considering detection ranges of unmanned planes at the same relative height. On the basis of considering disaster levels, unmanned aerial vehicle performance and the influence of low-altitude wind on the energy consumption of an unmanned aerial vehicle battery, a position location model of a navigation release unmanned aerial vehicle is established, a differential evolution mechanism is introduced, and an improved binary bat algorithm is applied to obtain a better result.

Secondly, the method comprises the following steps: and performing unmanned aerial vehicle task allocation on the search points allocated to each address selection position by using an NSGA-II algorithm, thereby completing navigation-unmanned aerial vehicle collaborative search and rescue task allocation. Compared with the past research, the unmanned aerial vehicle performance and the operation environment influence the energy consumption of the unmanned aerial vehicle, and the unmanned aerial vehicle task allocation research practicability and accuracy are greatly improved in the unmanned aerial vehicle task allocation problem.

Drawings

Fig. 1 is a diagram of a GA-UAV (navigation-unmanned aerial vehicle) cooperative search and rescue operation scene.

Fig. 2 is a schematic diagram of unmanned aerial vehicle release position addressing.

FIG. 3(a) is a schematic diagram of the results of an allocation that considers only cost objectives; fig. 3(b) is a diagram illustrating the distribution result after considering the balance of tasks.

Fig. 4 is a thermodynamic diagram of the population in qiang autonomous state 6, tiban.

Fig. 5 is an effect diagram of the visual range of the unmanned aerial vehicle.

Fig. 6 is a comparison graph of the iterative effects of BBA (binary bat algorithm) and IBBA (differential evolution mechanism binary bat algorithm).

Fig. 7 is a schematic diagram of the address selection result of the release point of unmanned aerial vehicle # 53 and the distribution result of the corresponding search and rescue point.

Fig. 8(a) shows the result at pop ═ 30; pareto solutions obtained at Gen 1000/500/300/200; fig. 8(b) shows the peak value at pop ═ 30; pareto solutions obtained at Gen 1000/500/300/200; fig. 8(c) is a graph represented in Gen 1000; (iii) poppy 100/50/30/20, pareto's diagram obtained below; fig. 8(d) is expressed in Gen 1000; the pareto solution obtained under the condition of pop 100/50/30/20.

FIG. 9 is a cross variation diagram.

FIG. 10 is a multiple variation graph.

FIG. 11 is a flow chart of the method of the present invention.

Detailed Description

The invention is described in further detail below with reference to the accompanying drawings.

According to the method, for each position of the unmanned aerial vehicle released in navigation, namely the allocated search and rescue point, the factors such as the performance and the environment of the unmanned aerial vehicle are considered, a multi-objective optimization model for minimizing the cost for completing tasks, the using number of the unmanned aerial vehicles and the total time for completing the tasks is established, and the NSGA-II algorithm is used for solving. As shown in fig. 1. The grey route is a navigation flight route, the grey route passes through the five-pointed star nodes successively and is an unmanned aerial vehicle release point, a low-altitude suspension search task is implemented at each search and rescue point (hollow point) of the oval dotted line task area where the unmanned aerial vehicle is located, and influences of terrain and low-altitude wind on battery energy consumption of the unmanned aerial vehicle in the task area need to be considered.

Navigation release unmanned aerial vehicle position point optimization model

Model assumptions

The position of the alternative navigation hovering point and the target searching point of the unmanned aerial vehicle are determined, and the distance between the nodes is calculated according to the Euclidean distance.

The influence of low-altitude wind on the energy consumption of the unmanned aerial vehicle battery and the conversion of the terrain around the flight to the unmanned aerial vehicle are considered, and a certain redundancy is reserved for the fine track planning in the later period.

And determining the detection time of the target search point according to disaster grades, and converting the detection time into the flight distance consumption of the unmanned aerial vehicle according to the endurance time, the flight speed, the hovering time and the flight time.

The target search and rescue points are randomly generated according to population density distribution, and the search and rescue height is 500m higher than the terrain of the position.

And each target search and rescue point has one unmanned aerial vehicle for searching.

The unmanned aerial vehicle is influenced by wind and terrain when flying, and is only influenced by wind when hovering, and is irrelevant to the terrain.

The acquisition cost of the unmanned aerial vehicle is not considered, the speeds of the unmanned aerial vehicles starting from all the nodes are consistent, and the range cost of the unmanned aerial vehicle is in direct proportion to the flying distance of the unmanned aerial vehicle (including hovering energy consumption conversion, terrain flying around and wind energy consumption conversion).

The problem of repeated recycling of the unmanned aerial vehicle is not considered, and the full electric quantity is obtained before releasing.

According to the minimum safety height requirement standard of navigation in mountainous areas, the relative ground height of the position of releasing the unmanned aerial vehicle is not less than 600 m.

Model building

And (4) considering the performance parameters of the unmanned aerial vehicle, and respectively solving the influence coefficient of low-altitude wind on the energy consumption of the battery and the flying and hovering energy consumption ratio of the rotor unmanned aerial vehicle. The maximum hovering time of the full-electric-quantity unmanned aerial vehicle under the windless condition is set to be T_hoverThe maximum time that can be flown at speed v is T_vThe maximum flight time under the influence of 3-level low-altitude wind on the energy consumption of the unmanned aerial vehicle battery is T_windThen, the influence coefficients of the flying and hovering energy consumption ratio of the unmanned aerial vehicle and the low-altitude wind on the battery energy consumption are respectively expressed as formulas (1) and (2):

wherein gamma is the energy consumption ratio of flying and hovering of the rotor unmanned aerial vehicle, and alpha is the influence coefficient of low-altitude wind on the energy consumption of the battery.

In the assignment of the flight mission of the drone, the problem of energy consumption due to the actual obstacle avoidance is considered, and the flight coverage of the drone in the worst case is considered, under the assumption that a fully loaded drone must arrive at an energy state of 2/3 and reserves of 1/3 for the return trip are reserved, but these assumptions can also be flexibly adjusted. And setting the obstacle avoidance energy consumption reservation rate beta, and estimating and setting according to the actual terrain condition of the mountainous area.

The objective function is that the total search and rescue cost Z is minimum:

constraint conditions are as follows:

y_jk≤x_j，j∈M，k∈N (6)

S_jk≤S，j∈M，k∈N (8)

y_jk＝{0，1}，j∈M，k∈N (9)

x_j＝{0，1}，j∈M (10)

the set M is a position set of alternative navigation release unmanned aerial vehicles, M is {1, 2.. multidot.m }, the set N is a target search point set of the unmanned aerial vehicles, and N is {1, 2.. multidot.n }. Parameter c is the cost per unit flight distance of the drone, F_jService cost for releasing the position j, number of positions l for planned navigation releasing unmanned aerial vehicles, d_jkIs the Euclidean distance, t, from the release position j to the target search and rescue point k_kThe hovering time of the unmanned aerial vehicle at a target searching point k, v is the speed of the unmanned aerial vehicle flying at a constant speed, beta is the obstacle avoidance energy consumption reservation rate, and S is the maximum driving mileage of the unmanned aerial vehicle (calculated by the cruising time of the unmanned aerial vehicle); s_jkAnd (4) representing the total voyage (including wind, terrain damage and unmanned aerial vehicle hovering conversion) consumed from the jth navigation suspension point to the kth search and rescue point. Decision variable x_jWhether the alternative center j is selected or not is shown, if the alternative center j is selected, the value is 1, and if not, the value is 0; decision variable y_jkAnd the judgment result represents whether the search and rescue point k is served by the alternative center j, if so, the judgment result is 1, and if not, the judgment result is 0.

Equation (3) is an objective function, which represents the total cost minimization, including the total flight cost (including hover) and the total service cost; the formula (4) shows that each search and rescue point is only used for dispatching the unmanned aerial vehicle from one navigation suspension point to perform search and rescue; formula (5) shows that at most one release position point is selected from the alternative navigation suspension points to release the unmanned aerial vehicle; if the unmanned aerial vehicle search and rescue is dispatched from the navigation suspension point j at the search and rescue point k, j must be the selected suspension point; the total voyage consumed by the unmanned aerial vehicle of formula (7) comprises the sum of the flight voyage and the hover reduced voyage; the search and rescue range constraint of the unmanned aerial vehicle from the navigation suspension point to the search and rescue point in the formula (8) shows that the total flight distance consumed by the unmanned aerial vehicle from the navigation suspension point j to the search and rescue point k is not more than the maximum flight distance of the unmanned aerial vehicle; the decision variables y of equations (9) and (10)_jkAnd x_jIs a variable from 0 to 1.

Unmanned aerial vehicle search and rescue task allocation model

Model assumptions

The target search point of the unmanned aerial vehicle is determined, and the distance between nodes is calculated according to the Euclidean distance. Each unmanned aerial vehicle can search a plurality of target points, each target search and rescue point has one unmanned aerial vehicle for searching, but the energy consumption of the unmanned aerial vehicle does not exceed the maximum energy consumption of the unmanned aerial vehicle battery.

The speed of the unmanned aerial vehicle starting from each node is cruise speed, and the range cost of the unmanned aerial vehicle is in direct proportion to the flying distance of the unmanned aerial vehicle (including energy consumption conversion of factors such as hovering, obstacle avoidance and low-altitude wind). Fig. 2 is a schematic position diagram of the navigation release drone.

The problem of repeated recycling of unmanned aerial vehicle charging is not considered, and the full power is obtained before transmission.

Model building

Objective function

The objective function mainly comprises three parts. Not only need consider the expense that unmanned aerial vehicle search and rescue produced, still need the use quantity of comprehensive consideration unmanned aerial vehicle, the total task time of accomplishing is the minimum balanced rationality of unmanned aerial vehicle task etc. factor promptly.

First, the objective function is such that the total cost Z is₁And the minimum cost comprises the cost generated by the flight of the unmanned aerial vehicle and the cost generated by the hovering, wherein the cost is in direct proportion to the flight cost of the unmanned aerial vehicle, and the cost generated by the hovering is in direct proportion to the hovering time.

Wherein: s_ijRepresenting the voyage converted from energy consumption (including terrain and wind into cruise uniform-speed flight) between the point i and the point j of the unmanned aerial vehicle; decision variable y _ij1 if node i is visited before node j, otherwise 0; decision variable x_ijkRepresenting whether the unmanned plane K (K belongs to K) flies to a point j (i, j belongs to M U N) from i, if yes, the unmanned plane K is 1, otherwise, the unmanned plane K is 0; s_iRepresenting the range of the energy consumption (hovering converted into cruising constant-speed flight) of the unmanned aerial vehicle at the target search point i; the set P represents a set of search and rescue points and release positions of the drone (P ═ {0, 1,2, 3., P }, 0 represents a position of the drone to be released), the set K represents a set of the drone, and K ═ 1, 2., K }.

Secondly, the reality of the unmanned planeNumber of radicals Z₂There are limits to minimize the number of drones used (where node 0 represents the location where navigation releases a drone):

wherein: x is the number of_0jkIndicating whether drone K (K e K) flies from the addressing point to point j.

As shown in fig. 3(a), the distribution result considering only the cost target is shown, and fig. 3(b) is a distribution result showing the balance of the task. Finally, the balance of the tasks is considered, and the maximum time difference Z for each unmanned aerial vehicle to complete the tasks needs to be considered while the cost is considered₃And (3) reducing to the minimum, namely:

constraint conditions are as follows:

s_i＝αγt_iv,i∈N′ (22)

wherein:

the longest time for the search and rescue task in the unmanned aerial vehicle,

for the shortest search and rescue task in unmanned aerial vehicle, x_0kWhether the kth unmanned aerial vehicle starts from the address selection position or not is represented, S is a subset of N ', namely a set of search and rescue points distributed to one unmanned aerial vehicle, the set N' represents a target search point set of the unmanned aerial vehicle (the set P is a set formed by removing elements left by the release position), and the set T represents the time for the unmanned aerial vehicle to complete a taskSet, T ═ T₁,T₂,T₃,…,T_k,}。t_ijRepresenting the flight time of the drone between point i and point j (calculated at a constant speed in terms of euclidean distance and cruising speed), t_iFor the hovering time of the unmanned aerial vehicle at the target searching point i, S £ {0} represents a set formed by a union of a set of search and rescue points allocated to one unmanned aerial vehicle and the departure point of the unmanned aerial vehicle, and d_ijIs the Euclidean distance between the target search and rescue point (or the search and rescue point and the release position) i, j. Decision variable x_ikAnd indicating whether the node i is served by the unmanned plane k, if so, the node i is 1, and otherwise, the node i is 0. The meaning of other letters is completely consistent with the addressing model.

The constraint condition formula (14) ensures that each search point can only be searched by one unmanned aerial vehicle; the constraint condition formula (15) indicates that if the unmanned aerial vehicle searches for a search and rescue point, the unmanned aerial vehicle must release a position point through navigation; if the unmanned aerial vehicle does not pass through the navigation release position, any search and rescue point cannot be searched. The constraint (16) ensures the continuity of each route, i.e. the drone visiting node j must leave node j. Constraint equation (17) indicates that if there are drones flying from node i to node j, they will be searched by the same drone. The constraint equation (18) is a classical sub-loop cancellation constraint, ensuring that no sub-loop solution is generated. The constraint equation (19) is a frame number limit of the drone. The constraint condition formula (20) represents the battery energy consumption limit of the unmanned aerial vehicle, namely the maximum range limit, which is the sum of the electric energy consumed by flight and the electric energy consumed by hovering. Equations (21) and (22) define a conversion equation of the flight distance, the pure flight endurance is taken as a reference, terrain circling flight and low-altitude wind for the power consumption of the battery of the unmanned aerial vehicle are considered during flight, conversion is carried out according to beta and gamma, wind and hovering energy consumption are considered during suspension, and conversion is carried out according to alpha and gamma. Equation (23) defines the total time it takes for the kth drone to complete the task assigned to it. The constraint equations (24) - (26) define the value ranges of the variables.

Solution scheme

The invention mainly uses four algorithms to solve. Firstly, solving the classification problem of search and rescue points by using a principal component analysis method; then, calculating the detectable area of each search and rescue point by using a detection blind area algorithm, standardizing the result, and then correcting and grading by using a clustering method; then, an improved binary bat algorithm is used for solving the address selection scheme of the position of the navigation release unmanned aerial vehicle; and finally solving the task allocation problem of the unmanned aerial vehicle by using an NSGA-II algorithm.

Principal component analysis algorithm for solving search and rescue point grading problem

The disaster situation grade preliminary classification steps are as follows:

unmanned aerial vehicle detection blind area calculation algorithm

In complicated mountain region environment, unmanned aerial vehicle's detection scope can receive the topography influence, therefore has to consider unmanned aerial vehicle detection scope to the hierarchical influence of disaster, considers that the size of unmanned aerial vehicle blind area can make the hierarchical result of disaster more accurate, and the computational process of unmanned aerial vehicle detection blind area is as follows:

solving the problem of selecting the position of a navigation release unmanned aerial vehicle by improving the binary bat algorithm

The bat algorithm is used for the problem of address selection of the navigation release unmanned aerial vehicle, so that the calculation is convenient, the calculation efficiency is high, and the calculation result is more accurate. However, due to the defects of the bat Algorithm, namely, optimization is mainly carried out by means of information exchange among individuals, and an individual variation mechanism is lacked, so that the obtained solution is rapidly gathered to excellent individuals, and the solution falls into local optimum.

Coding scheme design and population initialization

And selecting l release positions from m alternative positions in the initialization population, wherein m is the dimension of each bat. For formula (7), let

α_k＝αγt_kv, k ∈ n. Wherein for the kth search and rescue point, t_kIf it is determined, then alpha is known_kIs a constant. Alpha is alpha₂Is obviously a constant. Then equation (7) can be simplified to equation (27)

S_jk＝α₂d_jk+α_k，j∈M,k∈N (27)

As can be seen from the above equation, the total cost of the drone due to hovering is related to the selection of the release location, so once the release location is determined, the search and rescue point can be assigned to the release location closest thereto using the principle of close-proximity assignment. If 6 release positions are randomly selected from 10 alternative positions, the selected serial number position is set to 1, and the rest are set to 0, one bat can be coded as [ 1110110011 ]]^TIt means that the release positions 1,2,3,5,6,9,10 are open, and the other release positions are closed. After the release position is determined, the search and rescue points are distributed nearby and meet the constraint condition, the task is distributed, otherwise, the bat stops evolving towards the direction, and the parent continues evolving towards other directions.

Definition of merit function

The smaller the evaluation function value, the better the individual, the magnitude of the evaluation function value is taken to be the value of the objective function, namely:

wherein: evaluation represents an Evaluation function.

An Improved Binary Bat Algorithm (IBBA) DE has more abundant mutation strategies, mutation operators of the Improved binary bat algorithm more effectively utilize population distribution characteristics, and mutation efficiency is higher. The combination of DE and BBA can overcome the defects of low convergence precision of BA (bat algorithm), easy falling into local optimum, and the like. It should be noted that the DE algorithm is also a continuous optimization algorithm, and thus is converted to binary DE.

DE has more abundant mutation strategies, and the mutation operator of DE utilizes the population distribution characteristics more effectively and has higher mutation efficiency. The combination of DE and BBA can overcome the defects of low convergence precision of BA, easy falling into local optimum and the like. It should be noted that the DE algorithm is also a continuous optimization algorithm, and thus is converted to binary DE.

(1) DE-based variants

1) Random vector DE

For the ith individual

Randomly selecting two other individuals in the population

And

and then adds its weighted difference vector to the current individual, the new individual can be expressed as:

wherein:

is the jth vector of the ith variant individual,

is the jth vector for the ith individual,

is the jth vector for the pth individual,

is the jth vector for the qth individual; if it is not

Then whether the vector is mutated or not is determined by the difference vector, and the vector remains the same or vice versa.

2) Optimal solution with random vector DE

Randomly selecting two individuals different from the current individual from the population; then, adding their weighted difference vectors to the current optimal individuals, and differentiating the mutated new individuals

Can be expressed as:

wherein

Is the jth vector of the current best individual. The mutation strategy utilizes the best individual information of the current population, and the variant individuals are distributed near the current best individual, so that the local search capability is strong. Performing the above two operations on each current individual, and then performing mutation on the current individual

And

to select the best individual

(2) Crossed male parent individuals

And new variant individuals

Cross formula (31)

Wherein:

represents the result of the j vector crossing of the ith individual of the t generation, rand_jIs [0, 1]]A random number in between; p_c∈[0，1]Is a crossA fork probability;

j

1,2, L is the dimension of the vector; j. the design is a square_rand1, 2.. L is to ensure that a new individual has at least one dimension from a variant individual.

(3) After selection of mutations and crossovers, new individuals are introduced

With the original individual

A comparison is made. The better individuals were selected as next generation individuals.

Improving the binary bat algorithm:

step1 population initialization, i.e., the bat spreads a set of initial solutions of size n in a random fashion in a d-dimensional space. Maximum pulse volume A₀Maximum pulse rate r₀Maximum number of iterations N_gen. Randomly initializing the positions of the bats, and searching the optimal solution x of all the bats according to the fitness value^*。

Step2 updating the search pulse frequency of bat as the formula, wherein f_minAnd f_maxMinimum and maximum search pulse frequency ranges, f, respectively_iIs the search pulse frequency, β, of bat i₁Is a uniformly distributed random number, beta₁∈[0,1]。

f_i＝f_min+(f_max-f_min)β₁ (32)

The standard bat algorithm solution is used for solving the optimization problem of a continuous function, so that a position updating formula and a speed updating formula of the bat need to be transformed, and the bat algorithm suitable for the discrete optimization problem is obtained. In reference, standard BA is improved, the objective function is minimized, and in order to maintain the optimization domain within two numbers of 0-1, the formulas of speed transformation and position transformation of bats are improved as follows:

in the above formula, the first and second carbon atoms are,

the logical operators are exclusive OR, inverted AND, and round means rounding.

Represents the ith bat individual of the tth generation,

representing the speed of the bat i in the t generation,

represents the ith bat individual of the t-1 th generation,

representing the speed of the bat i in the t-1 generation.

Step3, generating uniformly distributed random numbers rand, if rand>r (r is bat walk step length), then x is paired^*Random perturbation is performed to generate a new solution x_newAnd performing border crossing processing on the new solution according to the formula.

x_new＝∧(x_old,round(rand(1，d))) (35)

Wherein: rand (1, d) represents the generation of random numbers between d dimensions 0-1. x is the number of_oldIs the original solution.

Step4, generating uniformly distributed random numbers rand, if rand<A_iAnd f (x)_i)<f(x^*) Then the new solution generated by Step4 is accepted and the sum is updated according to equations (39) and (40), where α is₁Attenuation coefficient of sound volume, gamma₁For searching enhancement coefficients of frequency, pair

γ₁Greater than 0, has

t→∞。

Wherein:

the pulse emission loudness of the ith bat representing the t generation,

represents the pulse emission loudness of the ith bat of the t +1 th generation,

pulse emission rate of the ith bat individual of the initial population,

represents the pulse emission rate of the ith bat individual of the t +1 th generation, and t is the iteration number.

Step5 mutation and crossover

And Step6, sorting the fitness values of all bats to find out the current optimal solution and optimal value.

And Step7, repeating the steps of Step 2-Step 6 until the set optimal solution condition is met or the maximum iteration number is reached.

And Step8, outputting the global optimal value and the optimal solution.

Solving the problem of task allocation of the unmanned aerial vehicle by the NSGA-II Algorithm because the established task allocation model of the unmanned aerial vehicle is a multi-objective optimization model and accurate calculation is complex, the problem is solved by applying a Non-dominant sequencing Genetic Algorithm (NSGA-II) with elite strategies. The optimal solution of the algorithm for solving the task allocation problem of the unmanned aerial vehicle depends on the evolution mechanism of the algorithm. Compared with NSGA, the method mainly provides a rapid non-dominated sorting algorithm, introduces an elite strategy and adopts a congestion degree and congestion degree comparison operator. The NSGA-II algorithm reduces the complexity of the non-inferior ranking genetic algorithm, and has the advantages of higher running speed, better solution convergence and the like.

i population initialization

In consideration of the limitation of the distribution problem of the multi-unmanned aerial vehicle search task, a double-chromosome coding mode is used for coding, wherein a chromosome I (hereinafter referred to as I) represents a target sequence, and a chromosome II (hereinafter referred to as II) represents a cutting position of the target sequence on the chromosome I. In I, each gene represents an index of a search target, and the total base factor is N_T. In II, the value of any gene should not be smaller than that of the preceding gene, and the number of genes is (N)_U-1), whereby the target sequence in I is cleaved into N_UA sub-sequence. Take four unmanned aerial vehicles, 10 search and rescue points encode as an example:

table 1 chromosome I (3,8,5,1,4,2,6,10,9, 7); chromosome II (2,5,8) data

a) As shown in Table 1, II is (2,5,8), so that the gene in I (3,8,5,1,4,2,6,10,9,7) is divided into four subsequences, second, fifth and eighth genes after cleavage. Then, the four subsequences (3,8), (5,1,4), (2,6,10) and (9,7) represent the target sequences of

UAVs

1,2,3 and 4, respectively.

ii learning based on oppositions

By using opponent based learning strategies in task assignment, opponent populations are generated after initialization and mutation operations to increase the likelihood of finding better solutions. For [ a, b ]]Variable z in interval, logarithm thereof

Defined by formula (28):

for a point in space P ═ z₁,z₂,...,z_D) Opposite to the face point

Is a value of each dimension generated by calculating a relative value

In drone task allocation, I is a multidimensional point considered to be the opposite of the computation. For example, I is (1,7,8,4,5,3,6, 2). The lower and upper limits for all genes are 1 and 8, respectively. The opposite of I is (8,2,1,5,4,6,3, 7).

(3) Fast non-dominated sorting

For each individual in the population P, firstly, the gene is decoded to obtain the task allocation result of the unmanned aerial vehicle, and according to each objective function value, two parameters n are obtained for each individual i_iAnd s_i(n_iTo govern the number of solutions of an individual i in a population, s_iA set of solutions governed by an individual i) to be layered and ordered to obtain different levels of pareto fronts. The above operation is repeated until all individuals are set to the leading edge. (4) Crowdedness distance L (i) calculation formula of crowdedness comparison operator individual i

Wherein: z_kFor the purpose of the k-th objective function,

is the maximum value of the k-th objective function,

is the minimum of the kth objective function.

After fast non-dominated sorting and crowding calculation, each individual of the population has two attributes: non-dominant rank i from non-dominant rank_rankAnd degree of congestion i_dAnd setting a congestion degree comparison operator according to the two attributes, and if any one of the following conditions is satisfied for the individual i and the individual j, winning the individual i:

the non-dominant layer of the i individual is superior to the non-dominant layer of the j individual, i_rank＜j_rank，j_rankRepresents the non-dominant order of the jth individual;

i and j have the same rank, while the crowding distance of the individual i is larger, i.e. i_rank＝j_rankAnd i is_d＝jd，j_dIndicating the congestion level of the jth individual.

(5) Selection and elite retention

After the population is initialized and a new population is obtained through each iteration, a proper male parent is selected by using a binary bidding method to carry out crossing and mutation operations. And after crossing and mutation, the population is subjected to opposite learning, and the size of the population is expanded. Generating a new population Q of the t generation_tAnd parent population P_tMerging, performing non-dominant sorting and congestion degree calculation, reserving excellent individuals in parents and descendants, and entering the next iteration.

(6) Crossing

The crossover operator is only applied to chromosome I, while chromosome II is kept unchanged in the crossover process, and the crossover rate is P₁. This work uses a partial mapping crossover operator in which a portion of one parent gene is swapped with a portion of another parent gene, and the remaining genes are replicated or regenerated by mapping.

First, two cut points were randomly selected as: a point between the 2 nd and 5 th genes; the point between the 4 th and 5 th genes. Then, two mapping parts (3,4,5) and (5,1,2) are determined, and mappings 3-5, 4-1 and 5-2 are also defined. Second, the mapped portion in parent I (parent II) is copied to descendant II (descendant I). Thereafter, the remaining genes of the two offspring are filled in by replicating the genes of the corresponding parents or regenerating by mapping. As shown in fig. 9, for example, by directly replicating the first gene of parent I, the 2 nd gene of progeny I is 2. But the gene (i.e., 2) is already present. Thus, the first element of descendant I is reset to 3 according to mappings 3-5 and 5-2, and so on, and descendant I is (4, 3,5, 1,2, 6). In a similar manner, progeny II may be generated as (6, 2,3,4, 5, 1).

(7) Multiple variation

For chromosome I, there are four total operation modes of mutation, namely, maintenance, turnover, exchange and sliding, four offspring can be generated at one time, and the mutation rate is P₂. As shown in FIG. 10, two locations (e.g., 2 and 5) on chromosome I are first randomly generated. There are three variant operations:

for chromosome II, the initial solution is set to a one-machine-one task point, and then needs to be at (N)_U-1) randomly selecting a position within the range, and setting the value of the position as the value of the gene adjacent to the position to obtain the variation result of the chromosome II.

Numerical experiment

In the patent, Wenchuan earthquake of 12 days in 5 months in 2008 is taken as an example for performing calculation analysis, and six counties (Wenchuan county, Luzhong county, Ririken county, Xiaojin county, Heihe county and Panpan county) in Qiang autonomous State of Kangwa of Sichuan province are carried on the earthquake zone.

Data acquisition

The data required to be acquired comprise population density data, disaster index data, elevation data of disaster areas and relevant data of unmanned aerial vehicle models. 2008-year population grid data and administrative boundary data of each researched disaster-stricken county are respectively obtained from a national earth system scientific data sharing platform (http:// www.geodata.cn /). 8 indexes such as total population, area weighted average intensity, death and missing number, death and missing rate of thousands of people and the like of each county (city and district) are obtained. Elevation data for the region of interest is downloaded from google earth. The data of the Wenchuan earthquake disaster in 2008 are shown in Table 2.

TABLE 2 disaster-stricken data of 8-level Daseis Sichuan counties and cities in Wenchuan 2008

b) The invention selects three different unmanned aerial vehicle models for analysis, and the performance data of the unmanned aerial vehicle required to be used is shown in table 3:

table 3 performance parameters of the drone used in the present invention

The type A unmanned aerial vehicle used by the invention is a rotor type unmanned aerial vehicle Keweitai X6L (double batteries), the maximum endurance time is 120min, the calculation is carried out at the flying speed of 15m/s, and the maximum voyage is 108 km. The model B is Kobutai X6M (double cell), and the model C is Kobutai Z6M (double cell). The parameters and algorithm parameter settings for the solution model are shown in table 4.

TABLE 4 model analysis parameter settings

Simulation of target search and rescue points

After an earthquake occurs, disaster-affected points are selected randomly, the rescue work can have higher value by using the unmanned aerial vehicle to carry out image real-time return in places with denser population distribution, and in some places, the population is very rare, the rescue value is not very high, so that the density degree of the points is determined according to the population distribution when target rescue points are generated randomly, and the rescue work is more efficient.

As shown in fig. 4, in the population thermodynamic diagram of qiang autonomous state of the tibetan dam, disaster-affected points are randomly selected from points with high population density to serve as search areas in the rescue process of the unmanned aerial vehicle. Combining the search and rescue capacity of an unmanned aerial vehicle, simulating 100 search and rescue points, wherein the number of disaster affected points of Wenchuan county (No. 1-13), Luchuan county (No. 14-26), Yun county (No. 27-40), Xiaojin county (No. 41-58), Heihui county (No. 59-72) and Pan county (No. 73-100) is respectively as follows: table 5 shows the data of the simulated disaster site location parts, which are 13, 14, 18, 14, and 28.

Table 5 simulation disaster-affected point concrete coordinates

Ranking of target search and rescue points

(1) Preliminary grading based on disaster indicators

The main component analysis of SPSS software is adopted to analyze the disaster indicator, and the obtained KMO (Kaiser-Meyer-Olkin) and Butterest test results show that the correlation coefficient unit matrix cannot be the unit matrix (sig value is less than or equal to 0.05), so that a correlation system exists among variables, so that the main component analysis or factor analysis can be carried out, the KMO statistic is 0.694 (more than 0.6), and the condition is suitable for the factor analysis. And extracting 2 principal component analysis problems according to the characteristic root of more than 1 or the cumulative variance contribution rate of more than 80%. Thus, the original 8 indexes are converted into 2 comprehensive indexes, and the two comprehensive indexes can explain 71.472% of the information in the original 8 indexes.

The expressions for the principal components can be given from the factorial load matrix.

Expression of the first principal component:

a second principal component expression:

in the formula (I), the compound is shown in the specification,

representing the variables after normalization of the original variables.

The weight of the calculation of the composite score is the variance contribution rate after rotation, so the composite factor score is: 0.38126 factor 1 score +0.33346 factor 2 score. Thus, disaster comprehensive scores of each county are obtained and ranked as shown in table 6:

TABLE 6 correlation score and ranking for each administration unit

Averaging the comprehensive scores, then sorting according to an averaging processing result value Q, and dividing disaster grades into four types: if Q is more than or equal to 2, the disaster area is a serious disaster area (code number is '1'), including Wenchuan county and North county; if Q is more than or equal to 1 and less than 2, the disaster area is a serious disaster area (the code is '2'), including Mianzhu city, Shi\37025city and Qingchuan county; if Q is more than or equal to 0 and less than 1, the disaster area is a heavier disaster area (the code number is 3), including eight counties (cities and districts) such as the Mao county; if Q is less than 0, it is a general disaster area (code number is "4"), including other counties.

(2) Modifying the classification according to the detection range of the unmanned aerial vehicle

The influence that unmanned aerial vehicle's flight shooting can receive mountain body obstacle is considered, further brings the shooting range into hierarchical index. Fig. 5 is a diagram showing simulation effects of the detection range of a search and rescue point, in which a white portion is a visible region, the value of the point is 1, a black point in a circular region is a blind region, the value of the point is 0, and the point outside the circular region is not in the detection range.

The method comprises the steps of simulating detection ranges of 100 search and rescue points by taking the point where the search and rescue point of the unmanned aerial vehicle is located and the detection range of 6 kilometers as a standard, obtaining the detection range (the number of 1 s in a corresponding matrix in a statistical chart) of each point to obtain a visual range index in a table 10, carrying out cluster analysis to obtain corresponding grading results, and listing the grading results in the following table, wherein the grades of disasters are changed to a certain extent, and part of final grading results are shown in the table 7

Table 7: grading result table

After disaster conditions of different areas are classified, disaster condition grade attributes of all target rescue points are obtained, and the target rescue points are hoveredThe shooting time is set to different time respectively from high level to low level, and the hovering time of the unmanned aerial vehicle at the suspension point is set as t_iI is 1,2,3,4, i indicates a rank, t_iThe values of (A) are respectively set to 6min, 5.5min, 5min and 4.5 min. Navigation release unmanned aerial vehicle position addressing based on improved binary system bat algorithm

The release positions are provided with 100 alternatives, according to the lowest flying height standard of the navigable mountainous area, the release positions are selected to be located 600m above the obstacle at least, the release positions are selected to be located 600m above the altitude of the release points for releasing the unmanned aerial vehicle, the navigation flying safety is guaranteed, the longitude and latitude coordinates of the alternative points are consistent with those of the search and rescue points, and the altitude is increased by 100m on the basis of the search and rescue points (the relative altitude is 500 m).

(1) Population initialization

First, model A was analyzed, and models B and C were similar. Firstly, generating a full 1 sequence with the length of 100, in order to make the following iteration make the number of the open positions less and less, randomly taking one of the positions as 0, then one of the bats can be [ 011 … 1], and generating the rest bats in the same way, wherein the number of the release positions is not less than 18.

(2) Calculation results

The operation result is obtained by solving according to a Binary Bat Algorithm (BBA) and a binary bat algorithm (IBBA) introduced into a differential evolution mechanism, and the specific result of 8 times of the operation program is shown in Table 8.

TABLE 8 BBA and IBBA solution results

Fig. 6 is a comparison curve of the third operation result of the BBA algorithm and the fourth operation result of the IBBA algorithm (the total cost of the addressing result obtained by both are relatively close to the average value): it can be seen from the figure that compared with the BBA algorithm, the IBBA algorithm improves the defect of premature convergence, the obtained result is more excellent, and under the condition of ensuring that all the same parameters are consistent, according to the 8-time operation results, although the operation time of the IBBA is increased, the total cost of the IBBA algorithm compared with the BBA algorithm is reduced by 4.39% according to the calculation of the average value of the IBBA objective function. Similarly, the model B and the model C are respectively used for carrying out the example analysis, and the IBBA algorithm is obtained, and the total cost is respectively reduced by 6.62 percent and 7.53 percent compared with the BBA. The results of site selection operation for the three models are shown in table 9:

table 9: operating results table

In the results of the three models, the total cost of IBBA is reduced by 6.18 percent on average compared with the result obtained by BBA, which shows that the improved algorithm is obviously improved in the solution quality.

Taking the operation result of IBBA for further research, wherein the corresponding solution is as follows: [ 0000001000001001100000001000010000000110010000000000100000001000000000100110010000000000101000000001 ], namely, 18 positions of 7, 13, 16, 17, 25, 30, 38, 39, 42, 53, 61, 71, 74, 75, 78, 89, 91, 100 and the like are opened as positions for releasing the unmanned aerial vehicle by navigation, wherein the search and rescue point allocation result of the position 53 is as shown in fig. 7 (wherein the opened release positions are marked as pentagons, and the search and rescue points are marked as solid circles):

the results of assigning search and rescue points to each release position are shown in table 10:

table 1018 unmanned aerial vehicle release point addressing results and corresponding task allocation results

Unmanned aerial vehicle task allocation

(1) Data selection: the unmanned aerial vehicle task allocation algorithm selects the release position No. 53 for analysis, and the position is responsible for 18 search points of 7, 13, 16, 17, 25, 30, 38, 39, 42, 53, 61, 71, 74, 75, 78, 89, 91, 100 and the like.

(2) Population initialization: firstly, population initialization is needed, because battery energy consumption constraint is satisfied, for simple calculation, an initial solution is set to be one point, namely each unmanned aerial vehicle searches for one search and rescue point, and total 13 search points exist, then 13 unmanned aerial vehicles are used for the initial solution, and chromosome II codes are [ 12345678910111213 ]. Chromosome I generates the initial population by random means. The parameter settings are shown in table 3.

(3) Solving the result

The results obtained in 8 experiments with a Pop of 30, a Gen of 1000/500/300/200 and a Gen of 1000, respectively, and a pareto front are shown in fig. 9, and the pareto solutions obtained at Pop of 30 and Gen of 1000/500/300/200 are shown in fig. 8(a) and (b), and the pareto solutions obtained at Gen of 1000 and Pop of 100/50/20/30 are shown in fig. 8(c) and (d). From the operation results, it is understood that a large number of pareto solutions can be obtained when the population size is about 30 and the number of iterations is about 1000.

Thus comparing the results obtained for a single target with the results obtained for multiple targets. Two pareto solutions and two sets of solutions obtained by iteration of genetic algorithm solving a single target are taken, and the result pair is shown in table 11:

TABLE 11 comparison of Multi-target model and Single-target model results

The distribution result obtained by analyzing the single-target solution obviously has little difference in the number of the used unmanned aerial vehicles compared with the pareto solution obtained by the multi-target model, and the single-target model does not consider the balance of tasks, so that the difference of the task completion time among the unmanned aerial vehicles is large, but the cost is obviously superior. As can be seen from the results, the results of the multi-objective solution have no cost advantage, but have obvious advantages in the time for completing the task. And for the multi-objective solution result, the required pareto solution can be obtained according to actual requirements. The results focusing on the minimum total search and rescue cost, the minimum number of search and rescue unmanned aerial vehicles and the minimum task difference of the unmanned aerial vehicles are respectively taken for sensitivity analysis, and are shown in table 12:

TABLE 12 Multi-objective model sensitivity analysis emphasizing objective results

TABLE 13 comparison analysis of the results of the multiple target models with emphasis on target (calculation from mean)

As can be seen from table 13, when the pareto solution is focused on the search and rescue cost, the cost values are all decreased slightly compared to the minimum focusing on the UAV number and the search and rescue time difference, and the maximum time difference of the UAV (unmanned aerial vehicle) mission is greatly increased, wherein the value of the cost target is not significantly decreased compared to the target focusing on the UAV usage number; when the UAV usage amount is weighted, the UAV usage amount value decreases compared to the other two objective functions, wherein the UAV usage amount value decreases less significantly than the cost objective; when the pareto solution is focused on the UAV search and rescue maximum time difference to be minimum, compared with the other two objective functions, the time difference values are all reduced to a certain extent, the reduction range is extremely large, and the values of the other two objective functions are all increased to a certain extent.

In actual search and rescue, three indexes of search and rescue time, cost and the number of unmanned aerial vehicles need to be comprehensively balanced, so that search and rescue work is more effective. Generally, on the basis of considering the index of minimum search and rescue balance time, the cost and the number of unmanned aerial vehicles are increased; however, when the number of the actual search and rescue unmanned aerial vehicles is very limited, the investment of time or cost needs to be increased; when the economic benefit is taken as the guide, the search and rescue time is increased.

Claims

1. An unmanned aerial vehicle task allocation method considering operation environment and performance in collaborative search and rescue is characterized by comprising the following steps:

(1) acquiring unmanned aerial vehicle performance data, including maximum endurance time, maximum hover time, flight endurance time under the influence of 3-level low-altitude wind on unmanned aerial vehicle battery energy consumption, a rise limit, a price, maximum charging times, terrain data of a researched area and main eight disaster indicators;

(3) the influence of flight endurance performance, disaster level, wind and terrain of the unmanned aerial vehicle on the endurance time of the unmanned aerial vehicle is considered, the minimum total search and rescue cost is taken as a target, and the problem of determining the position of the unmanned aerial vehicle released in navigation is modeled; improving a binary bat algorithm, introducing a differential evolution mechanism, solving the model, determining the position of the unmanned aerial vehicle released by navigation, and comparing and analyzing the position with a result obtained by the unmodified binary bat algorithm; the optimization model of the position points of the navigation release unmanned aerial vehicle is as follows:

considering the performance parameters of the unmanned aerial vehicle, respectively solving the influence coefficient of low-altitude wind on the energy consumption of the battery and the flying and hovering energy consumption ratio of the rotor unmanned aerial vehicle; the maximum hovering time of the full-electric-quantity unmanned aerial vehicle under the windless condition is set to be T_hoverThe maximum time that can be flown at speed v is T_vThe maximum flight time under the influence of 3-level low-altitude wind on the energy consumption of the unmanned aerial vehicle battery is T_windThen, the influence coefficients of the flying and hovering energy consumption ratio of the unmanned aerial vehicle and the low-altitude wind on the battery energy consumption are respectively expressed as formulas (1) and (2):

wherein gamma is the ratio of flight energy consumption to hovering energy consumption of the rotor unmanned aerial vehicle, and alpha is the influence coefficient of low-altitude wind on the energy consumption of the battery;

when the unmanned aerial vehicle flight task is assigned, the energy consumption problem caused by actual obstacle avoidance needs to be considered; setting the obstacle avoidance energy consumption reservation rate as beta, and estimating and setting according to the actual terrain condition of the mountainous area;

the objective function of the model is that the total search and rescue cost Z is minimum:

constraint conditions are as follows:

y_j′k≤x_j′，j′∈M，k∈N (6)

S_j′k≤S，j′∈M，k∈N (8)

y_j′k＝{0，1}，j′∈M，k∈N (9)

x_j′＝{0，1}，j′∈M (10)

the set M is a position set of alternative navigable release unmanned aerial vehicles, M is {1, 2,. multidot.m }, the set N is a target search and rescue point set of the unmanned aerial vehicles, and N is {1, 2,. multidot.n }; parameter c is the cost per flight distance of the drone, F_j′Releasing the service cost of the unmanned aerial vehicle position j' for the alternative navigation, l is the number of the unmanned aerial vehicle positions selected for the plan, d_j′kReleasing drone position j' to target search and rescue from alternative navigationEuclidean distance of point k, t_kThe hovering time of the unmanned aerial vehicle at a target search and rescue point k, v is the speed of the unmanned aerial vehicle flying at a constant speed, beta is the obstacle avoidance energy consumption reservation rate, and S is the maximum driving mileage of the unmanned aerial vehicle and is calculated according to the endurance time of the unmanned aerial vehicle; s_j′kRepresenting the total voyage required to be consumed from the jth alternative navigation release unmanned aerial vehicle position to the kth search and rescue point, including wind, terrain discount and unmanned aerial vehicle hovering discount; decision variable x_j′Whether the position j ' of the alternative navigation release unmanned aerial vehicle is selected or not is shown, the position j ' of the alternative navigation release unmanned aerial vehicle is 1 if selected, and the position j ' of the alternative navigation release unmanned aerial vehicle is 0 if not selected; decision variable y_j′kWhether the search and rescue point k is served by the position j' of the unmanned aerial vehicle released by the alternative navigation is represented, if so, the search and rescue point k is 1, and if not, the search and rescue point k is 0;

equation (3) is an objective function, representing the total cost minimum, including the total cost of flight, including hover and total cost of service; the formula (4) shows that each search and rescue point is only provided with an unmanned aerial vehicle for search and rescue by one alternative navigation release unmanned aerial vehicle position; formula (5) shows that at most one release position point is selected from alternative navigation release unmanned aerial vehicle positions to release the unmanned aerial vehicle; equation (6) shows that if the search and rescue point k is sent out by the alternative navigation release unmanned aerial vehicle position j ', j' must be the selected alternative navigation release unmanned aerial vehicle position; equation (7) represents that the total voyage consumed by the drone includes the sum of the flight voyage and the hover reduced voyage; the formula (8) represents that the unmanned aerial vehicle releases the search and rescue range constraint from the position of the unmanned aerial vehicle to the search and rescue point from the alternative navigation, and represents that the total voyage consumed by the unmanned aerial vehicle to release the position j' of the unmanned aerial vehicle from the alternative navigation to the search and rescue point k is not more than the maximum voyage of the unmanned aerial vehicle; equations (9) and (10) represent the decision variable y, respectively_j′kAnd x_j′Is a variable from 0 to 1;

(4) after the positions of the unmanned aerial vehicles released by navigation are determined, task allocation of the unmanned aerial vehicles is carried out on the target search and rescue points allocated to each position point: at each position for releasing the unmanned aerial vehicles, an allocation model is established for a target function by considering the search and rescue cost, the using number of the unmanned aerial vehicles and the balance of the unmanned aerial vehicle allocation tasks; the unmanned aerial vehicle search and rescue task allocation model is as follows: first, the objective function is such that the total cost Z is₁Minimum, including the costs incurred by flying the drone andthe cost of the flight distance consumption generated by hovering is proportional to the flight distance consumption of the unmanned aerial vehicle, and the cost of the flight distance consumption generated by hovering is proportional to the hovering time

Wherein: s_ijThe method comprises the steps that the voyage of the unmanned aerial vehicle between a search and rescue point i and a search and rescue point j is converted from energy consumption, wherein the voyage comprises terrain and wind, and the voyage is converted into cruise uniform-speed flight; decision variable y_ijIndicating that 1 if search and rescue point i is visited before search and rescue point j, and 0 otherwise; decision variable x_ijk′Whether the unmanned aerial vehicle K 'flies to a search and rescue point j from the search and rescue point i or not is represented, wherein K' belongs to K, i, j belongs to M and U N, if yes, the unmanned aerial vehicle K 'is 1, and if not, the unmanned aerial vehicle K' is 0; s_iRepresenting the voyage of the energy consumption of the unmanned aerial vehicle at the search and rescue point i, and converting hovering into cruising constant-speed flight; the set P represents a set formed by unmanned aerial vehicle search and rescue points and positions of the unmanned aerial vehicles released by navigation, wherein P is {0, 1,2,3,. and P }, 0 represents the positions of the unmanned aerial vehicles released by navigation, and the set K represents a set of the unmanned aerial vehicles;

secondly, the actual number of drones Z₂There is a limit, and so the number of drones used is to be minimized, where node 0 represents the location where navigation releases a drone:

wherein: x is the number of_ojk′Whether the unmanned aerial vehicle K 'flies to a search and rescue point j from the position where the unmanned aerial vehicle is released by navigation or not is represented, and K' belongs to K;

the balance of tasks is considered, the cost is considered, and the maximum time difference Z for each unmanned aerial vehicle to finish the tasks is also considered₃And (3) reducing to the minimum, namely:

constraint conditions are as follows:

s_i＝αγt_iv，i∈N′ (22)

wherein:

the longest time for the search and rescue task in the unmanned aerial vehicle,

shortest time for search and rescue task in unmanned aerial vehicle, x_ok′The set N' represents a target search and rescue point set of the unmanned aerial vehicles, namely a set P formed by removing elements left by release positions, the set T represents a time set for completing tasks of the unmanned aerial vehicles, and T is { T ═ T { (T } T₁，T₂，T₃，...，T_k’)；t_ijThe flight time of the unmanned plane between the point i and the point j is represented by uniform calculation according to the Euclidean distance and the cruising speed, and t is_iFor the hovering time of the unmanned aerial vehicle at a target search and rescue point i, S {0} represents a set formed by a union of a set of search and rescue points allocated to one unmanned aerial vehicle and a starting point of the unmanned aerial vehicle, and d_ijThe Euclidean distance is a target search and rescue point or a Euclidean distance between the search and rescue point and a release position i, j; decision variable x_ik′Indicating whether the search and rescue point i is served by the unmanned aerial vehicle k', if so, the search and rescue point i is 1, otherwise, the search and rescue point i is 0; the meanings of other letters are completely consistent with the addressing model;

and solving the multi-target model by using an NSGA-II algorithm to obtain a distribution result, and comparing and analyzing the distribution result with a result obtained by the single-target model.