Smartphone-based UAV task scheduling method for post-disaster rescue

Abstract

The present invention discloses a method for scheduling unmanned aerial vehicle (UAV) tasks based on a smartphone for post-disaster rescue, and belongs to the technical field of unmanned aerial vehicle (UAV). The present application proposes to use a smartphone as a mobile auxiliary computing device to help unmanned aerial vehicles (UAVs) complete a large number of intensive computing tasks in post-disaster rescue. The present application takes into account the overhead of UAVs and the utility of smartphones while designing a task offloading strategy. The proposed optimization algorithm can minimize the overhead of UAVs and improve the utility of smartphones. In addition, the task offloading method proposed in the present invention takes into account the diversity of tasks in real scenarios and can meet the service quality requirements of different tasks. Compared with traditional bat algorithms, the multi-objective mutation immune bat algorithm used in the present application has better global solutions and convergence. The present application flexibly uses weight vectors and multi-objective mutation immune bat algorithms, and can also be used for task offloading optimization in other scenarios.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles, and in particular relates to a method for dispatching unmanned aerial vehicle tasks based on a smart phone for post-disaster rescue. Background Art

[0002] In recent decades, the number of natural disasters has increased significantly. In addition to strengthening disaster prevention measures, post-disaster rescue operations are also very necessary. UAVs can be equipped with different types of modules, such as information collection modules, communication modules, computing modules, etc., and UAVs have the advantages of strong data collection capabilities, high maneuverability, and strong adaptability to complex environments. They are widely used in post-disaster rescue.

[0003] In order to ensure the flexibility and maneuverability of drones during flight, drones are usually restricted in weight and size to achieve lightweight and compact design, which also limits the capabilities of the computing module, communication module and battery power module equipped on drones. Therefore, when faced with massive real-time data, the processing capacity of drones is quite limited.

[0004] Existing drones have insufficient computing resources when faced with a large number of intensive computing tasks in post-disaster rescue. This will lead to reduced mission completion efficiency and extended mission completion time, which is unacceptable in disaster rescue missions. Summary of the invention

[0005] In order to solve the problems raised in the above background technology, the present invention provides a drone task scheduling method based on a smartphone for post-disaster rescue, so as to solve the problem of insufficient computing resources of existing drones when facing a large number of intensive computing tasks in post-disaster rescue.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The method for scheduling UAV tasks based on smartphones for post-disaster rescue includes the following steps:

[0008] S1: Construct a system model. The system model includes three types of entities: drones, smartphones, and post-disaster rescue points. n drones are evenly distributed throughout the disaster area. The set of drones is U = {1, 2, 3, …, n}. Each drone covers a preset disaster area with a coverage radius of r0. The flight altitude of the drone is h. Each drone has a task queue T, represented by T = {t1, t2, …, t i ,…,t I}, there are I tasks in the task queue, drones do not communicate with each other and their coverage areas do not overlap, multiple post-disaster rescue points are deployed in the disaster area, each post-disaster rescue point is equipped with a certain number of smartphones, and the set of smartphones in the post-disaster rescue point is M = {1, 2, 3, …, m};

[0009] S2: Model the post-disaster rescue points according to the disaster level and obtain the remaining waiting time Tr for individuals to get rescue at the post-disaster rescue points with their smartphones after the disaster occurs. m ;

[0010] S3: Describe the service quality requirements of UAV tasks through transmission energy consumption and data transmission delay;

[0011] S4: Calculate the path loss, transmission rate, transmission delay, transmission energy consumption from drone n to smartphone m, and the processing task t of drone n i The duration of the task t handled by drone n i The local data preprocessing energy consumption of smartphone m executing task t i The computing energy consumption of the smartphone and the energy consumption of receiving data;

[0012] S5: Calculate the total cost of tasks in the unloaded task queue of drone n and the total utility of the smartphone executing all unloaded tasks of drone n through the calculation data of S4;

[0013] S6: Based on the calculation data of S5 and the remaining waiting time data Tr in S2 m ,define the multi-objective optimization problem of minimizing the total cost of the UAV and maximizing the total utility of the smartphone execution under the ,quality of service requirements of different types of tasks;

[0014] S7: Combine the bat algorithm with the clonal immune algorithm to form a multi-objective mutation immune bat algorithm. The multi-objective mutation immune bat algorithm is used to solve the multi-objective optimization problem and output the final UAV task scheduling method.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] The present application innovatively proposes to use smart phones as mobile auxiliary computing devices to help drones complete a large number of intensive computing tasks in post-disaster rescue. The use of smart phones not only greatly increases the computing power of drones, but also makes full use of idle computing resources. It can help drones to handle a large number of application tasks in a timely manner in disaster situations where infrastructure may be damaged on a large scale, thereby improving the task completion rate and reducing the task completion time. In addition, the present application takes into account the overhead of drones and the utility of smart phones while designing the task offloading strategy. The proposed optimization algorithm can minimize the overhead of drones and improve the utility of smart phones. The task offloading method proposed in the present invention takes into account the diversity of tasks in real scenarios and can meet the service quality requirements of different tasks. Finally, compared with the traditional bat algorithm, the multi-objective mutation immune bat algorithm used in the present application has better global solutions and convergence. The weight vector and multi-objective mutation immune bat algorithm proposed in the present invention are flexible to use, and can be used not only in the optimization of the post-disaster rescue method in this article, but also in the task offloading optimization of other scenarios, and can obtain better performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of the system model of this application;

[0018] Figure 2(a) and Figure 2(b) are schematic diagrams of the overhead of the UAV (UVA) and the execution utility of the smartphone (SP) in mild and severe disaster scenarios, respectively, when using different algorithms under default parameters;

[0019] Figure 3(a) and Figure 3(b) are schematic diagrams showing the change in the execution utility of the smartphone (SP) and the change in the overhead of the drone (UVA) when the number of tasks changes in a mild disaster scenario, respectively;

[0020] Figure 4(a) and Figure 4(b) are schematic diagrams showing the changes in UAV overhead and the changes in the utility of smartphones (SP) when the number of smartphones (SP) changes under a mild disaster situation, respectively;

[0021] Figure 5(a) and Figure 5(b) are schematic diagrams of the overhead change of the UAV and the execution utility change of the smartphone (SP) when the multi-objective mutation immune bat algorithm (MOMIB) changes iteratively;

[0022] Figure 6(a) and Figure 6(b) are schematic diagrams of the overhead changes of the unmanned aerial vehicle (UAV) and the execution utility changes of the smartphone (SP), respectively, when the multi-objective mutation immune bat algorithm (MOMIB) changes iteratively in a real environment. DETAILED DESCRIPTION

[0023] In order to facilitate those skilled in the art to understand the technical content of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific examples. It should be understood that the specific examples described herein are only used to explain the present invention and are not used to limit the present invention.

[0024] Example 1

[0025] The method for scheduling UAV tasks based on smartphones for post-disaster rescue includes the following steps:

[0026] S1: Build a system model, such as Figure 1 As shown in the figure, the system model contains three types of entities, namely, drones (UVA), smartphones (SP) and disaster relief centers (DRC). Assume that n drones are evenly distributed in the entire disaster area. The set of drones is represented by U = {1, 2, 3, ..., n}. Each drone has a task queue, and the set of tasks in the queue is T = {t1, t2, ..., t i ,…,t I}, and each drone covers a certain range of ground area, the radius of the coverage area is r0, the flight altitude of the drone is set to h, assuming that the drones do not communicate with each other and the coverage area of ​​each drone does not overlap. Figure 1 In the above, each drone is responsible for collecting data in its coverage area. When there are many tasks or the amount of task data is large, the drone will offload part of the tasks to the smartphones within its coverage area to perform calculations. Usually, after a disaster occurs, multiple post-disaster relief points need to be deployed in the disaster area. The post-disaster relief points can provide different types of assistance to nearby residents according to the severity of the disaster, such as accommodation and relief supplies. There are a certain number of smartphones in each post-disaster relief point, which are carried by the people in the post-disaster relief point. For a drone n, suppose there are D post-disaster relief points within its range, and there are a certain number of smartphones in each post-disaster relief point. Suppose there are no post-disaster relief points. The set of smartphones in human-machine n is M = {1, 2, 3, ..., m}. In order to encourage smartphones to assist drones in performing unloading tasks, the system can provide incentives for the owners of smartphones who perform unloading tasks. According to the operating experience of P2P file systems, about 30% of users are willing to provide assistance for the common interests of the group. These users are sufficient to support the daily operation of the system. For relatively selfish smartphone users, the system can incentivize mobile phone carriers to complete tasks by issuing rewards, such as issuing shopping coupons, call coupons, etc. or cash rewards. In addition, the drone acts as a system controller to perform task offloading and multi-objective optimization tasks.

[0027] S2: Model the post-disaster rescue points according to the disaster level and obtain the remaining waiting time Tr for individuals to get rescue at the post-disaster rescue points with their smartphones after the disaster occurs. m; Since different disaster magnitudes cause different degrees of damage to the surface, the present invention takes earthquake disasters as an example and establishes two different levels for earthquake disasters, namely, mild disaster and severe disaster scenarios; Mild disaster: If the earthquake magnitude is less than level D, the earthquake will naturally cause damage to some infrastructure, but the damage to buildings is relatively light, and people can still live in their own homes. In this case, the post-disaster relief point only needs to distribute disaster relief supplies to people, which means that people can dynamically enter and exit the post-disaster relief point; Assume that the probability of entering each post-disaster relief point obeys a Poisson process with an average arrival rate of λ, and assume that each post-disaster relief point has only one service window and a service speed of μ. At this time, the process of people arriving at the post-disaster relief point follows the M / M / 1 queuing model. According to this model, the average waiting time T of people in the post-disaster relief point is w for:

[0028] Among them, T q and T s They represent the average queuing time and the average service time respectively. For smartphone m, its remaining waiting time Tr in the post-disaster rescue point m is: Tr m =T w -T m ;

[0029] Where T m is the time smartphone m has been waiting for service in the post-disaster relief point;

[0030] Severe disaster: When the magnitude is ≥ D, the earthquake causes serious damage to the city. In this case, the post-disaster relief point not only needs to distribute necessary survival supplies to the people nearby, but also needs to provide them with accommodation. In this case, the present invention considers that the smartphone is located in the post-disaster relief point and the position is relatively fixed. At this time, the remaining stay time of the smartphone m in the post-disaster relief point is described as: Tr m =∞.

[0031] S3: Describe the service quality requirements of UAV tasks through transmission energy consumption and data transmission delay;

[0032] Quality of Service (QoS) is an important metric for tasks. In actual scenarios, different types of tasks have different QoS requirements. Real-time tasks, video tasks, audio tasks, and image tasks have different QoS requirements. The QoS requirements of different types of tasks are shown in Table 1.

[0033] Table 1 QoS requirements for various applications

[0034]

[0035] Data transmission rate and data transmission delay are the most commonly used parameters to describe the QoS requirements of tasks. Since the transmission energy consumption is proportional to the data transmission rate, the present invention selects transmission energy consumption and data transmission delay to describe the QoS requirements of tasks. As shown in Table 2, the QoS levels of these two indicators are set from l1 to l4 according to the corresponding values ​​in Table 1, indicating that the QoS level is from strong to weak.

[0036] Table 2 QoS priorities for different task types

[0037]

[0038] For example, since image data has the lowest latency requirement among all data types in Table 1, the QoS latency level of the image task is set to l1.

[0039] To describe the QoS requirements of tasks, we define the task QoS vector. i , whose QoS vector is defined as η i ={η i，1 ,η i，2}, η i,1 and η i，2 They represent the weights of transmission energy consumption and data transmission delay respectively, and their calculation method is shown in Algorithm 1.

[0040] Algorithm 1: Calculate the weight coefficient of the QoS vector;

[0041] Input: T, P∈R 5×2 ,A i ∈[1,5],l1=1,l2=2,l3=3,l4=4;

[0042] Output: weight coefficient matrix;

[0043]

[0044]

[0045] where P∈R 5×2 Represents the QoS weight matrix, mapped from Table 2, A i Indicates the task type. The algorithm first inputs the task queue information and initializes the matrix-related variables. Then the two weight coefficients corresponding to different variables are calculated respectively (lines 3-6). After that, the two weights are normalized again to obtain the final weight coefficient. Finally, the obtained QoS weight coefficient value is added to the weight coefficient list.

[0046] In addition, the system gives a certain reward to the smartphone that performs the task. i , basic incentive R i For: R i =αSi ;

[0047] Where α is a constant, S i is the data size of a single task;

[0048] Punishment i For: i =D i / (D i -T i al );

[0049] Among them, D i is the maximum tolerable delay of the task, T i al Is to perform the task t i The total delay of the task, the reward of the smartphone performing the task re m，i for:

[0050] S4: Calculate the path loss, transmission rate, transmission delay, transmission energy consumption from drone n to smartphone m, and the processing task t of drone n i The duration of the task t handled by drone n i The local data preprocessing energy consumption of smartphone m executing task t i The computing energy consumption of the smartphone and the energy consumption of receiving data;

[0051] Assume that the position of drone n is Ln = [xn, yn, hn], where hn = h, and the position of smartphone m in the post-disaster rescue point is represented by Lm = [xm, ym, hm], where hm = 0, and the path loss from drone n to smartphone m is δ n，m for:

[0052] where ε ref，n Indicates the received power;

[0053] The transmission rate r from drone n to smartphone m n,m for:

[0054] Among them B n,m is the transmission bandwidth between the two, P n,m is the transmission power from drone n to smartphone m, σ 2 is the power spectral density of Gaussian noise;

[0055] The task t is transmitted between drone n and smartphone m i Transmission delay It is expressed as:

[0056] where s i For task t i The size of the drone n performs a certain degree of data preprocessing before sending the data to the smartphone m. The duration of data preprocessing for:

[0057] where f n is the CPU computing resource of drone n allocated to task preprocessing, c is the CPU resource required to process each bit of data, and I is the number of tasks in the task queue;

[0058] Transmission energy consumption from drone n to smartphone m for:

[0059] The energy consumption of local data preprocessing of drone n for this task for:

[0060] K n is the power coefficient of UAV n;

[0061] Smartphone m performs task t i Computational energy consumption for:

[0062] Where K m is the power factor of smartphone m, f m is the CPU cycle frequency of smartphone m used for data calculation;

[0063] Smartphones also generate energy consumption for receiving data for:

[0064] P m is the data receiving power of smartphone m.

[0065] S5: Calculate the total cost of tasks in the unloaded task queue of drone n and the total utility of the smartphone executing all unloaded tasks of drone n through the calculation data of S4;

[0066] Drone offloading mission i Total delay and energy consumption Respectively expressed as:

[0067] Considering the service quality requirements of the task, the cost of unloading the drone of this task is cos n,i for:

[0068] For drone n, the total overhead of offloading tasks in the task queue for:

[0069]

[0070] Smartphone m performs task t i Total energy consumption cos m，i It is expressed as:

[0071] Considering the cost and incentive of smartphones to perform tasks, smartphone m performs task t i The utility of U m,i For: m，i =re m，i / cos m，i ;

[0072] The total utility of smartphone m performing all offloaded tasks of drone n for:

[0073] S6: Based on the calculation data of S5 and the remaining waiting time data Tr in S2 m ,define the multi-objective optimization problem of minimizing the total cost of the UAV and maximizing the total utility of the smartphone execution under the ,quality of service requirements of different types of tasks;

[0074] Most studies only consider the performance of the task publisher when designing task scheduling, which makes it difficult for these methods to guarantee the interests of the task executors. In contrast, some studies focus on maximizing the interests of the task executors when allocating offload tasks, which may lead to low task completion rates and long task execution delays. The goal of this paper is to minimize the overhead of the drone while maximizing the execution utility of the smartphone. In addition, when modeling the overhead of the drone, the QoS requirements of different types of tasks in real environments are also considered.

[0075] The multi-objective optimization problem of the present invention is defined as:

[0076]

[0077] stC1:e m >e min , m∈M;

[0078] C2:D min =min{D1, D2, ..., D i}, i∈I;

[0079] C3:Tr m >D min , m∈M;

[0080] C4:

[0081] C5:

[0082] Among them, C1 represents the power limit of the smartphone SP. If the remaining power of the smartphone is less than emin, it will not be used to execute the task. C2 and C3 indicate that smartphones with remaining waiting time lower than the minimum delay requirement of the offloaded task cannot be used for task execution. C4 and C5 limit the number of tasks executed by each smartphone.

[0083] S7: Combine the bat algorithm with the clonal immune algorithm to form a multi-objective mutation immune bat algorithm. The multi-objective mutation immune bat algorithm is used to solve the multi-objective optimization problem and output the final UAV task scheduling method.

[0084] Pareto optimality is a good method for solving multi-objective optimization problems. It is usually used to achieve a good trade-off between multiple objectives. It can find an unimprovable equilibrium solution in multi-objective optimization. Therefore, based on the Pareto optimal idea and the multi-objective evolutionary algorithm framework, the present invention proposes a multi-objective mutation immune bat (multi-objective optimization problem, multi-objective mutation immune bat MOMIB) algorithm. It combines the bat algorithm (BA) with the clonal immune algorithm to solve the above optimization problem. As a classic multi-objective optimization algorithm, BA has fast convergence speed and good performance. However, due to the lack of sufficient mutation mechanism to continuously enrich the population, it is easy to fall into a local optimal solution during the optimization process. Therefore, the present invention introduces an immune algorithm with mutation ability into BA, and designs a population adjustment evaluation algorithm according to the characteristics of the problem, and finally obtains a multi-objective mutation immune bat algorithm. The algorithm can obtain a wider solution optimization domain while ensuring rapid convergence, thereby obtaining better performance and more targeted optimization capabilities.

[0085] In the first part of the multi-objective mutation immune bat algorithm, each bat is represented as a feasible solution. During the t-th generation optimization process, the i-th generation can be represented as a vector <x i (t), v i (t), fr i (t), A i (t), r i (t)>, where x i (t) = (x i1 (t), x i2 (t), ..., x ik (t), ..., x iD (t)) represents the position coordinates of the ith bat in the tth generation, vi (t)=(v i1 (t), v i2 (t), ..., v ik (t), ..., v iD (t)) represents the speed direction of the ith bat in the tth generation, fr i (t), A i (t) and r i (t) are the frequency, loudness and pulse emission frequency required for bat movement. At the beginning of the iteration, the frequency is first generated as:

[0086] fr i (t) = fr min +(fr max -fr min )·rand1;

[0087] Where rand1 is a random number in the range of (0,1), max and fr min They are the preset upper and lower frequency limits respectively;

[0088] Bat speed updated to: v ik (t+1)=v ik (t)+(x ik (t)-p k (t))·fr i (t);

[0089] In the above formula represents the optimal position in the history of the previous t generations of the population. Then, based on the random number rand2 distributed in (0,1), choose to execute the global update formula or the local update formula. When rand2<r i (t), the i-th bat performs a global search, otherwise, it performs a local search, where: The global update formula is x′ i (t+1)=x i (t)+v i (t+1); the local update formula is ε i is a random number uniformly distributed in (-1,1), is the average loudness of the bats in the tth generation;

[0090] During the position update process, a random number rand3 is generated, which is uniformly distributed in (0,1). i (t) and When both conditions are met, the position of the i-th bat is updated to Otherwise, the position of the i-th bat remains

[0091]

[0092] Finally, update the pulse emission frequency and loudness to: r i (t + 1)= r i (0).(1 - e -rt ); A i (t + 1)= βA i (t);

[0093] where both γ and β are preset parameters greater than 0, r i (0) and A i (0) are the initial values of the pulse emission frequency and loudness respectively, A i (0) is a random number belonging to [0, 1], r i (0)= 0.1, A i (t) and r i (t) represent the loudness and pulse emission frequency that need to be used during the movement of the bat respectively;

[0094] The second part of the multi - objective mutation immune bat algorithm is the immune algorithm. The population B = {b1, b2,... b n} can be obtained from the bat algorithm part and input into the immune algorithm part. During the immune cloning process, each bat in the population is regarded as an antibody. The detailed steps are as follows:

[0095] S7.1: Randomly select k (k < n) antibodies, denoted as the sub - population Sub1 = {b′1, b′2,..., b′ k} and then clone and replicate the sub - population to obtain a new population sub2. The cloning rule is as follows:

[0096] Sub2 = T c (Sub1)= [T c (b′1), T c (b′2),..., T c (b′ k )];

[0097] where T c (b′ i ) = q i × b′ i (i = 1, 2,..., k), q i is the cloning number of b′ i and is proportional to the fitness of b′ i ;

[0098] S7.2: Perform high - frequency mutation on each individual in the sub2 population. The mutation operator can adapt to evolution in terms of the number of iterations and individual fitness. For any individual b′ in sub2i =(b′ i1 , b′ i2 , ..., b′ ij , ..., b′ in ), perform the following mutation:

[0099]

[0100] The calculation formula of η(t) is:

[0101] Where b is a normal number, r1∈(0,1);

[0102] In early iterations, is small, making η(t)≈1, indicating that the mutation probability is large. In subsequent iterations, the population characteristics gradually approach the ideal state in evolution, and η(t)≈0, allowing local search in a small space. Further expansion of △ yields:

[0103] Where r2∈(0,1), τ is used to adjust the range of the search domain and takes the value in [2,5]. R in the above formula is expressed as:

[0104] Where fit(b′ i ) indicates antibody b′ i The fitness, fit max is the maximum fitness value, that is, the aggregation density of individuals;

[0105] S7.3: Select the best individuals from the cloned mutant individuals as the next generation;

[0106] The third part of the multi-target mutation immune bat is the population adjustment evaluation algorithm, which is as follows:

[0107] First, the obtained solution set is screened according to the constraints in the multi-objective optimization problem P, and the solutions that do not meet the constraints are eliminated. Since the data distribution of the two optimization indicators is significantly different, their distribution ranges are uniformly adjusted according to their data distribution, so that the importance weights of the two indicators in the final ranking are consistent;

[0108] Then, the adjusted weighted sum result is used as the ranking index. After iterating the immune part, the ranking index of each solution in the solution set is calculated and the solutions are ranked using the index.

[0109] Finally, the solution set is pruned according to the group size required for the iteration, and the current best optimized solution is retained for the next iteration until the iteration limit is reached or the optimal solution is obtained, and the final UAV task scheduling method is output.

[0110] Based on the wide distribution of smartphones and their increasingly powerful computing capabilities, this application innovatively proposes using smartphones as mobile auxiliary computing devices to help drones complete a large number of intensive computing tasks in post-disaster rescue. The use of smartphones not only greatly increases the computing power of drones, but also makes full use of idle computing resources. In disaster situations where infrastructure may be widely damaged, it can still help drones process a large number of application tasks in a timely manner, thereby improving the task completion rate and reducing the task completion time.

[0111] In addition, this application takes into account the overhead of drones and the utility of smartphones when designing the task offloading strategy. The proposed optimization algorithm can minimize the overhead of drones and improve the utility of smartphones. The task offloading method proposed in the present invention takes into account the diversity of tasks in real scenarios and can meet the service quality requirements of different tasks.

[0112] Finally, compared with the traditional bat algorithm, the multi-objective mutation immune bat algorithm used in this application has better global solution and convergence. The weight vector and multi-objective mutation immune bat algorithm proposed in the present invention are flexible to use. They can be used in the optimization of the post-disaster rescue method in this article, and can also be used for task offloading optimization in other scenarios, and can obtain better performance.

[0113] The present invention evaluates the performance of the proposed algorithm through simulation experiments. The system framework is constructed using the Python environment. The main simulation parameters are shown in Table 3.

[0114] Table 3 Parameter value settings in the experiment

[0115]

[0116]

[0117] In the experiment, the communication coverage of the UAV is set to 30m, and different methods are used to distribute SPs in the UAV coverage area according to different disaster scenarios. In the case of a mild disaster, a UAV can cover 2-3 DRCs, the arrival rate of SPs is 0.02 to 0.09 per second, and the distribution speed of survival supplies is 0.05 to 0.1 per second.

[0118] The baseline methods used for comparison are as follows: 1) STS (Stochastic Task Scheduling Algorithm): Randomly match unloaded tasks to random SPs; 2) ETP: Emergency Task Priority Scheduling Algorithm (ETP), which prioritizes high-urgency tasks (the smaller the remaining delay of the task, the more urgent it is) to be executed by the nearest SP. 3) Original bat algorithm (OB): This algorithm is the original bat algorithm without improvement.

[0119] 1. Experiment under default parameters: Figure 2 shows the final optimization results of different algorithms. It can be seen that the algorithm proposed in the present invention is far superior to STS. In addition, compared with the ETP algorithm using relatively monotonic optimization indicators, the MOMIB algorithm also achieves better performance. In addition, the performance of the algorithm proposed in the present invention is also higher than that of the traditional OB algorithm. The algorithm proposed in the present invention has the highest SP utility and the lowest UAV energy consumption overhead. This shows the effectiveness of the algorithm proposed in the present invention and reflects the good optimization ability of the MOMIB algorithm for multi-objective problems.

[0120] 2. Experiments on changing the number of tasks: While keeping the number of SPs unchanged, the number of tasks was changed, the number of SPs was set to 70, and the number of tasks was gradually increased from 35 to 60. Figure 3 shows the optimization of each algorithm for multiple objectives in a mild disaster, as well as the comparison with other algorithms. It can be seen that as the number of tasks increases, the overall overhead of the UAV increases accordingly, and the utility of SPs also increases relatively. This is because the more the number of tasks, the greater the unloading overhead of the drone, and the benefit of the smartphone performing the task will also increase accordingly. Compared with other comparison algorithms, the optimization algorithm proposed in the present invention can always optimize multiple objectives well, find non-inferior solutions, and obtain better optimization results.

[0121] 3. Experiments after changing the number of SPs: In this experiment, the number of tasks is set to 40, and the number of SPs is increased from 50 to 80. In Figure 4, the performance of the MOMIB algorithm in a mild disaster scenario is compared with the performance of other algorithms. It can be seen that as the number of smartphones increases, the corresponding computing resources are becoming more and more sufficient, the computing power of the system has been gradually increased, and the choice of smartphones for performing offload tasks is wider, so the energy consumption of drones in each algorithm is gradually reduced, and the utility of mobile phones is gradually improved. With the continuous increase of computing resources, the performance of MOMIB is always better than other comparison algorithms.

[0122] 4. Algorithm Iteration Convergence Experiment: As shown in Figure 5, we experimented with the iterative convergence of the MOMIB algorithm in two disaster environments. It can be seen that at the beginning of the iteration, the UAV overhead and the utility of the mobile phone have not yet achieved the best optimization effect. With the continuous iteration of the algorithm, the two optimization objectives are continuously optimized. By continuously optimizing the task allocation scheme, the UAV overhead is reduced, the utility of the SP is gradually improved, and convergence is basically achieved within the ninth cycle, reflecting the rapid convergence ability of the MOMIB algorithm. Figure 6 shows the iterative performance of MOMIB in a real environment. In the experiment of Figure 6, various applications are allocated at a ratio of 35%, 15%, 20%, 20%, and 10%. It can be seen that at this time, the MOMIB algorithm also achieves fast performance under different task ratios, further verifying the stability of the proposed algorithm.

Claims

1. A method for scheduling UAV tasks based on smartphones for post-disaster rescue, characterized in that: The following steps are involved: S1: Construct a system model. The system model includes three types of entities: drones, smartphones, and post-disaster rescue points. n drones are evenly distributed throughout the disaster area. The set of drones is U = {1, 2, 3, …, n}. Each drone covers a preset disaster area with a coverage radius of r0. The flight altitude of the drone is h. Each drone has a task queue T, represented by T = {t1, t2, …, t i ,…,t I }, there are I tasks in the task queue, drones do not communicate with each other and their coverage areas do not overlap, multiple post-disaster rescue points are deployed in the disaster area, each post-disaster rescue point is equipped with a certain number of smartphones, and the set of smartphones in the post-disaster rescue point is M = {1, 2, 3, …, m}; S2: Model the post-disaster rescue points according to the disaster level and obtain the remaining waiting time Tr for individuals to get rescue at the post-disaster rescue points with their smartphones after the disaster occurs. m ; S3: Describe the service quality requirements of UAV tasks through transmission energy consumption and data transmission delay; S4: Calculate the path loss, transmission rate, transmission delay, transmission energy consumption from drone n to smartphone m, and the processing task t of drone n i The duration of the task t handled by drone n i The local data preprocessing energy consumption of smartphone m executing task t i The computing energy consumption of the smartphone and the energy consumption of receiving data; S5: Calculate the total cost of tasks in the unloaded task queue of drone n and the total utility of the smartphone executing all unloaded tasks of drone n through the calculation data of S4; S6: Based on the calculation data of S5 and the remaining waiting time data Tr in S2 m ,define the multi-objective optimization problem of minimizing the total cost of the UAV and maximizing the total utility of the smartphone execution under the ,quality of service requirements of different types of tasks; S7: Combine the bat algorithm with the clonal immune algorithm to form a multi-objective mutation immune bat algorithm. The multi-objective mutation immune bat algorithm is used to solve the multi-objective optimization problem and output the final UAV task scheduling method.

2. The method for dispatching unmanned aerial vehicle tasks based on smartphones for post-disaster rescue according to claim 1 is characterized in that: In S2, two different disaster levels, namely, mild disaster and severe disaster, are established to model post-disaster relief points, specifically: When the disaster level is a mild disaster, assume that the probability of individuals entering each post-disaster relief point follows a Poisson process with an average arrival rate of λ, and assume that each post-disaster relief point has only one service window and the service speed is μ At this time, the process of people arriving at the post-disaster relief point follows the M / M / 1 queuing model, so the average waiting time of people at the post-disaster relief point is T w for: Among them, T q and T s They represent the average queuing time and the average service time respectively. For smartphone m, its remaining waiting time Tr in the post-disaster rescue point m for: Tr m =T w -T m ; Where T m is the time smartphone m has been waiting for service in the post-disaster relief point; When the disaster level is a severe disaster, assume that the smartphone is located in the post-disaster rescue point and its position is relatively fixed. At this time, the remaining waiting time Tr of the smartphone m in the post-disaster rescue point is m Described as: Tr m ＝∞。 3. The method for dispatching unmanned aerial vehicle tasks based on smartphones for post-disaster rescue according to claim 1 is characterized in that: In S3, the task t i The service quality vector is defined as η i ={η i，1 ,η i，2 }, η i,1 and η i，2 denote the weights of transmission energy consumption and data transmission delay respectively. In addition, the reward for smartphones to perform tasks is set; For task t i , basic incentive R i for: R i =αS i ; Where α is a constant, S i is the data size of a single task; Punishment i for: l i =D i / (D i -T i al ); Among them, D i is the maximum tolerable delay of the task, T i al Is to perform the task t i The total delay of the task, the reward of the smartphone performing the task re m，i for:

4. The method for dispatching unmanned aerial vehicle tasks based on smartphones for post-disaster rescue according to claim 1 is characterized in that: S4 is specifically: Assume that the position of drone n is Ln = [xn, yn, hn], where hn = h, and the position of smartphone m in the post-disaster rescue point is represented by Lm = [xm, ym, hm], where hm = 0, and the path loss from drone n to smartphone m is δ n，m for: where ε ref，n Indicates the received power; The transmission rate r from drone n to smartphone m n,m for: Among them B n,m is the transmission bandwidth between the two, P n,m is the transmission power from drone n to smartphone m, σ 2 is the power spectral density of Gaussian noise; The task t is transmitted between drone n and smartphone m i Transmission delay It is expressed as: where s i For task t i The size of the drone n performs a certain degree of data preprocessing before sending the data to the smartphone m. The duration of data preprocessing for: where f n is the CPU computing resource of drone n allocated to task preprocessing, c is the CPU resource required to process each bit of data, and I is the number of tasks in the task queue; Transmission energy consumption from drone n to smartphone m for: The energy consumption of local data preprocessing of drone n for this task for: K n is the power coefficient of UAV n; Smartphone m performs task t i Computational energy consumption for: Where K m is the power factor of smartphone m, f m is the CPU cycle frequency of smartphone m used for data calculation; Smartphones also generate energy consumption for receiving data for: P m is the data receiving power of smartphone m.

5. The method for dispatching drone tasks based on smartphones for post-disaster rescue according to any one of claims 3 or 4, characterized in that: S5 is specifically: Drone offloading mission i Total delay and energy consumption Respectively expressed as: Considering the service quality requirements of the task, the cost of unloading the drone of this task is cos n,i for: For drone n, the total overhead of offloading tasks in the task queue for: Smartphone m performs task t i Total energy consumption cos m，i It is expressed as: Considering the cost and incentive of smartphones to perform tasks, smartphone m performs task t i The utility of U m,i for: U m，i =re m，i / cos m，i ; The total utility of smartphone m performing all offloaded tasks of drone n for:

6. The method for dispatching unmanned aerial vehicle tasks based on smartphones for post-disaster rescue according to claim 5 is characterized in that: The multi-objective optimization problem is defined as: Among them, C1 represents the power limit of the smartphone SP. If the remaining power of the smartphone is less than emin, it will not be used to execute the task. C2 and C3 indicate that smartphones with remaining waiting time lower than the minimum delay requirement of the offloaded task cannot be used for task execution. C4 and C5 limit the number of tasks executed by each smartphone.

7. The method for dispatching unmanned aerial vehicle tasks based on smartphones for post-disaster rescue according to claim 6, characterized in that: S7 is specifically: In the first part of the multi-objective mutation immune bat algorithm, each bat is represented as a feasible solution. During the t-th generation optimization process, the i-th generation can be represented as a vector <x i (t), v i (t), fr i (t), A i (t), r i (t)>, where x i (t) = (x i1 (t), x i2 (t), ..., x ik (t), ..., x iD (t)) represents the position coordinates of the ith bat in the tth generation, v i (t)=(v i1 (t), v i2 (t), ..., v ik (t), ..., v iD (t)) represents the speed direction of the ith bat in the tth generation, fr i (t), A i (t) and r i (t) are the frequency, loudness and pulse emission frequency required for bat movement. At the beginning of the iteration, the frequency is first generated as: fr i (t)=fr min +(fr max -fr min )·rand1; Where rand1 is a random number in the range of (0,1), max and fr min They are the preset upper and lower frequency limits respectively; Bat speed updated to: v ik (t+1)=v ik (t)+(x ik (t)-p k (t))·fr i (t); In the above formula represents the optimal position in the history of the previous t generations of the population. Then, based on the random number rand2 distributed in (0,1), choose to execute the global update formula or the local update formula. When rand2<r i (t), the i-th bat performs a global search, otherwise, it performs a local search, where: The global update formula is x′ i (t+1)=x i (t)+v i (t+1); The local update formula is: ε i is a random number uniformly distributed in (-1,1), is the average loudness of the bats in the tth generation; During the position update process, a random number rand3 is generated, which is uniformly distributed in (0,1). i (t) and When both conditions are met, the position of the i-th bat is updated to Otherwise, the position of the i-th bat remains Finally, update the pulse emission frequency and loudness to: r i (t+1)=r i (0).(l-e -rt ); And i (t+1)=βA i (t); Among them, γ and β are preset parameters greater than 0, r i (0) and A i (0) are the initial values ​​of pulse emission frequency and loudness, A i (0) is a random number in [0,1], r i (0) = 0.1, A i (t) and r i (t) represent the loudness and pulse emission frequency required to be used during the movement of the bat; The second part of the multi-objective mutation immune bat algorithm is the immune algorithm, where the population B = {b1, b2, …b n } can be obtained from the bat algorithm part and input into the immune algorithm part. In the immune cloning process, each bat in the population is regarded as an antibody. The detailed steps are as follows: S7.1: Randomly select k (k < n) antibodies, denoted as subgroup Sub1 = {b′1, b′2,..., b′ k}, and then clone and replicate the subpopulation to obtain a new population sub2. The cloning rules are as follows: Sub2=T c (Sub1)=[T c (b1),T c (b2),...,T c (b k )]; Where T c (b′ i )=q i ×b′ i (i=1, 2, ..., k), q i b′ i The number of clones and b′ i is proportional to the fitness; S7.2: Perform high-frequency mutation on each individual in the sub2 population. The mutation operator can adapt to evolution in terms of the number of iterations and individual fitness. For any individual b′ in sub2 i =(b′ i1 , b′ i2 , ..., b′ ij , ..., b′ in ), perform the following mutation: The calculation formula of η(t) is: Where b is a positive constant, r1∈(0,1); In early iterations, is small, making η(t)≈1, indicating that the mutation probability is large. In subsequent iterations, the population characteristics gradually approach the ideal state in evolution, and η(t)≈0, allowing local search in a small space. Further expansion of △ yields: Where r2∈(0,1), τ is used to adjust the range of the search domain and takes the value in [2,5]. R in the above formula is expressed as: Where fit(b′ i ) indicates antibody b′ i The fitness, fit max is the maximum fitness value, that is, the aggregation density of individuals; S7.3: Select the best individuals from the cloned mutant individuals as the next generation; The third part of the multi-target mutation immune bat is the population adjustment evaluation algorithm, which is as follows: First, the obtained solution set is screened according to the constraints in the multi-objective optimization problem P, and the solutions that do not meet the constraints are eliminated. Since the data distribution of the two optimization indicators is significantly different, their distribution ranges are uniformly adjusted according to their data distribution, so that the importance weights of the two indicators in the final ranking are consistent; Then, the adjusted weighted sum result is used as the ranking index. After iterating the immune part, the ranking index of each solution in the solution set is calculated and the solutions are ranked using the index. Finally, the solution set is pruned according to the group size required for the iteration, and the current best optimized solution is retained for the next iteration until the iteration limit is reached or the optimal solution is obtained, and the final UAV task scheduling method is output.

Images