Unmanned aerial vehicle trajectory planning and task unloading decision-making method based on joint optimization

Abstract

The invention discloses an unmanned aerial vehicle track planning and task unloading decision-making method based on joint optimization, which comprises the following steps of: 1, optimizing an unloading decision of a flight unmanned aerial vehicle track and a mobile equipment task under a double-layer unmanned aerial vehicle cooperative communication network framework consisting of a centralized flight unmanned aerial vehicle and a group of distributed unmanned aerial vehicles; 2, obtaining an unloading decision and a dispatching decision of a mobile equipment task by using a Wolf-PHC-based method and a game theory-based mode; and 3, based on the obtained unloading decision and dispatching decision, applying a DDPG algorithm to solve the trajectory optimization problem of the flying unmanned aerial vehicle. The method can effectively solve the problems of limited computing resources and overlong task execution time under the scene that the electric quantity of the mobile equipment and the electric quantity of the unmanned aerial vehicles are limited, and finds an optimal unloading strategy through cooperation and trajectory optimization between the unmanned aerial vehicles, so that the total delay of task completion is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of unmanned aerial vehicle trajectory planning and task offloading, and specifically is a unmanned aerial vehicle trajectory planning and task offloading decision method based on joint optimization. Background Art

[0002] With the rapid development of mobile Internet, intelligent applications such as augmented reality, autonomous driving, and health monitoring are emerging. These new applications have several common characteristics, namely, low latency tolerance, high energy consumption, and computational intensiveness. Due to the limited computing power and energy budget of mobile devices, it is difficult to meet the computing needs of such complex applications. In this case, the contradiction between the various workloads of computationally intensive applications and the limited computing resources of mobile devices poses a challenge to user experience. Therefore, the tasks generated by mobile devices can be offloaded from the local to the multi-access edge computing server (MEC) located at the edge of the network. In contrast to mobile cloud computing, MEC can significantly reduce transmission latency because the MEC server is closer to the device than the cloud server. However, the fixed location of the MEC server limits the coverage range, making it impossible for mobile devices to directly access access points (APs) or MECs in areas such as mountains, forests, deserts, underwater, or rural areas. In addition, in environments such as disaster response, emergency rescue, or military exercises, some ground base stations (BS) may be damaged by natural disasters or military attacks, and it takes a lot of time to install new BSs or APs, and mobile devices in these areas cannot get help at this time.

[0003] Although drone systems with MEC servers have great potential and provide better and faster computing services than traditional fixed-location MEC systems, there are still some challenges in adopting this technology in the real world. First, drone systems that support MEC have lower computing energy than cloud servers and traditional MEC systems, resulting in a greater risk of crashing and causing property damage if the mobile device fails to complete its mission. Second, mobile devices on the ground usually present different distribution scenarios, and existing methods cannot ensure a stable connection between mobile devices and drones while ensuring flexibility; third, mobile devices in real-world scenarios usually have real-time tasks and a certain degree of mobility, which is not considered in existing scenarios. Summary of the invention

[0004] The present invention aims to solve the deficiencies of the above-mentioned prior art and proposes a UAV trajectory planning and task offloading decision method based on joint optimization, in order to find an optimal offloading strategy, thereby reducing the total delay in task completion.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme:

[0006] The multi-UAV layout and task offloading decision method based on multi-objective joint optimization of the present invention is characterized in that it is applied to a In a square ground area V, a vertex of the square ground area V is set as the origin o, two sides connected to the origin o are set as the X axis and the Y axis respectively, and a vertical line perpendicular to the XY plane and passing through the origin o is set as the Z axis, thereby establishing a three-dimensional coordinate system;

[0007] Deploy a distributed drone set U in the square ground area V = { , ,..., ,..., } and a flying drone To assist n mobile devices on the ground in a time slot set consisting of T time slots In the above example, represents the jth distributed UAV; [1, m]; m represents the total number of distributed drones; the jth distributed drone The position is recorded as ;in, represents the jth distributed UAV The coordinates projected onto the ground, Indicates the height of each distributed drone's hovering position; The position in the tth time slot is recorded as , Indicates flying drone The coordinates projected onto the ground in the tth time slot, Indicates flying drone The flight altitude will fly the drone The flight angle in the tth time slot is recorded as , the flight distance is recorded as ; Represents the flying drones in each time slot Maximum flight distance;

[0008] The position of n mobile devices in the tth time slot is recorded as ,in, represents the X-axis coordinate and Y-axis coordinate of the t-th time slot of the i-th mobile device, and the task set of the t-th time slot on the n mobile devices is recorded as ={ , , ..., , ..., },in, represents the task of the i-th mobile device in the t-th time slot; if the task is executed on the mobile device, it is recorded as local computing, and if the task is executed on the drone, it is recorded as edge computing; [1,n];

[0009] Let the i-th task in the t-th time slot The attribute is recorded as = ,in, represents the amount of data of the i-th mobile device in the t-th time slot; Indicates processing The number of CPU cycles required;

[0010] The UAV trajectory planning and task offloading decision method based on joint optimization is carried out in the following steps:

[0011] Step 1: Build a UAV trajectory planning and task offloading decision model based on the mobile device's mission and the UAV's location:

[0012] Step 1.1: Use equation (1) to calculate the task of the i-th mobile device in the t-th time slot: Latency calculated locally :

[0013] (1)

[0014] In formula (1), It is represented as the computing power of the i-th mobile device;

[0015] Step 1.2: Use equation (2) to calculate the task of the i-th mobile device in the t-th time slot: Energy consumption of local calculations :

[0016] (2)

[0017] In formula (2), It is the correlation coefficient with the CPU chip performance of the mobile device;

[0018] Step 1.3: Calculate the Mobile device tasks Unloading from the ground to Distributed drones The transmission delay ;

[0019] Step 1.4: Calculate the tasks of the i-th mobile device under different dispatch decisions In the jth distributed UAV Dispatch processing latency on ;

[0020] Step 1.5: Use equation (20) to calculate the task of the i-th mobile device in the t-th time slot Latency of computing at the edge ;

[0021] = + (20)

[0022] Step 1.6: Calculate the jth distributed drone in the tth time slot in edge computing Total energy consumption and flying drones Total energy consumption ;

[0023] Step 1.7: Use equation (26) to define the objective function W of the UAV trajectory planning and task offloading decision model: (26)

[0024] In formula (36), T represents the total number of time slots, Task offloading decisions for all mobile users, Task allocation decisions for mobile users, is the trajectory of the flying drone, Indicates the completion of the task of the i-th mobile device in the t-th time slot The delay is calculated from equation (27):

[0025] (27)

[0026] In formula (27), is the uninstall decision of the i-th mobile device, when =0 indicates the task of the i-th mobile device When executed locally, =j represents the task of the i-th mobile device Unload to the jth distributed UAV Execute on; is the indicator function;

[0027] Step 1.8: Use equations (28a) to (28e) to construct the constraints of the UAV trajectory planning and task offloading decision model:

[0028] (28a)

[0029] (28b)

[0030] (28c)

[0031] (28d)

[0032] In formula (28a) to formula (28e), represents the maximum energy of the i-th mobile device, represents the jth distributed UAV The maximum energy, Indicates flying drone Maximum energy;

[0033] Step 2: Define and initialize various parameters, including: the locations of n mobile devices, the task list, the locations of m distributed drones, and the starting position of the flying drone;

[0034] Step 3: Use the Wolf-PHC-based task offloading algorithm to generate the offloading strategy for all mobile devices in the tth time slot (t);

[0035] Step 4: According to the offloading strategy of all mobile devices in the tth time slot (t), and the number of drones flying in the tth time slot Location , and thus the dispatching strategy of mobile devices in the tth time slot is obtained by using the dispatching algorithm based on game theory ;

[0036] Step 5: Based on the state of the tth time slot , and use the DDPG algorithm to get the flying drone Action at time slot t ,in They represent the remaining power of the i-th mobile device in the t-th time slot and the j-th distributed UAV. Remaining battery and flying drone The remaining power;

[0037] Step 6: Calculate the reward for the tth time slot And the state of the t+1th time slot ,in, is the drone flying in the tth time slot The penalty constant for flying out of the target area. is the drone flying in the tth time slot The battery capacity loss penalty coefficient; represents the remaining power of the i-th mobile device in the t+1-th time slot; represents the jth distributed UAV in the t+1th time slot The remaining power; Indicates the number of drones flying in the t+1th time slot The remaining power; represents the position of the i-th mobile device in the t+1-th time slot, represents the task of the i-th mobile device in the t+1-th time slot The amount of data, Indicates processing the i-th mobile device task in the t+1-th time slot The number of CPU cycles;

[0038] Step 7: Judgement Whether it is achieved If yes, then the coordinates of all time slots constitute the flying drone Trajectory Otherwise, return to step 3 to continue.

[0039] The UAV trajectory planning and task offloading decision method based on joint optimization described in the present invention is also characterized in that the step 1.3 includes:

[0040] Step 1.3.1: Use equation (3) to calculate the Mobile devices and Distributed drones The probability of line-of-sight connection between :

[0041] (3)

[0042] In formula (3), C and D are two constants related to the propagation environment type. is the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot The Euclidean distance of is obtained by formula (4):

[0043] (4)

[0044] Step 1.3.2: Use equation (5) to calculate the mobile device and the jth distributed drone The probability of non-line-of-sight connection :

[0045] (5)

[0046] Step 1.3.3: Use equation (6) and equation (7) to calculate the mobile device and the jth distributed drone Line-of-sight link Loss and non-line-of-sight links Loss :

[0047] (6)

[0048] (7)

[0049] In formula (6) and formula (7), is the path loss exponent, and , represents the carrier frequency, is the speed of light, and Line-of-sight link and non-line-of-sight links The average additional loss of

[0050] Step 1.3.4: Use equation (8) to calculate the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot: The path loss between :

[0051] (8)

[0052] Step 1.3.5: Use equation (9) to calculate the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot: Channel gain between :

[0053] (9)

[0054] Step 1.3.6: Use equation (10) to calculate the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot: Uplink transmission speed between :

[0055] (10)

[0056] In formula (10), represents the average bandwidth allocated to the i-th mobile device, is the transmit power of the ith mobile device, is the Gaussian white noise power;

[0057] Step 1.3.7: Use equation (11) to calculate the Mobile device tasks Unloading from the ground to Distributed drones The transmission delay :

[0058] (11).

[0059] Further, the step 1.4 includes:

[0060] Step 1.4.1: Use equation (12) to calculate the jth distributed UAV in the tth time slot With flying drone Channel gain between :

[0061] (12)

[0062] In formula (12), is the jth distributed UAV in the tth time slot With flying drone The path loss between = + ,in, Line-of-sight link The additional attenuation coefficient is represents the jth distributed UAV in the tth time slot With flying drone The free space path loss between them is calculated and obtained by equation (13):

[0063] (13)

[0064] In formula (13), is the carrier frequency, is the speed of light, is the jth distributed UAV in the tth time slot and flying drones The Euclidean distance between them is obtained by formula (14):

[0065] (14)

[0066] Step 1.4.2: Use equation (15) to calculate the jth distributed UAV in the tth time slot and flying drones Uplink transmission speed between :

[0067] (15)

[0068] In formula (15), represents the average allocation to the jth distributed UAV bandwidth, is the jth distributed drone The transmission power;

[0069] Step 1.4.3: Define the The allocation decision for mobile devices is ,in, Indicates Mobile device tasks Assign to distributed drones for execution; Indicates that Assign to flying drone implement; Indicates that the Mobile device tasks Assign to flying drone ;in, represents the optimal allocation ratio allocated in the tth time slot;

[0070] Step 1.4.4: When When , the task of the i-th mobile device in the t-th time slot is calculated using formula (16): In the jth distributed UAV Dispatch processing latency on ;

[0071] (16)

[0072] In formula (16), is the jth distributed UAV in the tth time slot The computing resources allocated to the i-th mobile device are obtained by formula (17):

[0073] (17)

[0074] In formula (17), is the jth distributed drone The largest available computing resources, Indicates unloading to the jth distributed UAV The task set above; Indicates processing of the jth distributed UAV On Mobile device tasks The number of CPU cycles required;

[0075] when When , use formula (18) to calculate the task of the i-th mobile device in the t-th time slot In the jth distributed UAV Dispatch processing latency on ;

[0076] (18)

[0077] In formula (18), is the number of drones flying in the tth time slot Assigned to the jth distributed UAV computing resources;

[0078] when When , use formula (19) to calculate the task of the i-th mobile device in the t-th time slot In the jth distributed UAV Dispatch processing latency on ;

[0079] + (19).

[0080] Furthermore, the step 1.6 includes:

[0081] Step 1.6.1: Use equation (21) to calculate the jth distributed UAV in the tth time slot Total energy consumption :

[0082] (twenty one)

[0083] In formula (21), represents the jth distributed UAV in the tth time slot Process the i-th mobile device task The computing energy consumption, represents the jth distributed UAV in the tth time slot The task of the i-th mobile device Assign to flying drone The transmission energy consumption, represents the jth distributed UAV in the tth time slot The hovering energy consumption is calculated by equation (22a)-(22c):

[0084] (22a)

[0085] (22b)

[0086] (22c)

[0087] In formula (22a) to formula (22c), is the correlation coefficient with the CPU chip in the distributed drone, is the thrust proportional to the mass of the drone, is the jth distributed drone The power efficiency, is the number of rotors per distributed UAV, is the diameter of the rotor, is the air density, is the jth distributed drone The hovering time is calculated by formula (23):

[0088] (twenty three)

[0089] In formula (23), Indicates the number of Mobile device tasks The latency of computing at the edge;

[0090] Step 1.6.2: Use equation (24) to calculate the number of drones flying in the tth time slot Total energy consumption :

[0091] (twenty four)

[0092] In formula (24), Indicates the number of drones flying in the tth time slot A collection of tasks to be processed; represents the jth distributed UAV in the tth time slot The first Mobile device tasks Dispatched to flying drones The computational energy consumption of processing, Indicates the number of drones flying in the tth time slot The flight energy consumption is calculated by formula (25):

[0093] (25)

[0094] In formula (25), It's a flying drone Quality, It's a flying drone Flight time, is the number of drones flying in the tth time slot flight speed.

[0095] Furthermore, the step three includes:

[0096] Step 3.1: Initialize i=1, the current number of iterations is ,initialization =1;

[0097] Step 3.2: For the tth time slot The uninstall decision of the i-th mobile device in the iteration Associate a corresponding Q value, denoted as , and initialize =0;

[0098] Step 3.3: Use equation (29) to get the The uninstall decision of the i-th mobile device in the iteration The associated Q value :

[0099] (29)

[0100] In formula (29), Indicates the number of The i-th mobile device completes the task in the iteration The total delay of Indicates the number of The learning rate of the i-th mobile device in the iteration is obtained by formula (30):

[0101] (30)

[0102] In formula (30), represents the feasible offloading decision space of the i-th mobile device; is a lower learning rate, is a higher learning rate, and < , Indicates the number of The i-th mobile device executes the uninstallation strategy in the iteration The probability that =1, let =1 / (m+1), Indicates the number of The i-th mobile device executes the uninstallation strategy in the iteration The average probability of =1, let =1 / (m+1);

[0103] Step 3.4: Use equations (31a), (31b) and (32) to obtain the The i-th mobile device executes the uninstallation strategy in the iteration Probability and the tth time slot +1 iteration, the i-th mobile device executes the uninstallation strategy The average probability :

[0104] (31a)

[0105] (31b)

[0106] (32)

[0107] In formula (31a), formula (31b) and formula (32), It means that the i-th mobile device in the t-th time slot is Make uninstallation decision before iterations The number of times; Indicates the number of The uninstall decision of the i-th mobile device in the iteration The strategy adjustment parameters are represents the feasible offloading decision space of the i-th mobile device except An uninstall decision other than Indicates the number of Another uninstallation decision of the i-th mobile device in the iteration Associate a corresponding Q value, Indicates the number of In the next iteration, the i-th mobile device executes another uninstallation strategy probability;

[0108] Step 3.5: According to Get the first Uninstallation decision of the i-th mobile device in +1 iteration , and assign i+1 to i, and then return to step 3.2 and execute sequentially until i>n; thus, the number of all mobile users in the tth time slot is obtained. +1 Uninstallation decision in iteration ,judge and Are they equal? ​​If they are equal, it means As the offloading decision of all mobile devices in the tth time slot (t .

[0109] Furthermore, the step 4 includes:

[0110] Step 4.1: Calculate the optimal allocation ratio for the tth time slot using equation (33): :

[0111] (33)

[0112] Step 4.2: According to the optimal allocation ratio allocated in the tth time slot , get the dispatch decision of the i-th mobile device in the t-th time slot ;

[0113] Step 4.3: Initialize i=1 and randomly initialize the dispatching decisions of all mobile devices , and record the current dispatch decision as the optimal dispatch decision for the i-th mobile device in the t-th time slot ;

[0114] Step 4.4: Offloading decision at a fixed time slot t Under the premise of , the dispatch decision of all mobile devices in the tth time slot is calculated by formula (27): The corresponding cost function value is , and record it as the optimal cost ;

[0115] Step 4.5: Change the dispatch decision of the i-th mobile device in the t-th time slot when the dispatch decisions of other mobile devices in the t-th time slot do not change , get the changed dispatch strategy of the i-th mobile device in the t-th time slot , and calculate its corresponding cost function value ;

[0116] Step 4.6: If < , then the dispatch decision after the change of the i-th mobile device Assign to the optimal dispatch decision ; Otherwise, the optimal dispatch decision remain unchanged;

[0117] Step 4.7: After assigning i+1 to i, return to step 4.5 until i>n; thus obtaining the optimal dispatch decision for all mobile devices in the tth time slot And assign it to .

[0118] An electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the UAV trajectory planning and task offloading decision method, and the processor is configured to execute the program stored in the memory.

[0119] The present invention provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, which is characterized in that when the computer program is run by a processor, the steps of the UAV trajectory planning and task offloading decision-making method are executed.

[0120] Compared with the prior art, the present invention has the following beneficial effects:

[0121] 1. The present invention proposes a two-layer drone-assisted MEC network architecture. Specifically, users can select appropriate distributed drones for computing offload according to the current offload situation, and the distributed drones make task assignment decisions based on the positions of the flying drones in each time slot. Through resource sharing and collaboration between the upper and lower layers of drones, the wireless network performance in the MEC scenario is optimized, providing efficient computing capabilities under limited resources.

[0122] 2. The present invention studies the problems of joint offloading decision, task assignment, and drone trajectory planning in drone-assisted MEC collaboration. First, in the case of fixed trajectory action, wolf-PHC and game theory-based methods are used to obtain the optimal offloading strategy and task assignment strategy. Secondly, under the premise of the optimal offloading strategy and task assignment strategy, the DDPG algorithm is used to let the flying drone act as an agent, and the trajectory optimization is completed through coordination and cooperation between drones, thereby reducing the total delay in task completion and improving the user service quality.

[0123] 3. The present invention designs a dynamic real-time communication and data transmission scenario, taking into account the real-time tasks and certain mobility of mobile devices, which is closer to reality and improves the practicality of the solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0124] Figure 1 is a scene graph of the present invention;

[0125] Figure 2 is a flow chart of the method of the present invention;

[0126] Figure 3a This is a result diagram of the mobile device of the present invention in a randomly distributed scenario;

[0127] Figure 3b This is a result diagram of the skewed distribution scenario of the present invention;

[0128] Figure 3c This is a result diagram of the centralized distribution scenario of the present invention;

[0129] Figure 3d This is a result diagram of the decentralized distribution scenario of the present invention. DETAILED DESCRIPTION

[0130] In this embodiment, in order to solve the problem of limited computing resources and long task execution time in the scenario where both mobile devices and drones have limited power, a drone trajectory planning and task offloading decision method based on joint optimization is proposed. The method helps ground mobile devices to perform task calculations through the collaboration between two layers of drones. Once the computing resources of the distributed drones are insufficient, the computing tasks offloaded to the mobile devices will be assigned to the flying drones with more powerful computing power for execution. The scenario diagram is as follows: Figure 1 Specifically, the method is applied to a In the square ground area V, a vertex of the square ground area V is set as the origin o, two sides connected to the origin o are set as the X axis and the Y axis respectively, and a vertical line perpendicular to the XY plane and passing through the origin o is set as the Z axis, thereby establishing a three-dimensional coordinate system;

[0131] Deploy a distributed drone set U in a square ground area V = { , ,..., ,..., } and a flying drone To assist n mobile devices on the ground in a time slot set consisting of T time slots In the process of completing its own computing tasks, represents the jth distributed UAV; [1, m]; m represents the total number of distributed drones; the jth distributed drone The position of ;in, represents the jth distributed UAV The coordinates projected onto the ground, Indicates the height of each distributed drone's hovering position; The position in the tth time slot is recorded as , Indicates flying drone The coordinates projected onto the ground in the tth time slot, Indicates flying drone The flight altitude will fly the drone The flight angle in the tth time slot is recorded as , the flight distance is recorded as ; Represents the flying drones in each time slot Maximum flight distance;

[0132] The position of n mobile devices in the tth time slot is recorded as ,in, represents the X-axis coordinate and Y-axis coordinate of the t-th time slot of the i-th mobile device, and the task set of the t-th time slot on the n mobile devices is recorded as ={ , , ..., , ..., },in, represents the task of the i-th mobile device in the t-th time slot; if the task is executed on the mobile device, it is recorded as local computing, and if the task is executed on the drone, it is recorded as edge computing; [1,n];

[0133] Let the i-th task in the t-th time slot The attribute is recorded as = ,in, represents the amount of data of the i-th mobile device in the t-th time slot; Indicates processing The number of CPU cycles required.

[0134] In specific implementation, the UAV trajectory planning and task offloading decision method based on joint optimization is carried out in the following steps:

[0135] Step 1: Build a UAV trajectory planning and task offloading decision model based on the mobile device's mission and the UAV's location:

[0136] Step 1.1: Use equation (1) to calculate the task of the i-th mobile device in the t-th time slot: Latency calculated locally :

[0137] (1)

[0138] In formula (1), It is represented as the computing power of the i-th mobile device;

[0139] Step 1.2: Use equation (2) to calculate the task of the i-th mobile device in the t-th time slot: Energy consumption of local calculations :

[0140] (2)

[0141] In formula (2), It is a correlation coefficient with the CPU chip performance of the mobile device.

[0142] Step 1.3: Calculate the Mobile device tasks Unloading from the ground to Distributed drones The transmission delay ;

[0143] Step 1.3.1: Use equation (3) to calculate the Mobile devices and Distributed drones The probability of line-of-sight connection between :

[0144] (3)

[0145] In formula (3), C and D are two constants related to the propagation environment type. is the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot The Euclidean distance of is obtained by formula (4):

[0146] (4)

[0147] Step 1.3.2: Use equation (5) to calculate the mobile device and the jth distributed drone The probability of non-line-of-sight connection :

[0148] (5)

[0149] Step 1.3.3: Use equation (6) and equation (7) to calculate the mobile device and the jth distributed drone Line-of-sight link Loss and non-line-of-sight links Loss :

[0150] (6)

[0151] (7)

[0152] In formula (6) and formula (7), is the path loss exponent, and , represents the carrier frequency, is the speed of light, and Line-of-sight link and non-line-of-sight links The average additional loss.

[0153] Step 1.3.4: Use equation (8) to calculate the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot: The path loss between :

[0154] (8)

[0155] Step 1.3.5: Use equation (9) to calculate the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot: Channel gain between :

[0156] (9)

[0157] Step 1.3.6: Use equation (10) to calculate the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot: Uplink transmission speed between :

[0158] (10)

[0159] In formula (10), represents the average bandwidth allocated to the i-th mobile device, is the transmit power of the ith mobile device, is the Gaussian white noise power.

[0160] Step 1.3.7: Use equation (11) to calculate the Mobile device tasks Unloading from the ground to Distributed drones The transmission delay :

[0161] (11)

[0162] Step 1.4: Calculate the tasks of the i-th mobile device under different dispatch decisions In the jth distributed UAV Dispatch processing latency on .

[0163] Step 1.4.1: Use equation (12) to calculate the jth distributed UAV in the tth time slot With flying drone Channel gain between :

[0164] (12)

[0165] In formula (12), is the jth distributed UAV in the tth time slot With flying drone The path loss between = + ,in, Line-of-sight link The additional attenuation coefficient is represents the jth distributed UAV in the tth time slot With flying drone The free space path loss between them is calculated and obtained by equation (13):

[0166] (13)

[0167] In formula (13), is the carrier frequency, is the speed of light, is the jth distributed UAV in the tth time slot and flying drones The Euclidean distance between them is obtained by formula (14):

[0168] (14)

[0169] Step 1.4.2: Use equation (15) to calculate the jth distributed UAV in the tth time slot and flying drones Uplink transmission speed between :

[0170] (15)

[0171] In formula (15), represents the average allocation to the jth distributed UAV bandwidth, is the jth distributed drone of transmission power.

[0172] Step 1.4.3: Define the The allocation decision for mobile devices is ,in, Indicates Mobile device tasks Assign to distributed drones for execution; Indicates that Assign to flying drone implement; Indicates that the Mobile device tasks Assign to flying drone ;in, represents the optimal allocation ratio allocated in the tth time slot.

[0173] Step 1.4.4: When When , the task of the i-th mobile device in the t-th time slot is calculated using formula (16): In the jth distributed UAV Dispatch processing latency on ;

[0174] (16)

[0175] In formula (16), is the jth distributed UAV in the tth time slot The computing resources allocated to the i-th mobile device are obtained by formula (17):

[0176] (17)

[0177] In formula (17), is the jth distributed drone The largest available computing resources, Indicates unloading to the jth distributed drone The task set above; Indicates processing of the jth distributed UAV On Mobile device tasks The number of CPU cycles required.

[0178] when When , use formula (18) to calculate the task of the i-th mobile device in the t-th time slot In the jth distributed UAV Dispatch processing latency on ;

[0179] (18)

[0180] In formula (18), is the number of drones flying in the tth time slot Assigned to the jth distributed UAV computing resources;

[0181] when When , use formula (19) to calculate the task of the i-th mobile device in the t-th time slot In the jth distributed UAV Dispatch processing latency on ;

[0182] + (19)

[0183] Step 1.5: Use equation (20) to calculate the task of the i-th mobile device in the t-th time slot Latency of computing at the edge ;

[0184] = + (20)

[0185] Step 1.6: Calculate the jth distributed drone in the tth time slot in edge computing Total energy consumption and flying drones Total energy consumption :

[0186] Step 1.6.1: Use equation (21) to calculate the jth distributed UAV in the tth time slot Total energy consumption :

[0187] (twenty one)

[0188] In formula (21), represents the jth distributed UAV in the tth time slot Process the i-th mobile device task The computing energy consumption, represents the jth distributed UAV in the tth time slot The task of the i-th mobile device Assign to flying drone The transmission energy consumption, represents the jth distributed UAV in the tth time slot The hovering energy consumption is calculated by equation (22a)-(22c):

[0189] (22a)

[0190] (22b)

[0191] (22c)

[0192] In formula (22a) to formula (22c), is the correlation coefficient with the CPU chip in the distributed drone, is the thrust proportional to the mass of the drone, is the jth distributed drone The power efficiency, is the number of rotors per distributed UAV, is the diameter of the rotor, is the air density, is the jth distributed drone The hovering time is calculated by formula (23):

[0193] (twenty three)

[0194] In formula (23), Indicates the number of Mobile device tasks The latency calculated at the edge.

[0195] Step 1.6.2: Use equation (24) to calculate the number of drones flying in the tth time slot Total energy consumption :

[0196] (twenty four)

[0197] In formula (24), Indicates the number of drones flying in the tth time slot A collection of tasks to be processed; represents the jth distributed UAV in the tth time slot The first Mobile device tasks Dispatched to flying drones The computational energy consumption of processing, Indicates the number of drones flying in the tth time slot The flight energy consumption is calculated by formula (25):

[0198] (25)

[0199] In formula (25), It's a flying drone Quality, It's a flying drone Flight time, is the number of drones flying in the tth time slot flight speed.

[0200] Step 1.7: Use equation (26) to define the objective function W of the UAV trajectory planning and task offloading decision model: (26)

[0201] In formula (36), T represents the total number of time slots, Task offloading decisions for all mobile users, Task allocation decisions for mobile users, is the trajectory of the flying drone, Indicates the completion of the task of the i-th mobile device in the t-th time slot The delay is calculated from equation (27):

[0202] (27)

[0203] In formula (27), is the uninstall decision of the i-th mobile device, when =0 indicates the task of the i-th mobile device When executed locally, =j represents the task of the i-th mobile device Unload to the jth distributed UAV Execute on; is the indicator function.

[0204] Step 1.8: Use equations (28a) to (28e) to construct the constraints of the UAV trajectory planning and task offloading decision model:

[0205] (28a)

[0206] (28b)

[0207] (28c)

[0208] (28d)

[0209] In formula (28a) to formula (28e), represents the maximum energy of the i-th mobile device, represents the jth distributed UAV The maximum energy, Indicates flying drone Constraint (28a) ensures that the energy consumption of each mobile device is less than its maximum power; constraint (28b) ensures that the energy consumption of each distributed drone is less than its maximum power; constraint (28c) ensures that the energy consumption of the flying drone is less than its maximum power; constraint (28d) ensures that the flight distance of the flying drone in each time slot is less than its maximum allowed flight distance.

[0210] Step 2: Define and initialize various parameters, including: the locations of n mobile devices, the task list, the locations of m distributed drones, and the starting position of the flying drone;

[0211] Step 3: Use the Wolf-PHC-based task offloading algorithm to generate the offloading strategy for all mobile devices in the tth time slot (t);

[0212] Step 3.1: Initialize i=1, the current number of iterations is ,initialization =1;

[0213] Step 3.2: For the tth time slot The uninstall decision of the i-th mobile device in the iteration Associate a corresponding Q value, denoted as , and initialize =0.

[0214] Step 3.3: Use equation (29) to get the The uninstall decision of the i-th mobile device in the iteration The associated Q value :

[0215] (29)

[0216] In formula (29), Indicates the number of The i-th mobile device completes the task in the iteration The total delay of Indicates the number of The learning rate of the i-th mobile device in the iteration is obtained by formula (30):

[0217] (30)

[0218] In formula (30), represents the feasible offloading decision space of the i-th mobile device; is a lower learning rate, is a higher learning rate, and < , Indicates the number of The i-th mobile device executes the uninstallation strategy in the iteration The probability that =1, let =1 / (m+1), Indicates the number of The i-th mobile device executes the uninstallation strategy in the iteration The average probability of =1, let =1 / (m+1).

[0219] Step 3.4: Use equations (31a), (31b) and (32) to obtain the The i-th mobile device executes the uninstallation strategy in the iteration Probability and the tth time slot +1 iteration, the i-th mobile device executes the uninstallation strategy The average probability :

[0220] (31a)

[0221] (31b)

[0222] (32)

[0223] In formula (31a), formula (31b) and formula (32), It means that the i-th mobile device in the t-th time slot is Make uninstallation decision before iterations The number of times; Indicates the number of The uninstall decision of the i-th mobile device in the iteration The strategy adjustment parameters are represents the feasible offloading decision space of the i-th mobile device except An uninstall decision other than Indicates the number of Another uninstallation decision of the i-th mobile device in the iteration Associate a corresponding Q value, Indicates the number of In the next iteration, the i-th mobile device executes another uninstallation strategy probability.

[0224] Step 3.5: According to Get the first Uninstallation decision of the i-th mobile device in +1 iteration , and assign i+1 to i, and then return to step 3.2 and execute sequentially until i>n; thus, the number of all mobile users in the tth time slot is obtained. +1 Uninstallation decision in iteration ,judge and Are they equal? ​​If they are equal, it means As the offloading decision of all mobile devices in the tth time slot (t .

[0225] Step 4: According to the offloading strategy of all mobile devices in the tth time slot (t), and the number of drones flying in the tth time slot Location , and thus the dispatching strategy of mobile devices in the tth time slot is obtained by using the dispatching algorithm based on game theory ;

[0226] Step 4.1: In the tth time slot, due to the offloading decision (t) has been determined, the completion delay of the local computing task is known, and the offloading transmission delay of the edge computing task is also known. Therefore, it is only necessary to minimize the dispatch processing delay of the i-th mobile device. However, for the three dispatch cases, the dispatch ratio in the partial dispatch is an indeterminate number between (0,1), which complicates the calculation of scheduling processing delay in this case. The size depends on In addition, the amount of tasks that a single user needs to calculate and offload in time slot t is certain. When the amount of tasks remains unchanged, the amount of tasks offloaded to distributed drones and flying drones is inversely proportional, corresponding to and is also inversely proportional. According to the characteristics of increasing and decreasing functions in mathematics, Only when Therefore, we can first deduce the optimal allocation ratio for the tth time slot: , can be calculated using formula (33):

[0227] (33)

[0228] Step 4.2: According to the optimal allocation ratio allocated in the tth time slot , get the dispatch decision of the i-th mobile device in the t-th time slot .

[0229] Step 4.3: Initialize i=1 and randomly initialize the dispatching decisions of all mobile devices , and record the current dispatch decision as the optimal dispatch decision for the i-th mobile device in the t-th time slot ;

[0230] Step 4.4: Offloading decision at a fixed time slot t Under the premise of , the dispatch decision of all mobile devices in the tth time slot is calculated by formula (27): The corresponding cost function value is , and record it as the optimal cost ;

[0231] Step 4.5: Change the dispatch decision of the i-th mobile device in the t-th time slot when the dispatch decisions of other mobile devices in the t-th time slot do not change , get the changed dispatch strategy of the i-th mobile device in the t-th time slot , and calculate its corresponding cost function value .

[0232] Step 4.6: If < , then the dispatch decision after the change of the i-th mobile device Assign to the optimal dispatch decision ; Otherwise, the optimal dispatch decision remain unchanged;

[0233] Step 4.7: After assigning i+1 to i, return to step 4.5 until i>n; thus obtaining the optimal dispatch decision for all mobile devices in the tth time slot And assign it to .

[0234] Step 5: Based on the state of the tth time slot , and use the DDPG algorithm to get the flying drone Action at time slot t and the position of the flying drone in the t+1th time slot ,in They represent the remaining power of the i-th mobile device in the t-th time slot and the j-th distributed UAV. Remaining battery and flying drone remaining power.

[0235] like Figure 2 As shown, step five includes the following steps:

[0236] Step 5.1: Initialize the current network parameters of Actor and Critic , and the parameters of the target network , ;

[0237] Step 5.2: According to the state of the tth time slot and a deterministic policy function Select Action ,in, is to add random noise to each output action;

[0238] Step 5.3: Calculate the reward for the tth time slot And the state of the t+1th time slot ,in, is the drone flying in the tth time slot The penalty constant for flying out of the target area. is the drone flying in the tth time slot The battery capacity loss penalty coefficient; represents the remaining power of the i-th mobile device in the t+1-th time slot; represents the jth distributed UAV in the t+1th time slot The remaining power; Indicates the number of drones flying in the t+1th time slot The remaining power; represents the position of the i-th mobile device in the t+1-th time slot, represents the task of the i-th mobile device in the t+1-th time slot The amount of data, Indicates processing the i-th mobile device task in the t+1-th time slot The number of CPU cycles.

[0239] Step 5.4: In order to solve the problems of sample relevance and stability, the DDPG algorithm also introduces an experience replay mechanism to replay the experience gained from the interaction between the flying drone and the environment. Stored in the experience replay buffer Then randomly select The current network of Critic can be trained based on the current record and deterministic strategy function. calculate The value can be calculated using formula (34):

[0240] (34)

[0241] In formula (34), is the value function in the tth time slot, represents the Bellman equation, Indicates the current state of the Critic's network Execute actions when The reward function is represents the depreciation rate, Indicates that the current network of the Critic in the t+1th time slot is in state Execute the deterministic policy function The value function of .

[0242] Step 5.5: Next, the Actor target network will get the next state through approximate estimation The corresponding action value The target network of the Critic can approximate the state-action relationship at the next moment. The value can be calculated using formula (35):

[0243] (35)

[0244] In formula (35), is the current state in the tth time slot The target value of the value function, is the reward in the tth time slot, Indicates that the target network of the Critic in the t+1th time slot has parameters and status Deterministic Strategy The obtained value function.

[0245] Step 5.6: In order to stabilize the learning process and speed up the convergence, combine the Actor target network and the Critic target network. After that, the target network of the Critic will transmit it to the current network of the Critic to minimize its loss function , can be calculated using formula (36):

[0246] (36)

[0247] In formula (36), is the number of samples drawn, Indicates that the current network of the Critic in the tth time slot is and the status is Execute actions when The value function of .

[0248] Step 5.7: After updating the parameters of the Critic current network After that, use the gradient strategy to update the strategy parameters of the Actor's current network , can be calculated using formula (37):

[0249] (37)

[0250] In formula (37), To find the sign of the partial derivative, is the loss function, Represents the loss function right Find partial derivatives; Representation value function right Find the partial derivative, and , so we can apply the chain rule to continue to The partial derivative is recorded as .

[0251] Step 5.8: Use the soft update method to update the parameters of the two target networks, which can be calculated using equations (38) and (39):

[0252] (38)

[0253] (39)

[0254] In formulas (38)-(39), Soft update parameters.

[0255] Step 6: Judgement Whether it is achieved If yes, then the coordinates of all time slots constitute the flying drone Trajectory Otherwise, return to step 3 to continue.

[0256] The effectiveness of this method is tested through different distribution scenarios. Figure 3a , Figure 3b , Figure 3c and Figure 3d The flight trajectories of the flying UAV in four different distribution scenarios of mobile devices: random distribution, skewed distribution, concentrated distribution, and dispersed distribution. The MD trajectory represents the movement trajectory of the mobile device, which is displayed in black lines. The D-UAV location represents the hovering position of the distributed UAV, which is marked with a red five-pointed star. The F-UAV start location represents the initial position of the flying UAV, which is marked with a blue five-pointed star. The F-UAV trajectory represents the flight trajectory of the flying UAV, which is displayed with a green line. Figure 3a-3d In the figure, we can see that the flight trajectory of the flying drone is mainly concentrated between the distributed drones. This is because the mobile device will offload the task to the distributed drones in the first layer, so that in the second layer, only the communication between the flying drone and the distributed drones needs to be maintained, which greatly reduces the flight energy consumption of the flying drone.

[0257] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0258] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.

Claims

1. A multi-UAV layout and task offloading decision method based on multi-objective joint optimization, characterized in that: is applied to a In a square ground area V, a vertex of the square ground area V is set as the origin o, two sides connected to the origin o are set as the X axis and the Y axis respectively, and a vertical line perpendicular to the XY plane and passing through the origin o is set as the Z axis, thereby establishing a three-dimensional coordinate system; Deploy a distributed drone set U in the square ground area V = { , ,..., ,..., } and a flying drone To assist n mobile devices on the ground in a time slot set consisting of T time slots In the process of completing its own computing tasks, represents the jth distributed UAV; [1, m]; m represents the total number of distributed drones; the jth distributed drone The position of ;in, represents the jth distributed UAV The coordinates projected onto the ground, Indicates the height of each distributed drone's hovering position; The position in the tth time slot is recorded as , Indicates flying drone The coordinates projected onto the ground in the tth time slot, Indicates flying drone The flight altitude will fly the drone The flight angle in the tth time slot is recorded as , the flight distance is recorded as ; Represents the flying drones in each time slot Maximum flight distance; The position of n mobile devices in the tth time slot is recorded as ,in, represents the X-axis coordinate and Y-axis coordinate of the t-th time slot of the i-th mobile device, and the task set of the t-th time slot on the n mobile devices is recorded as ={ , , ..., , ..., },in, represents the task of the i-th mobile device in the t-th time slot; if the task is executed on the mobile device, it is recorded as local computing, and if the task is executed on the drone, it is recorded as edge computing; [1,n]; Let the i-th task in the t-th time slot The attribute is recorded as = ,in, represents the amount of data of the i-th mobile device in the t-th time slot; Indicates processing The number of CPU cycles required; The UAV trajectory planning and task offloading decision method based on joint optimization is carried out in the following steps: Step 1: Build a UAV trajectory planning and task offloading decision model based on the mobile device's mission and the UAV's location: Step 1.1: Use equation (1) to calculate the task of the i-th mobile device in the t-th time slot: Latency calculated locally : (1) In formula (1), It is represented as the computing power of the i-th mobile device; Step 1.2: Use equation (2) to calculate the task of the i-th mobile device in the t-th time slot: Energy consumption calculated locally : (2) In formula (2), It is the correlation coefficient with the CPU chip performance of the mobile device; Step 1.3: Calculate the Mobile device tasks Unloading from the ground to Distributed drones The transmission delay ; Step 1.4: Calculate the tasks of the i-th mobile device under different dispatch decisions In the jth distributed UAV Dispatch processing latency on ; Step 1.5: Use equation (20) to calculate the task of the i-th mobile device in the t-th time slot Latency of computing at the edge ; = + (20) Step 1.6: Calculate the jth distributed drone in the tth time slot in edge computing Total energy consumption and flying drones Total energy consumption ; Step 1.7: Use equation (26) to define the objective function W of the UAV trajectory planning and task offloading decision model: (26) In formula (36), T represents the total number of time slots, Task offloading decisions for all mobile users, Task allocation decisions for mobile users, is the trajectory of the flying drone, Indicates the completion of the task of the i-th mobile device in the t-th time slot The delay is calculated from equation (27): (27) In formula (27), is the uninstall decision of the i-th mobile device, when =0 indicates the task of the i-th mobile device When executed locally, =j represents the task of the i-th mobile device Unload to the jth distributed UAV Execute on; is the indicator function; Step 1.8: Use equations (28a) to (28e) to construct the constraints of the UAV trajectory planning and task offloading decision model: (28a) (28b) (28c) (28d) In formula (28a) to formula (28e), represents the maximum energy of the i-th mobile device, represents the jth distributed UAV The maximum energy, Indicates flying drone Maximum energy; Step 2: Define and initialize various parameters, including: the locations of n mobile devices, the task list, the locations of m distributed drones, and the starting position of the flying drone; Step 3: Use the Wolf-PHC-based task offloading algorithm to generate the offloading strategy for all mobile devices in the tth time slot (t); Step 4: According to the offloading strategy of all mobile devices in the tth time slot (t), and the number of drones flying in the tth time slot Location , and thus the dispatching strategy of mobile devices in the tth time slot is obtained by using the dispatching algorithm based on game theory ; Step 5: Based on the state of the tth time slot , and use the DDPG algorithm to get the flying drone Action at time slot t ,in They represent the remaining power of the i-th mobile device in the t-th time slot and the j-th distributed UAV. Remaining battery and flying drone The remaining power; Step 6: Calculate the reward for the tth time slot And the state of the t+1th time slot ,in, is the drone flying in the tth time slot The penalty constant for flying out of the target area. is the drone flying in the tth time slot The battery capacity loss penalty coefficient; represents the remaining power of the i-th mobile device in the t+1-th time slot; represents the jth distributed UAV in the t+1th time slot The remaining power; Indicates the number of drones flying in the t+1th time slot The remaining power; represents the position of the i-th mobile device in the t+1-th time slot, represents the task of the i-th mobile device in the t+1-th time slot The amount of data, Indicates processing the i-th mobile device task in the t+1-th time slot The number of CPU cycles; Step 7: Judgement Whether it is achieved If yes, then the coordinates of all time slots constitute the flying drone The trajectory Otherwise, return to step 3 to continue.

2. The method for UAV trajectory planning and task offloading decision-making based on joint optimization according to claim 1 is characterized in that: The step 1.3 comprises: Step 1.3.1: Use equation (3) to calculate the Mobile devices and Distributed drones The probability of line-of-sight connection between : (3) In formula (3), C and D are two constants related to the propagation environment type. is the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot The Euclidean distance of is obtained by formula (4): (4) Step 1.3.2: Use equation (5) to calculate the mobile device and the jth distributed drone The probability of non-line-of-sight connection : (5) Step 1.3.3: Use equation (6) and equation (7) to calculate the mobile device and the jth distributed drone Line-of-sight link Loss and non-line-of-sight links Loss : (6) (7) In formula (6) and formula (7), is the path loss exponent, and , represents the carrier frequency, is the speed of light, and Line-of-sight link and non-line-of-sight links The average additional loss of Step 1.3.4: Use equation (8) to calculate the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot: The path loss between : (8) Step 1.3.5: Use equation (9) to calculate the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot: Channel gain between : (9) Step 1.3.6: Use equation (10) to calculate the distance between the i-th mobile device and the j-th distributed UAV in the t-th time slot: Uplink transmission speed between : (10) In formula (10), represents the average bandwidth allocated to the i-th mobile device, is the transmit power of the ith mobile device, is the Gaussian white noise power; Step 1.3.7: Use equation (11) to calculate the Mobile device tasks Unloading from the ground to Distributed drones The transmission delay : (11)。 3. The UAV trajectory planning and task offloading decision method based on joint optimization according to claim 2 is characterized in that: The step 1.4 comprises: Step 1.4.1: Use equation (12) to calculate the jth distributed UAV in the tth time slot With flying drone Channel gain between : (12) In formula (12), is the jth distributed UAV in the tth time slot With flying drone The path loss between = + ,in, Line-of-sight link The additional attenuation coefficient is represents the jth distributed UAV in the tth time slot With flying drone The free space path loss between them is calculated and obtained by equation (13): (13) In formula (13), is the carrier frequency, is the speed of light, is the jth distributed UAV in the tth time slot and flying drones The Euclidean distance between them is obtained by formula (14): (14) Step 1.4.2: Use equation (15) to calculate the jth distributed UAV in the tth time slot and flying drones Uplink transmission speed between : (15) In formula (15), represents the average allocation to the jth distributed UAV bandwidth, is the jth distributed drone The transmission power; Step 1.4.3: Define the The allocation decision for mobile devices is ,in, Indicates Mobile device tasks Assign to distributed drones for execution; Indicates that Assign to flying drone implement; Indicates that the Mobile device tasks Assign to flying drone ;in, represents the optimal allocation ratio allocated in the tth time slot; Step 1.4.4: When When , the task of the i-th mobile device in the t-th time slot is calculated using formula (16): In the jth distributed UAV Dispatch processing latency on ; (16) In formula (16), is the jth distributed UAV in the tth time slot The computing resources allocated to the i-th mobile device are obtained by formula (17): (17) In formula (17), is the jth distributed drone The largest available computing resources, Indicates unloading to the jth distributed drone The task set above; Indicates processing of the jth distributed UAV On Mobile device tasks The number of CPU cycles required; when When , the task of the i-th mobile device in the t-th time slot is calculated using formula (18): In the jth distributed UAV Dispatch processing latency on ; (18) In formula (18), is the number of drones flying in the tth time slot Assigned to the jth distributed UAV computing resources; when When , use formula (19) to calculate the task of the i-th mobile device in the t-th time slot In the jth distributed UAV Dispatch processing latency on ; + (19)。 4. The method for UAV trajectory planning and task offloading decision-making based on joint optimization according to claim 3 is characterized in that: The step 1.6 comprises: Step 1.6.1: Use equation (21) to calculate the jth distributed UAV in the tth time slot Total energy consumption : (21) In formula (21), represents the jth distributed UAV in the tth time slot Process the i-th mobile device task The computing energy consumption, represents the jth distributed UAV in the tth time slot The task of the i-th mobile device Assign to flying drone The transmission energy consumption, represents the jth distributed UAV in the tth time slot The hovering energy consumption is calculated by equation (22a)-(22c): (22a) (22b) (22c) In formula (22a) to formula (22c), is the correlation coefficient with the CPU chip in the distributed drone, is the thrust proportional to the mass of the drone, is the jth distributed drone The power efficiency, is the number of rotors per distributed UAV, is the diameter of the rotor, is the air density, is the jth distributed drone The hovering time is calculated by formula (23): (23) In formula (23), Indicates the number of Mobile device tasks The latency of computing at the edge; Step 1.6.2: Use equation (24) to calculate the number of drones flying in the tth time slot Total energy consumption : (24) In formula (24), Indicates the number of drones flying in the tth time slot A collection of tasks to be processed; represents the jth distributed UAV in the tth time slot The first Mobile device tasks Dispatched to flying drones The computational energy consumption of processing, Indicates the number of drones flying in the tth time slot The flight energy consumption is calculated by formula (25): (25) In formula (25), It's a flying drone Quality, It's a flying drone Flight time, is the number of drones flying in the tth time slot flight speed.

5. The method for UAV trajectory planning and task offloading decision-making based on joint optimization according to claim 4 is characterized in that: The step three comprises: Step 3.1: Initialize i=1, the current number of iterations is ,initialization =1; Step 3.2: For the tth time slot The uninstall decision of the i-th mobile device in the iteration Associate a corresponding Q value, denoted as , and initialize =0; Step 3.3: Use equation (29) to get the The uninstall decision of the i-th mobile device in the iteration The associated Q value : (29) In formula (29), Indicates the number of The i-th mobile device completes the task in the iteration The total delay of Indicates the number of The learning rate of the i-th mobile device in the iteration is obtained by formula (30): (30) In formula (30), represents the feasible offloading decision space of the i-th mobile device; is a lower learning rate, is a higher learning rate, and < , Indicates the number of The i-th mobile device executes the uninstallation strategy in the iteration The probability that =1, let =1 / (m+1), Indicates the number of The i-th mobile device executes the uninstallation strategy in the iteration The average probability of =1, let =1 / (m+1); Step 3.4: Use equations (31a), (31b) and (32) to obtain the The i-th mobile device executes the uninstallation strategy in the iteration Probability and the tth time slot +1 iteration, the i-th mobile device executes the uninstallation strategy The average probability : (31a) (31b) (32) In formula (31a), formula (31b) and formula (32), It means that the i-th mobile device in the t-th time slot is Make uninstallation decision before iterations The number of times; Indicates the number of The uninstall decision of the i-th mobile device in the iteration The strategy adjustment parameters are represents the feasible offloading decision space of the i-th mobile device except An uninstall decision other than Indicates the number of Another uninstallation decision of the i-th mobile device in the iteration Associate a corresponding Q value, Indicates the number of In the next iteration, the i-th mobile device executes another uninstallation strategy probability; Step 3.5: According to Get the first Uninstallation decision of the i-th mobile device in +1 iteration , and assign i+1 to i, and then return to step 3.2 and execute sequentially until i>n; thus, the number of all mobile users in the tth time slot is obtained. +1 Uninstallation decision in iteration ,judge and Are they equal? ​​If they are equal, it means As the offloading decision of all mobile devices in the tth time slot (t .

6. The method for UAV trajectory planning and task offloading decision-making based on joint optimization according to claim 5 is characterized in that: The fourth step comprises: Step 4.1: Calculate the optimal allocation ratio for the tth time slot using equation (33): : (33) Step 4.2: According to the optimal allocation ratio allocated in the tth time slot , get the dispatch decision of the i-th mobile device in the t-th time slot ; Step 4.3: Initialize i=1 and randomly initialize the dispatching decisions of all mobile devices , and record the current dispatch decision as the optimal dispatch decision for the i-th mobile device in the t-th time slot ; Step 4.4: Offloading decision at a fixed time slot t Under the premise of , the dispatch decision of all mobile devices in the tth time slot is calculated by formula (27): The corresponding cost function value is , and record it as the optimal cost ; Step 4.5: Change the dispatch decision of the i-th mobile device in the t-th time slot when the dispatch decisions of other mobile devices in the t-th time slot do not change , get the changed dispatch strategy of the i-th mobile device in the t-th time slot , and calculate its corresponding cost function value ; Step 4.6: If < , then the dispatch decision after the change of the i-th mobile device Assign to the optimal dispatch decision ; Otherwise, the optimal dispatch decision remain unchanged; Step 4.7: After assigning i+1 to i, return to step 4.5 until i>n; thus obtaining the optimal dispatch decision for all mobile devices in the tth time slot And assign it to .

7. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the drone trajectory planning and task offloading decision method described in any one of claims 1-6, and the processor is configured to execute the program stored in the memory.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the UAV trajectory planning and task offloading decision method described in any one of claims 1-6 are executed.

Images