Internet-of-things side-end collaborative unmanned aerial vehicle position and resource joint optimization method

Abstract

The invention discloses an Internet of Things side-end collaborative unmanned aerial vehicle position and resource joint optimization method, in the aspect of region division, the region division range is adjusted through a graph theory-based normalized cut algorithm by using the computing power weight of a server and the task demand weight of a user, the server with the larger weight can influence a larger region, and the region division efficiency is improved. Users with high task requirements are more likely to select servers with high computing power; in the aspects of unmanned aerial vehicle deployment and resource allocation, a successive convex approximation algorithm is utilized, a complex non-convex problem is gradually converted into a series of convex optimization sub-problems which are easier to solve, the sub-problems are iteratively solved, a global or local optimal solution is gradually approximated, and higher-precision unmanned aerial vehicle position and resource allocation are obtained. According to the invention, the task is segmented and unloaded through an edge-end cooperation mechanism, and the mobile user, the unmanned aerial vehicle and the edge computing node share task processing together, so that the task can be efficiently completed and resource utilization can be optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles, and specifically relates to a method for jointly optimizing the position and resources of unmanned aerial vehicles in collaboration with the edge of the Internet of Things. Background Art

[0002] The Internet of Things (IoT) has developed rapidly due to its interconnectivity, intelligence, real-time, and scalability. With the rapid popularization of the Internet of Things, the number of IoT devices has increased dramatically. It is estimated that by 2030, there will be 500 billion IoT devices in use, and such a large number of IoT devices are supporting many attractive new applications, such as smart cities, smart homes, and telemedicine. However, these applications will generate many computationally intensive and latency-sensitive tasks, and these tasks are extremely dependent on the ability to quickly process data and extract useful information, which makes the traditional cloud-based data processing model no longer applicable. In order to perform these computationally intensive and latency-sensitive tasks, Mobile Edge Computing (MEC) has gradually come into people's view.

[0003] MEC is a technology that deploys computing and storage resources at the edge of mobile networks. MEC is able to deploy computing power, storage and network resources at the edge of the network, that is, near the point where users access the network, such as base stations, access points or user devices. This layout allows data to be processed and analyzed at the place of generation, rather than transmitted to a remote data center, thereby achieving rapid response and decision-making. MEC can bring many benefits to users, such as lower service latency, reduced network congestion and improved service quality. Although MEC is envisioned to solve computationally intensive and delay-sensitive tasks, it still faces many challenges, such as energy saving, guaranteed latency and computing cooperation, and MEC has limited resources compared to remote clouds. In addition, the tight coupling of communication and computing in MEC makes resource management a key issue for MEC. In particular, existing MEC technologies are not suitable for situations where the number of mobile users is explosively growing or network facilities are sparsely distributed. In view of this deficiency, using drone-supported wireless networks to assist MEC in improving connectivity between ground IoT devices has become an effective solution.

[0004] Drones were initially used in the military field, but are now widely used in the civilian field, such as traffic monitoring, search and rescue, environmental monitoring, battlefield communications, etc. In drone-assisted MEC networks, drones, as aerial base stations, can help ground wireless devices expand their communication range, thereby improving network coverage and communication quality. The high flexibility and rapid deployment capabilities of drones enable them to respond quickly to emergencies and provide efficient communication support in areas with complex geographical environments or limited infrastructure. Resource-constrained mobile devices can offload their computationally intensive and delay-sensitive tasks to drones with computing capabilities, which can not only effectively reduce the computing load of mobile devices, but also save their battery energy. Drones can provide efficient MEC services to mobile devices by connecting to edge devices, thereby reducing the delay time of task processing. In addition, drones can usually obtain better channel gain due to their high controllability and air superiority, which means lower transmission delay and higher transmission rate. In this way, drones can not only improve users' computing offload experience, but also significantly improve users' communication quality and service stability. This MEC approach using drones, combined with its flexibility, rapid deployment, and good channel gain, provides an efficient and reliable computing offload solution for mobile devices, greatly improving user experience and network performance.

[0005] Although the research prospects of drone-assisted edge collaborative IoT are broad, there are also many problems that need to be solved. Drones are highly flexible and can be flexibly deployed and adjusted according to communication scenarios and user needs. Therefore, how to reasonably deploy the location of drones is an important task and challenge in drone-assisted MEC systems. In addition, due to the small size of drones themselves, their battery capacity is limited and cannot support long-term flight and communication tasks. Only by allocating resources reasonably can we ensure service quality while reducing system energy consumption. In addition, with the technological development of mobile user devices, the computing power of mobile users has been improved. In drone-assisted MEC networks, how to reasonably use the computing resources of mobile users, drones, and edge computing nodes is also an important challenge. In summary, how to jointly optimize the location of drones and system resource allocation in drone-assisted MEC systems to achieve collaborative work among drones, edge devices, and mobile users is very challenging and of long-term significance. Summary of the invention

[0006] In order to overcome the shortcomings of the prior art, the present invention provides a method for joint optimization of drone positions and resources with edge-end collaboration in the Internet of Things. In terms of regional division, the computing power weight of the server and the task demand weight of the user are used to adjust the scope of regional division through a normalized cut algorithm based on graph theory. Servers with larger weights can affect larger areas, and users with higher task demands are more inclined to choose servers with strong computing power. In terms of drone deployment and resource allocation, a successive convex approximation algorithm is used to gradually transform complex non-convex problems into a series of convex optimization sub-problems that are easier to solve, and these sub-problems are iteratively solved to gradually approach the global or local optimal solution, thereby obtaining a more accurate drone position and resource allocation. The present invention divides and unloads tasks through an edge-end collaborative mechanism, and mobile users, drones, and edge computing nodes share task processing, which can ensure efficient task completion and optimize resource utilization.

[0007] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0008] Step 1: System initialization;

[0009] Step 1-1: Deployment of drones and edge devices;

[0010] Several ground IoT devices are deployed in the target area. UAVs are deployed in the air at the beginning of the mission as relay nodes between IoT devices and edge cloud. UAVs have certain computing resources and are equipped with modules for communication and data processing. Edge cloud servers are deployed at fixed locations on the ground to provide high-performance computing resources.

[0011] Step 1-2: Task generation and allocation;

[0012] Each IoT device generates computing tasks based on its sensing tasks. These tasks include data analysis that requires high computing power. The tasks are modeled as a triple W i =(L i ,C i ,λ i ), where L i Indicates the size of the input data for the processing task, C i Indicates the number of CPU cycles required to process 1 bit of data, λ i represents the task arrival rate; each task is decomposed into multiple subtasks, some of which are executed by drones, and other tasks are processed by the edge cloud after being relayed by drones;

[0013] Step 2: Regional division;

[0014] Step 2-1: Obtain the locations of all users in the system, the locations of edge servers, the computing capabilities of edge servers, i.e., weights, and the task requirements and weights of users, and construct a similarity matrix, in which the elements represent the similarity or dependency between nodes;

[0015] Step 2-2: Calculate the Laplacian matrix of the graph;

[0016] First, the degree matrix of the graph is calculated. This matrix is ​​a diagonal matrix. The elements on the diagonal of the matrix are the degree of each node, that is, the strength of the connection with other nodes. Then the Laplacian matrix of the graph is calculated. It is the difference between the degree matrix and the similarity matrix. The Laplacian matrix reflects the connection relationship between the nodes in the graph.

[0017] Step 2-3: Perform eigenvalue decomposition on the Laplacian matrix of the graph to obtain eigenvectors and eigenvalues; eigenvalue decomposition can map the nodes in the graph to a low-dimensional space to reflect the similarity between the nodes; select the smallest eigenvectors in the Laplacian matrix, which represent the embedded representation of the nodes in the low-dimensional space;

[0018] Step 2-4: Use the K-means clustering algorithm to cluster the feature vectors and divide them into multiple regions. Each region corresponds to a cluster; based on the clustering results, the nodes in the system, i.e. users, devices, and ECs, are divided into different regions;

[0019] Step 3: Task offloading and resource allocation;

[0020] Step 3-1: A ground-to-air uplink communication link is established between the drone and the IoT device to transmit the mission data of the IoT device; an air-to-ground downlink is established between the drone and the edge cloud server to forward the mission data to the edge cloud; the communication link adopts the free space path loss model. Considering the influence of the height and position of the drone, the uplink data transmission rate between each IoT device and the drone is calculated according to the Shannon capacity formula; therefore, the channel gain from mobile user i to the drone is described by the free space path loss model:

[0021]

[0022] Among them, g 0 It represents the channel power gain when the reference distance is 1m, d UAV,k represents the distance between mobile user k and the drone, ||·|| represents the Euclidean norm of the vector, represents the location of user k, Loc UAV Indicates the position of the drone;

[0023] The channel gain between the drone and the edge computing node j is expressed as:

[0024]

[0025] Among them, d UAV,j represents the distance between edge computing node j and the drone, represents the location of edge device j;

[0026] During the task offloading process, mobile users share bandwidth resources through the frequency division multiple access protocol. According to the Shannon capacity formula, the data transmission rate between mobile user k and the UAV is expressed as:

[0027]

[0028] in, represents the bandwidth allocated to mobile user k, represents the transmission power of mobile user k, σ 2 represents the noise power at the drone; it is assumed that the noise power of the drone is the same as that of the edge computing node;

[0029] The data transmission rate between the drone and the edge computing node j is expressed as:

[0030]

[0031] in, represents the bandwidth pre-allocated to edge computing node j, P t UAV represents the transmission power of the drone;

[0032] The total delay of the task offloading process is divided into the following six parts: (1) the computational delay of tasks that are completely processed locally; (2) the computational delay of tasks that need to be split and processed locally; (3) the transmission delay from the mobile user to the drone; (4) the computational delay of the drone; (5) the transmission delay from the drone to the edge computing node; (6) the computational delay of the edge computing node;

[0033] Computational latency of tasks processed entirely locally: All tasks are first calculated locally. If a task exceeds the maximum tolerable latency when executed locally, it will be considered to be offloaded to the drone for further processing. Therefore, the computational latency of tasks processed entirely locally is expressed as:

[0034]

[0035] Among them, L i,0 represents the input data size of the task that is completely processed locally, C i,0 Indicates the number of CPU cycles required to process 1 bit of task data in this part of the task. represents the computing resources of the mobile user that handles this part of the task;

[0036] The computational latency of tasks that need to be split and processed locally: Tasks whose local execution time exceeds the maximum tolerable latency will be split and offloaded. Some of these tasks will be executed locally. Therefore, the computational latency of this part is expressed as:

[0037]

[0038] in, Indicates the task split ratio that needs to be processed locally, L i,k represents the input data size of the task to be split, C i,k Indicates the number of CPU cycles required to process 1 bit of task data in this part of the task. represents the computing resources of the mobile user that handles this part of the task;

[0039] Transmission delay from mobile users to drones: The tasks that need to be split are offloaded to drones through transmission links. Assume that the transmission delay from mobile user k to drones is equal to the transmission delay of transmitting all tasks to drones. No task splitting is performed in this process:

[0040]

[0041] in, Indicates the data transmission rate from the mobile user to the UAV that needs to perform task segmentation;

[0042] The computational latency of the drone: The drone will decide whether the received task part should be processed locally on the drone or further offloaded to the ground edge computing node for processing; therefore, the computational latency of the drone is expressed as:

[0043]

[0044] in, Indicates the task split ratio that needs to be processed by the drone. represents the computing resources allocated by the UAV to mobile user k;

[0045] Transmission delay from the drone to the edge computing node: The drone further offloads the task to the edge computing node on the ground for processing to reduce computing delay. Therefore, the transmission delay from the drone to the edge computing node is expressed as:

[0046]

[0047] Among them, α kj Indicates the task split ratio that needs to be processed on the edge computing node.

[0048] Computational delay of edge computing node: After receiving the offloaded task data from the UAV, the edge computing node starts the computation process; therefore, the computational delay of the edge computing node processing the offloaded task from mobile user k to edge computing node j via the UAV is expressed as:

[0049]

[0050] in, represents the computing resources allocated to mobile user k by edge computing node j;

[0051] Therefore, the total delay of the system is expressed as:

[0052]

[0053] Computational energy consumption of drones: The power consumption of the drone CPU is modeled as where κ represents the effective switched capacitance that depends on the CPU architecture; therefore, the corresponding energy consumption of the drone when processing the task offloaded from mobile user k is given by the product of the power level and the computation time:

[0054]

[0055] Transmission energy consumption of UAV: ​​The transmission energy consumption of UAV when receiving task input data from mobile user k through uplink communication is expressed as:

[0056]

[0057] in, is the received power of the drone;

[0058] The transmission energy consumption when the UAV offloads part of the task of mobile user k to edge computing node j through uplink communication is expressed as:

[0059]

[0060] Therefore, the total energy consumption of the drone is expressed as:

[0061]

[0062] Step 3-2: Task splitting and offloading decision;

[0063] For each task generated by an IoT device, the system determines the task allocation ratio based on the task's computing requirements, the drone's computing power, and the edge cloud's resource status; the task allocation parameters control the task allocation ratio between the drone and the edge cloud; the system jointly optimizes the drone's location, uplink and downlink bandwidth allocation, task allocation ratio, and computing resource allocation to minimize the total service delay and the drone's energy consumption;

[0064] The objective function is given by:

[0065]

[0066] Among them, ρ>0 is a parameter that defines the relative weight of energy consumption and delay, C 1 , C 3 , C 4 , C 8 and C 9 Ensure that the link bandwidth, computing resource allocation of drones and edge computing nodes are non-negative and do not exceed the limit, C 2 Ensure that the tasks of mobile users are completely processed locally, in the UAV, and at the edge computing node. 5 , C 6 and C 7 Make sure the task split ratio is between 0 and 1;

[0067] Step 4: Optimization process of successive convex approximation algorithm;

[0068] Step 4-1: Initialization of the optimization problem;

[0069] First, the drone’s location, task allocation parameters, bandwidth allocation parameters, and computing resource allocation parameters are initialized to ensure that each IoT device can offload tasks in a timely manner;

[0070] Step 4-2: Convex approximation optimization iteration;

[0071] Since the optimization problem in the system is a non-convex problem, the successive convex approximation (SCA) algorithm is used to solve it. The SCA algorithm transforms the non-convex objective function and constraints into a series of convex sub-problems, which gradually converge to the global suboptimal solution through iteration. In each iteration, the system calculates the service delay of the task and the energy consumption of the drone based on the current resource allocation and location decision, and then adjusts the location of the drone and the allocation strategy of tasks and resources.

[0072] Step 4-3: Solve sub-problems;

[0073] In each iteration, the system needs to solve a convex optimization subproblem, and the optimization content includes:

[0074] (1) Drone position: Adjust the 3D coordinates of the drone to ensure the optimal communication channel with IoT devices and edge cloud;

[0075] (2) Communication bandwidth allocation: Allocate uplink and downlink bandwidth to increase data transmission rate and reduce task transmission delay;

[0076] (3) Task split ratio: Determine the optimal distribution ratio of tasks between drones and edge clouds to reduce overall computing latency;

[0077] (4) Computing resource allocation: Optimize the computing resource allocation of drones and edge clouds to ensure that computing tasks can be completed in the shortest time;

[0078] Step 5: Data processing and task completion;

[0079] Step 5-1: Calculation and data processing;

[0080] When tasks are offloaded to drones, the drones will process some simple tasks based on their computing resources. For tasks with a computing amount exceeding a set threshold, the system will forward them to the edge cloud for processing via the drones.

[0081] The processed data is relayed back to the IoT device via the drone, or directly stored in the edge cloud server for subsequent analysis as needed;

[0082] Step 5-2: Task completion and feedback;

[0083] When all tasks are processed, the system records the total delay of task execution and the energy consumption of the drone as evaluation indicators of system performance; if there are tasks that are not completed on time, the system will adjust the deployment location or resource allocation strategy of the drone based on the feedback of the SCA algorithm to ensure improved performance in subsequent task execution.

[0084] A computer program that enables a computer to execute the above-mentioned drone position and resource joint optimization method.

[0085] An electronic device comprises: a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the electronic device executes the above-mentioned drone position and resource joint optimization method.

[0086] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned method for jointly optimizing the position and resources of a drone.

[0087] A chip includes: a processor, which is used to call and run a computer program from a memory, so that a device equipped with the chip executes the above-mentioned drone position and resource joint optimization method.

[0088] A computer program product, the computer program product comprising a computer storage medium storing a computer program, the computer program comprising instructions executable by at least one processor, and the above-mentioned method for joint optimization of drone position and resources is implemented when the instructions are executed by the at least one processor.

[0089] The beneficial effects of the present invention are as follows:

[0090] 1) This invention proposes an edge-end collaboration system that focuses on meeting the computing needs of mobile users. The system uses algorithms to achieve close collaboration among mobile users, drones, and edge computing nodes, giving full play to their respective advantages and computing resources. Tasks are segmented and unloaded through the edge-end collaboration mechanism, and mobile users, drones, and edge computing nodes share task processing, ensuring efficient task completion and optimizing resource utilization.

[0091] 2) Considering the strict quality requirements of MEC services and the limited battery capacity of drones, the computing resources of drones, edge computing nodes, and bandwidth allocation of mobile users are jointly optimized through edge-end collaboration to make full use of heterogeneous resources in the system to minimize the system's service latency and energy consumption. Through this collaborative optimization, the system can allocate and utilize heterogeneous resources more efficiently, thereby reducing overall costs and improving service quality.

[0092] 3) Using regional division can effectively achieve load balancing, ensure the reasonable distribution of computing tasks, avoid overloading of certain servers or drones, and improve the overall system efficiency. At the same time, regional division can assign users to closer servers or drones to reduce communication delays. In addition, regional division based on the computing power of the server and the task requirements of the user can more efficiently utilize resources and optimize the allocation of tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0094] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0095] The present invention mainly considers the large number of user tasks in the area and the uneven resource allocation, divides the edge devices and users in the area into regions, and considers the edge collaborative processing tasks of drones, mobile users and edge computing nodes. The task offloading process is expressed as an optimization problem, and then a drone position and resource joint optimization algorithm based on successive convex approximation is proposed based on the idea of ​​successive convex approximation algorithm.

[0096] The technical solution of the present invention is achieved in this way:

[0097] 1. Technical Principle

[0098] In terms of regional division, the normalized cut algorithm based on graph theory is used to adjust the scope of regional division by utilizing the computing power weight of the server and the task requirement weight of the user. Servers with larger weights can affect larger areas, and users with higher task requirements tend to choose servers with strong computing power. In terms of drone deployment and resource allocation, the successive convex approximation algorithm is used to gradually transform complex non-convex problems into a series of convex optimization sub-problems that are easier to solve. These sub-problems are solved iteratively, gradually approaching the global or local optimal solution, and obtaining more accurate drone positions and resource allocation.

[0099] 2. Implementation plan

[0100] According to the above principle, the present invention is mainly a joint optimization algorithm of UAV position and resources based on successive convex approximation, which is implemented as follows:

[0101] Region division,First, the region division is performed based on the normalized cut algorithm based on graph theory,,the goal is to assign users to the most suitable edge servers according to the,task requirement weights of users and the computing power weights of servers.,By calculating the weighted distance between users and servers, users are assigned to,the server region with the best distance.

[0102] After the users are assigned, the region of each server (including its corresponding users) is determined. In each region, the SCA algorithm is used to optimize the location and task allocation of the drones. The goal of SCA is to minimize the total transmission delay from the user to the drone and edge server, as well as the energy consumption of the drone.

[0103] In each area, the position of the drone is first initialized, usually set at the center of the area, and the task allocation ratio is initialized to uniform distribution. Next, an optimization problem is constructed with the goal of minimizing latency and energy consumption. This optimization problem is a non-convex problem, which is solved by decomposing it into a series of convex subproblems using the Successive Convex Approximation (SCA) algorithm. The objective function calculates the transmission distance from the user to the drone and the server, the task allocation ratio, and the energy consumption. The drone position and task allocation ratio are updated through each iteration until the optimization result meets the convergence condition.

[0104] Example:

[0105] Reference Figure 1 , the implementation steps of this example are as follows:

[0106] Step 1, system initialization;

[0107] 1.1) Deployment of drones and edge devices. Several ground IoT devices (IoT devices) are deployed in the target area. These devices may not be able to communicate directly with the edge cloud server due to signal obstruction or geographical location. Drones are reasonably deployed in the air at the beginning of the mission as relay nodes between IoT devices and the edge cloud. Drones have certain computing resources and are equipped with modules for communication and data processing. Edge cloud servers are deployed at fixed locations on the ground to provide high-performance computing resources.

[0108] 1.2) Task generation and allocation, each IoT device generates computing tasks based on its sensing tasks, which may include data analysis that requires high computing power, such as video processing, real-time monitoring, etc. Each task can be decomposed into multiple subtasks, some of which are performed by drones, and other complex tasks are processed by the edge cloud after being relayed by drones.

[0109] Step 2, regional division;

[0110] 2.1) Obtain the locations of all users in the system, the locations of edge servers, the computing power (weight) of edge servers, and the task requirements (weight) of users, and construct a similarity matrix, in which the elements represent the similarity or dependency between nodes.

[0111] 2.2) Calculate the Laplacian matrix of the graph. First, calculate the degree matrix of the graph. This matrix is ​​a diagonal matrix. The elements on the diagonal of the matrix are the degree of each node, that is, the strength of its connection with other nodes. Then calculate the Laplacian matrix of the graph, which is the difference between the degree matrix and the similarity matrix. The Laplacian matrix reflects the connection relationship between the nodes in the graph.

[0112] 2.3) Perform eigenvalue decomposition on the Laplacian matrix of the graph to obtain eigenvectors and eigenvalues. Eigenvalue decomposition can map the nodes in the graph to a low-dimensional space, reflecting the similarity between the nodes. Select the smallest few eigenvectors in the Laplacian matrix, which will represent the embedded representation of the nodes in the low-dimensional space.

[0113] 2.4) Use the K-means clustering algorithm to cluster the feature vectors and divide them into multiple regions. Each region corresponds to a cluster. Based on the clustering results, the nodes in the system (users, devices, ECs, etc.) are divided into different regions.

[0114] Step 3: Task offloading and resource allocation;

[0115] 3.1) A ground-to-air uplink communication link is established between the drone and the IoT device to transmit the mission data of the IoT device; an air-to-ground downlink is established between the drone and the edge cloud server to forward the mission data to the edge cloud. The communication link adopts the free space path loss model. Considering the influence of the height and position of the drone, the uplink data transmission rate between each IoT device and the drone is calculated according to the Shannon capacity formula. Therefore, the channel gain from mobile user i to the drone can be described by the free space path loss model:

[0116]

[0117] Among them, g 0 It represents the channel power gain when the reference distance is 1m, d UAV,i represents the distance between mobile user i and the drone, and ||·|| represents the Euclidean norm of the vector. Similarly, the channel gain between the drone and the edge computing node j can be expressed as:

[0118]

[0119] Among them, d UAV,j Represents the distance between edge computing node j and the drone.

[0120] During the task offloading process, mobile users share bandwidth resources through the frequency division multiple access protocol. Therefore, according to the Shannon capacity formula, the data transmission rate between mobile user k and the drone can be expressed as:

[0121]

[0122] in, represents the bandwidth allocated to mobile user i, represents the transmission power of mobile user k, σ 2 Represents the noise power at the drone. For the convenience of calculation, this paper assumes that the noise power of the drone and the edge computing node is the same

[31] Similarly, the data transmission rate between the drone and the edge computing node j can be expressed as:

[0123]

[0124] in, represents the bandwidth pre-allocated to edge computing node j, P t UAV Indicates the transmission power of the drone.

[0125] In the present invention, mobile users will give priority to executing tasks locally. If the local computing delay exceeds the maximum tolerable delay, part of the task will be offloaded to the drone for execution. The drone will then determine whether part of the task is processed by the drone or further offloaded to the edge computing node on the ground for processing. Compared with the entire communication and computing delay, the decision time for splitting tasks is extremely short and can be ignored. In addition, due to the local computing delay of the mobile user, in many intensive applications, the output data of the calculation result is usually very small compared to the input data, so the delay in returning the calculation result to the mobile user will also be negligible.

[0126] In summary, the present invention divides the total latency of the task offloading process into the following six parts: (1) the computational latency of tasks that are completely processed locally; (2) the computational latency of tasks that need to be split and processed locally; (3) the transmission latency from the mobile user to the drone; (4) the computational latency of the drone; (5) the transmission latency from the drone to the edge computing node; and (6) the computational latency of the edge computing node.

[0127] Computational latency of tasks that are completely processed locally: All tasks will first be calculated locally. If a task exceeds the maximum tolerable latency when executed locally, it will be considered to be offloaded to the drone for further processing. Therefore, the computational latency of tasks that can be completely processed locally is expressed as:

[0128]

[0129] Among them, L i,0 represents the input data size of the task that is completely processed locally, C i,0 Indicates the number of CPU cycles required to process 1 bit of task data in this part of the task. Represents the computing resources of the mobile user that handles this part of the task.

[0130] The computational latency of tasks that need to be split and processed locally: Tasks whose local execution time exceeds the maximum tolerable latency will be split and offloaded. Some of these tasks will be executed locally. Therefore, the computational latency of this part is expressed as:

[0131]

[0132] in, Indicates the task split ratio that needs to be processed locally, L i,k represents the input data size of the task to be split, C i,k Indicates the number of CPU cycles required to process 1 bit of task data in this part of the task. Represents the computing resources of the mobile user that handles this part of the task.

[0133] Transmission delay from mobile users to drones: The tasks that need to be split will be offloaded to drones through the transmission link. Since some tasks will be executed locally on the mobile users, but in order to facilitate subsequent calculations, Between 0 and 1, the present invention assumes that the transmission delay from mobile user k to the drone is equal to the transmission delay of transmitting all tasks to the drone, and no task segmentation is performed in this process:

[0134]

[0135] in, Indicates the data transmission rate from mobile users to drones that require task segmentation.

[0136] The computational latency of the drone: The drone will decide whether the received task part should be processed locally in the drone or further offloaded to the ground edge computing node for processing. Therefore, the computational latency of the drone is expressed as:

[0137]

[0138] in, Indicates the task split ratio that needs to be processed by the drone. represents the computing resources allocated by the UAV to mobile user k.

[0139] Transmission delay from drone to edge computing node: The drone further offloads the task to the more powerful edge computing node on the ground for processing to reduce computing delay. Therefore, the transmission delay from drone to edge computing node is expressed as:

[0140]

[0141] Among them, α kj Indicates the task split ratio that needs to be processed on the edge computing node.

[0142] Computational delay of edge computing node: After receiving the unloaded task data from the UAV, the edge computing node starts the computation process. Therefore, the computational delay of the edge computing node processing the unloaded task from mobile user k to edge computing node j via the UAV is expressed as:

[0143]

[0144] in, It represents the computing resources allocated by edge computing node j to mobile user k.

[0145] Therefore, the total delay of the system is expressed as:

[0146]

[0147] In addition, in addition to considering system latency, the present invention also considers the impact of energy consumption on the system. Since the battery size of the drone is limited, it is very important to manage the energy consumption of the drone in order to ensure service availability. The present invention mainly focuses on the computing and transmission energy consumption of the drone, while ignoring the hovering power because it is irrelevant to the decision of the present invention.

[0148] Computational energy consumption of drones: This paper models the power consumption of the CPU in drones as where κ represents the effective switched capacitance that depends on the CPU architecture. Therefore, the corresponding energy consumption of the drone when processing the task offloaded from mobile user k is given by the product of power level and computation time:

[0149]

[0150] Transmission energy consumption of UAV: ​​The transmission energy consumption of UAV when receiving task input data from mobile user k through uplink communication is expressed as:

[0151]

[0152] in, is the received power of the UAV. In addition, the transmission energy consumption when the UAV offloads part of the task of mobile user k to edge computing node j through uplink communication is expressed as:

[0153]

[0154] Therefore, the total energy consumption of the drone is expressed as:

[0155]

[0156] 3.2) Task splitting and offloading decision;

[0157] For each task generated by the IoT device, the system determines the task allocation ratio based on the computing requirements of the task, the computing power of the drone, and the resource status of the edge cloud. The task allocation parameters control the allocation ratio of tasks between the drone and the edge cloud. The system jointly optimizes the drone's position, uplink and downlink bandwidth allocation, task allocation ratio, and computing resource allocation to minimize the total service delay and the drone's energy consumption. Therefore, the objective function is given by the following formula:

[0158]

[0159] Among them, ρ>0 is a parameter that defines the relative weight of energy consumption and delay, C 1 , C 3 , C 4 , C 8 and C 9Ensure that the link bandwidth, computing resource allocation of drones and edge computing nodes are non-negative and do not exceed the limit, C 2 Ensure that the tasks of mobile users are completely processed locally, in the UAV, and at the edge computing node. 5 , C 6 and C 7 Make sure the task split ratio is between 0 and 1.

[0160] Step 4, optimization process of successive convex approximation algorithm;

[0161] 4.1) Initialization of the optimization problem,The system first initializes the position of the drone, task allocation parameters,,bandwidth allocation parameters, and computing resource allocation parameters,to ensure that each IoT device can offload tasks in time.

[0162] 4.2) Convex approximation optimization iteration,Since the optimization problem in the system is a non-convex problem, the successive convex approximation (SCA) algorithm is used to solve it. The SCA algorithm transforms the non-convex objective function and constraints into a series of convex sub-problems, and gradually converges to the global suboptimal solution through iteration. In each iteration, the system calculates the service delay of the task and the energy consumption of the drone based on the current resource allocation and location decision, and then adjusts the location of the drone and the allocation strategy of tasks and resources.

[0163] 4.3) Solving sub-problems;

[0164] In each iteration, the system needs to solve a convex optimization subproblem, and the optimization content includes:

[0165] (1) Drone position: Adjust the 3D coordinates of the drone to ensure the optimal communication channel with IoT devices and edge cloud.

[0166] (2) Communication bandwidth allocation: Allocate reasonable uplink and downlink bandwidth to improve data transmission rate and reduce task transmission delay.

[0167] (3) Task splitting ratio: Determine the optimal distribution ratio of tasks between drones and edge clouds to reduce overall computing latency.

[0168] (4) Computing resource allocation: Optimize the computing resource allocation of drones and edge clouds to ensure that computing tasks can be completed in the shortest time possible.

[0169] Step 5: Data processing and task completion;

[0170] 5.1) Computation and Data Processing,When tasks are offloaded to drones, the drones will process some simple tasks based on their computing resources.,For tasks with large computational loads, the system forwards them to the edge cloud for processing,through drones.

[0171] The processed data will be relayed back to the IoT device via the drone, or directly stored in the edge cloud server for subsequent analysis as needed.

[0172] 5.2) Task completion and feedback, when all tasks are processed, the system records the total delay of task execution and the energy consumption of the drone as the evaluation index of system performance. If there are tasks that are not completed on time, the system will adjust the deployment location or resource allocation strategy of the drone based on the feedback of the SCA algorithm to ensure improved performance in subsequent task execution.

Claims

1. A method for joint optimization of drone position and resources with IoT edge collaboration, characterized in that: The steps include: Step 1: System initialization; Step 1-1: Deployment of drones and edge devices; Several ground IoT devices are deployed in the target area. UAVs are deployed in the air at the beginning of the mission as relay nodes between IoT devices and edge cloud. UAVs have certain computing resources and are equipped with modules for communication and data processing. Edge cloud servers are deployed at fixed locations on the ground to provide high-performance computing resources. Step 1-2: Task generation and allocation; Each IoT device generates computing tasks based on its sensing tasks. These tasks include data analysis that requires high computing power. The tasks are modeled as a triple W i =(L i ,C i ,λ i ), where L i Indicates the size of the input data for the processing task, C i Indicates the number of CPU cycles required to process 1 bit of data, λ i represents the task arrival rate; each task is decomposed into multiple subtasks, some of which are executed by drones, and other tasks are processed by the edge cloud after being relayed by drones; Step 2: Regional division; Step 2-1: Obtain the locations of all users in the system, the locations of edge servers, the computing capabilities of edge servers, i.e., weights, and the task requirements and weights of users, and construct a similarity matrix, in which the elements represent the similarity or dependency between nodes; Step 2-2: Calculate the Laplacian matrix of the graph; First, the degree matrix of the graph is calculated. This matrix is ​​a diagonal matrix. The elements on the diagonal of the matrix are the degree of each node, that is, the strength of the connection with other nodes. Then the Laplacian matrix of the graph is calculated. It is the difference between the degree matrix and the similarity matrix. The Laplacian matrix reflects the connection relationship between the nodes in the graph. Step 2-3: Perform eigenvalue decomposition on the Laplacian matrix of the graph to obtain eigenvectors and eigenvalues; eigenvalue decomposition can map the nodes in the graph to a low-dimensional space to reflect the similarity between the nodes; select the smallest eigenvectors in the Laplacian matrix, which represent the embedded representation of the nodes in the low-dimensional space; Step 2-4: Use the clustering algorithm K-means to cluster the feature vectors and divide them into multiple regions; each region corresponds to a cluster; based on the clustering results, the nodes in the system, i.e. users, devices, and ECs, are divided into different regions; Step 3: Task offloading and resource allocation; Step 3-1: A ground-to-air uplink communication link is established between the drone and the IoT device to transmit the mission data of the IoT device; an air-to-ground downlink is established between the drone and the edge cloud server to forward the mission data to the edge cloud; the communication link adopts the free space path loss model. Considering the influence of the height and position of the drone, the uplink data transmission rate between each IoT device and the drone is calculated according to the Shannon capacity formula; therefore, the channel gain from mobile user i to the drone is described by the free space path loss model: Where g0 represents the channel power gain when the reference distance is 1m, d UAV,k represents the distance between mobile user k and the drone, ||·|| represents the Euclidean norm of the vector, represents the location of user k, Loc UAV Indicates the position of the drone; The channel gain between the drone and the edge computing node j is expressed as: Among them, d UAV,j represents the distance between edge computing node j and the drone, represents the location of edge device j; During the task offloading process, mobile users share bandwidth resources through the frequency division multiple access protocol. According to the Shannon capacity formula, the data transmission rate between mobile user k and the UAV is expressed as: in, represents the bandwidth allocated to mobile user k, represents the transmission power of mobile user k, σ 2 represents the noise power at the drone; it is assumed that the noise power of the drone is the same as that of the edge computing node; The data transmission rate between the drone and the edge computing node j is expressed as: in, represents the bandwidth pre-allocated to edge computing node j, P t UAV represents the transmission power of the drone; The total delay of the task offloading process is divided into the following six parts: (1) the computational delay of tasks that are completely processed locally; (2) the computational delay of tasks that need to be split and processed locally; (3) the transmission delay from the mobile user to the drone; (4) the computational delay of the drone; (5) the transmission delay from the drone to the edge computing node; (6) the computational delay of the edge computing node; Computational latency of tasks processed entirely locally: All tasks are first calculated locally. If a task exceeds the maximum tolerable latency when executed locally, it will be considered to be offloaded to the drone for further processing. Therefore, the computational latency of tasks processed entirely locally is expressed as: Among them, L i,0 represents the input data size of the task that is completely processed locally, C i,0 Indicates the number of CPU cycles required to process 1 bit of task data in this part of the task. represents the computing resources of the mobile user that handles this part of the task; The computational latency of tasks that need to be split and processed locally: Tasks whose local execution time exceeds the maximum tolerable latency will be split and offloaded. Some of these tasks will be executed locally. Therefore, the computational latency of this part is expressed as: in, Indicates the task split ratio that needs to be processed locally, L i,k represents the input data size of the task to be split, C i,k Indicates the number of CPU cycles required to process 1 bit of task data in this part of the task. represents the computing resources of the mobile user that handles this part of the task; Transmission delay from mobile users to drones: The tasks that need to be split are offloaded to drones through transmission links. Assume that the transmission delay from mobile user k to drones is equal to the transmission delay of transmitting all tasks to drones. No task splitting is performed in this process: in, Indicates the data transmission rate from the mobile user to the UAV that needs to perform task segmentation; The computational latency of the drone: The drone will decide whether the received task part should be processed locally on the drone or further offloaded to the ground edge computing node for processing; therefore, the computational latency of the drone is expressed as: in, Indicates the task split ratio that needs to be processed by the drone. represents the computing resources allocated by the UAV to mobile user k; Transmission delay from the drone to the edge computing node: The drone further offloads the task to the edge computing node on the ground for processing to reduce computing delay. Therefore, the transmission delay from the drone to the edge computing node is expressed as: Among them, α kj Indicates the task split ratio that needs to be processed in the edge computing node; Computational delay of edge computing node: After receiving the offloaded task data from the UAV, the edge computing node starts the computation process; therefore, the computational delay of the edge computing node processing the offloaded task from mobile user k to edge computing node j via the UAV is expressed as: in, represents the computing resources allocated to mobile user k by edge computing node j; Therefore, the total delay of the system is expressed as: Computational energy consumption of drones: The power consumption of the drone CPU is modeled as where κ represents the effective switched capacitance that depends on the CPU architecture; therefore, the corresponding energy consumption of the drone when processing the task offloaded from mobile user k is given by the product of the power level and the computation time: Transmission energy consumption of UAV: ​​The transmission energy consumption of UAV when receiving task input data from mobile user k through uplink communication is expressed as: in, is the received power of the drone; The transmission energy consumption when the UAV offloads part of the task of mobile user k to edge computing node j through uplink communication is expressed as: Therefore, the total energy consumption of the drone is expressed as: Step 3-2: Task splitting and offloading decision; For each task generated by an IoT device, the system determines the task allocation ratio based on the task's computing requirements, the drone's computing power, and the edge cloud's resource status; the task allocation parameters control the task allocation ratio between the drone and the edge cloud; the system jointly optimizes the drone's location, uplink and downlink bandwidth allocation, task allocation ratio, and computing resource allocation to minimize the total service delay and the drone's energy consumption; The objective function is given by: Among them, ρ>0 is a parameter that defines the relative weight of energy consumption and delay. C1, C3, C4, C8 and C9 ensure that the link bandwidth, the computing resource allocation of the UAV and the edge computing node are non-negative and do not exceed the limit. C2 ensures that the tasks of mobile users are completely processed locally, in the UAV and at the edge computing node. C5, C6 and C7 ensure that the task split ratio is between 0 and 1. Step 4: Optimization process of successive convex approximation algorithm; Step 4-1: Initialization of the optimization problem; First, the drone’s location, task allocation parameters, bandwidth allocation parameters, and computing resource allocation parameters are initialized to ensure that each IoT device can offload tasks in a timely manner; Step 4-2: Convex approximation optimization iteration; Since the optimization problem in the system is a non-convex problem, the successive convex approximation (SCA) algorithm is used to solve it. The SCA algorithm transforms the non-convex objective function and constraints into a series of convex sub-problems, which gradually converge to the global suboptimal solution through iteration. In each iteration, the system calculates the service delay of the task and the energy consumption of the drone based on the current resource allocation and location decision, and then adjusts the location of the drone and the allocation strategy of tasks and resources. Step 4-3: Solve sub-problems; In each iteration, the system needs to solve a convex optimization subproblem, and the optimization content includes: (1) Drone position: Adjust the 3D coordinates of the drone to ensure the optimal communication channel with IoT devices and edge cloud; (2) Communication bandwidth allocation: Allocate uplink and downlink bandwidth to increase data transmission rate and reduce task transmission delay; (3) Task split ratio: Determine the optimal distribution ratio of tasks between drones and edge clouds to reduce overall computing latency; (4) Computing resource allocation: Optimize the computing resource allocation of drones and edge clouds to ensure that computing tasks can be completed in the shortest time; Step 5: Data processing and task completion; Step 5-1: Calculation and data processing; When tasks are offloaded to drones, the drones will process some simple tasks based on their computing resources. For tasks with a computing amount exceeding a set threshold, the system will forward them to the edge cloud for processing via the drones. The processed data is relayed back to the IoT device via the drone, or directly stored in the edge cloud server for subsequent analysis as needed; Step 5-2: Task completion and feedback; When all tasks are processed, the system records the total delay of task execution and the energy consumption of the drone as evaluation indicators of system performance; if there are tasks that are not completed on time, the system will adjust the deployment location or resource allocation strategy of the drone based on the feedback of the SCA algorithm to ensure improved performance in subsequent task execution.

2. A computer program, characterized in that The computer program enables a computer to execute the method as claimed in claim 1.

3. An electronic device, characterized in that: include: Processor and memory; The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the electronic device executes the method as claimed in claim 1.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method as claimed in claim 1 is implemented.

5. A chip, characterized in that: include: A processor, used to call and run a computer program from a memory, so that a device equipped with the chip executes the method as claimed in claim 1.

6. A computer program product, characterized in that The computer program product comprises a computer storage medium storing a computer program, wherein the computer program comprises instructions executable by at least one processor, and when the instructions are executed by the at least one processor, the method according to claim 1 is implemented.

Images